PLoS ONE

Brant-Zawadzki, Michael;Fridman, Deborah;Robinson, Philip A.;Zahn, Matthew;Chau, Clayton;German, Randy;Breit, Marcus;Bock, Jason R.;Hara, Junko

doi: 10.1371/journal.pone.0240006pmid: 33180782

Introduction SARS-CoV-2 has driven a pandemic crisis. Its hallmark is very high infectivity, pre-symptomatic transmission and asymptomatic prevalence which continue to fuel dramatic cumulative numbers of infections, hospitalizations, and deaths. To better understand the extent of undetected transmission, serological surveys in sampled cohorts identified antibodies from prior infection ranging from 57% prevalence in Bergamo—Italy’s epicenter [1], 20% in New York City [2], 5.2% in Kenya [3], down to 4.7% in Los Angeles County [4] and 2.8% in Santa Clara County [5], California. While prevalence of antibodies among healthcare workers, presumed at high risk for infection, has also been increasingly studied, the prevalence, sample size, and sampling methodology greatly varies. Garcia-Basteiro, et al reports the cumulative prevalence (IgG, IgA, or current positive rRT-PCR) of SARS-CoV-2 infection of 11.2% among 578 subjects in a large hospital in Spain [6]. The study of 28,792 healthcare works from Denmark identified 2.81% prevalence for IgM and 2.67% for IgG, and found higher prevalence in front-line workers specifically working with COVID-19 patients, compared to other front-line workers [7]. It also found that subjects younger than 30 years had the highest seroprevalence compared to those who are 30 years or older. Another study found 13.7% IgG prevalence among 40,329 healthcare workers in the greater New York city area [8] similar to the community prevalence in New York State (14.0%) [9], and sero-positivity was strongly associated with self-reported suspicion of prior COVID-19 exposure and prior positive PCR testing. Further determining such prevalence among healthcare workers in varied geographic areas and examining duration of antibody presence may help stratify the workforce for risk, establish better health place policies and procedures, and potentially better mitigate transmission across different healthcare settings. This article reports on initial sero-surveillance conducted among 2,992 healthcare workers at Hoag Memorial Hospital Presbyterian, a regional hospital system in Orange County, California, United States, during May and June, 2020. Methods Recruitment and enrollment The Institutional Review Board approval was obtained for this study from Providence St. Joseph Health (IRB # 2020000337). Study subjects were recruited by email notifications to the entire employee workforce (6000+ individuals) and the independent medical staff (1600+ physicians), and were enrolled during May and June, 2020. Their work locations included 2 main hospital campuses, 9 health centers, 13 urge care locations, and other clinical and administrative facilities all within approximately 20 miles radius. Informed consent was obtained in person originally (n = 2,934), then electronically after June 18, 2020 (n = 58). All consenting subjects were asked to answer a questionnaire (S1 Table), and a blood sample was collected. Questionnaire The questionnaire was developed by two of the physician authors, one of whom is an infectious disease specialist, within the guidelines of the human resources department of Hoag Hospital and the rules and regulations of the Hoag medical staff department. The goal of the questionnaire development was ease of response while allowing collection of basic information including demographics, job duties and locations, and potential outside exposures. In addition, the COVID-19 symptoms were selected based on the list of symptoms by CDC and other COVID-19-related publications at the time of questionnaire development. Using the reported job duties and locations, each subject was classified into a) high (e.g., MD, RN, PA, emergency care tech, ICU tech), b) medium (e.g., therapist, phlebotomist, medical tech), or c) low (e.g., administration, coding, billing, lab tech/scientist, IT) risk groups to approximate levels of direct exposure to COVID-19 patients. IgG antibodies to SARS-CoV-2 analysis A 5 ml peripheral draw venous blood sample was collected, at the time of in-person consent and within 7 days (M = 1.67, SD = 1.36) of electronic consent, from each subject into a gold top serum separator vacutainer tube (BD Medical). Samples were centrifuged within 2 hours of collection at 4500 RPM for 5 minutes (RCF 3060). Aliquots were analyzed with calibrated lots of Anti-SARS-CoV-2 IgG Reagent Pack on the VITROS® XT 7600 according to manufacturer’s instructions for use. SARS-CoV-2 spike protein coated on lots is the antigen used [10]. Positive and negative quality controls were run daily prior to sample analysis (Ortho Diagnostics Anti-SARS-CoV-2 IgG Control). At the time of writing, this IgG test was approved only for use under the Food and Drug Administration’s Emergency Use Authorization. Manufacture sensitivity and specificity claims for the Ortho Clinical Diagnostics VITROS Anti-SARS-CoV-2 IgG assay is 100% (407/407) negative agreement (95% CI: 99.1–100.0%) in 407 presumed SARS-CoV-2 antibody negative subjects and 87.5% (42/48) positive agreement (95% CI: 74.8–95.3%) in 48 PCR positive subjects with days from positive PCR ranging from 1 day to 22 days and days from onset of symptoms ranging from 12 to 32 days. In-house validation studies were conducted with 35 samples from subjects with a known positive SARS-CoV-2 PCR test a mean of 43 days out from positive PCR test date (range 38–48 days), and 50 samples from subjects with a known negative SARS-CoV-2 PCR test. Twenty-nine of 31 PCR samples were positive for SARS-CoV-2 IgG antibody. All 50 of the PCR negative samples were SARS-CoV-2 IgG antibody negative. Thus, sensitivity of 93.6% (95%CI: 78.6–99.2%) and specificity of 100% (95% CI: 92.9–100.0%) were calculated for the Ortho Diagnostics VITROS Anti-SARS-CoV-2 IgG assay in run our laboratory on the Ortho Clinical Diagnostics VITROS® XT 7600 automated instrument platform, and adopted in this study. Data analysis Demographic, occupational, and symptom factors were assessed for group differences between negative vs. positive for the presence of IgG antibodies. A Mann-Whitney U test was used for assessing group difference in age, and a series of Fisher’s exact tests were used for the remaining categorical factors; for group differences in race (a 7×2 table), the Mehta-Patel algorithm [11] was applied. A value of p < .05 was used for statistical significance. For all analyses, the Stata statistical software package, edition 15 [12], was used. Recruitment and enrollment The Institutional Review Board approval was obtained for this study from Providence St. Joseph Health (IRB # 2020000337). Study subjects were recruited by email notifications to the entire employee workforce (6000+ individuals) and the independent medical staff (1600+ physicians), and were enrolled during May and June, 2020. Their work locations included 2 main hospital campuses, 9 health centers, 13 urge care locations, and other clinical and administrative facilities all within approximately 20 miles radius. Informed consent was obtained in person originally (n = 2,934), then electronically after June 18, 2020 (n = 58). All consenting subjects were asked to answer a questionnaire (S1 Table), and a blood sample was collected. Questionnaire The questionnaire was developed by two of the physician authors, one of whom is an infectious disease specialist, within the guidelines of the human resources department of Hoag Hospital and the rules and regulations of the Hoag medical staff department. The goal of the questionnaire development was ease of response while allowing collection of basic information including demographics, job duties and locations, and potential outside exposures. In addition, the COVID-19 symptoms were selected based on the list of symptoms by CDC and other COVID-19-related publications at the time of questionnaire development. Using the reported job duties and locations, each subject was classified into a) high (e.g., MD, RN, PA, emergency care tech, ICU tech), b) medium (e.g., therapist, phlebotomist, medical tech), or c) low (e.g., administration, coding, billing, lab tech/scientist, IT) risk groups to approximate levels of direct exposure to COVID-19 patients. IgG antibodies to SARS-CoV-2 analysis A 5 ml peripheral draw venous blood sample was collected, at the time of in-person consent and within 7 days (M = 1.67, SD = 1.36) of electronic consent, from each subject into a gold top serum separator vacutainer tube (BD Medical). Samples were centrifuged within 2 hours of collection at 4500 RPM for 5 minutes (RCF 3060). Aliquots were analyzed with calibrated lots of Anti-SARS-CoV-2 IgG Reagent Pack on the VITROS® XT 7600 according to manufacturer’s instructions for use. SARS-CoV-2 spike protein coated on lots is the antigen used [10]. Positive and negative quality controls were run daily prior to sample analysis (Ortho Diagnostics Anti-SARS-CoV-2 IgG Control). At the time of writing, this IgG test was approved only for use under the Food and Drug Administration’s Emergency Use Authorization. Manufacture sensitivity and specificity claims for the Ortho Clinical Diagnostics VITROS Anti-SARS-CoV-2 IgG assay is 100% (407/407) negative agreement (95% CI: 99.1–100.0%) in 407 presumed SARS-CoV-2 antibody negative subjects and 87.5% (42/48) positive agreement (95% CI: 74.8–95.3%) in 48 PCR positive subjects with days from positive PCR ranging from 1 day to 22 days and days from onset of symptoms ranging from 12 to 32 days. In-house validation studies were conducted with 35 samples from subjects with a known positive SARS-CoV-2 PCR test a mean of 43 days out from positive PCR test date (range 38–48 days), and 50 samples from subjects with a known negative SARS-CoV-2 PCR test. Twenty-nine of 31 PCR samples were positive for SARS-CoV-2 IgG antibody. All 50 of the PCR negative samples were SARS-CoV-2 IgG antibody negative. Thus, sensitivity of 93.6% (95%CI: 78.6–99.2%) and specificity of 100% (95% CI: 92.9–100.0%) were calculated for the Ortho Diagnostics VITROS Anti-SARS-CoV-2 IgG assay in run our laboratory on the Ortho Clinical Diagnostics VITROS® XT 7600 automated instrument platform, and adopted in this study. Data analysis Demographic, occupational, and symptom factors were assessed for group differences between negative vs. positive for the presence of IgG antibodies. A Mann-Whitney U test was used for assessing group difference in age, and a series of Fisher’s exact tests were used for the remaining categorical factors; for group differences in race (a 7×2 table), the Mehta-Patel algorithm [11] was applied. A value of p < .05 was used for statistical significance. For all analyses, the Stata statistical software package, edition 15 [12], was used. Results Of an initial 2,992 samples recorded, subjects were excluded from analyses due to missing age (n = 3), gender (n = 14), race, (n = 31), occupation (n = 8), and symptoms (n = 12), resulting in a complete pool of 2,924 (Table 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Sample characteristics and group differences. https://doi.org/10.1371/journal.pone.0240006.t001 Antibody testing identified 31 positive cases (2,893 negative), resulting in an observed prevalence of 1.06% (exact binomial 95% CI = 0.72% - 1.50%). Accounting for test sensitivity of 93.6% and specificity of 100%, an adjusted prevalence of 1.13% (95% CI = 0.78% - 1.58%) was calculated, indicating 33 positive cases (negative = 2,891) after adjustment. Nonparametric tests for group differences were performed for demographics and five symptoms of COVID-19. Significant differences between observed negative and positive cases were found for age (z = 2.64, p = .008), race (p = .037), presence of fever (p < .001), and loss of smell (p < .001), but not for occupation (p = .710). Interestingly, of those with previously PCR confirmed diagnosis of COVID-19 (n = 11), 6 were antibody positive with 5 non-reactive. None of the non-reactive 5 had a history of hospitalization or severe illness. Discussion Our study found a significantly lower prevalence (1.06% observed prevalence) of SARS-CoV-2 antibody carriers among our healthcare workers compared to prior reports ranging from 2.6% to 13.7%. During this same period, prevalence of antibodies tested by physician order at our hospital laboratory was 3.87%. One possible explanation for the low seroconversion rate in our work force is a relatively low overall regional estimated prevalence of infections (~4.4%), as further evidenced by an average 104 patients per day in ICU and 330 cumulative deaths in Orange County (total population of 3.18 million) at the time of our study [13]. This hypothesis is supported by the considerably higher prevalence in healthcare workers and even higher community prevalence in New York [9], indicating that geographic consideration needs to be given when evaluating the infection and transmission risks among healthcare workers. Despite our relatively low community prevalence in the early stage of the pandemic, our institution had implemented stringent workforce education on personal hygiene, social distancing and appropriate PPE usage since January 2020 when we saw the first California and third US case, with hospital-wide protocols in patient triage and symptom surveillance. Such strategies may have heightened our healthcare workers’ awareness, urgency, and compliance with our policies, both at and outside work place, possibly contributing to the lower prevalence we have found in this study. In addition, recent research indicates presence of innate and cross-reactive adaptive T cell mediated immunity, which may lower susceptibility to COVID-19 infection in some individuals. One might speculate that greater frequency of exposure to such agents occurs in healthcare workers vs. the general population. Several studies have reported that such innate T cell immunity exists [14] with documented cross-reactivity to related corona virus species [15–17]. One can speculate that workplace exposure is more frequent for health care workers to such various coronavirus pathogens. A combination of all of the above factors may explain our findings. In our cohort, there were 11 cases with previously PCR confirmed diagnosis of COVID-19 with 5 non-reactive cases, which cannot be fully explained by antibody test sensitivity and specificity. Recent studies found a rapid decay of IgG antibody in patients with mild symptoms, with the possible span of 2–3 months [18, 19]. Of those 5 non-relative cases, 4 reported no or mild COVID-19 symptoms in the questionnaire, their confirmed COVID-19 results being up to 2 months prior to antibody tests. While this provides additional support for the recent findings [17, 18], further research in the larger cohort is needed, and whether such decrease in antibodies lowers immunity should be examined, given the extremely rare cases of re-infection being reported [20]. We will retest this same cohort at 8 weeks and 6 months, to better understand the dynamics of SARS-CoV-2 antibody prevalence and duration in healthcare workers. Supporting information S1 Table. Study questionnaire. https://doi.org/10.1371/journal.pone.0240006.s001 (DOCX) Acknowledgments We acknowledge our healthcare workers who have contributed to this study.

Bruder, Lukas;Pelisek, Jaroslav;Eckstein, Hans-Henning;Gee, Michael W.

doi: 10.1371/journal.pone.0242097pmid: 33211767

1 Introduction An abdominal aortic aneurysm (AAA) is a slowly progressing vascular disease, causing an enlargement of the infrarenal aorta and is considered pathological if the aortic diameter exceeds 30 mm [1]. AAA prevalence has been reported within a range of 1.2% to 3.3% in men older than 60 years based on several studies in western societies [2]. In most cases, AAAs develop asymptomatically over several years, but they can rapidly turn into a serious clinical emergency in case of rupture. More than 50% of patients with a ruptured AAA die before reaching the hospital [1] and perioperative mortality rates range from 40% to 60% [3]. To prevent such a disastrous scenario, the clinical guidelines from the US-based Society for Vascular Surgery recommend elective repair for AAA patients with an aortic diameter greater or equal to 55 mm, regular screening intervals for patients with smaller-sized AAAs and one-time screenings for AAAs in men and women above a certain age and based on established risk factors [1]. This maximum diameter recommendation is based on a risk assessment, where the risk of rupture is weighed against the mortality risk of an elective repair. While the latter risks are relatively well-known, aneurysm rupture is a complex biomechanical failure event. With the increasing use of endovascular repair (EVAR) over open surgical repair (OSR) [4], however, which can be attributed to the significant short term mortality benefit of EVAR (1.4% compared to 4.2%) [1], interventional risks have become a less important factor in the risk assessment process. Nonetheless, a biomechanical rupture risk assessment can provide an additional important piece of information. It enables the possibility to provide patient-specific screening guidelines, avoid unnecessary interventions [5] and support the clinical decision process for cases that are not covered by the clinical guidelines. The Society for Vascular Surgery’s 55 mm recommendation, e.g., only holds for patients “at low or acceptable surgical risk with a fusiform AAA” [1]. Furthermore, there are no clear or only weak recommendations for women with AAAs of size 50-54 mm, aneurysms with non-fusiform geometries, smaller AAAs [6], or patients at higher surgical risk. In addition to that, not all AAAs are suitable for EVAR, with higher complication rates for AAA cases that are not covered by the instructions for use [7]. Lastly, recent meta-studies (e.g. [8, 9]) on the long term outcomes of EVAR versus OSR could not detect any differences with regards to the all-cause mortality or even concluded in favor of OSR. In this paper, we present a highly personalized, probabilistic framework for the biomechanical quantification of AAA rupture risk. The framework builds upon a comprehensive database, consisting of tensile experiments that were carried out on 305 AAA tissue samples from 139 patients and corresponding non-invasively and clinically accessible patient data. The approach consistently incorporates the available statistical information in terms of probability distributions in order to account for patient-specific uncertainties about relevant vessel wall properties. We emphasize the importance of accounting for these uncertainties and demonstrate that this leads to a more accurate individualized rupture risk assessment as compared to deterministic approaches. Our work builds upon previous efforts by our group and collaborators regarding the biomechanical modeling and characterization of AAA in-vivo behavior [10–16], as well as several previous studies indicating that biomechanical indices are more accurate predictors for AAA rupture risk than the clinically established maximum diameter criterion [17–24]. In contrast to the approaches in [17–20, 22, 24], however, we advocate a probabilistic treatment to account for uncertain vessel wall properties. Our work thus goes along the lines of [21], but with the key difference that it includes the stiffness parameters of the AAA vessel wall as statistical quantities, uses patient-specific vessel wall properties and accounts for statistical correlations among these properties. The paper is organized as follows. Motivated by a failure-based criterion, our rupture risk index is formulated in Section 2.1 incorporating patient-specific statistical information. Section 2.2 defines the biomechanical AAA model and specifies the probabilistic regression model to obtain personalized vessel wall properties. In Section 2.3, a method for the efficient evaluation of the rupture risk index is proposed and in Section 3, the framework is applied on a total of 36 retrospective, diameter matched asymptomatic (group 1, n = 18) and known symptomatic/ruptured (group 2, n = 18) cohort of AAA patients. 2 Materials and methods 2.1 Failure-based probabilistic quantification of rupture risk 2.1.1 Rupture as an event of material failure. From a mechanical point of view, rupture represents an event of local material failure at a point x in the aneurysm wall, which motivates its definition via a failure function φ(x) and the failure criterion (1) We limit ourselves to stress-based failure and define rupture as an event where the local wall stress measure σ(x) exceeds the local wall strength σγ(x). This results in the failure function φ(x) = σ(x) − σγ(x), or the criterion (2) Using the equivalent von Mises stress σvm(x) as the local stress measure σ(x) and an assumed spatially constant wall strength σγ, this criterion can be evaluated as (3) where is the maximum von Mises stress . It is important to note that the above definition does not incorporate any aspect about failure over time. In order to be able to include time in the analysis, i.e. to make a statement about the risk of rupture in the next year, one would require knowledge about the future progression of the AAA for this patient, such as a model for the aneurysm growth and change in vessel wall properties. Since there is hardly any knowledge about these aspects, we limit the further discussion to a rupture risk assessment at the point of time of the acquired data. While there are sudden events like calcification-induced formation of saccular aneurysms, we assume that in most cases an AAA is a slowly progressing disease and thus our approach has, at least for the near future, sufficient predictive capability. 2.1.2 Existing criteria and rupture risk indices. Rupture risk estimation for AAAs has been an ongoing research topic over several decades, with many attempts to establish decision criteria for clinical practice. The maximum diameter criterion [1] still represents the most widely used criterion for decision making today. It is often justified by Laplace’s law, which states that the vessel wall stress is proportional to its diameter in spherical geometries. Based on this and with data obtained from several clinical studies, a very simple criterion, (4) has been formulated, relating the patient’s AAA diameter d to a critical maximum diameter dmax. While established in clinical practice and easy to apply using CT or ultrasound imaging, this criterion has often been criticized [25] and is an ongoing subject for discussion [6]. With growing computational resources and advances in the modeling of biomechanical material behavior, the simulation of patient-specific AAA models has been advanced by several research groups. Experiments on harvested AAA samples were able to reveal material parameters and failure properties. In addition with regression models [13, 26, 27] for the prediction of the individual wall strength, this enabled the definition of biomechanics-based indices [19, 20, 22, 28], such as the rupture potential index (RPI) (5) relating the von Mises stress to the wall strength. Furthermore, it could be shown [19, 20, 24] that these indices can be better rupture risk indicators than the maximum diameter criterion. Experimental testing [13, 26, 27] also revealed significant inter- and intra-patient variabilities in the mechanical properties of AAA tissue, motivating a probabilistic approach to rupture risk estimation [14, 16, 21] and resulting in the probabilistic rupture risk index (PRRI) [21] (6) where the authors used distributions for the wall thickness and wall strength that were fitted on cohort data published by our group [13]. 2.1.3 A novel probabilistic approach. In this work, we propose a novel failure-based, probabilistic rupture risk indicator that consistently incorporates all available statistical information and accounts for correlations among vessel wall properties. Fig 1 (left) illustrates the rationale for our approach, showing how part of the available data is directly involved in the estimation of the risk of rupture, while another part affects the evaluation of the computational model. In general, this data will be correlated, resulting in correlated quantities for the evaluation of rupture risk and necessitating a reformulation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Rationale for our novel formulation (left) and exemplary visualization of its estimation (right). The probability of rupture, , is calculated as the volume of the probability distribution within the triangular-shaped area marked in red. https://doi.org/10.1371/journal.pone.0242097.g001 To that end and recalling the rupture criterion from Eq (3), we can calculate the probability of rupture over the joint probability distribution as (7) where is the indicator function defined as (8) This formulation can be easily extended to, e.g., spatially varying vessel properties using Eqs (1) or (2) as failure events. Furthermore, it includes the PRRI in Eq (6) as a special case, when choosing . Lastly, it allows for a straightforward visual interpretation as illustrated in Fig 1 (right). The plot shows the joint probability distribution and visualizes the rupture event area in red. The blue area implies a high probability for the joint occurrence of the corresponding stress and strength values. The probability of rupture is simply the volume of this density within the triangular rupture event area. Thus, the larger the overlap between and the red area, the higher . 2.2 Data-informed patient-specific AAA models 2.1.1 Geometry creation from CT imaging and meshing. Patient-specific 3D AAA geometries are reconstructed via a semi-automatic segmentation process from CT imaging data using the software ScanIP (Synopsys, Mountain View, California) and based on a protocol as described in [12]. The minimal requirement for the spatial resolution of CT scans was 1 mm and for the slice thickness 3 mm. The upper boundary for the segmentation was the branching of the renal arteries and the lower boundary below the bifurcation at the iliac arteries. Due to the small thickness of the AAA wall, its low contrast and the limited resolution of the CT images, it is only possible to extract the blood lumen and intraluminal thrombus (ILT) geometries. After segmentation, the ILT geometry is exported as a surface model for meshing. In a next step, we use the software Trelis (csimsoft, American Fork, Utah) and bi-linear quadrilateral elements to mesh the abluminal ILT surface. From this surface mesh, the arterial wall layer is extruded with a specified, spatially constant thickness t, resulting in a tri-linear, single layer, hexahedral mesh for the AAA wall. Finally, linear tetrahedral elements are employed for the meshing of the complex ILT geometry and a layer of linear pyramid elements as a transition for mesh compatibility between AAA wall and thrombus. Element sizes were set to 1.6 mm, corresponding to the median of measured thicknesses of AAA wall specimens in our database and leading to hexahedral elements of shape 1.6 mm × 1.6 mm × t for the AAA wall. A mesh convergence study has been performed to assess that the chosen spatial mesh resolution is sufficient in the context of our application. The meshing procedure is also described in [29] in more detail. 2.2.2 Biomechanical modeling. Previous studies have shown that in order to accurately describe the biomechanical behavior of AAAs, a sufficient model complexity is required [20, 30], while results by [31, 32] indicate that also simpler models might be appropriate. For our purposes, we employ the finite deformation boundary value problem of nonlinear elasticity: (9) (10) (11) where Ω0 is the reference configuration of the AAA, u denotes the displacement field, F = I+ ∇u the deformation gradient, S the second Piola-Kirchhoff stress tensor and σ the Cauchy stress tensor. On the Neumann boundary γσ, i.e. the luminal ILT surface, an orthonormal load is applied, with the pressure value p and the unit outward surface normal n in the current configuration. Furthermore, at the proximal and distal end surfaces of the AAA model, Γu, we employ a Robin-type boundary condition with spring supports following [29, 33]. The stiffness parameter ks is per unit reference area and set to 100 kPa/mm in this study, while N is the unit outward surface normal in the reference configuration. To model the constitutive behavior of the ILT, we use the strain energy function proposed in [34] (12) and a linearly decreasing stiffness c from the luminal to the abluminal ILT surface [12]. and are the first and second invariants of the modified right Cauchy-Green deformation tensor , with C = FT F and J = det(F) [29]. The strain energy function employed for the AAA wall material is [35, 36] (13) with stiffness parameters α and β. Both strain energy functions are equipped with an additive volumetric component (14) including the bulk modulus (15) with parameters and for the employed ILT and wall material models and a Poisson’s ratio of ν = 0.48 [12]. To obtain a pressurized in vivo configuration of the AAA, the MULF prestressing method [10, 11] is used, where the applied load corresponds to the mean arterial pressure (MAP = 1/3 systolic pressure + 2/3 diastolic pressure). From this prestressed configuration, the pressure is raised by 50% to simulate elevated blood pressure conditions [21]. Following [19] and for comparability reasons, the values for the systolic and diastolic pressures were set to 121 mmHg and 87 mmHg for all cases, respectively, resulting in a MAP of 98.33 mmHg. With the finite element discretization from Section 2.2.1, a nonlinear system of equations is obtained, which is solved using an in-house finite element code. We note that in this study we neglect the effect of calcifications in the AAA for simplicity and assume constant vessel wall thickness t and stiffness parameters α and β throughout the aneurysm. Furthermore, we evaluate the maximum von Mises stress as the 99th percentile of the von Mises stress field in the aneurysm. For the remainder of this work, we will use the parameter to quantity of interest (QoI) map (16) with parameter vector and QoI to denote the forward problem. Thus, calculating for one realization of t, α and β will involve one evaluation of the nonlinear finite element model. 2.2.3 Patient database. The modeling of patient-specific vessel wall properties here is based on data that has been collected during several research projects between 2008 and 2017 on the mechanobiological behavior of AAAs [13, 15]. The study was approved by the ethics committee of the University Hospital rechts der Isar, Technical University of Munich. AAA patients undergoing elective OSR (including emergency repair due to rupture) at the University Hospital rechts der Isar in Munich, Germany, were added to the database, whenever it was possible to extract tissue samples for mechanical testing. Apart from anamnesis and CT imaging data, hemograms were evaluated and one or more AAA tissue samples harvested during OSR. These samples were mechanically and histologically investigated, resulting in an exhaustive retrospective AAA database. Further information on data collection and experimental testing can be found in [13, 15]. To date, the database contains a total number of 305 entries from an equal number of tissue samples that were collected from 139 patients. The data can be split into two groups. Invasive properties (cf. Table 1), denoted as , are properties, which have been determined retrospectively from AAA tissue samples and cannot be obtained for a prospective patient by using clinically established methods. They are, however, essential for the biomechanical modeling and simulation of AAAs and the calculation of the probability of rupture using Eq (7). Non-invasive properties (cf. Table 2), denoted by ξ, on the other hand, can be determined with standard methods in the clinic. The subrenal diameter in Table 2 is measured directly below the renal arteries. If the aneurysm reached the renal arteries, the aortic diameter between the celiac artery and the superior mesenteric artery minus 2.5 mm was used instead [12]. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Invasive properties represent key vessel wall characteristics for a biomechanical rupture risk assessment. https://doi.org/10.1371/journal.pone.0242097.t001 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Non-invasive properties overview. https://doi.org/10.1371/journal.pone.0242097.t002 Based on correlations between the invasive and non-invasive properties [13], the goal is to construct a statistical model for the patient-individualized prediction of vessel wall properties Θ(ξ) for a prospective new patient with non-invasive properties ξ. While this process is described in Section 2.2.4, a preprocessing step for the dataset is essential, since values are missing both in the invasive and non-invasive properties for several cases in our database. Moreover, the relatively small number of available data, but high number of non-invasive properties, requires a feature selection process to identify the most important properties in ξ. Similar to [15], we conduct the following preprocessing steps. Non-invasive features in ξ, where more than 30% of the data points had missing values and patients with more than 30% of missing features were excluded and all other missing non-invasive properties imputed with the corresponding median value across the population. As a consequence, the four parameters calcium, high-sensitivity C-reactive protein (hsCRP), creatine kinase and fibrinogen were disregarded. Afterwards, all non-invasive features were normalized. Based on a correlation analysis using Spearman’s rank correlation coefficient (cf. S1 Table), the total number of features was reduced to a final selection of 8 variables: maximum AAA diameter, maximum thrombus thickness, AAA length, subrenal diameter, thrombocytes, hemoglobin, mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV). The restriction was done using a sequential forward selection algorithm similar to [15]. In an attempt to keep the number of non-invasive parameters small, we iteratively added the highest correlating non-invasive parameters to the GP model (see Section 2.2.4) until no further improvement in the leave-one-out cross-validation (LOOCV) scores could be observed. We note, however, that this does not imply that other non-invasive features such as sex, medication or anamnesis parameters do not have an influence on the biomechanical properties of the AAA wall. The resulting dataset , that was used for the analysis in Section 3, consisted of ndata = 251 data points from 113 individual patients and is available as supplementary information to this study (cf. S2 and S3 Tables). 2.2.4 Prediction of invasive vessel wall properties. Previous approaches to create models for the AAA wall thickness, stiffness parameters or strength were either deterministic [12, 26], based on cohort statistics [21], or did not account for correlations among the vessel wall quantities [15]. In the following, we make use of a multivariate Gaussian process regression model [37–39] to address these shortcomings and achieve the following desiderata: Patient-specific modeling: obtain personalized estimates for the vessel wall quantities Θ based on correlations with the non-invasive properties ξ of a specific, prospective patient. Probabilistic treatment: take into account the uncertainties in the predictions for Θ (do not ignore statistical information). Dependencies: model the correlations among the invasive properties Θ in order to obtain a more accurate probabilistic description and avoid physically implausible parameter configurations. As a result, given the non-invasive properties ξ of a prospective AAA patient, the logarithm (acting as a positivity constraint) of the corresponding prediction Θ(ξ) will follow a multivariate Gaussian distribution with predicted mean μlogΘ and covariance matrix Σlog Θ, i.e. (17) As we will see in Section 3.2, our approach leads to more accurate estimates for Θ and also a lower variance in the predictions. All relevant details regarding this model are provided in Appendix A.1. 2.3 A Kriging surrogate model for the maximum stress 2.3.1 Estimating the probability of rupture. Since the calculation of the probability of rupture from Eq (7) using the high-fidelity, nonlinear finite element model from Section 2.2.2 is infeasible for a clinical application, we propose a Kriging surrogate model to speed up computations [40–42]. The surrogate model will effectively serve as a proxy for the maximum von Mises stress in the AAA vessel wall (cf. Eq (16)) and allows to make computationally cheap predictions at an arbitrary combination of θ = [t, α, β]T, i.e. (18) with the predicted mean and standard deviation , respectively. For all relevant details, we refer to Appendix A.2. The high-fidelity model can then be simply approximated as , allowing for a direct Monte Carlo estimation of the probability of rupture (19) where (20) and Θi ∼ p(log Θ), i = 1…neval. 2.3.2 An active learning approach to training. The Kriging surrogate training process is carried out under the following two demands: As few as possible high-fidelity model evaluations. Ensure that the Kriging model is accurate where necessary. To that end, we adopt and extend the Active Learning-MacKay (ALM) strategy from [43] and choose points for high-fidelity model evaluations such as to minimize a density- and stress-weighted predictive standard deviation objective function (21) where p(log Θ) is the patient-specific probability distribution for the invasive model parameters Θ = [t, α, β, σγ]T from the regression model in Section 2.2.4. The reasoning behind this choice follows from the ALM approach, where only the predictive standard deviations are considered in the objective function. In our case, we are equipped with a probability distribution, p(log Θ), so we can attribute a higher weight to the more probable regions in Θ. Additionally, we pay special attention to points in the input space, where the predicted maximum von Mises stresses are high to ensure the surrogate model accurately replicates the full model in these regions. The problem of choosing an appropriate point θnext for evaluation results in the optimization problem (22) which is approximated by creating a grid over the input space, calculating using the Kriging surrogate and determining (23) The next evaluation point θnext = [tnext, αnext, βnext]T can then simply be extracted from Θnext. During the active learning, we monitor the average (24) and stop the training process, when there are no more significant changes in with an increasing number of high-fidelity model evaluations. 2.1 Failure-based probabilistic quantification of rupture risk 2.1.1 Rupture as an event of material failure. From a mechanical point of view, rupture represents an event of local material failure at a point x in the aneurysm wall, which motivates its definition via a failure function φ(x) and the failure criterion (1) We limit ourselves to stress-based failure and define rupture as an event where the local wall stress measure σ(x) exceeds the local wall strength σγ(x). This results in the failure function φ(x) = σ(x) − σγ(x), or the criterion (2) Using the equivalent von Mises stress σvm(x) as the local stress measure σ(x) and an assumed spatially constant wall strength σγ, this criterion can be evaluated as (3) where is the maximum von Mises stress . It is important to note that the above definition does not incorporate any aspect about failure over time. In order to be able to include time in the analysis, i.e. to make a statement about the risk of rupture in the next year, one would require knowledge about the future progression of the AAA for this patient, such as a model for the aneurysm growth and change in vessel wall properties. Since there is hardly any knowledge about these aspects, we limit the further discussion to a rupture risk assessment at the point of time of the acquired data. While there are sudden events like calcification-induced formation of saccular aneurysms, we assume that in most cases an AAA is a slowly progressing disease and thus our approach has, at least for the near future, sufficient predictive capability. 2.1.2 Existing criteria and rupture risk indices. Rupture risk estimation for AAAs has been an ongoing research topic over several decades, with many attempts to establish decision criteria for clinical practice. The maximum diameter criterion [1] still represents the most widely used criterion for decision making today. It is often justified by Laplace’s law, which states that the vessel wall stress is proportional to its diameter in spherical geometries. Based on this and with data obtained from several clinical studies, a very simple criterion, (4) has been formulated, relating the patient’s AAA diameter d to a critical maximum diameter dmax. While established in clinical practice and easy to apply using CT or ultrasound imaging, this criterion has often been criticized [25] and is an ongoing subject for discussion [6]. With growing computational resources and advances in the modeling of biomechanical material behavior, the simulation of patient-specific AAA models has been advanced by several research groups. Experiments on harvested AAA samples were able to reveal material parameters and failure properties. In addition with regression models [13, 26, 27] for the prediction of the individual wall strength, this enabled the definition of biomechanics-based indices [19, 20, 22, 28], such as the rupture potential index (RPI) (5) relating the von Mises stress to the wall strength. Furthermore, it could be shown [19, 20, 24] that these indices can be better rupture risk indicators than the maximum diameter criterion. Experimental testing [13, 26, 27] also revealed significant inter- and intra-patient variabilities in the mechanical properties of AAA tissue, motivating a probabilistic approach to rupture risk estimation [14, 16, 21] and resulting in the probabilistic rupture risk index (PRRI) [21] (6) where the authors used distributions for the wall thickness and wall strength that were fitted on cohort data published by our group [13]. 2.1.3 A novel probabilistic approach. In this work, we propose a novel failure-based, probabilistic rupture risk indicator that consistently incorporates all available statistical information and accounts for correlations among vessel wall properties. Fig 1 (left) illustrates the rationale for our approach, showing how part of the available data is directly involved in the estimation of the risk of rupture, while another part affects the evaluation of the computational model. In general, this data will be correlated, resulting in correlated quantities for the evaluation of rupture risk and necessitating a reformulation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Rationale for our novel formulation (left) and exemplary visualization of its estimation (right). The probability of rupture, , is calculated as the volume of the probability distribution within the triangular-shaped area marked in red. https://doi.org/10.1371/journal.pone.0242097.g001 To that end and recalling the rupture criterion from Eq (3), we can calculate the probability of rupture over the joint probability distribution as (7) where is the indicator function defined as (8) This formulation can be easily extended to, e.g., spatially varying vessel properties using Eqs (1) or (2) as failure events. Furthermore, it includes the PRRI in Eq (6) as a special case, when choosing . Lastly, it allows for a straightforward visual interpretation as illustrated in Fig 1 (right). The plot shows the joint probability distribution and visualizes the rupture event area in red. The blue area implies a high probability for the joint occurrence of the corresponding stress and strength values. The probability of rupture is simply the volume of this density within the triangular rupture event area. Thus, the larger the overlap between and the red area, the higher . 2.1.1 Rupture as an event of material failure. From a mechanical point of view, rupture represents an event of local material failure at a point x in the aneurysm wall, which motivates its definition via a failure function φ(x) and the failure criterion (1) We limit ourselves to stress-based failure and define rupture as an event where the local wall stress measure σ(x) exceeds the local wall strength σγ(x). This results in the failure function φ(x) = σ(x) − σγ(x), or the criterion (2) Using the equivalent von Mises stress σvm(x) as the local stress measure σ(x) and an assumed spatially constant wall strength σγ, this criterion can be evaluated as (3) where is the maximum von Mises stress . It is important to note that the above definition does not incorporate any aspect about failure over time. In order to be able to include time in the analysis, i.e. to make a statement about the risk of rupture in the next year, one would require knowledge about the future progression of the AAA for this patient, such as a model for the aneurysm growth and change in vessel wall properties. Since there is hardly any knowledge about these aspects, we limit the further discussion to a rupture risk assessment at the point of time of the acquired data. While there are sudden events like calcification-induced formation of saccular aneurysms, we assume that in most cases an AAA is a slowly progressing disease and thus our approach has, at least for the near future, sufficient predictive capability. 2.1.2 Existing criteria and rupture risk indices. Rupture risk estimation for AAAs has been an ongoing research topic over several decades, with many attempts to establish decision criteria for clinical practice. The maximum diameter criterion [1] still represents the most widely used criterion for decision making today. It is often justified by Laplace’s law, which states that the vessel wall stress is proportional to its diameter in spherical geometries. Based on this and with data obtained from several clinical studies, a very simple criterion, (4) has been formulated, relating the patient’s AAA diameter d to a critical maximum diameter dmax. While established in clinical practice and easy to apply using CT or ultrasound imaging, this criterion has often been criticized [25] and is an ongoing subject for discussion [6]. With growing computational resources and advances in the modeling of biomechanical material behavior, the simulation of patient-specific AAA models has been advanced by several research groups. Experiments on harvested AAA samples were able to reveal material parameters and failure properties. In addition with regression models [13, 26, 27] for the prediction of the individual wall strength, this enabled the definition of biomechanics-based indices [19, 20, 22, 28], such as the rupture potential index (RPI) (5) relating the von Mises stress to the wall strength. Furthermore, it could be shown [19, 20, 24] that these indices can be better rupture risk indicators than the maximum diameter criterion. Experimental testing [13, 26, 27] also revealed significant inter- and intra-patient variabilities in the mechanical properties of AAA tissue, motivating a probabilistic approach to rupture risk estimation [14, 16, 21] and resulting in the probabilistic rupture risk index (PRRI) [21] (6) where the authors used distributions for the wall thickness and wall strength that were fitted on cohort data published by our group [13]. 2.1.3 A novel probabilistic approach. In this work, we propose a novel failure-based, probabilistic rupture risk indicator that consistently incorporates all available statistical information and accounts for correlations among vessel wall properties. Fig 1 (left) illustrates the rationale for our approach, showing how part of the available data is directly involved in the estimation of the risk of rupture, while another part affects the evaluation of the computational model. In general, this data will be correlated, resulting in correlated quantities for the evaluation of rupture risk and necessitating a reformulation. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Rationale for our novel formulation (left) and exemplary visualization of its estimation (right). The probability of rupture, , is calculated as the volume of the probability distribution within the triangular-shaped area marked in red. https://doi.org/10.1371/journal.pone.0242097.g001 To that end and recalling the rupture criterion from Eq (3), we can calculate the probability of rupture over the joint probability distribution as (7) where is the indicator function defined as (8) This formulation can be easily extended to, e.g., spatially varying vessel properties using Eqs (1) or (2) as failure events. Furthermore, it includes the PRRI in Eq (6) as a special case, when choosing . Lastly, it allows for a straightforward visual interpretation as illustrated in Fig 1 (right). The plot shows the joint probability distribution and visualizes the rupture event area in red. The blue area implies a high probability for the joint occurrence of the corresponding stress and strength values. The probability of rupture is simply the volume of this density within the triangular rupture event area. Thus, the larger the overlap between and the red area, the higher . 2.2 Data-informed patient-specific AAA models 2.1.1 Geometry creation from CT imaging and meshing. Patient-specific 3D AAA geometries are reconstructed via a semi-automatic segmentation process from CT imaging data using the software ScanIP (Synopsys, Mountain View, California) and based on a protocol as described in [12]. The minimal requirement for the spatial resolution of CT scans was 1 mm and for the slice thickness 3 mm. The upper boundary for the segmentation was the branching of the renal arteries and the lower boundary below the bifurcation at the iliac arteries. Due to the small thickness of the AAA wall, its low contrast and the limited resolution of the CT images, it is only possible to extract the blood lumen and intraluminal thrombus (ILT) geometries. After segmentation, the ILT geometry is exported as a surface model for meshing. In a next step, we use the software Trelis (csimsoft, American Fork, Utah) and bi-linear quadrilateral elements to mesh the abluminal ILT surface. From this surface mesh, the arterial wall layer is extruded with a specified, spatially constant thickness t, resulting in a tri-linear, single layer, hexahedral mesh for the AAA wall. Finally, linear tetrahedral elements are employed for the meshing of the complex ILT geometry and a layer of linear pyramid elements as a transition for mesh compatibility between AAA wall and thrombus. Element sizes were set to 1.6 mm, corresponding to the median of measured thicknesses of AAA wall specimens in our database and leading to hexahedral elements of shape 1.6 mm × 1.6 mm × t for the AAA wall. A mesh convergence study has been performed to assess that the chosen spatial mesh resolution is sufficient in the context of our application. The meshing procedure is also described in [29] in more detail. 2.2.2 Biomechanical modeling. Previous studies have shown that in order to accurately describe the biomechanical behavior of AAAs, a sufficient model complexity is required [20, 30], while results by [31, 32] indicate that also simpler models might be appropriate. For our purposes, we employ the finite deformation boundary value problem of nonlinear elasticity: (9) (10) (11) where Ω0 is the reference configuration of the AAA, u denotes the displacement field, F = I+ ∇u the deformation gradient, S the second Piola-Kirchhoff stress tensor and σ the Cauchy stress tensor. On the Neumann boundary γσ, i.e. the luminal ILT surface, an orthonormal load is applied, with the pressure value p and the unit outward surface normal n in the current configuration. Furthermore, at the proximal and distal end surfaces of the AAA model, Γu, we employ a Robin-type boundary condition with spring supports following [29, 33]. The stiffness parameter ks is per unit reference area and set to 100 kPa/mm in this study, while N is the unit outward surface normal in the reference configuration. To model the constitutive behavior of the ILT, we use the strain energy function proposed in [34] (12) and a linearly decreasing stiffness c from the luminal to the abluminal ILT surface [12]. and are the first and second invariants of the modified right Cauchy-Green deformation tensor , with C = FT F and J = det(F) [29]. The strain energy function employed for the AAA wall material is [35, 36] (13) with stiffness parameters α and β. Both strain energy functions are equipped with an additive volumetric component (14) including the bulk modulus (15) with parameters and for the employed ILT and wall material models and a Poisson’s ratio of ν = 0.48 [12]. To obtain a pressurized in vivo configuration of the AAA, the MULF prestressing method [10, 11] is used, where the applied load corresponds to the mean arterial pressure (MAP = 1/3 systolic pressure + 2/3 diastolic pressure). From this prestressed configuration, the pressure is raised by 50% to simulate elevated blood pressure conditions [21]. Following [19] and for comparability reasons, the values for the systolic and diastolic pressures were set to 121 mmHg and 87 mmHg for all cases, respectively, resulting in a MAP of 98.33 mmHg. With the finite element discretization from Section 2.2.1, a nonlinear system of equations is obtained, which is solved using an in-house finite element code. We note that in this study we neglect the effect of calcifications in the AAA for simplicity and assume constant vessel wall thickness t and stiffness parameters α and β throughout the aneurysm. Furthermore, we evaluate the maximum von Mises stress as the 99th percentile of the von Mises stress field in the aneurysm. For the remainder of this work, we will use the parameter to quantity of interest (QoI) map (16) with parameter vector and QoI to denote the forward problem. Thus, calculating for one realization of t, α and β will involve one evaluation of the nonlinear finite element model. 2.2.3 Patient database. The modeling of patient-specific vessel wall properties here is based on data that has been collected during several research projects between 2008 and 2017 on the mechanobiological behavior of AAAs [13, 15]. The study was approved by the ethics committee of the University Hospital rechts der Isar, Technical University of Munich. AAA patients undergoing elective OSR (including emergency repair due to rupture) at the University Hospital rechts der Isar in Munich, Germany, were added to the database, whenever it was possible to extract tissue samples for mechanical testing. Apart from anamnesis and CT imaging data, hemograms were evaluated and one or more AAA tissue samples harvested during OSR. These samples were mechanically and histologically investigated, resulting in an exhaustive retrospective AAA database. Further information on data collection and experimental testing can be found in [13, 15]. To date, the database contains a total number of 305 entries from an equal number of tissue samples that were collected from 139 patients. The data can be split into two groups. Invasive properties (cf. Table 1), denoted as , are properties, which have been determined retrospectively from AAA tissue samples and cannot be obtained for a prospective patient by using clinically established methods. They are, however, essential for the biomechanical modeling and simulation of AAAs and the calculation of the probability of rupture using Eq (7). Non-invasive properties (cf. Table 2), denoted by ξ, on the other hand, can be determined with standard methods in the clinic. The subrenal diameter in Table 2 is measured directly below the renal arteries. If the aneurysm reached the renal arteries, the aortic diameter between the celiac artery and the superior mesenteric artery minus 2.5 mm was used instead [12]. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Invasive properties represent key vessel wall characteristics for a biomechanical rupture risk assessment. https://doi.org/10.1371/journal.pone.0242097.t001 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Non-invasive properties overview. https://doi.org/10.1371/journal.pone.0242097.t002 Based on correlations between the invasive and non-invasive properties [13], the goal is to construct a statistical model for the patient-individualized prediction of vessel wall properties Θ(ξ) for a prospective new patient with non-invasive properties ξ. While this process is described in Section 2.2.4, a preprocessing step for the dataset is essential, since values are missing both in the invasive and non-invasive properties for several cases in our database. Moreover, the relatively small number of available data, but high number of non-invasive properties, requires a feature selection process to identify the most important properties in ξ. Similar to [15], we conduct the following preprocessing steps. Non-invasive features in ξ, where more than 30% of the data points had missing values and patients with more than 30% of missing features were excluded and all other missing non-invasive properties imputed with the corresponding median value across the population. As a consequence, the four parameters calcium, high-sensitivity C-reactive protein (hsCRP), creatine kinase and fibrinogen were disregarded. Afterwards, all non-invasive features were normalized. Based on a correlation analysis using Spearman’s rank correlation coefficient (cf. S1 Table), the total number of features was reduced to a final selection of 8 variables: maximum AAA diameter, maximum thrombus thickness, AAA length, subrenal diameter, thrombocytes, hemoglobin, mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV). The restriction was done using a sequential forward selection algorithm similar to [15]. In an attempt to keep the number of non-invasive parameters small, we iteratively added the highest correlating non-invasive parameters to the GP model (see Section 2.2.4) until no further improvement in the leave-one-out cross-validation (LOOCV) scores could be observed. We note, however, that this does not imply that other non-invasive features such as sex, medication or anamnesis parameters do not have an influence on the biomechanical properties of the AAA wall. The resulting dataset , that was used for the analysis in Section 3, consisted of ndata = 251 data points from 113 individual patients and is available as supplementary information to this study (cf. S2 and S3 Tables). 2.2.4 Prediction of invasive vessel wall properties. Previous approaches to create models for the AAA wall thickness, stiffness parameters or strength were either deterministic [12, 26], based on cohort statistics [21], or did not account for correlations among the vessel wall quantities [15]. In the following, we make use of a multivariate Gaussian process regression model [37–39] to address these shortcomings and achieve the following desiderata: Patient-specific modeling: obtain personalized estimates for the vessel wall quantities Θ based on correlations with the non-invasive properties ξ of a specific, prospective patient. Probabilistic treatment: take into account the uncertainties in the predictions for Θ (do not ignore statistical information). Dependencies: model the correlations among the invasive properties Θ in order to obtain a more accurate probabilistic description and avoid physically implausible parameter configurations. As a result, given the non-invasive properties ξ of a prospective AAA patient, the logarithm (acting as a positivity constraint) of the corresponding prediction Θ(ξ) will follow a multivariate Gaussian distribution with predicted mean μlogΘ and covariance matrix Σlog Θ, i.e. (17) As we will see in Section 3.2, our approach leads to more accurate estimates for Θ and also a lower variance in the predictions. All relevant details regarding this model are provided in Appendix A.1. 2.1.1 Geometry creation from CT imaging and meshing. Patient-specific 3D AAA geometries are reconstructed via a semi-automatic segmentation process from CT imaging data using the software ScanIP (Synopsys, Mountain View, California) and based on a protocol as described in [12]. The minimal requirement for the spatial resolution of CT scans was 1 mm and for the slice thickness 3 mm. The upper boundary for the segmentation was the branching of the renal arteries and the lower boundary below the bifurcation at the iliac arteries. Due to the small thickness of the AAA wall, its low contrast and the limited resolution of the CT images, it is only possible to extract the blood lumen and intraluminal thrombus (ILT) geometries. After segmentation, the ILT geometry is exported as a surface model for meshing. In a next step, we use the software Trelis (csimsoft, American Fork, Utah) and bi-linear quadrilateral elements to mesh the abluminal ILT surface. From this surface mesh, the arterial wall layer is extruded with a specified, spatially constant thickness t, resulting in a tri-linear, single layer, hexahedral mesh for the AAA wall. Finally, linear tetrahedral elements are employed for the meshing of the complex ILT geometry and a layer of linear pyramid elements as a transition for mesh compatibility between AAA wall and thrombus. Element sizes were set to 1.6 mm, corresponding to the median of measured thicknesses of AAA wall specimens in our database and leading to hexahedral elements of shape 1.6 mm × 1.6 mm × t for the AAA wall. A mesh convergence study has been performed to assess that the chosen spatial mesh resolution is sufficient in the context of our application. The meshing procedure is also described in [29] in more detail. 2.2.2 Biomechanical modeling. Previous studies have shown that in order to accurately describe the biomechanical behavior of AAAs, a sufficient model complexity is required [20, 30], while results by [31, 32] indicate that also simpler models might be appropriate. For our purposes, we employ the finite deformation boundary value problem of nonlinear elasticity: (9) (10) (11) where Ω0 is the reference configuration of the AAA, u denotes the displacement field, F = I+ ∇u the deformation gradient, S the second Piola-Kirchhoff stress tensor and σ the Cauchy stress tensor. On the Neumann boundary γσ, i.e. the luminal ILT surface, an orthonormal load is applied, with the pressure value p and the unit outward surface normal n in the current configuration. Furthermore, at the proximal and distal end surfaces of the AAA model, Γu, we employ a Robin-type boundary condition with spring supports following [29, 33]. The stiffness parameter ks is per unit reference area and set to 100 kPa/mm in this study, while N is the unit outward surface normal in the reference configuration. To model the constitutive behavior of the ILT, we use the strain energy function proposed in [34] (12) and a linearly decreasing stiffness c from the luminal to the abluminal ILT surface [12]. and are the first and second invariants of the modified right Cauchy-Green deformation tensor , with C = FT F and J = det(F) [29]. The strain energy function employed for the AAA wall material is [35, 36] (13) with stiffness parameters α and β. Both strain energy functions are equipped with an additive volumetric component (14) including the bulk modulus (15) with parameters and for the employed ILT and wall material models and a Poisson’s ratio of ν = 0.48 [12]. To obtain a pressurized in vivo configuration of the AAA, the MULF prestressing method [10, 11] is used, where the applied load corresponds to the mean arterial pressure (MAP = 1/3 systolic pressure + 2/3 diastolic pressure). From this prestressed configuration, the pressure is raised by 50% to simulate elevated blood pressure conditions [21]. Following [19] and for comparability reasons, the values for the systolic and diastolic pressures were set to 121 mmHg and 87 mmHg for all cases, respectively, resulting in a MAP of 98.33 mmHg. With the finite element discretization from Section 2.2.1, a nonlinear system of equations is obtained, which is solved using an in-house finite element code. We note that in this study we neglect the effect of calcifications in the AAA for simplicity and assume constant vessel wall thickness t and stiffness parameters α and β throughout the aneurysm. Furthermore, we evaluate the maximum von Mises stress as the 99th percentile of the von Mises stress field in the aneurysm. For the remainder of this work, we will use the parameter to quantity of interest (QoI) map (16) with parameter vector and QoI to denote the forward problem. Thus, calculating for one realization of t, α and β will involve one evaluation of the nonlinear finite element model. 2.2.3 Patient database. The modeling of patient-specific vessel wall properties here is based on data that has been collected during several research projects between 2008 and 2017 on the mechanobiological behavior of AAAs [13, 15]. The study was approved by the ethics committee of the University Hospital rechts der Isar, Technical University of Munich. AAA patients undergoing elective OSR (including emergency repair due to rupture) at the University Hospital rechts der Isar in Munich, Germany, were added to the database, whenever it was possible to extract tissue samples for mechanical testing. Apart from anamnesis and CT imaging data, hemograms were evaluated and one or more AAA tissue samples harvested during OSR. These samples were mechanically and histologically investigated, resulting in an exhaustive retrospective AAA database. Further information on data collection and experimental testing can be found in [13, 15]. To date, the database contains a total number of 305 entries from an equal number of tissue samples that were collected from 139 patients. The data can be split into two groups. Invasive properties (cf. Table 1), denoted as , are properties, which have been determined retrospectively from AAA tissue samples and cannot be obtained for a prospective patient by using clinically established methods. They are, however, essential for the biomechanical modeling and simulation of AAAs and the calculation of the probability of rupture using Eq (7). Non-invasive properties (cf. Table 2), denoted by ξ, on the other hand, can be determined with standard methods in the clinic. The subrenal diameter in Table 2 is measured directly below the renal arteries. If the aneurysm reached the renal arteries, the aortic diameter between the celiac artery and the superior mesenteric artery minus 2.5 mm was used instead [12]. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Invasive properties represent key vessel wall characteristics for a biomechanical rupture risk assessment. https://doi.org/10.1371/journal.pone.0242097.t001 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Non-invasive properties overview. https://doi.org/10.1371/journal.pone.0242097.t002 Based on correlations between the invasive and non-invasive properties [13], the goal is to construct a statistical model for the patient-individualized prediction of vessel wall properties Θ(ξ) for a prospective new patient with non-invasive properties ξ. While this process is described in Section 2.2.4, a preprocessing step for the dataset is essential, since values are missing both in the invasive and non-invasive properties for several cases in our database. Moreover, the relatively small number of available data, but high number of non-invasive properties, requires a feature selection process to identify the most important properties in ξ. Similar to [15], we conduct the following preprocessing steps. Non-invasive features in ξ, where more than 30% of the data points had missing values and patients with more than 30% of missing features were excluded and all other missing non-invasive properties imputed with the corresponding median value across the population. As a consequence, the four parameters calcium, high-sensitivity C-reactive protein (hsCRP), creatine kinase and fibrinogen were disregarded. Afterwards, all non-invasive features were normalized. Based on a correlation analysis using Spearman’s rank correlation coefficient (cf. S1 Table), the total number of features was reduced to a final selection of 8 variables: maximum AAA diameter, maximum thrombus thickness, AAA length, subrenal diameter, thrombocytes, hemoglobin, mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV). The restriction was done using a sequential forward selection algorithm similar to [15]. In an attempt to keep the number of non-invasive parameters small, we iteratively added the highest correlating non-invasive parameters to the GP model (see Section 2.2.4) until no further improvement in the leave-one-out cross-validation (LOOCV) scores could be observed. We note, however, that this does not imply that other non-invasive features such as sex, medication or anamnesis parameters do not have an influence on the biomechanical properties of the AAA wall. The resulting dataset , that was used for the analysis in Section 3, consisted of ndata = 251 data points from 113 individual patients and is available as supplementary information to this study (cf. S2 and S3 Tables). 2.2.4 Prediction of invasive vessel wall properties. Previous approaches to create models for the AAA wall thickness, stiffness parameters or strength were either deterministic [12, 26], based on cohort statistics [21], or did not account for correlations among the vessel wall quantities [15]. In the following, we make use of a multivariate Gaussian process regression model [37–39] to address these shortcomings and achieve the following desiderata: Patient-specific modeling: obtain personalized estimates for the vessel wall quantities Θ based on correlations with the non-invasive properties ξ of a specific, prospective patient. Probabilistic treatment: take into account the uncertainties in the predictions for Θ (do not ignore statistical information). Dependencies: model the correlations among the invasive properties Θ in order to obtain a more accurate probabilistic description and avoid physically implausible parameter configurations. As a result, given the non-invasive properties ξ of a prospective AAA patient, the logarithm (acting as a positivity constraint) of the corresponding prediction Θ(ξ) will follow a multivariate Gaussian distribution with predicted mean μlogΘ and covariance matrix Σlog Θ, i.e. (17) As we will see in Section 3.2, our approach leads to more accurate estimates for Θ and also a lower variance in the predictions. All relevant details regarding this model are provided in Appendix A.1. 2.3 A Kriging surrogate model for the maximum stress 2.3.1 Estimating the probability of rupture. Since the calculation of the probability of rupture from Eq (7) using the high-fidelity, nonlinear finite element model from Section 2.2.2 is infeasible for a clinical application, we propose a Kriging surrogate model to speed up computations [40–42]. The surrogate model will effectively serve as a proxy for the maximum von Mises stress in the AAA vessel wall (cf. Eq (16)) and allows to make computationally cheap predictions at an arbitrary combination of θ = [t, α, β]T, i.e. (18) with the predicted mean and standard deviation , respectively. For all relevant details, we refer to Appendix A.2. The high-fidelity model can then be simply approximated as , allowing for a direct Monte Carlo estimation of the probability of rupture (19) where (20) and Θi ∼ p(log Θ), i = 1…neval. 2.3.2 An active learning approach to training. The Kriging surrogate training process is carried out under the following two demands: As few as possible high-fidelity model evaluations. Ensure that the Kriging model is accurate where necessary. To that end, we adopt and extend the Active Learning-MacKay (ALM) strategy from [43] and choose points for high-fidelity model evaluations such as to minimize a density- and stress-weighted predictive standard deviation objective function (21) where p(log Θ) is the patient-specific probability distribution for the invasive model parameters Θ = [t, α, β, σγ]T from the regression model in Section 2.2.4. The reasoning behind this choice follows from the ALM approach, where only the predictive standard deviations are considered in the objective function. In our case, we are equipped with a probability distribution, p(log Θ), so we can attribute a higher weight to the more probable regions in Θ. Additionally, we pay special attention to points in the input space, where the predicted maximum von Mises stresses are high to ensure the surrogate model accurately replicates the full model in these regions. The problem of choosing an appropriate point θnext for evaluation results in the optimization problem (22) which is approximated by creating a grid over the input space, calculating using the Kriging surrogate and determining (23) The next evaluation point θnext = [tnext, αnext, βnext]T can then simply be extracted from Θnext. During the active learning, we monitor the average (24) and stop the training process, when there are no more significant changes in with an increasing number of high-fidelity model evaluations. 2.3.1 Estimating the probability of rupture. Since the calculation of the probability of rupture from Eq (7) using the high-fidelity, nonlinear finite element model from Section 2.2.2 is infeasible for a clinical application, we propose a Kriging surrogate model to speed up computations [40–42]. The surrogate model will effectively serve as a proxy for the maximum von Mises stress in the AAA vessel wall (cf. Eq (16)) and allows to make computationally cheap predictions at an arbitrary combination of θ = [t, α, β]T, i.e. (18) with the predicted mean and standard deviation , respectively. For all relevant details, we refer to Appendix A.2. The high-fidelity model can then be simply approximated as , allowing for a direct Monte Carlo estimation of the probability of rupture (19) where (20) and Θi ∼ p(log Θ), i = 1…neval. 2.3.2 An active learning approach to training. The Kriging surrogate training process is carried out under the following two demands: As few as possible high-fidelity model evaluations. Ensure that the Kriging model is accurate where necessary. To that end, we adopt and extend the Active Learning-MacKay (ALM) strategy from [43] and choose points for high-fidelity model evaluations such as to minimize a density- and stress-weighted predictive standard deviation objective function (21) where p(log Θ) is the patient-specific probability distribution for the invasive model parameters Θ = [t, α, β, σγ]T from the regression model in Section 2.2.4. The reasoning behind this choice follows from the ALM approach, where only the predictive standard deviations are considered in the objective function. In our case, we are equipped with a probability distribution, p(log Θ), so we can attribute a higher weight to the more probable regions in Θ. Additionally, we pay special attention to points in the input space, where the predicted maximum von Mises stresses are high to ensure the surrogate model accurately replicates the full model in these regions. The problem of choosing an appropriate point θnext for evaluation results in the optimization problem (22) which is approximated by creating a grid over the input space, calculating using the Kriging surrogate and determining (23) The next evaluation point θnext = [tnext, αnext, βnext]T can then simply be extracted from Θnext. During the active learning, we monitor the average (24) and stop the training process, when there are no more significant changes in with an increasing number of high-fidelity model evaluations. 3 Results 3.1 Framework summary Based on our retrospective AAA database of non-invasive and invasive data pairs and a multi-output Gaussian process model fitted to this dataset (cf. Section 2.2.4), the necessary steps to estimate the probability of rupture for a prospective patient are: Step 1: Data generation in the clinic: CT imaging, determination of the non-invasive parameters ξ from Table 2 Step 2: Geometry creation: segmentation and meshing of the AAA geometry Step 3: Model specification: modeling of the invasive properties Θ(ξ) using the multi-output Gaussian process model from Sections 2.2.4 and A.1. Step 4: Surrogate training: fitting of the Kriging model using active learning Step 5: Post-processing: estimating the probability of rupture While CT imaging is essential for geometry creation, the rupture risk analysis can also be carried out if no non-invasive properties ξ are available for a prospective patient by using cohort statistics (cf. Model 1, Section 3.2) without personalization. The computational procedure is summarized in Algorithm 1. In practice, it has proven feasible to choose ninit = 8 (where it makes sense to include the predicted mean μlog Θ in the set of initial samples), ngrid = neval = 10, 000 and tol = 1.0 × 10−4. Algorithm 1 Calculating the probability of rupture 1: Input: Input uncertainties p(logΘ(ξ)), simulation model , tol, ninit, ngrid, neval 2: Set 3: Generate ninit samples and calculate 4: Train a Kriging surrogate using the training data 5: Create a grid over the input space and calculate (cf. Eq (24)) 6: while do 7:  Determine θnext using Eq (23) and calculate 8:   Update the Kriging model with the new data point and calculate 9:  Set iter = iter + 1 10: end while 11: Generate neval samples and calculate according to Eq (19) using the Kriging surrogate 12: Output: 3.2 Regression model benchmark Before demonstrating the framework in full detail, a brief comparison between the multi-output Gaussian process regression model (cf. Section 2.2.4) with existing probabilistic modeling approaches used in the context of AAA rupture risk is provided. To that end, we employ leave-one-out-cross-validation (LOOCV) on our dataset (cf. Section 2.2.3) to test the predictive capabilities of three different models for p(logΘ): Model 1: assuming all variables are log-normally distributed and independent, the joint distribution (25) is obtained, where the means and variances are calculated across the whole population using the dataset , that is (26) with κ ∈ {t, α, β, σγ}. This corresponds to the approach chosen in [21]. Model 2: by training single-output Gaussian processes for each output variable separately following [15], the same decomposition of Gaussian distributions as in Eq (25) is obtained, however, with means and variances predicted individually for each patient. Model 3: our proposed multi-output Gaussian process (cf. Eq (17)). In addition to the mean of the patient standardized mean square error (PSMSE) [15], we also report the mean of the patient predictive entropy (PPE), (27) where is the entropy of the distribution p(logΘ) and a measure of uncertainty or variance for multivariate distributions. With regards to the different measures, it is desirable for both PSMSE and PPE to be small, corresponding to a model which is accurate and produces low-variance estimates. For conciseness, values for the mean of the PSMSE are averaged over the four predictive variables Θ. We refer to [15] for an exhaustive discussion of the LOOCV and calculation of the PSMSE. The obtained results for the three models are shown in Table 3. We note that our proposed model (Model 3) was able to consistently achieve the lowest scores, although the differences are rather small. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Leave-one-out-cross-validation (LOOCV) results for the three probabilistic models. https://doi.org/10.1371/journal.pone.0242097.t003 3.3 Framework demonstration for AAA Pat17 To illustrate the application of our proposed framework we demonstrate all steps in detail below, following the outline as presented in Section 3.1. We assume we are provided with CT imaging data and non-invasive properties ξ for one specific prospective AAA (Step 1), referred to as Pat17 in the following. Fig 2 shows the AAA as seen via CT imaging (I), a 3D rendering of the segmentation result (II) as well as the generated finite element mesh (III) (Step 2). The mesh consists of 117, 218 finite elements and 93, 840 nodal degrees of freedom, with an approximate element size of 1.6 mm. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. AAA Pat17 as seen via CT imaging (I), a 3D rendering of the segmentation result (II), the generated finite element mesh (III) and a visualization of the von Mises stress field corresponding to the mean μlogΘ of the predictive distribution p(logΘ) for that AAA (IV). https://doi.org/10.1371/journal.pone.0242097.g002 Table 4 shows the relevant 8 non-invasive properties ξ that are used by the regression model (cf. Section 2.2.4) to obtain the predictive distribution p(logΘ(ξ)), which is specific to Pat17. Along with that, means and standard deviations based on all 113 patients in are provided. Based on this data, we can predict the mean μlogΘ and covariance ΣlogΘ for this patient (Step 3). The obtained distribution is visualized in Fig 3 and the predictive means and standard deviations are provided in Table 5 along with reference values from the cohort. The entropy of p(logΘ) is 3.3050 and thus slightly lower than the LOOCV mean (cf. Table 3). Highest correlations among the invasive properties for Pat17 can be found between t and σγ (), β and σγ (), t and β (ρt,β = −0.1966) as well as α and β (ρα,β = 0.1413). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 4. Non-invasive properties ξ for AAA Pat17 as well as cohort means and standard deviations (based on all 113 patients in ) for comparison. https://doi.org/10.1371/journal.pone.0242097.t004 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Visualization of the predictive distribution p(logΘ) transformed to the physical parameter range for AAA Pat17. Plots (I)-(VI) show 2D marginal distributions over all possible parameter combinations between t, α, β and σγ. Highest correlations are observed between t and σγ (), β and σγ (), t and β (ρt,β = −0.1966) as well as α and β (ρα,β = 0.1413). https://doi.org/10.1371/journal.pone.0242097.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 5. Predicted means and standard deviations for the invasive properties of AAA Pat17 along with cohort values over all ndata = 251 samples for comparison. https://doi.org/10.1371/journal.pone.0242097.t005 Given p(logΘ), the forward model (cf. Eq (16)) for Pat17 is defined. The probability of rupture for this AAA is approximated using a Kriging surrogate model (Step 4). Fig 2 (IV) provides a visualization of the von Mises stresses corresponding to μlogΘ, the mean parameter combination of p(logΘ). Fig 4 shows the decrease of the objective function over the number of iterations on the left as well as a comparison of the Kriging-based approximate distribution together with a Monte Carlo reference calculated using 10, 000 samples on the right. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Left: Decrease of the objective function over the number of training iterations, where the first training iteration corresponds to the Kriging surrogate after ninit = 8 model evaluations. 11 model evaluations were used for the surrogate creation. Right: Estimated Kriging-based distribution along with a Monte Carlo reference. All densities were calculated using kernel density estimation with Gaussian kernels based on 10, 000 samples of the maximum von Mises stress . https://doi.org/10.1371/journal.pone.0242097.g004 Lastly, the probability of rupture can be estimated using the Kriging surrogate (Step 5), which amounts to 0.47% for Pat17 (cf. Fig 5 for a visualization). We stress that this value must not be compared to the operative risks associated with OSR or EVAR in order to make decisions. Rather, it needs to be put into context with results for other AAA patients that have been computed using the same methodology, which is discussed below in Section 3.4. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Visualization of for all AAAs in group 1. https://doi.org/10.1371/journal.pone.0242097.g005 3.4 Comparative case-control study using diameter matched groups To test the efficacy of the framework as a rupture risk indicator and to compare it with existing biomechanical indices, we consider diameter matched groups of asymptomatic (group 1, n = 18) and known symptomatic/ruptured (group 2, n = 18) AAA patients from our database. The groups were chosen such that their maximum diameter mean and standard deviation approximately match (group 1: 62.17±7.18 mm, group 2: 63.06±7.56 mm), rendering a differentiation between the groups based on the maximum diameter criterion ineffective. For a detailed overview regarding the selection of the two groups, we refer to Table 6. After preprocessing of our original dataset (cf. Section 2.2.3), we restricted the cohort to AAAs with a maximum diameter between 50 and 80 mm in order to obtain an intermediate-sized group of patients. As a result, 64 patients remained, of which 47 had asymptomatic and 17 had symptomatic or ruptured AAAs. The latter were put into one group, since symptomatic AAAs are known to be at an elevated risk of rupture [44]. The reason for the much lower number of symptomatic/ruptured AAAs is that these AAAs often have very large diameters (>80mm). We included AAA patients from a previous case-control study by our group [19], which examined 13 asymptomatic and 12 symptomatic AAA patients. Finally, we manually selected 18 asymptomatic and 18 symptomatic/ruptured patients based on the following criteria: Find two groups with the best match in diameter. Preferably include cases where non-invasive data is available and thus patient-specific invasive properties can be predicted. Disregard cases, where CT images are not available or lack a sufficient image quality to create simulation models. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 6. Overview: Selection process for the diameter matched groups. https://doi.org/10.1371/journal.pone.0242097.t006 Detailed information for all AAAs of both groups is provided in Tables 7 and 8 and a visualization of their rupture risk indices, , in Figs 5 and 6 (cf. Appendix A.3). No patient had known connective tissue disorders. For 10 out of 18 AAAs in group 1 and for 9 out of 18 AAAs in group 2 we had non-invasive data and were thus able to use the multi-output regression model to determine a personalized input density p(logΘ). For the remaining 8 (group 1) and 9 (group 2) AAAs, we used cohort statistics, i.e. Model 1 from Section 3.2. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Visualization of for all AAAs in group 2. https://doi.org/10.1371/journal.pone.0242097.g006 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 7. Group 1 (asymptomatic, 18 ♂, 0 ♀) overview and obtained results for , RPI, PRRI and . https://doi.org/10.1371/journal.pone.0242097.t007 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 8. Group 2 (symptomatic/ruptured, 13♂, 5 ♀) overview and obtained results for , RPI, PRRI and . https://doi.org/10.1371/journal.pone.0242097.t008 We apply our framework to all 36 AAAs using an individual prospective scenario, i.e. before starting the analysis for one AAA, this patient is removed from the database, while the other 35 AAAs are included. In order to provide a comparison of with other biomechanical indices, we calculate the following additional quantities: Maximum von Mises stress at the input parameter mean (neglects any statistical information): (28) Rupture potential index [28] at the input parameter mean (neglects any statistical information, but takes into account the wall strength): (29) Probabilistic rupture risk index [21] (takes into account cohort-based uncertainties in the wall thickness and wall strength according to Model 1, Section 3.2): (30) Comprehensive results for all patients are listed in Tables 7 and 8 (cf. Appendix A.3). The average number of high-fidelity model evaluations to train the Kriging surrogate was 11. Based on these results and to evaluate the performance of the individual quantities, we provide: Relative mean and median differences between group 1 and group 2 (cf. Table 9). Boxplots for both groups (cf. Fig 7). Receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) (cf. Fig 8) [45]. Computed true positive rates (TPR), false positive rates (FPR) and corresponding threshold values are provided for as supplementary information (cf. S4 Table). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Boxplots comparing dmax, , RPI, PRRI and for the asymptomatic and symptomatic/ruptured group. The plots illustrate the interquartile range (green and red color) including the sample median as well as the first and third quartiles. Whiskers indicate minimum and maximum values and black dots represent all values from the respective group. https://doi.org/10.1371/journal.pone.0242097.g007 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 9. Relative mean and median differences (in %) of dmax, , RPI, PRRI and between the asymptomatic and the symptomatic/ruptured group. https://doi.org/10.1371/journal.pone.0242097.t009 3.1 Framework summary Based on our retrospective AAA database of non-invasive and invasive data pairs and a multi-output Gaussian process model fitted to this dataset (cf. Section 2.2.4), the necessary steps to estimate the probability of rupture for a prospective patient are: Step 1: Data generation in the clinic: CT imaging, determination of the non-invasive parameters ξ from Table 2 Step 2: Geometry creation: segmentation and meshing of the AAA geometry Step 3: Model specification: modeling of the invasive properties Θ(ξ) using the multi-output Gaussian process model from Sections 2.2.4 and A.1. Step 4: Surrogate training: fitting of the Kriging model using active learning Step 5: Post-processing: estimating the probability of rupture While CT imaging is essential for geometry creation, the rupture risk analysis can also be carried out if no non-invasive properties ξ are available for a prospective patient by using cohort statistics (cf. Model 1, Section 3.2) without personalization. The computational procedure is summarized in Algorithm 1. In practice, it has proven feasible to choose ninit = 8 (where it makes sense to include the predicted mean μlog Θ in the set of initial samples), ngrid = neval = 10, 000 and tol = 1.0 × 10−4. Algorithm 1 Calculating the probability of rupture 1: Input: Input uncertainties p(logΘ(ξ)), simulation model , tol, ninit, ngrid, neval 2: Set 3: Generate ninit samples and calculate 4: Train a Kriging surrogate using the training data 5: Create a grid over the input space and calculate (cf. Eq (24)) 6: while do 7:  Determine θnext using Eq (23) and calculate 8:   Update the Kriging model with the new data point and calculate 9:  Set iter = iter + 1 10: end while 11: Generate neval samples and calculate according to Eq (19) using the Kriging surrogate 12: Output: 3.2 Regression model benchmark Before demonstrating the framework in full detail, a brief comparison between the multi-output Gaussian process regression model (cf. Section 2.2.4) with existing probabilistic modeling approaches used in the context of AAA rupture risk is provided. To that end, we employ leave-one-out-cross-validation (LOOCV) on our dataset (cf. Section 2.2.3) to test the predictive capabilities of three different models for p(logΘ): Model 1: assuming all variables are log-normally distributed and independent, the joint distribution (25) is obtained, where the means and variances are calculated across the whole population using the dataset , that is (26) with κ ∈ {t, α, β, σγ}. This corresponds to the approach chosen in [21]. Model 2: by training single-output Gaussian processes for each output variable separately following [15], the same decomposition of Gaussian distributions as in Eq (25) is obtained, however, with means and variances predicted individually for each patient. Model 3: our proposed multi-output Gaussian process (cf. Eq (17)). In addition to the mean of the patient standardized mean square error (PSMSE) [15], we also report the mean of the patient predictive entropy (PPE), (27) where is the entropy of the distribution p(logΘ) and a measure of uncertainty or variance for multivariate distributions. With regards to the different measures, it is desirable for both PSMSE and PPE to be small, corresponding to a model which is accurate and produces low-variance estimates. For conciseness, values for the mean of the PSMSE are averaged over the four predictive variables Θ. We refer to [15] for an exhaustive discussion of the LOOCV and calculation of the PSMSE. The obtained results for the three models are shown in Table 3. We note that our proposed model (Model 3) was able to consistently achieve the lowest scores, although the differences are rather small. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Leave-one-out-cross-validation (LOOCV) results for the three probabilistic models. https://doi.org/10.1371/journal.pone.0242097.t003 3.3 Framework demonstration for AAA Pat17 To illustrate the application of our proposed framework we demonstrate all steps in detail below, following the outline as presented in Section 3.1. We assume we are provided with CT imaging data and non-invasive properties ξ for one specific prospective AAA (Step 1), referred to as Pat17 in the following. Fig 2 shows the AAA as seen via CT imaging (I), a 3D rendering of the segmentation result (II) as well as the generated finite element mesh (III) (Step 2). The mesh consists of 117, 218 finite elements and 93, 840 nodal degrees of freedom, with an approximate element size of 1.6 mm. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. AAA Pat17 as seen via CT imaging (I), a 3D rendering of the segmentation result (II), the generated finite element mesh (III) and a visualization of the von Mises stress field corresponding to the mean μlogΘ of the predictive distribution p(logΘ) for that AAA (IV). https://doi.org/10.1371/journal.pone.0242097.g002 Table 4 shows the relevant 8 non-invasive properties ξ that are used by the regression model (cf. Section 2.2.4) to obtain the predictive distribution p(logΘ(ξ)), which is specific to Pat17. Along with that, means and standard deviations based on all 113 patients in are provided. Based on this data, we can predict the mean μlogΘ and covariance ΣlogΘ for this patient (Step 3). The obtained distribution is visualized in Fig 3 and the predictive means and standard deviations are provided in Table 5 along with reference values from the cohort. The entropy of p(logΘ) is 3.3050 and thus slightly lower than the LOOCV mean (cf. Table 3). Highest correlations among the invasive properties for Pat17 can be found between t and σγ (), β and σγ (), t and β (ρt,β = −0.1966) as well as α and β (ρα,β = 0.1413). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 4. Non-invasive properties ξ for AAA Pat17 as well as cohort means and standard deviations (based on all 113 patients in ) for comparison. https://doi.org/10.1371/journal.pone.0242097.t004 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Visualization of the predictive distribution p(logΘ) transformed to the physical parameter range for AAA Pat17. Plots (I)-(VI) show 2D marginal distributions over all possible parameter combinations between t, α, β and σγ. Highest correlations are observed between t and σγ (), β and σγ (), t and β (ρt,β = −0.1966) as well as α and β (ρα,β = 0.1413). https://doi.org/10.1371/journal.pone.0242097.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 5. Predicted means and standard deviations for the invasive properties of AAA Pat17 along with cohort values over all ndata = 251 samples for comparison. https://doi.org/10.1371/journal.pone.0242097.t005 Given p(logΘ), the forward model (cf. Eq (16)) for Pat17 is defined. The probability of rupture for this AAA is approximated using a Kriging surrogate model (Step 4). Fig 2 (IV) provides a visualization of the von Mises stresses corresponding to μlogΘ, the mean parameter combination of p(logΘ). Fig 4 shows the decrease of the objective function over the number of iterations on the left as well as a comparison of the Kriging-based approximate distribution together with a Monte Carlo reference calculated using 10, 000 samples on the right. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Left: Decrease of the objective function over the number of training iterations, where the first training iteration corresponds to the Kriging surrogate after ninit = 8 model evaluations. 11 model evaluations were used for the surrogate creation. Right: Estimated Kriging-based distribution along with a Monte Carlo reference. All densities were calculated using kernel density estimation with Gaussian kernels based on 10, 000 samples of the maximum von Mises stress . https://doi.org/10.1371/journal.pone.0242097.g004 Lastly, the probability of rupture can be estimated using the Kriging surrogate (Step 5), which amounts to 0.47% for Pat17 (cf. Fig 5 for a visualization). We stress that this value must not be compared to the operative risks associated with OSR or EVAR in order to make decisions. Rather, it needs to be put into context with results for other AAA patients that have been computed using the same methodology, which is discussed below in Section 3.4. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Visualization of for all AAAs in group 1. https://doi.org/10.1371/journal.pone.0242097.g005 3.4 Comparative case-control study using diameter matched groups To test the efficacy of the framework as a rupture risk indicator and to compare it with existing biomechanical indices, we consider diameter matched groups of asymptomatic (group 1, n = 18) and known symptomatic/ruptured (group 2, n = 18) AAA patients from our database. The groups were chosen such that their maximum diameter mean and standard deviation approximately match (group 1: 62.17±7.18 mm, group 2: 63.06±7.56 mm), rendering a differentiation between the groups based on the maximum diameter criterion ineffective. For a detailed overview regarding the selection of the two groups, we refer to Table 6. After preprocessing of our original dataset (cf. Section 2.2.3), we restricted the cohort to AAAs with a maximum diameter between 50 and 80 mm in order to obtain an intermediate-sized group of patients. As a result, 64 patients remained, of which 47 had asymptomatic and 17 had symptomatic or ruptured AAAs. The latter were put into one group, since symptomatic AAAs are known to be at an elevated risk of rupture [44]. The reason for the much lower number of symptomatic/ruptured AAAs is that these AAAs often have very large diameters (>80mm). We included AAA patients from a previous case-control study by our group [19], which examined 13 asymptomatic and 12 symptomatic AAA patients. Finally, we manually selected 18 asymptomatic and 18 symptomatic/ruptured patients based on the following criteria: Find two groups with the best match in diameter. Preferably include cases where non-invasive data is available and thus patient-specific invasive properties can be predicted. Disregard cases, where CT images are not available or lack a sufficient image quality to create simulation models. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 6. Overview: Selection process for the diameter matched groups. https://doi.org/10.1371/journal.pone.0242097.t006 Detailed information for all AAAs of both groups is provided in Tables 7 and 8 and a visualization of their rupture risk indices, , in Figs 5 and 6 (cf. Appendix A.3). No patient had known connective tissue disorders. For 10 out of 18 AAAs in group 1 and for 9 out of 18 AAAs in group 2 we had non-invasive data and were thus able to use the multi-output regression model to determine a personalized input density p(logΘ). For the remaining 8 (group 1) and 9 (group 2) AAAs, we used cohort statistics, i.e. Model 1 from Section 3.2. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Visualization of for all AAAs in group 2. https://doi.org/10.1371/journal.pone.0242097.g006 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 7. Group 1 (asymptomatic, 18 ♂, 0 ♀) overview and obtained results for , RPI, PRRI and . https://doi.org/10.1371/journal.pone.0242097.t007 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 8. Group 2 (symptomatic/ruptured, 13♂, 5 ♀) overview and obtained results for , RPI, PRRI and . https://doi.org/10.1371/journal.pone.0242097.t008 We apply our framework to all 36 AAAs using an individual prospective scenario, i.e. before starting the analysis for one AAA, this patient is removed from the database, while the other 35 AAAs are included. In order to provide a comparison of with other biomechanical indices, we calculate the following additional quantities: Maximum von Mises stress at the input parameter mean (neglects any statistical information): (28) Rupture potential index [28] at the input parameter mean (neglects any statistical information, but takes into account the wall strength): (29) Probabilistic rupture risk index [21] (takes into account cohort-based uncertainties in the wall thickness and wall strength according to Model 1, Section 3.2): (30) Comprehensive results for all patients are listed in Tables 7 and 8 (cf. Appendix A.3). The average number of high-fidelity model evaluations to train the Kriging surrogate was 11. Based on these results and to evaluate the performance of the individual quantities, we provide: Relative mean and median differences between group 1 and group 2 (cf. Table 9). Boxplots for both groups (cf. Fig 7). Receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) (cf. Fig 8) [45]. Computed true positive rates (TPR), false positive rates (FPR) and corresponding threshold values are provided for as supplementary information (cf. S4 Table). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Boxplots comparing dmax, , RPI, PRRI and for the asymptomatic and symptomatic/ruptured group. The plots illustrate the interquartile range (green and red color) including the sample median as well as the first and third quartiles. Whiskers indicate minimum and maximum values and black dots represent all values from the respective group. https://doi.org/10.1371/journal.pone.0242097.g007 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 9. Relative mean and median differences (in %) of dmax, , RPI, PRRI and between the asymptomatic and the symptomatic/ruptured group. https://doi.org/10.1371/journal.pone.0242097.t009 4 Discussion The obtained values for the relative mean and median differences in Table 9 confirm that group 1 and group 2 are indistinguishable based on the maximum diameter criterion. While the relative differences are higher for and RPI, PRRI and in particular our proposed index feature a significantly larger mean and median difference. Recalling that the maximum diameter, dmax, is one important non-invasive parameter in our framework (cf. Section 2.2.3), we emphasize that its influence has been rendered ineffective through the study design. A similar trend as in Table 9 can be observed in Fig 7, with RPI and PRRI providing a slightly better separation between the two groups than , while for the interquartile ranges of the two groups are non-overlapping. Finally, in Fig 8 we can observe that outperforms the remaining classifiers and achieves the best performance among all quantities in terms of the AUC score, followed by PRRI, RPI and . We further note that from the 18 patients in the symptomatic/ruptured group, 11 had ruptured AAAs (Pat19, Pat23, Pat24, Pat26, Pat27, Pat28, Pat29, Pat30, Pat32, Pat34, Pat35). The mean scores for the 11 ruptured AAAs is 6.57 and thus slightly higher than the mean 5.89 for the 7 symptomatic AAAs. To summarize our key observations: The maximum diameter criterion, by design, clearly fails to separate the two groups in all our comparisons. The proposed index consistently achieves the best separation. The results indicate that the more statistical information taken into account, the better the capability to distinguish between group 1 and group 2. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Receiver operating characteristic (ROC) curves showing true positive rates (TPR) over false positive rates (FPR) and area under the ROC curve (AUC) scores for dmax, , RPI, PRRI and . https://doi.org/10.1371/journal.pone.0242097.g008 Before translating these findings into any clinical application, however, there are several limitations that have to be kept in mind. First, this is a non-randomized, retrospective case-control study with a relatively small cohort size (group 1: n = 18, group 2: n = 18) and the database described in Section 2.2.3. Second, there was no matching based on other risk factors such as sex, age or family history, which could be a confounder. Third, since we only have access to electively repaired or symptomatic/ruptured AAAs for mechanical testing, the mean diameters of the two groups (group 1: 62.17 mm, group 2: 63.03 mm) are larger than the Society for Vascular Surgery’s decision criterion for elective repair (55 mm) [1]. In the future, due to the increasing use of EVAR, it will be even harder to obtain representative tissue samples from AAAs of relevant size for a database. As a result, caution is advised when interpreting the results presented here for smaller AAAs, e.g. of size 45 − 55 mm. Furthermore, all discussed approaches are unable to make any prediction about the future development of the AAA, such that the rupture risk assessment only holds for the point in time of data generation. In addition to that, the biomechanical model does not take into account factors like growth, calcifications and surrounding organs, which might be important for the analysis. 5 Conclusion We presented a novel data-informed, highly personalized, probabilistic framework for the quantification of abdominal aortic aneurysm (AAA) rupture risk and demonstrated competitive performance in comparison to existing approaches. Our framework results in the calculation of a rupture risk index, , which can be introduced as a relevant additional piece of information in the clinical decision process for AAA cases that are not or not unambiguously covered by existing guidelines and recommendations. In view of our results it is suggested to incorporate personalized, or at least cohort-based, statistical information and choose a probabilistic approach for the biomechanical rupture risk assessment. Deterministic indices were shown to be less accurate and do not account for possible sensitivities due to uncertain vessel wall quantities. In order to advance this framework to a clinical application, several further aspects need to be examined. Challenges lie especially in the fully automatic segmentation of the CT imaging data, which at the moment requires manual steps by a trained expert and can be time consuming. In view of the limitations discussed in Section 4, a larger, randomized study with risk factor matched groups is desirable to confirm this study’s findings regarding its clinical use. Future work will also address how further model parameters such as the blood pressure influence the rupture risk index and whether this quantity should be treated probabilistically as well. Lastly, to be able to make predictions over time, it is required to incorporate AAA growth [46, 47] into the framework and analyze its effect on the biomechanical rupture risk assessment. A Appendix A.1 Multi-output Gaussian process regression The data generation process for logΘ is assumed to underly a function that is contaminated by additive Gaussian noise, such that (31) It is further postulated that the vector follows a multivariate Gaussian distribution with the positive semi-definite covariance matrix and it is assumed that , with the diagonal matrix S and noise levels . Demanding that every entry of the vector corresponds to the same zero mean Gaussian process with covariance function k(ξ, ξ′), i.e. (32) can be expressed as a multivariate Gaussian process [39] (33) As a result, the collection follows a matrix-variate Gaussian distribution (34) with the covariance matrix K and entries Kij = k(ξi, ξj), modeling the covariance between two inputs ξi and ξj. Expressing the matrix Gaussian distribution as a multivariate Gaussian distribution and incorporating the additive noise ϵ, one obtains (35) where ⊗ denotes the Kronecker product and the ndata × ndata identity matrix. For our purposes, we choose the covariance function (36) with hyperparameters ζ1, ζ2, ζ3 and ζ4. Following [37–39], the matrix Ω is parameterized via the entries Lij of a Cholesky decomposition Ω = LLT. Together with the noise parameters from the matrix S, this results in the hyperparameter vector (37) where and . The predicted mean for an arbitrary point ξ⋆ becomes (38) and the predicted covariance (39) where k⋆ denotes the vector of covariance function evaluations between ξ⋆ and the data , i.e. (40) Finally, the log marginal likelihood is (41) and can be optimized with respect to its hyperparameters ζ. A.2 Kriging surrogate incorporating explicit basis functions Kriging can be regarded as a special case of a Gaussian process, where data points are assumed noise-free to interpolate the high-fidelity model at the provided high-fidelity evaluations. To find an adequate function for this purpose, we use a Kriging interpolation model that incorporates explicit basis functions as described in [42]. Using such a model, it is possible to exactly represent functions that can be described by the provided basis. Ensuring positive predictions via a log transformation, we approximate the high-fidelity model as (42) where is a zero mean Gaussian process with covariance function k(θ, θ′) and h(θ) denotes the chosen basis functions with coefficients η. A simple squared exponential kernel (43) is chosen, where the matrix is diagonal, leading to the vector of hyperparameters ζ = [ζ1, ζ2, ζ3, ζ4]T, with . Furthermore, trilinear basis functions, i.e. (44) are employed. Assuming a Gaussian prior for the coefficients, , this results in the Gaussian process (45) The dependence on the prior parameters b and B can be resolved, if a vague prior for η is chosen, i.e. if the limiting case is considered, where B−1 approaches the zero matrix 0. In that case, the predicted mean for a new point θ⋆ becomes (46) and the predicted variance (47) where , r = h⋆ − HK−1 k⋆ and is the vector of neval model evaluations. K is the data covariance matrix between all training data points such that (48) Moreover, k⋆ is a vector with the covariances between training and test points, i.e. (49) H a matrix containing vectors h(θ) at all training data points and h⋆ = h(θ⋆). It is interesting to note, how the terms in Eqs (46) and (47) consist of a contribution from the zero mean Gaussian process predicted mean and variance, and , and additional terms involving the provided basis functions, respectively. Finally, the marginal log likelihood is (50) where A = HK−1 HT C = K−1 HT A−1 HK−1 and m is the rank of HT. Maximizing the marginal log likelihood, optimal values for the hyperparameters ζ of the Kriging covariance model (cf. Eq (43)) and for the provided and can be determined. A.1 Multi-output Gaussian process regression The data generation process for logΘ is assumed to underly a function that is contaminated by additive Gaussian noise, such that (31) It is further postulated that the vector follows a multivariate Gaussian distribution with the positive semi-definite covariance matrix and it is assumed that , with the diagonal matrix S and noise levels . Demanding that every entry of the vector corresponds to the same zero mean Gaussian process with covariance function k(ξ, ξ′), i.e. (32) can be expressed as a multivariate Gaussian process [39] (33) As a result, the collection follows a matrix-variate Gaussian distribution (34) with the covariance matrix K and entries Kij = k(ξi, ξj), modeling the covariance between two inputs ξi and ξj. Expressing the matrix Gaussian distribution as a multivariate Gaussian distribution and incorporating the additive noise ϵ, one obtains (35) where ⊗ denotes the Kronecker product and the ndata × ndata identity matrix. For our purposes, we choose the covariance function (36) with hyperparameters ζ1, ζ2, ζ3 and ζ4. Following [37–39], the matrix Ω is parameterized via the entries Lij of a Cholesky decomposition Ω = LLT. Together with the noise parameters from the matrix S, this results in the hyperparameter vector (37) where and . The predicted mean for an arbitrary point ξ⋆ becomes (38) and the predicted covariance (39) where k⋆ denotes the vector of covariance function evaluations between ξ⋆ and the data , i.e. (40) Finally, the log marginal likelihood is (41) and can be optimized with respect to its hyperparameters ζ. A.2 Kriging surrogate incorporating explicit basis functions Kriging can be regarded as a special case of a Gaussian process, where data points are assumed noise-free to interpolate the high-fidelity model at the provided high-fidelity evaluations. To find an adequate function for this purpose, we use a Kriging interpolation model that incorporates explicit basis functions as described in [42]. Using such a model, it is possible to exactly represent functions that can be described by the provided basis. Ensuring positive predictions via a log transformation, we approximate the high-fidelity model as (42) where is a zero mean Gaussian process with covariance function k(θ, θ′) and h(θ) denotes the chosen basis functions with coefficients η. A simple squared exponential kernel (43) is chosen, where the matrix is diagonal, leading to the vector of hyperparameters ζ = [ζ1, ζ2, ζ3, ζ4]T, with . Furthermore, trilinear basis functions, i.e. (44) are employed. Assuming a Gaussian prior for the coefficients, , this results in the Gaussian process (45) The dependence on the prior parameters b and B can be resolved, if a vague prior for η is chosen, i.e. if the limiting case is considered, where B−1 approaches the zero matrix 0. In that case, the predicted mean for a new point θ⋆ becomes (46) and the predicted variance (47) where , r = h⋆ − HK−1 k⋆ and is the vector of neval model evaluations. K is the data covariance matrix between all training data points such that (48) Moreover, k⋆ is a vector with the covariances between training and test points, i.e. (49) H a matrix containing vectors h(θ) at all training data points and h⋆ = h(θ⋆). It is interesting to note, how the terms in Eqs (46) and (47) consist of a contribution from the zero mean Gaussian process predicted mean and variance, and , and additional terms involving the provided basis functions, respectively. Finally, the marginal log likelihood is (50) where A = HK−1 HT C = K−1 HT A−1 HK−1 and m is the rank of HT. Maximizing the marginal log likelihood, optimal values for the hyperparameters ζ of the Kriging covariance model (cf. Eq (43)) and for the provided and can be determined. Supporting information S1 Table. Correlation analysis. Computed correlations between all non-invasive and invasive AAA vessel wall properties using Spearman’s rank correlation coefficient. https://doi.org/10.1371/journal.pone.0242097.s001 (CSV) S2 Table. Non-invasive data. Dataset containing the eight non-invasive AAA vessel wall properties: maximum AAA diameter, maximum thrombus thickness, AAA length, subrenal diameter, thrombocytes, hemoglobin, mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV). All values are provided in their respective unit as specified in Table 2. The first column is a consecutive number for the individual patients and allows for the identification of data that was collected from the same patient. https://doi.org/10.1371/journal.pone.0242097.s002 (CSV) S3 Table. Invasive data. Dataset containing the four invasive AAA vessel wall properties: wall thickness, alpha stiffness, beta stiffness, wall strength. All values are provided in their respective unit as specified in Table 1. The first column is a consecutive number for the individual patients and allows for the identification of data that was collected from the same patient. https://doi.org/10.1371/journal.pone.0242097.s003 (CSV) S4 Table. ROC values. Computed true positive rates (TPR), false positive rates (FPR) and corresponding threshold values for . https://doi.org/10.1371/journal.pone.0242097.s004 (CSV)

Gruber, Maria;König, Daniel;Holzhäuser, Julika;Castillo, Deirdre Maria;Blüml, Victor;Jahn, Rebecca;Leser, Carmen;Werneck-Rohrer, Sonja;Werneck, Harald

doi: 10.1371/journal.pone.0242518pmid: 33212471

Objective Perceived parental influence on diet in early adolescence in the context of the parental relationship had previously not been studied in a clinical sample. The aim of this study was to investigate a possible association between eating disorders and characteristics of the relationship with parents and the parental feeding practices in early adolescence. Methods 21 female adolescents and young adults with an eating disorder (ED)–bulimia nervosa or anorexia nervosa–and 22 females without eating disorder (healthy control; HC), aged between 16 and 26, were assessed via self-report questionnaires for problematic eating behaviour, relationship with parents, perceptions of parent’s feeding practices at the age of 10–13 years and personality. Statistical evaluation was performed by means of group comparisons, effect sizes, regression analyses and mediator analyses. Results Adolescent and young adult females with ED reported more fears/overprotection and rejection/neglect by their mothers and less self-responsibility in terms of eating behaviour during adolescence than did the HC. The relationship with the fathers did not differ significantly. Females who perceived more cohesion, rejection/neglect and fears/overprotection by the mother were more likely to suffer from an ED. Rejection/neglect by both parents were associated with less self-acceptance of the young females with even stronger effect sizes for the fathers than the mothers. Harm prevention in the young females was a partial mediator between fears/overprotection and the drive for thinness. Conclusions The parental relationship is partly reflected in the self-acceptance and self-responsibility in eating of the adolescent and young females, both of them are particularly affected in EDs. Stressors in the parent-child relationship should be targeted in treatment of eating disorders. Nutritional counselling for parents might be useful in early adolescence. Introduction Feeding and eating disorders (ED) are characterized by disturbances of eating and other behaviour, leading to changes in food consumption and impairing physical health or psychosocial functioning: Anorexia nervosa (AN) is defined by restrictions in energy intake in relation to need, resulting in significantly low weight. Bulimia nervosa (BN) describes recurring episodes of binge eating and inappropriate compensatory behaviour to prevent weight gain. In both disorders, self-evaluation is affected and strongly influenced by the way one experiences one’s body shape and weight [1]. ED mainly affect females and exhibit a significant increase of incidence during adolescence [2]. Lifetime prevalence of AN and BN for females is reported to be 2.9% (0.1% among males of the same age brackets) [3]. A longitudinal cohort study on children, adolescents and young adults in Germany [4] found that 21.9% of the 11 to 17 years old even showed symptoms of an ED. Furthermore, the rates of incidence as well as prevalence of AN among adolescents have been increasing significantly in recent years [5] particularly in the highly vulnerable period before or at the beginning of adolescence [6]. Pubertal development, with its physical changes, often precedes or accompanies adolescence and early maturing girls are reported to be at special risk from disordered eating [7]. Adolescence involves several major developmental tasks such as adapting to physical changes, creating one's own identity, defining goals for the future or learning self-regulation of emotions which present stressful situations for young girls [2]. The bio-psycho-social model for the development of psychiatric disorders [8], representing the theoretical background of this study, postulates, that the genetic vulnerability for developing a certain psychiatric disorder, together with the intrapersonal stress and the environmental stress factors (family, friends, schooling situation), may lead to the outbreak of a disease. Since adolescence represents a period of the life cycle characterised by progressive independence from parents and family members, it may be particularly valuable to examine the extent to which family factors influence the behavioural and emotional wellbeing of the adolescents and consequently to the risk of developing an ED [2]. Subjectively perceived stressors in the parental relationship are hypothesized to increase the risk for psychosocial maladaptation and the occurrence of EDs [9]. Furthermore, the development of abilities to regulate emotions and impulses is influenced by the parental relationship [10] and the capacity to self-regulate was shown to be negatively affected by stronger parental control [11]. Attachment theory provides a comprehensive framework for understanding the psychological factors and family characteristics contributing to individual development in adolescence [12]. Early interactions and experiences with parents are internalized and used to regulate self-esteem. According to the attachment theory, secure attachment with availability, understanding and responsiveness of caregivers leads to a healthy self-esteem, while negative or inconsistent parenting behaviour is more likely to lead to problems in self-assessment [12]. The relationship between children and their parents instinctively shows different forms of bidirectional communications in different contexts and aims. Especially the feeding situation seems to play an important role in child-parent attachment. It has been shown that during feeding, when the relationship between children and parents is not able to coordinate and share sensitive interactions, the infant will not be able to learn the self-regulation of affects and may show maladaptive emotional and behavioural symptoms over time [13]. These parenting practices and their influence on children's emotional as well as food self-regulation abilities [14] are believed to play an important role in the development and maintenance of EDs [10, 14]. Parental overprotection during childhood is thought to lead to increased awareness of stress and mental health problems in adulthood [15]. Both, fathers and mothers of persons with EDs, are reported to overprotect their children [16]. However, while females with AN more frequently than their healthy siblings reported to have experienced maternal overprotection in their childhood, no such difference was reported for females with BN [17]. Contradictory, parents of patients suffering of AN [18–20] or BN [19] are postulated to be perceived as less caring, highly disengaged, poorly interwoven, rigid and with less cohesion and communication qualities, in comparison to parents of the respective peer group. It was also previously demonstrated, that persons with EDs experienced strong entanglement within their families [21]. To better understand parental impact on adolescents’ eating disorders, it is necessary to address the influence of parental feeding practices on eating behaviour. The current study addresses the limitations in the existing literature in several ways: Female patients with eating disorders (ED) and healthy controls (HC) were asked for both, their recent relationship with their parents and for their memories of parental feeding practices at early adolescence, retrospectively. The first aim was to examine female adolescent and young adults with ED as well as the HC group for their contemporary perception of the relationships with their mother or father. Dysfunctional relationships with parents in females with ED are reported in literature, but previous data are contradictory. We expected the adolescent and young females with ED to report a significantly worse relationship with their mother or father than the HC. The second aim was to explore a possible mediating role of the personality traits (“self-directedness”, “self-acceptance” and “harm avoidance”) on the maternal relationship and symptoms of eating disorders (“drive for thinness” and “body dissatisfaction”). Recent findings suggest self-regulating skills, such as self-compassion, reduce the effects of perfectionism on the development of ED [22]. Therefore, in the present study we anticipated a possible positive influence of personality traits on maternal relationship and eating disorder symptoms. The third aim was to retrospectively evaluate the perceived parental influence on eating behaviour at the age of 10–13 years. Previous studies investigating the connection between the influence of parents’ feeding practices and anorexic or bulimic EDs have primarily focused on early childhood and are still under-represented in research [23]. To the best of our knowledge, no previous study exists investigating the perceived parental’ feeding practices in early adolescence in the context of the parental relationship in a clinical sample of females diagnosed with ED. However, controlling food-related parenting practices were associated with adolescent disordered eating behaviours in population-based samples [24], and restrictive parental feeding practices were shown to lead to disordered eating behaviour in children in longitudinal studies [25, 26]. We expected remembered parental feeding practices in terms of “feeding responsibility”, “monitoring”, “pressure to eat” and “restriction” to be significantly more frequent in females with EDs than in HC. Methods Subjects and procedures Study design. This quantitative cross-sectional study aimed at examining the current situation and memories of the early adolescence in female adolescents and young adults with and without eating disorders (ED) using questionnaires. The data was collected in the period from September 2016 to February 2017. Although no longitudinally collected data was available, cause-effect conclusions could be drawn by using a structural modelling strategy [27]. Ethics and consent to participate. This study was approved by the ethics committee of the Medical University of Vienna (reference number: 1691/2016). Written consent by the participants, after thorough information about the study, was acquired in advance in all cases. Adolescents between 14 and 18 years of age are of legal age of consent in Austria. Thus, young females’ consent in our study population of older than 16 years, was sufficient–and in accordance with the Ethics Committee–and parents did not have to be informed about participation in the study. Study setting. Female adolescents and young adults with ED were recruited from inpatient and outpatient care of the Clinical Division of Social Psychiatry at the Department for Psychiatry and Psychotherapy, and of the Department for Child and Adolescent Psychiatry of the Medical University of Vienna. The healthy control group (HC) was recruited via circle of acquaintances. Study sample. Female adolescents and young adults with eating disorder. Females with an eating disorder (ED) were included if they a) were between 16 and 26 years of age b) were diagnosed with either the ICD-10 diagnosis bulimia nervosa (BN) (ICD-10: F50.2) or anorexia nervosa (AN) (ICD-10: F50.00)—restrictive type or AN—bulimic type (ICD-10: F50.01). The diagnoses were established by experienced clinicians at the recruitment institutions. Healthy controls. Inclusion criteria for the healthy control group (HC) were a) the willingness to participate and b) sufficient German language skills. Exclusion criteria for the HC were a) the existence of a current or past diagnosis of an ED, b) values above the normal range in at least two of the three scales: “drive for thinness”, “bulimia”, “body dissatisfaction” of the Eating Disorder Inventory-2 (EDI-2) [28]. When recruiting the control group, a similar age distribution as in the group of females with eating disorders was aimed at, but no age matching was performed in the analysis. Measurements Demographic data of the participants were collected by means of a study specific questionnaire. Data about age, weight, diagnosed mental health disorders according to ICD-10 [29], participation and duration of psychological/psychiatric treatment, housing situation with/without parents and school degree were collected. The symptoms of EDs were investigated using the frequently used Eating Disorder Inventory 2 (EDI-2; German Version) [28]. The EDI-2 is a revision of the EDI-1 [30], in which three subscales have been added. In the original English version of the EDI-2, the internal consistencies (Cronbach's Alphas) for the eight original scales ranged from 0.80 to 0.91, of which three were used in this study and a Principal Component Analysis supported the factor structure of the original EDI, but not that of the three additional scales [31] not used in this study. For our study, the validated German version of the questionnaire was used, which contains 91 items on a 6-level rating scale representing 11 scales. The German version achieved an Cronbach's Alpha between α = 0.73 and α = 0.93 in a sample of anorexic and bulimic patients, the test-retest reliability ranged from rtt = 0.81 to rtt = 0.89 [28]. In the current study, the three subscales (23 items) with the most relevant content for the symptoms of eating disorders were applied and Cronbach’s Alpha of these scales reached α = 0.91 for “drive for thinness”, α = 0.95 for “bulimia” and α = 0.93 for “body dissatisfaction” in the study population. The quality of the relationship with the parents was measured with the Parental-Representation-Screening-Questionnaire (PRSQ); German Version [Elternbildfragebogen für Kinder- und Jugendliche (EBF-KJ)] for children and adolescents [32]. This German questionnaire, which was applied in this study in its original version, consists of 36 items answered on a 5-level Likert scale and contains three resource-scales (“cohesion”, “identification”, “autonomy”), five risk-scales (“conflicts”, “rejection/ neglect”, “punishment”, “emotional burden”, “fears/overprotection”) and one additional scale “aid”. The Cronbach's Alpha of this validated version were between α = 0.60 and α = 0.85 for students and between α = 0.73 and α = 0.90 for patients [32]. Due to the quantitative evaluation of our study, items with an open response form were not used and for reasons of content, three subscales were not included for analysis of this study (25 instead of 36 items were evaluated). In the current sample, following reliabilities were measured related to the mother or the father: “cohesion” α = 0.92 or α = 0.89, “identification” α = 0.80 or α = 0.87, “autonomy” α = 0.69 or α = 0.81, “conflicts” α = 0.85 or α = 0.83, “punishment” α = 0.82 or α = 0.54, “rejection/neglect” α = 0.90 or α = 0.81, “emotional burden” α = 0.77 or α = 0.84, “fears/overprotection” α = 0.84 or α = 0.85 and “aid” α = 0.73 or α = 0.56. The Child Feeding Questionnaire for adolescents (CFQ-A) [33] which is an adaption from the CFQ for parents of adolescents [34] and the CFQ for parents of children ranging in age from about 2 to 11 years [35], was used to assess the perceptions of the feeding practices that parents/caregivers used when participants were between 10 and 13 years of age. The Cronbach's Alpha of the English original version of the CFQ-A were α = 0.68 for the scale “responsibility”, α = 0.90 for “monitoring”, α = 0.63 for “pressure to eat” and α = 0.85 for “restriction” [33]. For the purpose of this study, the CFQ-A questionnaire was translated into German language. The 19 items rated on a Likert scale are reflected on four subscales with following internal consistencies in the present sample: “perceived feeding responsibility” (after elimination of item 1) α = 0.80, “monitoring” α = 0.90, “pressure to eat” α = 0.84 and “restriction” α = 0.87. The Junior Temperament and Character Inventory 12–18 R [36] provides an evaluation of personality in adolescents. The questionnaire consists of 103 items, which are answered on a Likert scale. The German original version of the questionnaire showed internal consistencies between α = 0.79 and α = 0.85 [36]. Only several contently relevant scales were evaluated. In the current study an internal consistency of α = 0.90 for “harm avoidance”, α = 0.91 for “self-directedness” and α = 0.90 for “self-acceptance” was calculated. Statistical analysis The data was evaluated using IBM SPSS Statistics 24. The parental relationship of females with ED and HC were calculated using the t-test for heterogeneous variances (Welch test) and effect sizes according to Cohen. To examine the association of an ED with the relationship with the parent, a binary logistic regression with a stepwise backward method according to Wald was calculated in the total sample (ED and HC). Nagelkerke’s R2 was used to check the classification quality. A value > 40% can be regarded as good. The relationship between “rejection/neglect” and “self-acceptance” was calculated using the Spearman correlation (two-sided testing). The mediator analysis was calculated in the total sample (ED and HC) and therefore PROCESS 2.16.3 according to Hayes for SPSS® was applied with the criteria “drive for thinness” in relation to maternal relationship quality. The parental influence on diet at the age of 10–13 years for females with ED and HC was compared bilaterally using the Welch test. A power calculation was performed on the basis of the assumption of a difference of 4.5 points between females with ED and HC on the subscale “rejection/neglect” of the Parental-Representation-Screening-Questionnaire (PRSQ) and a confidence level of 95%. Mean values of school kids with normal parental representation (M 49 ± 4.8) and a psychiatric patient group with normal and abnormal parental representations (M 53.5) were derived from literature [9]. This showed that to obtain a power of 80% for the detection of significant differences, a minimum sample size of 18 patients per group was required. Subjects and procedures Study design. This quantitative cross-sectional study aimed at examining the current situation and memories of the early adolescence in female adolescents and young adults with and without eating disorders (ED) using questionnaires. The data was collected in the period from September 2016 to February 2017. Although no longitudinally collected data was available, cause-effect conclusions could be drawn by using a structural modelling strategy [27]. Ethics and consent to participate. This study was approved by the ethics committee of the Medical University of Vienna (reference number: 1691/2016). Written consent by the participants, after thorough information about the study, was acquired in advance in all cases. Adolescents between 14 and 18 years of age are of legal age of consent in Austria. Thus, young females’ consent in our study population of older than 16 years, was sufficient–and in accordance with the Ethics Committee–and parents did not have to be informed about participation in the study. Study setting. Female adolescents and young adults with ED were recruited from inpatient and outpatient care of the Clinical Division of Social Psychiatry at the Department for Psychiatry and Psychotherapy, and of the Department for Child and Adolescent Psychiatry of the Medical University of Vienna. The healthy control group (HC) was recruited via circle of acquaintances. Study sample. Female adolescents and young adults with eating disorder. Females with an eating disorder (ED) were included if they a) were between 16 and 26 years of age b) were diagnosed with either the ICD-10 diagnosis bulimia nervosa (BN) (ICD-10: F50.2) or anorexia nervosa (AN) (ICD-10: F50.00)—restrictive type or AN—bulimic type (ICD-10: F50.01). The diagnoses were established by experienced clinicians at the recruitment institutions. Healthy controls. Inclusion criteria for the healthy control group (HC) were a) the willingness to participate and b) sufficient German language skills. Exclusion criteria for the HC were a) the existence of a current or past diagnosis of an ED, b) values above the normal range in at least two of the three scales: “drive for thinness”, “bulimia”, “body dissatisfaction” of the Eating Disorder Inventory-2 (EDI-2) [28]. When recruiting the control group, a similar age distribution as in the group of females with eating disorders was aimed at, but no age matching was performed in the analysis. Study design. This quantitative cross-sectional study aimed at examining the current situation and memories of the early adolescence in female adolescents and young adults with and without eating disorders (ED) using questionnaires. The data was collected in the period from September 2016 to February 2017. Although no longitudinally collected data was available, cause-effect conclusions could be drawn by using a structural modelling strategy [27]. Ethics and consent to participate. This study was approved by the ethics committee of the Medical University of Vienna (reference number: 1691/2016). Written consent by the participants, after thorough information about the study, was acquired in advance in all cases. Adolescents between 14 and 18 years of age are of legal age of consent in Austria. Thus, young females’ consent in our study population of older than 16 years, was sufficient–and in accordance with the Ethics Committee–and parents did not have to be informed about participation in the study. Study setting. Female adolescents and young adults with ED were recruited from inpatient and outpatient care of the Clinical Division of Social Psychiatry at the Department for Psychiatry and Psychotherapy, and of the Department for Child and Adolescent Psychiatry of the Medical University of Vienna. The healthy control group (HC) was recruited via circle of acquaintances. Study sample. Female adolescents and young adults with eating disorder. Females with an eating disorder (ED) were included if they a) were between 16 and 26 years of age b) were diagnosed with either the ICD-10 diagnosis bulimia nervosa (BN) (ICD-10: F50.2) or anorexia nervosa (AN) (ICD-10: F50.00)—restrictive type or AN—bulimic type (ICD-10: F50.01). The diagnoses were established by experienced clinicians at the recruitment institutions. Healthy controls. Inclusion criteria for the healthy control group (HC) were a) the willingness to participate and b) sufficient German language skills. Exclusion criteria for the HC were a) the existence of a current or past diagnosis of an ED, b) values above the normal range in at least two of the three scales: “drive for thinness”, “bulimia”, “body dissatisfaction” of the Eating Disorder Inventory-2 (EDI-2) [28]. When recruiting the control group, a similar age distribution as in the group of females with eating disorders was aimed at, but no age matching was performed in the analysis. Measurements Demographic data of the participants were collected by means of a study specific questionnaire. Data about age, weight, diagnosed mental health disorders according to ICD-10 [29], participation and duration of psychological/psychiatric treatment, housing situation with/without parents and school degree were collected. The symptoms of EDs were investigated using the frequently used Eating Disorder Inventory 2 (EDI-2; German Version) [28]. The EDI-2 is a revision of the EDI-1 [30], in which three subscales have been added. In the original English version of the EDI-2, the internal consistencies (Cronbach's Alphas) for the eight original scales ranged from 0.80 to 0.91, of which three were used in this study and a Principal Component Analysis supported the factor structure of the original EDI, but not that of the three additional scales [31] not used in this study. For our study, the validated German version of the questionnaire was used, which contains 91 items on a 6-level rating scale representing 11 scales. The German version achieved an Cronbach's Alpha between α = 0.73 and α = 0.93 in a sample of anorexic and bulimic patients, the test-retest reliability ranged from rtt = 0.81 to rtt = 0.89 [28]. In the current study, the three subscales (23 items) with the most relevant content for the symptoms of eating disorders were applied and Cronbach’s Alpha of these scales reached α = 0.91 for “drive for thinness”, α = 0.95 for “bulimia” and α = 0.93 for “body dissatisfaction” in the study population. The quality of the relationship with the parents was measured with the Parental-Representation-Screening-Questionnaire (PRSQ); German Version [Elternbildfragebogen für Kinder- und Jugendliche (EBF-KJ)] for children and adolescents [32]. This German questionnaire, which was applied in this study in its original version, consists of 36 items answered on a 5-level Likert scale and contains three resource-scales (“cohesion”, “identification”, “autonomy”), five risk-scales (“conflicts”, “rejection/ neglect”, “punishment”, “emotional burden”, “fears/overprotection”) and one additional scale “aid”. The Cronbach's Alpha of this validated version were between α = 0.60 and α = 0.85 for students and between α = 0.73 and α = 0.90 for patients [32]. Due to the quantitative evaluation of our study, items with an open response form were not used and for reasons of content, three subscales were not included for analysis of this study (25 instead of 36 items were evaluated). In the current sample, following reliabilities were measured related to the mother or the father: “cohesion” α = 0.92 or α = 0.89, “identification” α = 0.80 or α = 0.87, “autonomy” α = 0.69 or α = 0.81, “conflicts” α = 0.85 or α = 0.83, “punishment” α = 0.82 or α = 0.54, “rejection/neglect” α = 0.90 or α = 0.81, “emotional burden” α = 0.77 or α = 0.84, “fears/overprotection” α = 0.84 or α = 0.85 and “aid” α = 0.73 or α = 0.56. The Child Feeding Questionnaire for adolescents (CFQ-A) [33] which is an adaption from the CFQ for parents of adolescents [34] and the CFQ for parents of children ranging in age from about 2 to 11 years [35], was used to assess the perceptions of the feeding practices that parents/caregivers used when participants were between 10 and 13 years of age. The Cronbach's Alpha of the English original version of the CFQ-A were α = 0.68 for the scale “responsibility”, α = 0.90 for “monitoring”, α = 0.63 for “pressure to eat” and α = 0.85 for “restriction” [33]. For the purpose of this study, the CFQ-A questionnaire was translated into German language. The 19 items rated on a Likert scale are reflected on four subscales with following internal consistencies in the present sample: “perceived feeding responsibility” (after elimination of item 1) α = 0.80, “monitoring” α = 0.90, “pressure to eat” α = 0.84 and “restriction” α = 0.87. The Junior Temperament and Character Inventory 12–18 R [36] provides an evaluation of personality in adolescents. The questionnaire consists of 103 items, which are answered on a Likert scale. The German original version of the questionnaire showed internal consistencies between α = 0.79 and α = 0.85 [36]. Only several contently relevant scales were evaluated. In the current study an internal consistency of α = 0.90 for “harm avoidance”, α = 0.91 for “self-directedness” and α = 0.90 for “self-acceptance” was calculated. Statistical analysis The data was evaluated using IBM SPSS Statistics 24. The parental relationship of females with ED and HC were calculated using the t-test for heterogeneous variances (Welch test) and effect sizes according to Cohen. To examine the association of an ED with the relationship with the parent, a binary logistic regression with a stepwise backward method according to Wald was calculated in the total sample (ED and HC). Nagelkerke’s R2 was used to check the classification quality. A value > 40% can be regarded as good. The relationship between “rejection/neglect” and “self-acceptance” was calculated using the Spearman correlation (two-sided testing). The mediator analysis was calculated in the total sample (ED and HC) and therefore PROCESS 2.16.3 according to Hayes for SPSS® was applied with the criteria “drive for thinness” in relation to maternal relationship quality. The parental influence on diet at the age of 10–13 years for females with ED and HC was compared bilaterally using the Welch test. A power calculation was performed on the basis of the assumption of a difference of 4.5 points between females with ED and HC on the subscale “rejection/neglect” of the Parental-Representation-Screening-Questionnaire (PRSQ) and a confidence level of 95%. Mean values of school kids with normal parental representation (M 49 ± 4.8) and a psychiatric patient group with normal and abnormal parental representations (M 53.5) were derived from literature [9]. This showed that to obtain a power of 80% for the detection of significant differences, a minimum sample size of 18 patients per group was required. Results Sample description A total of 52 females were included in the study. Of these, 30 participants were included into the healthy control group (HC), while 22 adolescent and young adults were included into the eating disorders group (ED). In the group of participants in whom clinicians diagnosed an eating disorder (ED) according to the criteria of the ICD-10, one participant had to be excluded due to lack of data and 21 persons could be subjected to further analyses. In the healthy control (HC) group, 8 participants were excluded as they were classified as “at risk of developing an ED” due to symptoms of disordered eating (according to EDI-2). The data of 22 HCs could be further analysed. Due to missing values, the sample size varied in the individual calculations. The age of the study participants was between 16 and 26 years with a similar distribution in both groups (HC: M = 18.55 ± 2.67; ED: M = 18.86 ± 2.72). The Body Mass Index (BMI) was M = 22.63 for the HC group and M = 18.30 for the ED group. A BMI less than 18.50 was classified as underweight. In the ED group, 17 (80.9%) females had been diagnosed with anorexia nervosa while four (19.1%) females had been diagnosed with bulimia nervosa. One person from the ED-group also suffered from depression. The onset of the ED was between 10 and 19 years (M = 15 ± 1.97). Of the participants with an ED, 21 (100%) were currently in psychological or psychiatric treatment. Persons in the HC group did not suffer from any diagnosed mental health disorder. Relationship with parents. With regard to the quality of the relationship with the mother, significant differences were revealed between the groups (ED n = 21, HC n = 22) for the scales “rejection/neglect” (t(24.75) = - 2.13, p = 0.043, d = - 0.65) and “fears/overprotection” (t(38.72) = - 3.01, p = 0.005, d = - 0.92). The results are demonstrated in Table 1. No significant differences could be demonstrated for the relationship with the mother for each group in “cohesion” (ED and HC) (t(40.32) = 0.95, p = 0.394, d = 0) 0.29), “autonomy”, (t(40.96) = 0.67, p = 0.504, d = 0.20), “conflicts” (t(40.85) = - 0.83, p = 0.410, d = - 0.25), and “emotional burden” (t(39.94) = 0.37, p = 0.714, d = 0.11). There was no significant difference in any scale between these two groups in terms of the quality of the relationship with the father. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Differences in relationship quality with the mother between females without eating disorders (HC) and females with eating disorders (ED). https://doi.org/10.1371/journal.pone.0242518.t001 Association of an eating disorder: Relationship with the mother. The variables “conflicts”, “emotional burden”, “autonomy”, “cohesion”, “rejection/neglect” and “fears/overprotection” of the mother were investigated by means of binary logistic regression (n = 42; see Table 2). It was found that young females who perceived more “cohesion”, “rejection/neglect” and “fears/overprotection” by their mother more often suffered from an ED. The model explained 43.9% of the variance. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Predictors for the presence of an eating disorder. https://doi.org/10.1371/journal.pone.0242518.t002 Mediators between maternal relationship and symptoms of eating disorders. The mediation model tries to identify a possible process that underlies the observed relationship between the independent variable “relationship with the mother” and the dependent variable “symptoms of eating disorders” via the inclusion of a third hypothetical mediator variable “personality traits”. In this study, analyses were only carried out for variables with strong correlations in previous calculations. For eating disorder symptoms, a separate analysis was done with the criteria "body dissatisfaction" and "drive for thinness" since it was not reasonable to combine these two scales to a common scale. Since there was no significant association between “autonomy” and the mediator “self-directedness” (r = 0.08, p = 0.627, n = 42), nor between “autonomy” and the criteria “body dissatisfaction” (r = - 0.08, p = 0.586, n = 43) / “drive for thinness” (r = 0.00, p = 0.995, n = 43), there can be no mediation and therefore, no analysis was carried out. The applied scales were tested for z-standardization before the analysis and with regard to the procedural requirements, the normal distribution of the data could generally be assumed on the basis of the validity of the central limit theorem for sample sizes ≥ 30. The mediator analysis (n = 42) found a direct and an indirect effect of “fears/overprotection” on the "drive for thinness" via anxious personality traits ("harm avoidance"). The direct effect measures the extent to which “fears/overprotection” changes when “drive for thinness” increases and the mediator variable remains unaltered. In contrast, the indirect effect measures the extent to which “drive for thinness” changes when “fears/overprotection” is held fixed and “harm avoidance” changes by the amount it would have changed had “fears/overprotection” increased by one unit. Part of the effect was mediated indirectly through the mediator “harm avoidance” (β = 0.14) and part of the effect was directly mediated from “fears/overprotection” to “drive for thinness” (β = 0.41), together they had a medium effect (β = 0.55). Fig 1 shows the relationships embedded in the model and gives the statistical parameters of the effects. The effect of “fears/overprotection” on “body dissatisfaction” was not mediated by “harm avoidance” (β = 0.07). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Mediator model fears/overprotection—harm avoidance—drive for thinness. *p<0.05. n = 42. β = standardized regression coefficient, indirect effect β .14, boot SE 0.08, BootLLCI (t) 0.01 a, BootULCI (p) 0.36 a, aconfidence interval 0 not included, therefore effect significant Bootstrap: 5000. https://doi.org/10.1371/journal.pone.0242518.g001 Parental feeding practices at the age of 10–13 years. The scales “monitoring” (t(39.41) = - 0.78, p = 0.443, d = - 0.24), “pressure to eat”(t(33.03) = - 1.55, p = 0.131, d = - 0.47) and “restriction” (t(35.35) = - 1.55, p = 0.129, d = - 0.47) showed no significant differences. Only the scale “perceived feeding responsibility” showed a significant difference between the two groups (t(39.67) = - 2.20, p = 0.034, d = - 0.67). Female adolescents and young adults with an ED reported that they had less personal responsibility at the age of 10–13 years for what and how much they ate in comparison to females without an ED (see Table 3). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Comparison of perceived parental influence on diet at the age of 10–13 for females without eating disorder (HC) and females with eating disorder (ED). https://doi.org/10.1371/journal.pone.0242518.t003 Relationship between parental rejection/neglect and self-acceptance. A significantly negative correlation between the scale “rejection/neglect” of the mother and “self-acceptance” of the test subjects could be determined (rs = - 0.35, p = 0.022, n = 42). There was also a significant negative correlation between the fathers’ “rejection/neglect” and “self-acceptance” (rs = - 0.47, p = 0.002, n = 42). This means that more rejection/neglect by both the mother and the father was accompanied by less self-acceptance of young females with an even greater influence of the fathers’ rejection/neglect. Sample description A total of 52 females were included in the study. Of these, 30 participants were included into the healthy control group (HC), while 22 adolescent and young adults were included into the eating disorders group (ED). In the group of participants in whom clinicians diagnosed an eating disorder (ED) according to the criteria of the ICD-10, one participant had to be excluded due to lack of data and 21 persons could be subjected to further analyses. In the healthy control (HC) group, 8 participants were excluded as they were classified as “at risk of developing an ED” due to symptoms of disordered eating (according to EDI-2). The data of 22 HCs could be further analysed. Due to missing values, the sample size varied in the individual calculations. The age of the study participants was between 16 and 26 years with a similar distribution in both groups (HC: M = 18.55 ± 2.67; ED: M = 18.86 ± 2.72). The Body Mass Index (BMI) was M = 22.63 for the HC group and M = 18.30 for the ED group. A BMI less than 18.50 was classified as underweight. In the ED group, 17 (80.9%) females had been diagnosed with anorexia nervosa while four (19.1%) females had been diagnosed with bulimia nervosa. One person from the ED-group also suffered from depression. The onset of the ED was between 10 and 19 years (M = 15 ± 1.97). Of the participants with an ED, 21 (100%) were currently in psychological or psychiatric treatment. Persons in the HC group did not suffer from any diagnosed mental health disorder. Relationship with parents. With regard to the quality of the relationship with the mother, significant differences were revealed between the groups (ED n = 21, HC n = 22) for the scales “rejection/neglect” (t(24.75) = - 2.13, p = 0.043, d = - 0.65) and “fears/overprotection” (t(38.72) = - 3.01, p = 0.005, d = - 0.92). The results are demonstrated in Table 1. No significant differences could be demonstrated for the relationship with the mother for each group in “cohesion” (ED and HC) (t(40.32) = 0.95, p = 0.394, d = 0) 0.29), “autonomy”, (t(40.96) = 0.67, p = 0.504, d = 0.20), “conflicts” (t(40.85) = - 0.83, p = 0.410, d = - 0.25), and “emotional burden” (t(39.94) = 0.37, p = 0.714, d = 0.11). There was no significant difference in any scale between these two groups in terms of the quality of the relationship with the father. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Differences in relationship quality with the mother between females without eating disorders (HC) and females with eating disorders (ED). https://doi.org/10.1371/journal.pone.0242518.t001 Association of an eating disorder: Relationship with the mother. The variables “conflicts”, “emotional burden”, “autonomy”, “cohesion”, “rejection/neglect” and “fears/overprotection” of the mother were investigated by means of binary logistic regression (n = 42; see Table 2). It was found that young females who perceived more “cohesion”, “rejection/neglect” and “fears/overprotection” by their mother more often suffered from an ED. The model explained 43.9% of the variance. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Predictors for the presence of an eating disorder. https://doi.org/10.1371/journal.pone.0242518.t002 Mediators between maternal relationship and symptoms of eating disorders. The mediation model tries to identify a possible process that underlies the observed relationship between the independent variable “relationship with the mother” and the dependent variable “symptoms of eating disorders” via the inclusion of a third hypothetical mediator variable “personality traits”. In this study, analyses were only carried out for variables with strong correlations in previous calculations. For eating disorder symptoms, a separate analysis was done with the criteria "body dissatisfaction" and "drive for thinness" since it was not reasonable to combine these two scales to a common scale. Since there was no significant association between “autonomy” and the mediator “self-directedness” (r = 0.08, p = 0.627, n = 42), nor between “autonomy” and the criteria “body dissatisfaction” (r = - 0.08, p = 0.586, n = 43) / “drive for thinness” (r = 0.00, p = 0.995, n = 43), there can be no mediation and therefore, no analysis was carried out. The applied scales were tested for z-standardization before the analysis and with regard to the procedural requirements, the normal distribution of the data could generally be assumed on the basis of the validity of the central limit theorem for sample sizes ≥ 30. The mediator analysis (n = 42) found a direct and an indirect effect of “fears/overprotection” on the "drive for thinness" via anxious personality traits ("harm avoidance"). The direct effect measures the extent to which “fears/overprotection” changes when “drive for thinness” increases and the mediator variable remains unaltered. In contrast, the indirect effect measures the extent to which “drive for thinness” changes when “fears/overprotection” is held fixed and “harm avoidance” changes by the amount it would have changed had “fears/overprotection” increased by one unit. Part of the effect was mediated indirectly through the mediator “harm avoidance” (β = 0.14) and part of the effect was directly mediated from “fears/overprotection” to “drive for thinness” (β = 0.41), together they had a medium effect (β = 0.55). Fig 1 shows the relationships embedded in the model and gives the statistical parameters of the effects. The effect of “fears/overprotection” on “body dissatisfaction” was not mediated by “harm avoidance” (β = 0.07). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Mediator model fears/overprotection—harm avoidance—drive for thinness. *p<0.05. n = 42. β = standardized regression coefficient, indirect effect β .14, boot SE 0.08, BootLLCI (t) 0.01 a, BootULCI (p) 0.36 a, aconfidence interval 0 not included, therefore effect significant Bootstrap: 5000. https://doi.org/10.1371/journal.pone.0242518.g001 Parental feeding practices at the age of 10–13 years. The scales “monitoring” (t(39.41) = - 0.78, p = 0.443, d = - 0.24), “pressure to eat”(t(33.03) = - 1.55, p = 0.131, d = - 0.47) and “restriction” (t(35.35) = - 1.55, p = 0.129, d = - 0.47) showed no significant differences. Only the scale “perceived feeding responsibility” showed a significant difference between the two groups (t(39.67) = - 2.20, p = 0.034, d = - 0.67). Female adolescents and young adults with an ED reported that they had less personal responsibility at the age of 10–13 years for what and how much they ate in comparison to females without an ED (see Table 3). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Comparison of perceived parental influence on diet at the age of 10–13 for females without eating disorder (HC) and females with eating disorder (ED). https://doi.org/10.1371/journal.pone.0242518.t003 Relationship between parental rejection/neglect and self-acceptance. A significantly negative correlation between the scale “rejection/neglect” of the mother and “self-acceptance” of the test subjects could be determined (rs = - 0.35, p = 0.022, n = 42). There was also a significant negative correlation between the fathers’ “rejection/neglect” and “self-acceptance” (rs = - 0.47, p = 0.002, n = 42). This means that more rejection/neglect by both the mother and the father was accompanied by less self-acceptance of young females with an even greater influence of the fathers’ rejection/neglect. Relationship with parents. With regard to the quality of the relationship with the mother, significant differences were revealed between the groups (ED n = 21, HC n = 22) for the scales “rejection/neglect” (t(24.75) = - 2.13, p = 0.043, d = - 0.65) and “fears/overprotection” (t(38.72) = - 3.01, p = 0.005, d = - 0.92). The results are demonstrated in Table 1. No significant differences could be demonstrated for the relationship with the mother for each group in “cohesion” (ED and HC) (t(40.32) = 0.95, p = 0.394, d = 0) 0.29), “autonomy”, (t(40.96) = 0.67, p = 0.504, d = 0.20), “conflicts” (t(40.85) = - 0.83, p = 0.410, d = - 0.25), and “emotional burden” (t(39.94) = 0.37, p = 0.714, d = 0.11). There was no significant difference in any scale between these two groups in terms of the quality of the relationship with the father. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Differences in relationship quality with the mother between females without eating disorders (HC) and females with eating disorders (ED). https://doi.org/10.1371/journal.pone.0242518.t001 Association of an eating disorder: Relationship with the mother. The variables “conflicts”, “emotional burden”, “autonomy”, “cohesion”, “rejection/neglect” and “fears/overprotection” of the mother were investigated by means of binary logistic regression (n = 42; see Table 2). It was found that young females who perceived more “cohesion”, “rejection/neglect” and “fears/overprotection” by their mother more often suffered from an ED. The model explained 43.9% of the variance. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Predictors for the presence of an eating disorder. https://doi.org/10.1371/journal.pone.0242518.t002 Mediators between maternal relationship and symptoms of eating disorders. The mediation model tries to identify a possible process that underlies the observed relationship between the independent variable “relationship with the mother” and the dependent variable “symptoms of eating disorders” via the inclusion of a third hypothetical mediator variable “personality traits”. In this study, analyses were only carried out for variables with strong correlations in previous calculations. For eating disorder symptoms, a separate analysis was done with the criteria "body dissatisfaction" and "drive for thinness" since it was not reasonable to combine these two scales to a common scale. Since there was no significant association between “autonomy” and the mediator “self-directedness” (r = 0.08, p = 0.627, n = 42), nor between “autonomy” and the criteria “body dissatisfaction” (r = - 0.08, p = 0.586, n = 43) / “drive for thinness” (r = 0.00, p = 0.995, n = 43), there can be no mediation and therefore, no analysis was carried out. The applied scales were tested for z-standardization before the analysis and with regard to the procedural requirements, the normal distribution of the data could generally be assumed on the basis of the validity of the central limit theorem for sample sizes ≥ 30. The mediator analysis (n = 42) found a direct and an indirect effect of “fears/overprotection” on the "drive for thinness" via anxious personality traits ("harm avoidance"). The direct effect measures the extent to which “fears/overprotection” changes when “drive for thinness” increases and the mediator variable remains unaltered. In contrast, the indirect effect measures the extent to which “drive for thinness” changes when “fears/overprotection” is held fixed and “harm avoidance” changes by the amount it would have changed had “fears/overprotection” increased by one unit. Part of the effect was mediated indirectly through the mediator “harm avoidance” (β = 0.14) and part of the effect was directly mediated from “fears/overprotection” to “drive for thinness” (β = 0.41), together they had a medium effect (β = 0.55). Fig 1 shows the relationships embedded in the model and gives the statistical parameters of the effects. The effect of “fears/overprotection” on “body dissatisfaction” was not mediated by “harm avoidance” (β = 0.07). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Mediator model fears/overprotection—harm avoidance—drive for thinness. *p<0.05. n = 42. β = standardized regression coefficient, indirect effect β .14, boot SE 0.08, BootLLCI (t) 0.01 a, BootULCI (p) 0.36 a, aconfidence interval 0 not included, therefore effect significant Bootstrap: 5000. https://doi.org/10.1371/journal.pone.0242518.g001 Parental feeding practices at the age of 10–13 years. The scales “monitoring” (t(39.41) = - 0.78, p = 0.443, d = - 0.24), “pressure to eat”(t(33.03) = - 1.55, p = 0.131, d = - 0.47) and “restriction” (t(35.35) = - 1.55, p = 0.129, d = - 0.47) showed no significant differences. Only the scale “perceived feeding responsibility” showed a significant difference between the two groups (t(39.67) = - 2.20, p = 0.034, d = - 0.67). Female adolescents and young adults with an ED reported that they had less personal responsibility at the age of 10–13 years for what and how much they ate in comparison to females without an ED (see Table 3). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Comparison of perceived parental influence on diet at the age of 10–13 for females without eating disorder (HC) and females with eating disorder (ED). https://doi.org/10.1371/journal.pone.0242518.t003 Relationship between parental rejection/neglect and self-acceptance. A significantly negative correlation between the scale “rejection/neglect” of the mother and “self-acceptance” of the test subjects could be determined (rs = - 0.35, p = 0.022, n = 42). There was also a significant negative correlation between the fathers’ “rejection/neglect” and “self-acceptance” (rs = - 0.47, p = 0.002, n = 42). This means that more rejection/neglect by both the mother and the father was accompanied by less self-acceptance of young females with an even greater influence of the fathers’ rejection/neglect. Discussion The aim of the study was to investigate a possible association between eating disorders (ED) and subjective stress in the relationship with parents as well as retrospective parental influence on eating behaviour at the early adolescence from the perspective of affected young females. Adolescents and young adult females with ED reported more fears/overprotection and rejection/neglect by their mothers and less self-responsibility regarding eating behaviour in adolescence than healthy young females. The relationship with fathers did not differ significantly. It was shown that young women who perceived more cohesion, rejection/neglect and fears/overprotection from their mothers had an increased chance of suffering from an eating disorder. Rejection or neglect by the mothers and the fathers was associated with lower self-acceptance among young females, while the influence of father's rejection/neglect on self-confidence was even greater. According to the attachment theory, a well-functioning relationship with parents is crucial for children to learn self-acceptance and independent emotion regulation in adolescence–both of which are known to influence the development of eating disorders in adolescence [13]. Previous research have focused on parental functioning in families of patients with ED, with a particular emphasis on the links between the psychopathological symptoms of mothers and daughters. However, only few and often contradictory findings exist on whether there are specific maladaptive psychological profiles characterizing the family structure as a whole when it includes adolescent and young adult females with anorexia nervosa (AN) or bulimia nervosa (BN) [37]. In our study, we expected to find an association of EDs and problematic family functioning. Furthermore, we expected females with an ED to have experienced significantly more rejection/neglect or fears/overprotection from their mothers than females without an ED. While rejection/neglect or fears/overprotection are contrary relationship patterns, our results indicate that mothers of patients with ED are experienced in different extremes between entanglement and rejection–in part depending on the type of ED. In previous literature, adolescents with BN were reported to having had a significantly more dysfunctional family background than those with AN [11]. Patients with BN perceived their families as badly organized, and reported the presence of high levels of family conflicts [37]. These conflicts may have led to perceived rejection/neglect from parents. In contrast, in families of adolescents diagnosed with AN, higher conflict avoidance and enmeshment was found. This means that over-concern for others with a loss of autonomous development could lead to fears/overprotection [37]. Literature, however, remains inconclusive as both a tendency towards parental overprotection in EDs [38] as well as no connection between overprotection and EDs have previously been demonstrated [39]. Our results are in agreement with previous literature where a general tendency towards more perceived parental rejection in persons with an ED was postulated [16, 18, 19]. The results regarding the father-daughter relationship and eating disorders are of particular interest. A recent review [40] identified a gap in the literature regarding this topic and called for research to further investigate this dyadic relationship. The father-daughter relationship is important during adolescence as it plays a role in the development of autonomy and self-esteem, both of them are involved in eating disorder psychopathology [40]. In our study, it could be shown that rejection or neglect by both mother and father was accompanied by less self-acceptance on the part of the young females. No direct connection could be found between the quality of the relationship with the father and the presence of an ED. However, an indirect effect of fathers on the symptoms of EDs was shown, since subjectively perceived rejection or neglect by the father was accompanied by less self-acceptance on the part of the young females. The paternal influence on self-acceptance was even greater than that of the mothers. Interestingly, a recent study found that changes in the perceived quality of the father–adolescent daughter attachment relationship were only associated with daughters’ self-esteem (not to that of the sons’) [41]. As attachment theory indicates, exploration and security represent two sides of a secure attachment style [12], and, according to traditional role models, fathers are seen as more independence encouraging (while mothers serve as a haven of safety). But some changes to the father–daughter attachment can be observed when girls approach adolescence. In adolescence, daughters may increasingly start to value their fathers’ independence-encouraging behaviour. This makes them more susceptible for shortcomings in the perceived quality of the father–child attachment relationship [12, 40]. Harm avoidance as an adolescent and young adult females’ personality trait in part mediated the direct influence of the fears/overprotection of the mothers on aspects of the ED symptoms of the young females. Parents exercised some control over the young females’ drive for thinness through experienced overprotection, which was influenced by their own control in the form of harm avoidance. Stenbæk [15] found a similar influence of the personality trait harm avoidance, which had a mediating effect on the direct influence of parental overprotection and perceived stress or susceptibility to psychological symptoms in healthy adults. It has also been shown that self-regulating skills, such as self-compassion, moderate the impact of perfectionism on disordered eating [22]. In this study, in advance, we expected that female adolescents and young adults with ED will report about more experienced and stricter regulations on dieting than HC. Interestingly, the female adolescent and young adults with EDs reported significantly less feeding responsibility in early adolescence compared to the females in the HC group, but other characteristics of controlling parental feeding practices like food restrictions, monitoring during meal and the pressure to eat did not show significant difference in both groups. Previous studies had shown restrictive parental feeding practices to increase the likelihood of disordered eating [25, 26]. If excessive parental control over food consumption is applied in raising children, the development of self-regulatory capacities could be negatively influenced [25]. Stringent maternal restriction of food resulted in eating psychopathology such as extreme weight control in daughters [24]. Although we found no such effect of controlling feeding practices such as restriction, pressure to eat and monitoring in our study, we did find an effect on feeding responsibility, which is the opposite of controlling practices of the parents by informing about the freedom to make dietary decisions on behalf of the adolescent females. Strengths and limitations A strength of our paper is the innovative study design investigating the perceived parental’ feeding practices in early adolescence in the context of the relationship with the parents in a clinical sample of female adolescent and young adults diagnosed with EDs as well as in healthy participants (HC). To the best of our knowledge, no examination of perceived parental feeding practices, as evaluated in this study adapting the Child Feeding Questionnaire for Adolescents [35], in females with valid diagnoses of EDs according to ICD-10 [29] has yet been conducted by clinicians at ED institutions. Previous studies were carried out in non-clinical population and they mainly investigated the relationship between parental feeding practices and eating disorders psychopathology [35, 42] The DSM-5 [1] has recently also included binge eating disorder (BED) as an autonomous diagnosis characterized by recurrent episodes of eating fits accompanied by a feeling of loss of control but lower interest in body shape or weight. In the present study, we state that the parental relationship is mediated through self-assessment and that self-evaluation is extremely influenced by the symptoms of the eating disorder (such as body shape or weight). Compared to AN and BN, body shape and weight do not seem to play such a central role in regulating self-confidence in the BED, so the BED was not included. Some limitations do, however, need to be addressed: The cross-sectional design of the present study does not allow causal interpretation and results need to be interpreted with causation. Although the healthy control (HC) group was composed with regard to a similar age distribution as in the group of female adolescents and young adults with eating disorders (ED), no age matching was performed in the analysis. Furthermore, although a sample size calculation based on group differences of main characteristics had been performed, a relatively small sample size may be interpreted another limitation of our study as only moderate and clear effects, but no small effects may be recorded [43] and results with very high standard deviations should be treated with caution. However, the significant and reliable differences and effect sizes found despite these limitations underline the relevance of the results. Eight people were excluded from the control group due to symptoms of an eating disorder above the threshold, which means that almost one third were affected by the disordered eating behaviour. Therefore, the results of females with ED were not compared with the baseline values of a group representative of the population. The retrospective collection of the data on perceived parental feeding behaviour in early adolescence of females offers limitations due to possible distortions of the remembered influence [44]. The retrospective measurement is another possible weakness of the study, as we cannot rule out possible recall biases involved in the reports of the adolescent and young adult females aged between 18–19 years making statements about parental feeding practices when they were 10–13 years of age. Another limitation is that the Child Feeding Questionnaire for adolescents (CFQ-A) was only translated into German language for this study. To our knowledge there exists no German validated version and therefore we have provided the psychometric properties of our translated version. While the questionnaire was first translated into German by multiple authors and then back translated into an English version for comparability, the translation was not previously validated. In this study, based on these observations, we assume that female adolescents and young adults with ED have experienced more or stricter regulations on dieting than HC. As our study does not include longitudinal data, only retrospectively assessed recall may be interpreted in our cross-sectional questionnaire study. Previous literature, however, has shown, that cross-sectional studies may infer information of possible influencing factors[27, 45] Nonetheless, results need to be interpreted with caution. Strengths and limitations A strength of our paper is the innovative study design investigating the perceived parental’ feeding practices in early adolescence in the context of the relationship with the parents in a clinical sample of female adolescent and young adults diagnosed with EDs as well as in healthy participants (HC). To the best of our knowledge, no examination of perceived parental feeding practices, as evaluated in this study adapting the Child Feeding Questionnaire for Adolescents [35], in females with valid diagnoses of EDs according to ICD-10 [29] has yet been conducted by clinicians at ED institutions. Previous studies were carried out in non-clinical population and they mainly investigated the relationship between parental feeding practices and eating disorders psychopathology [35, 42] The DSM-5 [1] has recently also included binge eating disorder (BED) as an autonomous diagnosis characterized by recurrent episodes of eating fits accompanied by a feeling of loss of control but lower interest in body shape or weight. In the present study, we state that the parental relationship is mediated through self-assessment and that self-evaluation is extremely influenced by the symptoms of the eating disorder (such as body shape or weight). Compared to AN and BN, body shape and weight do not seem to play such a central role in regulating self-confidence in the BED, so the BED was not included. Some limitations do, however, need to be addressed: The cross-sectional design of the present study does not allow causal interpretation and results need to be interpreted with causation. Although the healthy control (HC) group was composed with regard to a similar age distribution as in the group of female adolescents and young adults with eating disorders (ED), no age matching was performed in the analysis. Furthermore, although a sample size calculation based on group differences of main characteristics had been performed, a relatively small sample size may be interpreted another limitation of our study as only moderate and clear effects, but no small effects may be recorded [43] and results with very high standard deviations should be treated with caution. However, the significant and reliable differences and effect sizes found despite these limitations underline the relevance of the results. Eight people were excluded from the control group due to symptoms of an eating disorder above the threshold, which means that almost one third were affected by the disordered eating behaviour. Therefore, the results of females with ED were not compared with the baseline values of a group representative of the population. The retrospective collection of the data on perceived parental feeding behaviour in early adolescence of females offers limitations due to possible distortions of the remembered influence [44]. The retrospective measurement is another possible weakness of the study, as we cannot rule out possible recall biases involved in the reports of the adolescent and young adult females aged between 18–19 years making statements about parental feeding practices when they were 10–13 years of age. Another limitation is that the Child Feeding Questionnaire for adolescents (CFQ-A) was only translated into German language for this study. To our knowledge there exists no German validated version and therefore we have provided the psychometric properties of our translated version. While the questionnaire was first translated into German by multiple authors and then back translated into an English version for comparability, the translation was not previously validated. In this study, based on these observations, we assume that female adolescents and young adults with ED have experienced more or stricter regulations on dieting than HC. As our study does not include longitudinal data, only retrospectively assessed recall may be interpreted in our cross-sectional questionnaire study. Previous literature, however, has shown, that cross-sectional studies may infer information of possible influencing factors[27, 45] Nonetheless, results need to be interpreted with caution. Conclusions As an eating problem manifests and develops, the signs of disordered eating are enormously stressful for the whole family and it is natural for parents to react with grave concerns. Feeding practices and relationships might be drastically altered in profound ways. A recent meta-analysis underlines difficulties in establishing a therapeutic alliance and highlights especially the risk that professionals, adolescents, and parents will not converse about treatment, although such a dialogue appears to be an essential component in the construction of a therapeutic alliance [46]. As the results of the current study also underline, prevention and treatment efforts with focus on nutritional counselling of parents and stressors in the parent-child relationship should be targeted in particular at young females aged 10–19, at the highest incidence of anorexia and bulimia [44]. The established family therapy of adolescents with EDs [42], which works on dysfunctional parental relationship, and the psychotherapy of adults with EDs, which focuses on self-regulatory mechanism [10], therapeutic relationship and personality development, would also prove useful against the background of the present results. Knowledge about parental influence on eating behaviour and psychosocial development in the context of EDs may be useful in parent counselling and prevention work so that parents can be increasingly seen as a resource in the prevention and treatment of EDs. There is a need for further research with a prospective design and more objective measures like observation by investigators to allow for statements about parental feeding practices/relationship as protective or risk factors in the development of eating disorders.

Editors, The PLOS ONE

doi: 10.1371/journal.pone.0242853pmid: 33206728

Following the publication of this article [1], concerns were raised regarding the results presented in Figs 1, 4, 5, 6, and 7. Specifically, In Fig 1B, the FACS results presented for the Oral tumor biopsy (HPV+ve) Verapamil (-ve) appear more similar to the FACS results for the Oral tumor biopsy (HPV+ve) Verapamil (+ve) than would be expected from independent FACS samples. The authors have indicated that inadvertent errors have been made during the assembly of the figure panels, and that the raw underlying data to support the FACS results presented in Fig 1B are no longer available. Similarities were observed between the following FACS results presented in this article [1] and FACS results depicting different experimental conditions in an article previously published in Frontiers in Oncology [2]. The authors have indicated that the FACS data presented in the latter article is incorrect, but they have not provided raw underlying data to support the FACS results presented in this article: ○. The UPCI:SCC131 (HPV-ve) Verapamil(-ve) panel of Fig 1A [1] appears similar to the UPCI:SCC84 (HPV-ve) Verapamil (-ve) panel of Fig 1 [2]. ○. The Oral tumor biopsy (HV-ve) Verapamil (-ve) panel of Fig 1B [1] appears similar to the UPCI:SCC131 (HPV-ve) Verapamil (-ve) panel of Fig 1 [2]. ○. The Oral tumor biopsy (HV-ve) Verapamil (+ve) panel of Fig 1B [1] appears similar to the UPCI:SCC131 (HPV-ve) Verapamil (+ve) of Fig 1 [2]. Similarities have been detected between results presented in the western blot panels presented in Fig 4B and Fig 5A. The authors indicated that these similarities are the result of inadvertent errors that occurred during figure preparation: ○. The UD-SCC2 Oct 4 panel of Fig 4B appears similar to the UPCI-SCC 84 (HPV-) p52 and the UPCI-SCC 84 (HPV-) Rel-B panels of Fig 5A. In addition, the second (NSP) and third (SP) lanes of the forementioned panels appear similar to the second (NSP) and third (SP) lane of the UD SCC-2 (HPV 16+) p52 panel of Fig 5A. ○. The UPCI:SCC131 Sox-2 panel of Fig 4B appears similar to the UD SCC-2 (HPV 16+) c-Rel panel of Fig 5A and the UPCI-SCC 84 (HPV-) c-Rel panel of Fig 5A. ○. In Fig 5A, the second (NSP) lane of the UD SCC-2 (HPV 16+) Rel-B panel appears similar to the first (Parental) lane of the UPCI-SCC 84 (HPV-) p65 panel. Irregularities have been detected in the background signal of results presented in Figs 4B, 5A, 6A, 7A, and 7C. The authors indicated that the irregularities are the result of gel splicing conducted to ensure the panels demonstrated pertinent results only: ○. A vertical irregularity was detected between the first and the second lane of the UPCI:SCC131 Oct 4 panel of Fig 4B. ○. Vertical irregularities were detected between the first and the second lanes of the UD SCC-2 (HPV 16+) p50, p52, and Rel-B panels of Fig 5A. ○. Horizontal and vertical irregularities were detected around the second lane of the UPCI-SCC 131 (HPV-) p52 panel and the second lane of the UPCI-SCC 84 (HPV-) p65 panel of Fig 5A. ○. A horizontal irregularity was detected between the second and the third lane of the HPV-ve OSCC biopsy p50 panel of Fig 6A. ○. A vertical irregularity was detected directly below the bands presented in the (ii) panel of Fig 7A. ○. Discrepancies were detected in the background surrounding the results presented in the second and third lane of the NSP panel of Fig 7C (ii). The underlying raw FACS data and the individual level data for the various graphs presented in this article were not made publicly available, although the Data Availability statement states that all relevant data are within the manuscript and its Supporting Information files. The authors provided underlying data for some but not all the blots presented in this article, however the data received by the journal were not sufficient to resolve the concerns summarised above. In light of the concerns affecting multiple figures that question the integrity of the reported results, the PLOS ONE Editors retract this article. BCD agrees with the retraction decision. NB, MY, and DM either did not respond directly or could not be reached.

Zhang, Chenguang;Dai, Guangzheng;Pazo, Emmanuel Eric;Xu, Ling;Wu, Xianwei;Zhang, Hongda;Lin, Tiezhu;He, Wei

doi: 10.1371/journal.pone.0241630pmid: 33216749

Objective To compare the accuracy of five kinds of intraocular lens calculation formulas (SRK/T, Haigis, Hoffer Q, Holladay and Barrett Universal Ⅱ) in cataract patients with steep curvature cornea ≥ 46.0 diopters. Methods This is a retrospective study of cataract phacoemulsification combined with intraocular lens implantation in patients with steep curvature cornea (corneal curvature ≥ 46D). The refractive prediction errors of IOL power calculation formulas (SRK/T, Haigis, Holladay, Hoffer Q, and Barrett Universal II) using User Group for Laser Interference Biometry (ULIB) constants were evaluated and compared. Objective refraction results were assessed at one month postoperatively. According to axial length (AL), all patients were divided into three groups: short AL group (<22mm), normal AL group (>22 to ≤24.5mm) and long AL group (>24.5mm). Calculate the refractive error and absolute refractive error (AE) between the actual postoperative refractive power and the predicted postoperative refractive power. The covariance analysis was used for the comparison of five formulas in each group. The correlation between the absolute refractive error and AL from every formula were analyzed by Pearson correlation test, respectively. Result Total 112 eyes of 83 cataract patients with steep curvature cornea were collected. The anterior chamber depth (ACD) was a covariate in the short AL group in the covariance analysis of absolute refractive error (P<0.001). The SRK/T and Holladay formula had the lowest mean absolute error (MAE) (0.47D), there were statistically significant differences in MAE between the five formulas for short AL group (P = 0.024). The anterior chamber depth had no significant correlation in the five calculation formulas in the normal AL group and long AL group (P = 0.521, P = 0.609 respectively). In the normal AL group, there was no significant difference in MAE between the five calculation formulas (P = 0.609). In the long AL group, Barrett Universal II formula had the lowest MAE (0.35), and there were statistically significant differences in MAE between the five formulas (P = 0.012). Over the entire AL range, the Barrett Universal II formula had the lowest MAE and the highest percentage of eyes within ± 0.50 D, ± 1.00 D, and ± 1.50 D (69.6%, 93.8%, and 98.2% respectively). Conclusion Compared to SRK/T, Haigis, Hoffer Q, and Holladay, Barrett Universal Ⅱ formula is more accurate in predicting the IOL power in the cataract patients with steep curvature cornea ≥ 46.0 diopters. Introduction Cataract surgery is one of the most commonly performed procedures in the increasingly ageing population. Postoperative refractive error is one of the primary cause of postoperative visual outcome dissatisfaction among patients. Thus, accurate IOL power calculations have become extremely important. The calculation formula of artificial crystal has gradually developed from the first generation of theoretical formula to the Hill-radial basis function formula [1]. Yet there is still considerable debate about which formula provides the most accurate refractive prediction. Some scholars found that Hoffer Q performed best for eyes shorter than 22.0 mm [2], while most modern theoretical IOL formulas perform well for eyes with the normal eye axis lengths (22.0–24.5 mm) [3], Barrett Universal Ⅱ formula is considered to be most accurate in the long axis lengths [4, 5]. Borish and Duke-Elder classification of myopia as an optical system suggests 3 forms of myopia. Myopia due to increased axial length of the eye (axial myopia), refractive myopia and curvature myopia [6]. Although, AL is considered to be most important parameter in most modern-day formulae as it can changes the IOL power by nearly 2.5 to 3 times, more so in short eyes than in longer ones [7]. K readings are considered the second most important factor necessary for accurate IOL power calculations. Changes in K reading can alter the IOL power in a ratio of nearly 1:1 ratio [8]. As for whether the corneal curvature variation will affect the measurement of intraocular lens power, especially in the case of large corneal curvature, there is a paucity of research in their area. Therefore, this study assessed cataract patients with steep corneal curvature (corneal curvature≥ 46.0D). Five formulas were used to calculate the refractive power of the IOLs, namely the SRK / T [9], Haigis [10], Hoffer Q [11], Holladay [12], and Barrett Universal Ⅱ [13] formulas. Although, the accuracy of IOL power calculation formulas for highly myopic eyes have been assessed by Zhang et al. [5] on patients’ keratometric value ranging from 40.36 to 48.19 D (mean: 43.61±2.22 D), our current study includes a higher range of keratometric values (range: 46.00 to 51.03 D; mean: 47.31±1.08 D). Therefore, the purpose of this study is to assess the predictive accuracy of various IOL power formulas in eyes with steep corneal curvature. Patients and methods The study was approved by the Ethics Committee of the He Eye Specialists Hospitals, Shenyang on January 2020 and all research was performed in accordance with the Declaration of Helsinki. This was a retrospective study of cataract phacoemulsification combined with intraocular lens implantation at He Eye Specialists Hospitals, Shenyang included patient data from January to June 2019 with steep corneal curvature (corneal curvature ≥ 46D) from the central hospital database. Cataract extraction with IOL implantation was performed at He Eye Specialists Hospitals, Shenyang China by the practical senior surgeons. Cases from January 2019 to June 2019, were reviewed. The inclusion criteria were as follows: (1) biometric measurements determined by PCI (IOLMaster, Software V5.4 and above, Carl Zeiss Meditec, Inc., Dublin, CA, USA); (2) cataract surgery performed by phacoemulsification and in-the-bag IOL implantation; (3) use of the Hoya IOL (Hoya-PS AF-1 Series, Model PC-60AD, Hoya, Japan); and (4) 2.75 mm clear corneal incisions located temporally or superiorly. The exclusion criteria were as follows: (1) patients with a history of previous intraocular surgery or intraoperative or postoperative complications; (2) preexisting ocular diseases that may influence postoperative refraction, including keratoconus, corneal scarring, endothelial dystrophy, retinal detachment, and macular edema; (3) patients undergone prior refractive surgical procedures including refractive laser correction; (4) patients with follow-up of less than 1 month. In order to accurately measure all parameters required for IOL power calculation (SRN>2.0, including AL, anterior chamber depth (ACD), and keratometric (K) (both K 1 and K 2) values were collected for the back calculation of formulas. We selected cataract patients with corneal curvature ≥46D, without pterygium or keratoconus. Patients with intraoperative complications such as posterior capsule rupture, lens nucleus dislocation into the vitreous cavity, sulcus, or sutured lens were also excluded from the analysis. All data included for the final analysis was anonymized and coded in serial numbers. The authors of this current study were not allowed access to information that could identify individual participants during or after data collection. Eyes were divided into three groups according to their AL. The three groups consisted of short axis group (AL≤22mm), normal axis group (22mm <AL≤24.5mm) and long axis group (AL> 24.5mm). The IOL calculation results of SRK/T, Haigis, Hoffer Q, and Holladay formulas were obtained from IOL Master 500 and the calculation results of the Barrett Universal Ⅱ formula were obtained online. The objective post-operative refraction was assessed 1 month postoperatively. The prediction error is defined as the actual postoperative spherical equivalent (SE) that was calculated by each IOL formula minus the predicted postoperative SE. Thus, a positive predicted error in refraction indicates a hyperopic result than the predicted refraction. The absolute refractive error (AE) also were calculated as well as the percentage of eyes that had a predicted refractive error of within ±0.50 D, ± 0.75 D, and ±1.00 D. we assessed correlations between AL and AE. Statistical comparisons between formula absolute errors were performed using repeated measures analysis of variance (ANOVA). The relationship between the AE and AL was analyzed by Pearson correlation test. Statistical significance was defined as P< 0.05. Results The final analysis consisted of 112 eyes from 83 patients with steep corneal curvature (corneal curvature ≥46D) who underwent cataract surgery. Of the 83 patients, 64 (77.11%) were female, and the ages ranged from 45 to 91 years (mean = 70.04 ± 9.75 years). The ALs of the study population ranged from 19.83 to 31.92 mm (mean = 23.30±2.37 mm), the preoperative corneal curvatures from 46.00 to 51.03 D (mean = 47.31±1.08 D), and the preoperative anterior depths from 2.06 to 4.11 mm (mean = 2.97±0.44 mm) (Table 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Study population characteristics. https://doi.org/10.1371/journal.pone.0241630.t001 Table 2 shows the results of five formulas for different AL groups. The MAE varied with respect to AL with all 5 formulas. The anterior chamber depth was a covariate in the short axis eye group in the covariance analysis of absolute refractive error (p <0.001), and had no significant correlation in the five calculation formulas in the normal AL group and long AL group (P = 0.521, P = 0.609 respectively). There were statistically significant differences in MAE between the 5 formulas for the short AL group and long AL group (P = 0.024, P = 0.012 respectively). The Haigis formula has the highest MAE (0.77D). The SRK/T and Holladay formula has the lowest MAE (0.47D) and is close to the Barrett formula (0.48D). In the normal AL group, the ANOVA showed the difference between the 5 formulas was not statistically significant (P = 0.608). In the normal AL group and long AL group, the lowest MAE was achieved with the Barrett Universal II formula (0.41D, 0.35D respectively). In the whole sample, the lowest MAE was achieved once more with the Barrett Universal II formula. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Mean absolute predicted error by five formulas. https://doi.org/10.1371/journal.pone.0241630.t002 Table 3 shows in the short eye axis group, the mean refractive errors of the five formulas are all positive. In the SRK/T formula, the mean refractive error of each group is also positive. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Mean predicted error by five formulas. https://doi.org/10.1371/journal.pone.0241630.t003 Table 4 shows the percentage of eyes whose predicted refractive outcome for each formula are within ± 0.50 D, ± 0.75 D, or ± 1.00 D. In short AL group, the Hoffer Q formula had the highest percentage of eyes within ± 0.50 D, ± 1.00 D, and ± 1.50 D (70.0%, 90.2%, and 97.6% respectively). In long AL group and the entire AL range, the Barrett Universal II formula had the highest percentage of eyes within ± 0.50 D, ± 1.00 D, and ± 1.50 D. In the normal axis group, the refractive error of most eyes within ± 1.00 D. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 4. Percentage of eyes within predicted error ranges. https://doi.org/10.1371/journal.pone.0241630.t004 There is a positive correlation between AE and AL of SRK/T (r = 0.247, r2 = 0.061; P = 0 .009) (Fig 1), Hoffer Q (r = 0.388, r2 = 0.151,P<0.001) (Fig 2)and Holladay (r = 0.412, r2 = 0.170; P<0.001) (Fig 3), but not for Haigis (P = 0.428) (Fig 4), and Barrett Universal II formula (P = 0.278) (Fig 5). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Correlations between axial length and absolute error. The associations between axial length and absolute error were analyzed using absolute errors derived from SRK/T formula (r = 0.247, r2 = 0.061; P = 0 .009). https://doi.org/10.1371/journal.pone.0241630.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Correlations between axial length and absolute error. The associations between axial length and absolute error were analyzed using absolute errors derived from Hoffer Q formula (r = 0.388, r2 = 0.151; P<0.001). https://doi.org/10.1371/journal.pone.0241630.g002 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Correlations between axial length and absolute error. The associations between axial length and absolute error were analyzed using absolute errors derived from Holladay formula (r = 0.412, r2 = 0.170; P<0.001). https://doi.org/10.1371/journal.pone.0241630.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Correlations between axial length and absolute error. The associations between axial length and absolute error were analyzed using absolute errors derived from Haggis formula (r = -0.076, r2 = 0.006; P = 0.428). https://doi.org/10.1371/journal.pone.0241630.g004 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Correlations between axial length and absolute error. The associations between axial length and absolute error were analyzed using absolute errors derived from Barrett Universal II formula (r = -0.103, r2 = 0.011; P = 0.278). https://doi.org/10.1371/journal.pone.0241630.g005 Discussion Althought, the accuracy of IOL power calculation formulas for highly myopic eyes have been assessed by Zhang et al. [5] (mean K: 43.61±2.22 D), our current study includes a higher range of keratometric values (range: 46.00 to 51.03 D; mean K: 47.31±1.08 D). Our study did not include keratoconic eyes but we had eyes with high corneal curvature, in general IOL power calculation formulas are assumed for normal eyes. Studies such as Watson et al. [14] have reported that while hyperopic error was low in eyes with mild (K < 48.00 D) and moderate (K = 48.00–55.00 D) corneal curvature eyes following cataract surgery, However, errors can be unpredictable in severe steepening of the cornea. Studies have found that in normal eyes, the refractive outcomes predicted by modern formulas are within ± 0.5 D of the actual postoperative refractive outcomes in about 75% of eyes with general curvature cornea [15–17]. However, in this study, all the eyes are with steep curvature cornea, most formulas only achieved a ±0.5 D prediction accuracy in 70% of eyes with the short AL group and normal AL group. In the long AL group, the Barrett Universal II formula achieved a ±0.5 D predictive accuracy in 76.9% of eyes, with most other formulas only achieving around 50% accuracy. Over the entire AL range, the Barrett Universal II was the most accurate formula by a significant margin, had a highest percentage of eyes with prediction errors between ±0.5 D, ±1.0 D, and ±1.50 D than the other 4 formulas assessed. Cooke et al. [15] found Barrett Universal II had the highest percentage of eyes with prediction errors between± 0.5 D and ±1.0 D than other formulas in all eyes. In this study, the percentage of eyes within ±0.5 D in most formulas is much lower than the reported in normal eyes, probably because the relationship between corneal curvature, ACD and IOL position changes, thus reducing the accuracy of formula in predicting the effective lens position. In short AL group, the average refractive errors of the five formulas are negative, the mean refractive error of the Haigis formula was the highest (-0.65), and the Holladay formula was the lowest (-0.01D), which is close to the Barrett Universal II formula (-0.08D). All 5 formulas assessed in this study resulted in mean myopic predicted errors. The same conclusion can be found in the study by Kane et al. [17] For patients in short AL, the choice of intraocular lens power can be appropriately small. In long AL group, except for the SRK / T formula, the mean refractive error of the other four formulas are all positive, indicates that such patients have hyperopic predicted errors. Chen et al. [18] found that most eyes with an AL of >33 mm presented with postoperative hyperopia of +2.0 D to +3.0 D. For patients with long ALs, especially with extremely long ALs, the reserved intraocular lens power should be increased. Gavin et al. [2] analyzed the British short eyes patients and found that the Hoffer Q formula is more accurate. Aristodemou et al. [19] showed that HofferQ performed the best on AL from 20.00 to 20.99 mm, and Hoffer Q and Holladay 1 performed best on AL from 21.00 to 21.49 mm. Day et al. [20] reported that in patients with AL ≤ 22mm, the Hoffer Q, Holladay, and Haigis formulas have best accuracy, and the SRK / T formula has a large error. In our study, in short group, the Holladay and SRK/T formulas having the lowest MAE followed by Barrett Universal II and Hoffer Q formulas. The Haigis formulas had the highest MAE. There were statistically significant differences in MAE between the 5 formulas (P = 0.024). Similar to Kane et al. [17] we also found that short ALs, the Holladay 1 formula had the lowest MAE followed by SRK/T, Holladay 2, and Barrett Universal II formulas, while Haigis and Hoffer Q formulas had the higher MAE. In the normal axis group, the lowest MAE was achieved with the Barrett Universal II and Holladay formulas (0.41), but the difference between formulas was not statistically significant. The most accurate IOL power formula in the normal AL group was not clearly defined. Most researchers believe that the accuracy of various formulas is similar. Reitblat et al. [3] study on Holladay 1, SRK/T, Hoffer Q, Haigis, Holladay 2, Barrett Universal II and Olsen formulas in this AL range did not find any formula to be more accurate. In our study, we found that the most accurate formula was Barrett Universal II followed by SRK/T, Haigis, Holladay and Hoffer Q in long AL group. There were statistically significant differences in MAE between the 5 formulas (P = 0.012). Zhang Y et al. [5] found through comparison and analysis of patients with AL of 26.00 mm or above, Barrett Universal II formula has the lowest absolute refractive error compared to Haigis, Hoffer Q, Holladay, and SRK / T formulas. Over the entire AL range, Barrett Universal II formula has the lowest absolute refractive error and was the most accurate formula than the other 4 formulas. Melles et al. [16] found that the MAE of Barrett Universal II formula in each eye axis group is the lower. Barrett Universal II formula is still more accurately with steep corneal curvatures. Kendrick et al. [21] found that Barrett Universal II formula was superior in eyes with stage I (corneal powers of 48 D or lower) and stage II (corneal power between 48.01 D and 53 D) keratoconus. The results are the same as our research. Our study has several limitations. First, this study is retrospective, and a randomized controlled study may provide more detailed clinical information. Second, the sample size was small, especially for long axis. However, this is the first paper to present the IOLs calculation in the eyes with steep curvature corneal curvature ≥ 46.0 diopters. Additionally, modern formulas such as Holladay 2 and Oslen were not available at our disposal and therefore we were unable to assess their performance. Since the Olsen is a ray-tracing thick-lens formula and includes variables such as the AL, K value, ACD, lens thickness, and age of the patient for calculation [22], it theoretically can perform at par with that Barrett Universal Ⅱ formula [4]. In conclusion, Barrett Universal Ⅱ formula is more accurate than other calculation formulas in predicting the IOL degrees in cataract patients with steep curvature cornea, and the impact of AL is less compared to others. Further research is warranted using modern IOL formulas in order to understand the impact of steep curvature cornea on post-surgical refractive outcomes in cataract and clear lens extraction patients.

Sjögren, Karl-Göran;Olalde, Iñigo;Carver, Sophie;Allentoft, Morten E.;Knowles, Tim;Kroonen, Guus;Pike, Alistair W. G.;Schröter, Peter;Brown, Keri A.;Brown, Kate Robson;Harrison, Richard J.;Bertemes, Francois;Reich, David;Kristiansen, Kristian;Heyd, Volker

doi: 10.1371/journal.pone.0241278

Chang, Shao-Hsuan;Zhou, Dong;Eliasy, Ashkan;Li, Yi-Chen;Elsheikh, Ahmed

doi: 10.1371/journal.pone.0240724pmid: 33147249

1. Introduction Research in corneal cross-linking (CXL) has gradually developed a tool that became the first-line treatment for treating keratoconus. The prevalence of this disease ranges from 1 in 375 in the general population [1]). This technique creates new covalent cross-links between molecules and extracellular matrix within stroma to strengthen and stabilize the structure of cornea. Collagen cross-linking using the conventional Dresden protocol is initially thought to affect the mechanical properties, ultrastructure, hydrodynamic and enzymatic behavior of the cornea [2–7]. To assess the effectiveness of this treatment, biomechanical characterization provides quantitative measurements to assess the degree of cross-linking. Many studies reported a significant increase in the corneal stiffness through experimental studies on cross-linked corneas measured by tensile or inflation tests [6,8–10]. However, various range of stiffness is proposed due to the variations in methodologies used for experimental settings. In addition, the estimated tangent modulus using tensile test is considerably higher than the measurement from inflation, which can vary by the order of magnitude. Therefore, in this study, a unique inflation test rig was designed and developed which allowed characterisation of intact porcine eye globes for more accurate measurement and predictability of CXL. The stiffness of the cornea was determined by its nonlinear geometry and material behavior. The material behavior was highly related to its micro- and ultra-structure, including the hydration, spatial fibril density and arrangement [11]. Uniaxial tensile test is the most commonly used technique for the measurement of corneal stiffness. However, the mechanical measurements may be inaccurate due to its technique destructive specimen preparation with disrupted fibril orientations, inconsideration of corneal curvature, and non-uniform stress distribution while applied the stretching force along the corneal strip [12–14]. Therefore, inflation test has been developed and attempted to address this issue by mimicking the in vivo loading conditions of the eye, which is considered to be more reliable and closely related to in vivo conditions than uniaxial tensile test. It is expected to produce the average behavior of intact corneas due to the stromal anisotropy resulting from the preferred collagen fibril orientation. Inflation test analyses the degree of extension of the cornea in response to the change in IOP. During the test, a variety of monitoring techniques including the use of laser or DIC can be applied to track displacements [15–17]. For post-test analysis, the finite element (FE) modelling technique is employed to construct numerical models of whole corneas or eye globes, and the material stress-strain relationships can be adjusted until the predicted surface deformations of the models matched those observed experimentally. This technique has provided an accurate means of determining the tissue’s tangent modulus [18,19]. Due to the non-uniform curvature with variable thickness, both the external and internal geometries of eye globe were obtained to generate the corneal model. A model developed for the cornea using entire eye globe allowed appropriate realistic displacement at the limbus, the displacements at the limbus were tracked during the experiment and then introduced at the boundary of the corneal FE model. This approach of performing a cornel only model was to establish a biomechanical model that can mimic the functional response of a real eye with the reduced geometrical complexity and increased efficiency of the computational calculation. Paired eye globes were used to examine the intact corneal behaviors, with one as the test sample with CXL treatment and the other as its non-treated control in PBS. Through inverse analysis, non-linear material behaviours were better understood with the effect of Dresden cross-linking protocol on stiffness of the tissue. This study was also the first one to consider the stiffness of cornea as a whole following CXL. These results may potentially supervise the further modelling of CXL treatment in order for the accurate surgical prediction, which may allow clinicians to estimate the severity of keratoconus and accomplish the direct comparison with the refractive outcomes in practice. 2. Method 2.1. Specimen and preparation Seven paired fresh porcine eyes were obtained from a local abattoir (Morphets, Tan house farm, Widnes) and tested within 6–9 hours after death. Soft muscular tissue was removed with surgical scissors. The superior direction was marked and the eye globe was placed in a customized compartment for accurate needle insertion through the posterior pole. The internal eye components were removed through the posterior pole using a 14G needle. The needle was then lightly glued around the posterior pole and the intra-ocular cavity was washed with 5 to 6 ml PBS (Sigma, Dorset, United Kingdom). The outer surface of the globe was continually kept hydrated by applying PBS every 2–5 minutes. Random speckles were applied on the globe by lightly spraying a waterproof and fast drying black paint to facilitate deformation tracking in post-analysis. The prepared specimen was then placed into a custom-designed eye chamber filled with PBS, and transferred onto the inflation rig (Fig 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Inflation test equipment. (A) inflation setup with front cover removed, (B) camera array showing angles between cameras and distance from cameras to eye chamber (placed in the centre). https://doi.org/10.1371/journal.pone.0240724.g001 2.2. Dresden protocol The right eyes of the pared specimens were prepared for CXL prior to inflation test. The procedure was performed following conventional Dresden protocol [20], with the anterior surface of the corneas applied with 5 mL of 0.1% riboflavin in dextran at 3-minute intervals treating for 30 minutes. UVA (370 nm) illumination at 3 mW/cm2 (Opto Xlink; Mehra Eyetech Pvt. Ltd., Delhi, India) was then performed for further 30 minutes. Topical dosing of riboflavin drops was continued during the irradiation. 2.3. Test rig The inflation test rig provides full-field observation of ocular response to uniform intraocular pressure (IOP) changes. The physical test equipment is fully bespoke having been designed and built in-house (Fig 1). The equipment features closed loop control software written in LabVIEW (version 10.0.1, RRID:SCR_014325) to regulate IOP while collecting real-time data by triggering cameras to take pictures of the globe. The obtained images are used for measurement of deformation across the globe. The specimen was clamped in a horizontally placed eye chamber with high precision real-time laser (LK-2001, Keyence, UK) pointing towards the apical displacement. An array of six high resolution digital cameras (18.0 megapixels, 550D, Canon, Tokyo, Japan) surrounding the eye chamber and a pressure adjusting tank was placed vertically to inflate the eye while taking synchronous images. The camera setup shown in Fig 1B allows an angle of 25° within each pair and an angle of 120° between each set. 2.4. Testing control and protocol A custom-built LabVIEW software was used to tightly control the pressure. The experiments started by 3 pre-conditioning cycles. The pre-conditioning cycles were to ensure the eye was sitting comfortably on the needle, and the tissue behavior was repeatable [15] An initial pressure of 2.5 mmHg was used to balance the external pressure applied by PBS in the pressure chamber, and was therefore considered a zero-pressure point for the inflation test. Specimens were loaded to a maximum internal load at a medium rate of 0.55 mmHg/s for each cycle. During each cycle the eye was allowed to relax for a period of 2 minutes which was obtained experimentally to allow tissue to fully recover to its relaxation state. The behavior of specimen in the final loading cycle was used for post-analysis. 2.5. Thickness measurement After the experiment was completed, the eye was removed from the test rig and dissected into anterior and posterior parts. Eight meridian profiles of discrete thickness measurements were selected as shown in Fig 2. The thickness at each desired point on each meridian line was determined using an in-house developed Thickness Measurement Device (TMD) (LTA-HS, Newport, Oxfordshire, UK) which was developed by the Biomechanical Engineering group to measure the thickness of biological tissue. A vertical measurement probe was located at a height of about 30 mm above the centre point of the support. The probe moved down with a controlled velocity until it reached the surface of the tissue. By precisely knowing the original distance between the initial position of probe and the surface of support, the measured value was recorded as the thickness of the tissue. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Thickness measurement was performed on each sample along the eight meridian lines with four points measured per line. https://doi.org/10.1371/journal.pone.0240724.g002 2.6. Geometric modelling To decrease the geometrical complexity and understand the effect of CXL treatment on corneas where the application of interest is, we built up a corneal-only model by excluding the sclera part from a whole globe model. In this corneal model, the orphan mesh of geometry was constructed with Abaqus 6.13 (Dassault Systèmes Simulia Corp., Rhode Island, USA) using bespoke software. The 2592 elements with 8611 total nodes adopted the hybrid and quadratic type with triangular cross-section (C3D15H), which were arranged in 12 rings across the cornea surface and 3 layers through the thickness. Corneal apex was restrained against displacement in X- and Y-directions, whereas limbus was restrained in the X-, Y-, and Z-direction. The intraocular pressure was distributed on the posterior surface of the cornea. The apical displacement of the entire cornea was extracted by the displacement of corneal apex minus the average displacement of limbus in the anterior-posterior direction. 2.7. Deformation measurement by Digital Image Correlation (DIC) The image profiles obtained were analyzed using a 2D DIC method named Particle Image Velocimetry (PIV) to obtain deformations on the surface of the eye (Fig 3) [21,22]. PIV compares an un-deformed and deformed image pairs of specimen surface which was speckled to present the local displacements within the selected subsets. Three discrete locations including corneal apex and limbus were measured from each camera (Fig 3B). As only cornea was considered in the study, the cornea deformation was calculated by subtracting the average displacement of limbus in the anterior-posterior direction from the displacement of corneal apex. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. A demonstration of Particle Image Velocimetry (PIV). (A) screen capture of the manual tagging of desired points. (B) result of deformed tagged points (corneal apex and limbus). mm/pixel = 1. https://doi.org/10.1371/journal.pone.0240724.g003 2.8. Determining the corneal material properties An in-house built software that uses Particle Swarm Optimization (PSO) as an optimization strategy was developed in Matlab (RRID:SCR_001622) to conduct the inverse analysis optimization due to its success in the engineering applications [23–25]. PSO evaluates the fitness of the apical displacement between simulation and experiment and iterates over the different values of material parameters to decrease the error until the best fitness appears. The material constitutive model chosen to demonstrate the material behavior of ocular tissue during loading was Ogden model as presented in Eq 2.1, utilized in a number of previous studies on soft tissues [26,27]. (2.1) where W is the strain energy density; are the deviatoric principal stretches, ; λi are the principal stretches; J denotes the determinant of deformation gradient and describes the change of material volume; the second term was ignored as the cornea tissue was incompressible (J = 1). αi and μi are material parameters; N is the function order and N = 1 was used in this study. The Ogden material model order one relies on two parameters of μ (shear modulus) and α (strain hardening exponent) to define the non-linear material behavior. The use of first order material model (N = 1) reduced the complexity of optimization and thus the computational cost as a result of less variables. The values of material parameters α and μ represented the output of the inverse modelling process that resulted in the highest fitness of simulation against inflation experiment. Therefore, the objective function was to minimize the root mean squared (RMS) of deformation, which was calculated as shown in Eq 2.2: (2.2) where M is the number of measurement locations; N is the number of pressure levels; δi,j is the deformation at each particular pressure level i and location j. The design optimization process adjusts the value of μ and α within the constitutive model while setting a wide lower and upper boundary range (lower boundary = [0.005, 50]; upper boundary = [0.2, 200]). The error limit of RMS was set as 10%, which terminated the optimization once the error is lower than the limit. With these parameters, stress and strain could then be extracted from the numerical modelling results. The uniaxial-mode stress was calculated through obtained μ and α in Table 2, based on the previously described method [28] and then tangent modulus was calculated numerically from the gradient of the resulting stress-strain curve by Eq 2.3: (2.3) Where σ is stress and ε is strain. The strain difference Δε in this study is 0.2%, the corresponding stress values at each strain value were presented as shown in Fig 6. 2.9. Statistical analysis The statistical evaluation was performed using SPSS software version 18.0 (IBM Corp. USA, RRID:SCR_002865). Results are expressed as means ± standard deviation (SD) and significant differences are calculated using one-way analysis of variance (ANOVA) with Turkey’s HSD post-hoc test. Significance differences accepted where p < 0.05. 2.1. Specimen and preparation Seven paired fresh porcine eyes were obtained from a local abattoir (Morphets, Tan house farm, Widnes) and tested within 6–9 hours after death. Soft muscular tissue was removed with surgical scissors. The superior direction was marked and the eye globe was placed in a customized compartment for accurate needle insertion through the posterior pole. The internal eye components were removed through the posterior pole using a 14G needle. The needle was then lightly glued around the posterior pole and the intra-ocular cavity was washed with 5 to 6 ml PBS (Sigma, Dorset, United Kingdom). The outer surface of the globe was continually kept hydrated by applying PBS every 2–5 minutes. Random speckles were applied on the globe by lightly spraying a waterproof and fast drying black paint to facilitate deformation tracking in post-analysis. The prepared specimen was then placed into a custom-designed eye chamber filled with PBS, and transferred onto the inflation rig (Fig 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Inflation test equipment. (A) inflation setup with front cover removed, (B) camera array showing angles between cameras and distance from cameras to eye chamber (placed in the centre). https://doi.org/10.1371/journal.pone.0240724.g001 2.2. Dresden protocol The right eyes of the pared specimens were prepared for CXL prior to inflation test. The procedure was performed following conventional Dresden protocol [20], with the anterior surface of the corneas applied with 5 mL of 0.1% riboflavin in dextran at 3-minute intervals treating for 30 minutes. UVA (370 nm) illumination at 3 mW/cm2 (Opto Xlink; Mehra Eyetech Pvt. Ltd., Delhi, India) was then performed for further 30 minutes. Topical dosing of riboflavin drops was continued during the irradiation. 2.3. Test rig The inflation test rig provides full-field observation of ocular response to uniform intraocular pressure (IOP) changes. The physical test equipment is fully bespoke having been designed and built in-house (Fig 1). The equipment features closed loop control software written in LabVIEW (version 10.0.1, RRID:SCR_014325) to regulate IOP while collecting real-time data by triggering cameras to take pictures of the globe. The obtained images are used for measurement of deformation across the globe. The specimen was clamped in a horizontally placed eye chamber with high precision real-time laser (LK-2001, Keyence, UK) pointing towards the apical displacement. An array of six high resolution digital cameras (18.0 megapixels, 550D, Canon, Tokyo, Japan) surrounding the eye chamber and a pressure adjusting tank was placed vertically to inflate the eye while taking synchronous images. The camera setup shown in Fig 1B allows an angle of 25° within each pair and an angle of 120° between each set. 2.4. Testing control and protocol A custom-built LabVIEW software was used to tightly control the pressure. The experiments started by 3 pre-conditioning cycles. The pre-conditioning cycles were to ensure the eye was sitting comfortably on the needle, and the tissue behavior was repeatable [15] An initial pressure of 2.5 mmHg was used to balance the external pressure applied by PBS in the pressure chamber, and was therefore considered a zero-pressure point for the inflation test. Specimens were loaded to a maximum internal load at a medium rate of 0.55 mmHg/s for each cycle. During each cycle the eye was allowed to relax for a period of 2 minutes which was obtained experimentally to allow tissue to fully recover to its relaxation state. The behavior of specimen in the final loading cycle was used for post-analysis. 2.5. Thickness measurement After the experiment was completed, the eye was removed from the test rig and dissected into anterior and posterior parts. Eight meridian profiles of discrete thickness measurements were selected as shown in Fig 2. The thickness at each desired point on each meridian line was determined using an in-house developed Thickness Measurement Device (TMD) (LTA-HS, Newport, Oxfordshire, UK) which was developed by the Biomechanical Engineering group to measure the thickness of biological tissue. A vertical measurement probe was located at a height of about 30 mm above the centre point of the support. The probe moved down with a controlled velocity until it reached the surface of the tissue. By precisely knowing the original distance between the initial position of probe and the surface of support, the measured value was recorded as the thickness of the tissue. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Thickness measurement was performed on each sample along the eight meridian lines with four points measured per line. https://doi.org/10.1371/journal.pone.0240724.g002 2.6. Geometric modelling To decrease the geometrical complexity and understand the effect of CXL treatment on corneas where the application of interest is, we built up a corneal-only model by excluding the sclera part from a whole globe model. In this corneal model, the orphan mesh of geometry was constructed with Abaqus 6.13 (Dassault Systèmes Simulia Corp., Rhode Island, USA) using bespoke software. The 2592 elements with 8611 total nodes adopted the hybrid and quadratic type with triangular cross-section (C3D15H), which were arranged in 12 rings across the cornea surface and 3 layers through the thickness. Corneal apex was restrained against displacement in X- and Y-directions, whereas limbus was restrained in the X-, Y-, and Z-direction. The intraocular pressure was distributed on the posterior surface of the cornea. The apical displacement of the entire cornea was extracted by the displacement of corneal apex minus the average displacement of limbus in the anterior-posterior direction. 2.7. Deformation measurement by Digital Image Correlation (DIC) The image profiles obtained were analyzed using a 2D DIC method named Particle Image Velocimetry (PIV) to obtain deformations on the surface of the eye (Fig 3) [21,22]. PIV compares an un-deformed and deformed image pairs of specimen surface which was speckled to present the local displacements within the selected subsets. Three discrete locations including corneal apex and limbus were measured from each camera (Fig 3B). As only cornea was considered in the study, the cornea deformation was calculated by subtracting the average displacement of limbus in the anterior-posterior direction from the displacement of corneal apex. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. A demonstration of Particle Image Velocimetry (PIV). (A) screen capture of the manual tagging of desired points. (B) result of deformed tagged points (corneal apex and limbus). mm/pixel = 1. https://doi.org/10.1371/journal.pone.0240724.g003 2.8. Determining the corneal material properties An in-house built software that uses Particle Swarm Optimization (PSO) as an optimization strategy was developed in Matlab (RRID:SCR_001622) to conduct the inverse analysis optimization due to its success in the engineering applications [23–25]. PSO evaluates the fitness of the apical displacement between simulation and experiment and iterates over the different values of material parameters to decrease the error until the best fitness appears. The material constitutive model chosen to demonstrate the material behavior of ocular tissue during loading was Ogden model as presented in Eq 2.1, utilized in a number of previous studies on soft tissues [26,27]. (2.1) where W is the strain energy density; are the deviatoric principal stretches, ; λi are the principal stretches; J denotes the determinant of deformation gradient and describes the change of material volume; the second term was ignored as the cornea tissue was incompressible (J = 1). αi and μi are material parameters; N is the function order and N = 1 was used in this study. The Ogden material model order one relies on two parameters of μ (shear modulus) and α (strain hardening exponent) to define the non-linear material behavior. The use of first order material model (N = 1) reduced the complexity of optimization and thus the computational cost as a result of less variables. The values of material parameters α and μ represented the output of the inverse modelling process that resulted in the highest fitness of simulation against inflation experiment. Therefore, the objective function was to minimize the root mean squared (RMS) of deformation, which was calculated as shown in Eq 2.2: (2.2) where M is the number of measurement locations; N is the number of pressure levels; δi,j is the deformation at each particular pressure level i and location j. The design optimization process adjusts the value of μ and α within the constitutive model while setting a wide lower and upper boundary range (lower boundary = [0.005, 50]; upper boundary = [0.2, 200]). The error limit of RMS was set as 10%, which terminated the optimization once the error is lower than the limit. With these parameters, stress and strain could then be extracted from the numerical modelling results. The uniaxial-mode stress was calculated through obtained μ and α in Table 2, based on the previously described method [28] and then tangent modulus was calculated numerically from the gradient of the resulting stress-strain curve by Eq 2.3: (2.3) Where σ is stress and ε is strain. The strain difference Δε in this study is 0.2%, the corresponding stress values at each strain value were presented as shown in Fig 6. 2.9. Statistical analysis The statistical evaluation was performed using SPSS software version 18.0 (IBM Corp. USA, RRID:SCR_002865). Results are expressed as means ± standard deviation (SD) and significant differences are calculated using one-way analysis of variance (ANOVA) with Turkey’s HSD post-hoc test. Significance differences accepted where p < 0.05. 3. Results By providing the pressures from posterior pole to the corneal apex with respect to IOP, the measurements of displacement indicate the response of the eye globes. The thickness variations across the whole cornea were demonstrated in Table 1. Four points per line across the cornea were identified as the central cornea, central-peripheral, peripheral-limbus, and limbus. The mean thickness measurements following CXL treatment from central to limbus were 0.95 ± 0.08 mm, 1.03 ± 0.07 mm, 1.03 ± 0.06 mm and 1.07 ± 0.05 mm. Comparing to the control eyes, the measurements were recorded as 1.34 ± 0.09 mm (central, p = 0.003), 1.19 ± 0.05 mm (central-peripheral, p = 0.02), 1.17 ± 0.06 mm (peripheral-limbus, p = 0.02) and 1.14 ± 0.02 mm (limbus, p = 0.06) accordingly. The result demonstrated a statistically significant reduction in thickness in the central and peripheral areas (29±8% and 13±7%, respectively) which were affected by CXL. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Average thickness measurements across the whole cornea at four different points. The thickness was measured after the inflation test. https://doi.org/10.1371/journal.pone.0240724.t001 To compare the controlled and cross-linked specimens, the material representations have been derived for corneal regions. It can be considered by the numerical parameters α and μ, in which μ is relating to the initial shear modulus and α to the non-linearity. Material behavior was compared firstly by the values of optimized material parameters α and μ which can reproduce the displacement curves of cornea apex. The inverse analysis resulted in a RMS error of 5.58 ± 1.79% (approximately 17 μm), which showed that the simulation closely matched the experimental results. The values of material parameters α and μ for all specimens using the inverse modelling procedure were generated and shown in Table 2. The average values of μ were 0.02 ± 0.012 and 0.01 ± 0.002 (p = 0.157), and the average values of α were 94.3 ± 28.9 and 65.1 ± 15.9 (p = 0.037) in CXL and PBS control, respectively. The specimens following CXL showed an increase of 45.4 ± 28.9% and 66.1 ± 110.9% in the values of α and μ. The greater values of material parameters indicated a stiffer material behavior, resulting in 27.9 ± 9.5% less displacement of cornea apex in CXL (307 ± 65 μm) than PBS (437 ± 63 μm) specimens at the IOP of 27.25 mmHg (Fig 4). The FE model provided a match of the displacements obtained from the experiment. Experimental and numerical results were examined up to an IOP of 27.5 mmHg, which demonstrated the progressive increase in corneal apex displacement as shown in Figs 4 and 5. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. IOP-displacement curves of experimental and numerical results for each individual paired eye. https://doi.org/10.1371/journal.pone.0240724.g004 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Finite element model of a tested porcine cornea. Image viewed from the side of (A) control and (B) CXL eyes. https://doi.org/10.1371/journal.pone.0240724.g005 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Optimised material parameters α and μ obtained for all specimens from the corneal region using inverse modelling procedure. https://doi.org/10.1371/journal.pone.0240724.t002 The corneal stress and strain curves were then extracted from the FE models and tangent modulus was calculated from the resulting stress-strain behavior. All specimens demonstrated the nonlinear behavior with an initial low tangent modulus increasing gradually under higher stress. The curve of tangent modulus (Et) versus stress (σ) for each cornea was shown in Fig 6A. The overall stiffening effect was demonstrated by the ratio of the tangent modulus larger than 1 at all stress levels (Et Cross-linking / Et Control PBS) (Fig 6B). The average ratio values stayed between 1.4 and 1.5 throughout the inflating with IOP. Comparisons performed on tangent modulus ratios at a stress of 0.03 MPa close to the physiological level [29], a 43% ± 24% increase in tangent modulus was observed in the corneas treated with the Dresden protocol (Et Cross-linking: 2.48 ± 0.69 vs Et Control PBS: 1.73 ± 0.40, p = 0.029). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. The ratio of tangent modulus of paired samples. (A) the tangent modulus vs stress behaviour of right and left eye from 7 paired eyes, (B) the ratio of tangent modulus between control and CXL (n = 7). Values from each individual pair eye tested are indicated by gray lines. Average stiffening ratio (mean ± SD) indicated by the bold line and error bars. The red dashed line represented the value of 1. Gray shaded region represents 0.03 MPa. https://doi.org/10.1371/journal.pone.0240724.g006 4. Discussion In this study, porcine eyes were used to investigate the biomechanical behavior over the entire cornea, whole eye globes were used in generating the corneal FE model. The inverse FE method used to optimize material behavior parameters was found to provide an adequate fit between the experimental and numerical pressure-displacement behavior following CXL treatment. A significant increase was found in stiffness after CXL and a 29% reduction in corneal apical rise with increase IOP during inflation. It has been reported that CXL significantly increased the stiffness of porcine corneas by about 42% when subjected to high pressure (300 mmHg) using inflation, but no significant difference was observed under physiological range of pressure (15 mmHg) [30]. A possible explanation could be due to the too low loading stress to observe the changes in polar distribution of fibril networks (reflecting straightening of crimp or reorientation of lamellae). The stiffening effect of cross-linked porcine cornea obtained using inflation in this study has been demonstrated to be relatively small compared to that from tensile test [6,31]. The difference in mechanical response depends on the experimental strain rate, the regions analyzed, tissues’ anisotropy, and the constitutive models presented. Due to the anisotropic nature of the corneal tissue, the corneal strength is explained not only through the variations in the structure of collagen and its interactions with extracellular matrix, but also through the orientation of collagen fibrils according to the direction of the load. Collagen fibrils re-orientate themselves to the direction of the applied load, which further stiffen the tissue behavior [32]. The procedure involved in tensile test has some inherent deficiencies such as non-uniform stress distribution across the curved corneal strips, which have reduced the reliability of this method [12]. It has been reported that longitudinal fibrils are found in regions supporting tensile loads, and transverse fibrils corresponds to regions under compressive loading or loading in orthogonal direction [32,33]. Tensile and inflation testing follow different loading orientations in measurement of corneal mechanical properties. Tensile testing induces a change in collagen fibrils’ alignment towards the load direction while inflation testing ensures the tissue to behave closer to the in-vivo conditions. Additionally, the mathematical analysis of the present inflation testing was built on a number of assumptions of corneal material which was modelled as a homogenous and hyper-elastic material properties. The analysis in the current study did not take into account the variation of material stiffness between the corneal epithelium, endothelium and stroma [34], and the preferred orientation of collagen fibrils [35]. For these reasons, the analysis is expected to produce the characteristic behaviors of intact corneas which were considered as a whole in numerical simulations process. Previous inflation study on trephination specimens which showed a rise in the apex as a function of increased IOP was very similar to the observations in porcine corneas [12]. Although the current study cannot rule out a contribution of the entire eye motion to the apical radius displacement, the behaviour which was observed in corneal apical rise was consistent in 7 pairs of eye globes. It was reported that non-cross-linked corneas did not fully return to the initial apical position after pressurized. However, cross-linked corneas tended to return to the original values both for apical position and corneal thickness, showing a more elastic behaviour than then non-cross-linked cornea [6]. The current study did not examine the hysteresis properties before and after CXL, but the results are consistent with the cross-linked cornea being stiffer. Limitations and future work Although this study used a corneal model instead of a whole eye model, which was thought to be crucial to establish a biomechanical model mimicking the refractive function of a real eye [36–38]. The full parametric characterization of human corneal deformation as a function of pressure will be valuable to enhance the predictability of FE modelling of the cornea and ultimately the predictability of the procedure. One limitation of the study was the boundary condition put on the limbus as fixed. This condition restricted the expansion of limbus in the simulation, which could underestimate the stiffness of cornea in both cross-linking treated and untreated eyes due to the externally introduced stiffness of the boundary condition. As the study chose the apex point in the calibration of material parameters, the effect of limbus expansion was considered minor. However, it was suggested to quantify the effect in our further study. Open questions such as the apparent anisotropy of the intact porcine cornea in the biomechanical response and in response to treatment are yet to be confirmed in humans and of interest in pathologic or keratoconic corneas. Keratoconus is regarded as a degenerative disease affecting the corneal collagen networks, in which a degeneration of the collagen fibril structure and an increased propensity of fibril sliding could give rise to altered macroscopic morphology [39,40]. Therefore, using different methodologies may have profound impacts on the mechanical outcome measured, especially for the comparison of normal and diseased tissue. Further work will be needed to evaluate the effects on keratoconic corneas instead of normal corneas. Limitations and future work Although this study used a corneal model instead of a whole eye model, which was thought to be crucial to establish a biomechanical model mimicking the refractive function of a real eye [36–38]. The full parametric characterization of human corneal deformation as a function of pressure will be valuable to enhance the predictability of FE modelling of the cornea and ultimately the predictability of the procedure. One limitation of the study was the boundary condition put on the limbus as fixed. This condition restricted the expansion of limbus in the simulation, which could underestimate the stiffness of cornea in both cross-linking treated and untreated eyes due to the externally introduced stiffness of the boundary condition. As the study chose the apex point in the calibration of material parameters, the effect of limbus expansion was considered minor. However, it was suggested to quantify the effect in our further study. Open questions such as the apparent anisotropy of the intact porcine cornea in the biomechanical response and in response to treatment are yet to be confirmed in humans and of interest in pathologic or keratoconic corneas. Keratoconus is regarded as a degenerative disease affecting the corneal collagen networks, in which a degeneration of the collagen fibril structure and an increased propensity of fibril sliding could give rise to altered macroscopic morphology [39,40]. Therefore, using different methodologies may have profound impacts on the mechanical outcome measured, especially for the comparison of normal and diseased tissue. Further work will be needed to evaluate the effects on keratoconic corneas instead of normal corneas. 5. Conclusion The current study has provided experimental data of the significant changes in corneal thickness and apical rise with increased IOP after CXL. The comparisons of stiffness and analysis of inverse FE modelling presented in this study provided important information relating to the effectiveness of CXL in biomechanical behavior across the corneas, which can be useful in applications where the prediction of the modifications of CXL protocol in corneal material stiffness is required. Although only corneal model was performed in this study, the experimental data are valuable input parameters in FE models that will allow a better understanding and increased predictability of the CXL treatment.

Jia, Peter;Caspari, Gino;Betts, Alison;Mohamadi, Bahaa;Balz, Timo;Cong, Dexin;Shen, Hui;Meng, Qi

doi: 10.1371/journal.pone.0240739pmid: 33147229

1. Introduction Western Xinjiang is a land of mountains where the narrow ridge of the Tianshan is split in two by the Yili River. The high steep-sided peaks are permanently glaciated, forming numerous rivers cutting deep valleys down to the desert below. The upper slopes below the snowline are forested, with open patches of grassland. With the exception of the Yili Valley, rainfall is low, and the land is not well suited to agriculture, although dry farming is possible in some particularly favourable places. Irrigation agriculture is practiced today at the foot of the mountains, exploiting permanently flowing rivers fed by snow and glacial meltwater. Historically, the main economic strategy has been transhumant pastoralism, taking advantage of the extreme ranges of altitude and the rich seasonal grassland pastures. This is still the case today where Kazakhs and Mongols, who have historically migrated into this region, herd sheep, goat, cattle and horses in a round of seasonal movement. Archaeologically, this lifestyle can be traced back in various forms to the Bronze Age when the first herders moved into the western Tianshan from the Eurasian steppe. Transhumant pastoralism is characterized by regular movement of herders and their livestock between fixed points to exploit the seasonal availability of pastures [1]. In western Xinjiang, annual movement is vertical and typically fairly short range, generally from 20–200 km. Although the herds provide the main basis of the economy for most pastoralists in the region, almost all of them engage to a greater or lesser extent in limited agriculture, depending on the environment in which they operate. Contemporary pastoralists in the lush Yili Valley cultivate quite large numbers of fields, both for fodder and for cash or food crops, but in the more arid Bortala Valley, agriculture is relatively limited. Going back into prehistory, the Eurasian Bronze Age is widely characterised as being based on a pastoral economy, but little work has been carried out on the precise nature of that economy beyond identification of the animals and plants that supported it. In particular, few studies have been undertaken on the nature of seasonal movement in relation to economic management, a notable exception being the work of Frachetti [2] in the Semirech’ye region of eastern Kazakhstan. The aim of this paper is to reconstruct models of seasonal movement for Bronze Age pastoralists in western Xinjiang. The archaeological evidence, while extensive, is insufficient to determine the structure of seasonal movements with any degree of precision, nor is the reasoning behind choices for camp sites clearly identifiable. To model these for the archaeological record, this study uses data drawn from ethnographic studies and analysis of modern snow and grass cover. Fieldwork permission (No. 350) was issued by the Chinese State Bureau of Relics to D. Cong (3196027). 2. The Bronze Age in Xinjiang The first Bronze Age peoples appeared in Xinjiang in the 3rd millennium BCE, mobile pastoralists superseding, and probably blending with, the last communities practicing age-old lifestyles of hunting, fishing and foraging who had dominated the wider region since the Palaeolithic. Bronze Age groups first appeared in the northern part of the Inner Asian Mountain Corridor bordering western Xinjiang when the eastward spread of Yamnaya peoples from beyond the Ural Mountains created the rise of a local Early Bronze Age culture, the Afanasievo, in the Altai Mountains. Afanasievo people in the Altai-Sayan region are genetically indistinguishable from Yamnaya populations, confirming an eastward expansion across the steppe [3]. Afanasievo groups are also linked to the oldest known evidence for dairy consumption in the eastern Eurasian steppe, suggesting that human migration into the Altai was associated with the introduction of domestic livestock [4]. Although a few Afanasievo burials have recently been reported in western Xinjiang [5–7], the first strong evidence for Bronze Age groups there appears with the Qiemu’erqieke (Chemurchek, Khemtseg) population in the north of Xinjiang [8] and Xiaohe/Gumugou to the south [9]. The Qiemu’erqieke culture can be dated broadly from around the mid-3rd millennium BCE, with the tradition as a whole lasting probably up until c. 1700 BCE [10]. Xiaohe/Gumugou dates also from the mid-3rd millennium to the mid-2nd millennium BCE [9]. Qiemu’erqieke and Xiaohe/Gumugou emerged out of Altaic/Mongolian and east Eurasian populations [11–13], while a third early group, the Tianshanbeilu culture, appearing in the oasis of Hami at the eastern end of the Tianshan, had its ancestry to the east in early oasis farming populations in Gansu and the Hexi Corridor [10, 14]. By the early 2nd millennium BCE, new Eurasian agro-pastoralists moved in from the west, occupying some of the same lands as the southerly expansion of Qiemu’erqieke peoples. They spread into the western hills and mountains of Xinjiang, along the Tianshan towards the east, and south into the Pamirs, close to the western end of the Tibetan Plateau. This group shared broad affinity with the loosely defined Andronovo complex that appeared widely across Eurasia in the later Bronze Age [15–17], more specifically the eastern Federovo variant [18]. A long-term project to conduct a detailed study of Andronovo groups in Xinjiang is in progress under the leadership of a team from the Institute of Archaeology of the Chinese Academy of Social Science (CASS) and the University of Sydney, Australia. The team has been working in Wenquan County, in the Bortala River area, in the far west of Xinjiang [18, 19], particularly at the occupation site and associated burial grounds of Adunqiaolu. The Upper Bortala River, also known as the Wenquan (Hot-Springs) River, originates between the Alatao and Biezhentao Mountains in the western Tianshan, flowing eastward through Wenquan (Hot Springs) County (Fig 1). It then joins the Daheyanzi River and runs out into Aibi (Ebinur) Lake. In the upper reaches of the Bortala River, the river bed lies close to the foothills of the Biezhentao Mountains on the south bank and on the north bank is separated from the Alatao Mountains by a wide band of gently sloping alluvial fans at an elevation of about 2000 m a.s.l. Behind the alluvial fans are low hills, backed in turn by the mountains. The alluvial fans form important areas of open pasture in the semi-arid steppe, drained by a dense network of seasonal streams. The east-west orientation of the valley provides the north bank with maximum exposure to seasonal sunlight. Altitude ranges from c. 1000 in the riverbed to c. 4000 m a.s.l. in the mountains. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. A general view of the Bortala valley looking to the northwest from the town of Wenquan (A. Betts). https://doi.org/10.1371/journal.pone.0240739.g001 Located far from any oceans, the region has a temperate, semi-arid, continental climate. The annual average temperature is around 3.6°C. Temperature extremes range from a maximum of 37.2°C to a minimum of -35.9°C [20]. Generally, the lowest temperature of the year occurs in January, with an average of -15°C, while in July the mean temperature is around 23°C [21]. Rainfall is low with an annual average precipitation rate of c. 200mm, mainly concentrated from May to July, when the region experiences about 60% of the annual total. The other months are much dryer, with monthly precipitation between November and March at usually less than 10mm. The vegetation types in Wenquan county are determined largely by the region’s distinct zones of elevation [22]. In the mountain foothills steppe vegetation is the most common, composed mainly of stipa grasses, including S. caucasica, S. grandis, S. orientalis and S. sareptana. Plants including sagebrush (Artemisia frigida), fetusca grasses (F. ovina, F. rupicola), and Seriphidium gracilescens are also commonly included in the steppe community. Areas adjacent to and above the steppes are covered with meadow vegetation, with the dominant plants represented by artemesia (Artemisia subulate), sedge (Carex stenocarpa) and annual meadowgrass (Poa annua). Alpine vegetation composed of cushion-like rock jasmine (Androsace) dominates on the higher slopes. Forests of Asian spruce (Picea schrenkiana) alternate with steppes and meadows, principally on the northern sides of the mountains. Around the Adunqiaolu site, stipa steppe is found widely and constitutes the main landscape. Spruce forest is rare, while in the river valley bottoms poplar (Populus) and willow (Salix) are common. As is generally the case with prehistoric pastoralist populations, Bronze Age remains include many cemeteries, but in the Bortala Valley there is also clear evidence of settlement through the presence of stone house footings of distinctive plan. Bronze Age stone-based habitation sites might be far more widespread than previously assumed. This type of monument extends into northern Xinjiang [23, 24], Kazakhstan [25], the Russian Altai and Tuva Republic, although there is an ongoing debate whether smaller versions are connected to seasonal habitation [26] or should rather be interpreted within a ritualistic framework [27, 28]. Archaeologically, investment in construction of stone houses is often taken to indicate permanent settlement, but this is clearly not the case in the Bortala Valley. A body of evidence shows that the houses were used on a rotating seasonal basis; investment in the construction of permanent buildings was worthwhile because the houses were used regularly over many years. This can be proven in various ways, one of the most valuable of which is the recent ethnographic record. Despite the fact that the modern pastoral populations in the region have tents, they construct stone or timber houses which they occupy seasonally, making use of their tents only in the warmer months when they move more widely in search of good pasture. As is the case today, the Andronovo people appear to have constructed houses in differing locations according to the cycles of the seasonal round: for winter, for summer and for the various mid-seasons at either end of the year. Bronze Age houses have been found at high altitudes around 3000 m a.s.l. near the summer campsites of local herders in locations that would be uninhabitable in winter due to deep snow cover and sub-zero temperatures (2016 test pit excavation, unpublished database). Large residential structures have been identified on mountain slopes at the relatively high altitude of 2300 m a.s.l. next to the winter campsites of the modern herding community at Adunqiaolu [18, 29]. As part of the work of the CASS and University of Sydney team in the Bortala Valley, several field seasons have been devoted to study of recent camp sites and herding practices in order to gain a deeper understanding of Bronze Age land use patterns and economic decision making. 3. Seasonal movement of modern pastoralists 3.1. Seasonal movement Traditional subsistence economies in the valley comprise seasonal transhumant pastoralism, combined with limited cultivation. The modern vertical transhumance pattern is driven by the seasonally oscillating availability of pastureland resources at different altitudes. Transhumance usually includes four major movements correlated with changes through the seasons. There is an annual migration between three seasonal campsites; the local term for the traditional pattern of animal management is “four seasons and three locations”. The herders move their animals following the four seasons, spring, summer, autumn and winter, to provide animals with the best grazing. “Three locations” means that the herders use only three major campsites within three pasturelands over the four seasons, using the same campsite for both spring and autumn. Each campsite is constructed differently, according to the specific needs of the season or seasons in which it is used (Fig 2). The herders move to the spring camps in mid to late March to prepare for lambing. The spring and autumn camps are located in the same place, usually on an open flat low-lying area at around 900–1200 m a.s.l. In these milder seasons the location of adequate grazing takes priority over access to water which may be some distance away. In the low-lying areas the spring grass appears earlier than elsewhere and lasts longer into the autumn. The spring/autumn camp is also near arable land suitable for growing crops or hay, and most herders have a block of land which is used primarily for the cultivation of fodder crops to supplement winter grazing. In early May the move is made to the summer camp when high altitude pastures first start growing. Here the animals are kept until late September. The summer pastures generally lie around the tree line at 3200 m a.s.l. In late September the herds are brought down again to the lowlands to the autumn/spring camp. At this low-lying camp the grazing is available as late as possible into the cold season. In early December the last move of the year is made to the winter camp. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Differences in seasonal camps in the Wenquan area. https://doi.org/10.1371/journal.pone.0240739.g002 Variations on this type of seasonal pastoral transhumance can be found throughout the Eurasian steppe, in Central Asia, Iran and in the Near East [30–33]. The choice of location and number of seasonal camp sites may differ according to local climate and landscape. Cribb [33, 34] describes a similar pattern of three seasonal camps used over four seasons among western Iranian herders in Luristan. In highland Asia, herders on the Tibetan plateau and some pastoralists along the Himalayan rim may only have two seasonal campsites, for summer and winter respectively [34]. In the Altai Mountains, Kazakh herders may have more than three locations, depending on local conditions [35]. Over many generations of herding practice, local people in the Bortala Valley have developed a deep understanding of how best to use the landscape through the changing seasons (Figs 3–5). For the herders of Wenquan, the best winter pastures are in mid-altitude locations at Adunqiaolu and Kazan, while the summer pastures are located at high elevations in Husta, Harnur, Anji and around Sarimu Lake. Autumn/spring pastures are low-lying, near the river. The location of each seasonal camp is selected according to specific conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Transhumant routes among the local herders in Wenquan. Map created using ArcGIS Pro 2.5.0. Single use licence issued to Meng Qi by ESRI Australia Pty Ltd (Environmental Systems Research Institute). We acknowledge the use of Esri base maps as noted in the image. https://doi.org/10.1371/journal.pone.0240739.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Kazak herders in the Yili Valley migrating to their winter camp at Kazan in Wenquan in mid- November 2016 (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g004 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Transhumance from the summer pastures at Harnur to the autumn camp in the river valley in early September 2018 (D. Cong). https://doi.org/10.1371/journal.pone.0240739.g005 Given the extremely cold climate during the winter within high latitude regions such as the Tianshan, the most important issue for local herders is to find a good location to stay during the winter. Among pastoralist societies, winter is a particularly difficult period for both humans and animals, not only due to the cold climate but also because there is a higher risk of natural disasters. Two major winter threats (dzud) [36, 37] are exceptionally heavy snowfall called the “white disaster”, and drought caused by very little snowfall, called the “black disaster” by herders all over the Eurasian steppe [30–33]. Heavy snowfall can cause severe problems for herders since thick snow cover can prevent the animals from accessing natural grazing. If there is insufficient fodder in storage, the animals may die in large numbers. Heavy snowfall can also cut off all road connections, leaving the herders and animals in the winter pastures isolated from major population centres. If the animals become weak and sick the herders have no access to veterinary treatment and medications. By contrast, in the case of a “black disaster”, there may be insufficient snow to provide water for both humans and animals. Over hundreds of years of experience, local herders have selected a range of different areas as optimal locations for winter pasturelands. Winter camps need to be placed in a south-facing spot sheltered by a small hill, or in a south-facing depression in a mountain valley. The shelter, and south-facing aspect exposed to maximum winter sunlight, limit the amount of snow lying on the pastureland. Sheep need less than 15 cm depth of snow accumulation to access grass under the snow. However, some snow is required so that it can be collected to provide a water supply for both humans and animals. These winter pasturelands could also be used to graze animals in other seasons as well, but this is not always the case. For instance, the grassland at Adunqiaolu about 2000 m a.s.l. is a favoured winter pastureland and although there is also good grass there in summer the local herding community conserves this by a ban on herding in this area except in winter. The Adunqiaolu pastureland is optimally suited for winter camps (Fig 6). The strong wind along the valley prevents the snow from drifting and it generally lies less than 15 cm deep, while the entire area faces south towards the sun. There are also small hills sporadically distributed across the slope, forming sheltered spots and depressions suitable for winter camps. The area has long been used in winter and today more than a hundred camps are located in this area, making it one of the best winter pasturelands in Wenquan County. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Modern winter camp at Adunqiaolu with little snow cover on the adjacent pasture. The camp faces south towards the sun with a small hill behind (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g006 For the summer camp there is a quite different set of criteria. The site must have an open flat area to build the camp and a sheep fold. It should be cool during summer, near a fresh water supply such as a spring, a stream or a river, and there needs to be a good supply of high-quality grazing. Glacial meltwater in summer contributes to increased water flow in the rivers and streams and to the appearance of numerous springs at higher elevations [25]. In the upper Bortala Valley, these conditions for summer pastureland are usually found up in the mountain areas, up to 2600–3000 m a.s.l. in places such as Harnur, Angji and the grassland surrounding Sarimu Lake (Fig 7). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Wenquan herder Bangba’s summer camp at Big Angji, a high-altitude summer pastureland (2700 m a.s.l.) (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g007 For spring and autumn, the Bortala valley pastoralists generally select a single location for both camps (Figs 8 and 9). These campsites are usually down in the river valley at a lower altitude than the winter and summer camps so that spring comes earlier, and autumn lasts longer than in the other areas. Autumn and spring are also key periods for animal reproduction, in order to time lambing to coincide with spring. Partitions are built within the sheepfolds to isolate the rams with the mature ewes in autumn and to separate out the pregnant ewes, lactating ewes and young lambs during the spring. Otherwise, rams are kept isolated from the rest of the flock for most of the year except during the winter after the ewes are already pregnant. The flat land near the river may also be used for fodder production (hay) for winter. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Traditional spring/autumn campsite for local herders. The site was abandoned after the herders moved to a newly constructed government funded house with an animal fold in a permanently settled village (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g008 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 9. Intermediate campsite used during the spring–summer transhumance (1800 m. a.s.l.). The circular dark patch of soil surrounded by a fence is the animal fold. The small log-built hut sees only seasonal use for the herders to stay (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g009 However, not all local herding patterns follow the “four seasons and three locations” since even within the small section of valley around Wenquan County area pasture and climate are highly diversified. For some areas such as the Husta grassland which stands at a relative low altitude (1600–1800 m a.s.l.) on the southern slope of the Alatao mountains with a moderate climate and good pasture from spring through summer into autumn [38], herders need to move only twice, once in spring up to the Husta pastureland and again in autumn to winter camps at Adunqiaolu. By contrast, some herders using relatively high altitude (2900 m a.s.l. or above) summer pastureland, such as Bangba’s (a local Mongolian herder) summer pastures, Big and Small Angji on the high slopes of the Tianshan (Fig 7), may need several short-stay intermediate campsites and pasturelands between the spring and summer pastures. Speed of movement is dictated by increasing temperatures and growth of grass during the spring. Exploiting several short-stay pastures one by one, the herders can move slowly up to the high pastures at Angji by the time full summer arrives. Such intermediate campsites are also used during the summer-autumn transhumance. When the temperature starts to fall on the high mountain pastureland in autumn, Bangba moves to the lower campsite (Fig 9), and then on again to the autumn camp when the temperature becomes too cold for the animals. 3.2 Ethnographic analogies The application of ethnographic analogy to archaeological data is often perceived as problematic [e.g. 39, 40] and merits some discussion in relation to this study. In this case, there are two parts to the analogy which should be considered separately. Firstly, there are some basic constants. The landscape of the Bortala Valley has not changed substantially between the Bronze Age and the present day. Latitude and altitude remain the same and the sun still stands in the south. The needs of the domestic animals, sheep, goat, cattle and horses, while making allowances for modern breeds, remain overall very similar. There are climatic differences, as discussed below, but they are not so extensive as to radically change the types of economic options available to subsistence occupants of the valley. The second part of the analogy concerns socio-cultural patterning. Modern day herders operate in the context of an urban based market economy and within a centralised state. Bronze Age agro-pastoralists functioned as extended familial or sub-tribal groups within an unconstrained landscape. These differences influence decision making, particularly in the spring and autumn camps. The spring and autumn camps are primarily selected on the basis of early and late grazing, but today proximity to roads, transport and markets is also a factor. Access is a factor in all modern seasonal choices. Where in the Bronze Age support in times of difficulty would only have been available from near neighbours in local pastures, today regular contact is needed at all times of year between the town and the camps. Critics of ethnographic analogies also recognise these two sides to the problem. A key issue is that cultures may diverge sharply in their responses to given economic-ecological constraints [48]. Hawkes [41] suggests that the further removed the subject of the analogy is from physical, natural constraints, the greater the ‘human’ element and thus the greater the probability of culturally diverse practices. However, we argue that in this case the degree of removal is slight. The approach taken in this paper is supported by Gould [42] who argues that if archaeologists are able to identify physical and biological limiting factors imposing essentially invariant constraints on human behaviour, it is then possible to formulate hypotheses about the behaviour of past populations under the same or very closely similar conditions. Wylie [39] notes that this relates to a very narrow set of circumstances, “in those rare limiting cases where the reconstructed behaviour is, by nature, a direct and exclusive consequence of impinging ecological or material conditions”, but she suggests that in these cases the inferences may be raised “to the level of deductive security”. Gould too agrees that such instances are rare, such as “where a particularly restrictive natural environment limits the options for survival”. We argue that the choice of seasonal encampments is, and was in the Bronze Age, governed by very specific ecological and environmental constraints which limit the influence of cultural differences to a degree that modern ethnographic data may be used to construct hypotheses about Bronze Age seasonal patterns of movement. 3.1. Seasonal movement Traditional subsistence economies in the valley comprise seasonal transhumant pastoralism, combined with limited cultivation. The modern vertical transhumance pattern is driven by the seasonally oscillating availability of pastureland resources at different altitudes. Transhumance usually includes four major movements correlated with changes through the seasons. There is an annual migration between three seasonal campsites; the local term for the traditional pattern of animal management is “four seasons and three locations”. The herders move their animals following the four seasons, spring, summer, autumn and winter, to provide animals with the best grazing. “Three locations” means that the herders use only three major campsites within three pasturelands over the four seasons, using the same campsite for both spring and autumn. Each campsite is constructed differently, according to the specific needs of the season or seasons in which it is used (Fig 2). The herders move to the spring camps in mid to late March to prepare for lambing. The spring and autumn camps are located in the same place, usually on an open flat low-lying area at around 900–1200 m a.s.l. In these milder seasons the location of adequate grazing takes priority over access to water which may be some distance away. In the low-lying areas the spring grass appears earlier than elsewhere and lasts longer into the autumn. The spring/autumn camp is also near arable land suitable for growing crops or hay, and most herders have a block of land which is used primarily for the cultivation of fodder crops to supplement winter grazing. In early May the move is made to the summer camp when high altitude pastures first start growing. Here the animals are kept until late September. The summer pastures generally lie around the tree line at 3200 m a.s.l. In late September the herds are brought down again to the lowlands to the autumn/spring camp. At this low-lying camp the grazing is available as late as possible into the cold season. In early December the last move of the year is made to the winter camp. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. Differences in seasonal camps in the Wenquan area. https://doi.org/10.1371/journal.pone.0240739.g002 Variations on this type of seasonal pastoral transhumance can be found throughout the Eurasian steppe, in Central Asia, Iran and in the Near East [30–33]. The choice of location and number of seasonal camp sites may differ according to local climate and landscape. Cribb [33, 34] describes a similar pattern of three seasonal camps used over four seasons among western Iranian herders in Luristan. In highland Asia, herders on the Tibetan plateau and some pastoralists along the Himalayan rim may only have two seasonal campsites, for summer and winter respectively [34]. In the Altai Mountains, Kazakh herders may have more than three locations, depending on local conditions [35]. Over many generations of herding practice, local people in the Bortala Valley have developed a deep understanding of how best to use the landscape through the changing seasons (Figs 3–5). For the herders of Wenquan, the best winter pastures are in mid-altitude locations at Adunqiaolu and Kazan, while the summer pastures are located at high elevations in Husta, Harnur, Anji and around Sarimu Lake. Autumn/spring pastures are low-lying, near the river. The location of each seasonal camp is selected according to specific conditions. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Transhumant routes among the local herders in Wenquan. Map created using ArcGIS Pro 2.5.0. Single use licence issued to Meng Qi by ESRI Australia Pty Ltd (Environmental Systems Research Institute). We acknowledge the use of Esri base maps as noted in the image. https://doi.org/10.1371/journal.pone.0240739.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 4. Kazak herders in the Yili Valley migrating to their winter camp at Kazan in Wenquan in mid- November 2016 (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g004 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 5. Transhumance from the summer pastures at Harnur to the autumn camp in the river valley in early September 2018 (D. Cong). https://doi.org/10.1371/journal.pone.0240739.g005 Given the extremely cold climate during the winter within high latitude regions such as the Tianshan, the most important issue for local herders is to find a good location to stay during the winter. Among pastoralist societies, winter is a particularly difficult period for both humans and animals, not only due to the cold climate but also because there is a higher risk of natural disasters. Two major winter threats (dzud) [36, 37] are exceptionally heavy snowfall called the “white disaster”, and drought caused by very little snowfall, called the “black disaster” by herders all over the Eurasian steppe [30–33]. Heavy snowfall can cause severe problems for herders since thick snow cover can prevent the animals from accessing natural grazing. If there is insufficient fodder in storage, the animals may die in large numbers. Heavy snowfall can also cut off all road connections, leaving the herders and animals in the winter pastures isolated from major population centres. If the animals become weak and sick the herders have no access to veterinary treatment and medications. By contrast, in the case of a “black disaster”, there may be insufficient snow to provide water for both humans and animals. Over hundreds of years of experience, local herders have selected a range of different areas as optimal locations for winter pasturelands. Winter camps need to be placed in a south-facing spot sheltered by a small hill, or in a south-facing depression in a mountain valley. The shelter, and south-facing aspect exposed to maximum winter sunlight, limit the amount of snow lying on the pastureland. Sheep need less than 15 cm depth of snow accumulation to access grass under the snow. However, some snow is required so that it can be collected to provide a water supply for both humans and animals. These winter pasturelands could also be used to graze animals in other seasons as well, but this is not always the case. For instance, the grassland at Adunqiaolu about 2000 m a.s.l. is a favoured winter pastureland and although there is also good grass there in summer the local herding community conserves this by a ban on herding in this area except in winter. The Adunqiaolu pastureland is optimally suited for winter camps (Fig 6). The strong wind along the valley prevents the snow from drifting and it generally lies less than 15 cm deep, while the entire area faces south towards the sun. There are also small hills sporadically distributed across the slope, forming sheltered spots and depressions suitable for winter camps. The area has long been used in winter and today more than a hundred camps are located in this area, making it one of the best winter pasturelands in Wenquan County. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 6. Modern winter camp at Adunqiaolu with little snow cover on the adjacent pasture. The camp faces south towards the sun with a small hill behind (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g006 For the summer camp there is a quite different set of criteria. The site must have an open flat area to build the camp and a sheep fold. It should be cool during summer, near a fresh water supply such as a spring, a stream or a river, and there needs to be a good supply of high-quality grazing. Glacial meltwater in summer contributes to increased water flow in the rivers and streams and to the appearance of numerous springs at higher elevations [25]. In the upper Bortala Valley, these conditions for summer pastureland are usually found up in the mountain areas, up to 2600–3000 m a.s.l. in places such as Harnur, Angji and the grassland surrounding Sarimu Lake (Fig 7). Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 7. Wenquan herder Bangba’s summer camp at Big Angji, a high-altitude summer pastureland (2700 m a.s.l.) (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g007 For spring and autumn, the Bortala valley pastoralists generally select a single location for both camps (Figs 8 and 9). These campsites are usually down in the river valley at a lower altitude than the winter and summer camps so that spring comes earlier, and autumn lasts longer than in the other areas. Autumn and spring are also key periods for animal reproduction, in order to time lambing to coincide with spring. Partitions are built within the sheepfolds to isolate the rams with the mature ewes in autumn and to separate out the pregnant ewes, lactating ewes and young lambs during the spring. Otherwise, rams are kept isolated from the rest of the flock for most of the year except during the winter after the ewes are already pregnant. The flat land near the river may also be used for fodder production (hay) for winter. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 8. Traditional spring/autumn campsite for local herders. The site was abandoned after the herders moved to a newly constructed government funded house with an animal fold in a permanently settled village (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g008 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 9. Intermediate campsite used during the spring–summer transhumance (1800 m. a.s.l.). The circular dark patch of soil surrounded by a fence is the animal fold. The small log-built hut sees only seasonal use for the herders to stay (P. Jia). https://doi.org/10.1371/journal.pone.0240739.g009 However, not all local herding patterns follow the “four seasons and three locations” since even within the small section of valley around Wenquan County area pasture and climate are highly diversified. For some areas such as the Husta grassland which stands at a relative low altitude (1600–1800 m a.s.l.) on the southern slope of the Alatao mountains with a moderate climate and good pasture from spring through summer into autumn [38], herders need to move only twice, once in spring up to the Husta pastureland and again in autumn to winter camps at Adunqiaolu. By contrast, some herders using relatively high altitude (2900 m a.s.l. or above) summer pastureland, such as Bangba’s (a local Mongolian herder) summer pastures, Big and Small Angji on the high slopes of the Tianshan (Fig 7), may need several short-stay intermediate campsites and pasturelands between the spring and summer pastures. Speed of movement is dictated by increasing temperatures and growth of grass during the spring. Exploiting several short-stay pastures one by one, the herders can move slowly up to the high pastures at Angji by the time full summer arrives. Such intermediate campsites are also used during the summer-autumn transhumance. When the temperature starts to fall on the high mountain pastureland in autumn, Bangba moves to the lower campsite (Fig 9), and then on again to the autumn camp when the temperature becomes too cold for the animals. 3.2 Ethnographic analogies The application of ethnographic analogy to archaeological data is often perceived as problematic [e.g. 39, 40] and merits some discussion in relation to this study. In this case, there are two parts to the analogy which should be considered separately. Firstly, there are some basic constants. The landscape of the Bortala Valley has not changed substantially between the Bronze Age and the present day. Latitude and altitude remain the same and the sun still stands in the south. The needs of the domestic animals, sheep, goat, cattle and horses, while making allowances for modern breeds, remain overall very similar. There are climatic differences, as discussed below, but they are not so extensive as to radically change the types of economic options available to subsistence occupants of the valley. The second part of the analogy concerns socio-cultural patterning. Modern day herders operate in the context of an urban based market economy and within a centralised state. Bronze Age agro-pastoralists functioned as extended familial or sub-tribal groups within an unconstrained landscape. These differences influence decision making, particularly in the spring and autumn camps. The spring and autumn camps are primarily selected on the basis of early and late grazing, but today proximity to roads, transport and markets is also a factor. Access is a factor in all modern seasonal choices. Where in the Bronze Age support in times of difficulty would only have been available from near neighbours in local pastures, today regular contact is needed at all times of year between the town and the camps. Critics of ethnographic analogies also recognise these two sides to the problem. A key issue is that cultures may diverge sharply in their responses to given economic-ecological constraints [48]. Hawkes [41] suggests that the further removed the subject of the analogy is from physical, natural constraints, the greater the ‘human’ element and thus the greater the probability of culturally diverse practices. However, we argue that in this case the degree of removal is slight. The approach taken in this paper is supported by Gould [42] who argues that if archaeologists are able to identify physical and biological limiting factors imposing essentially invariant constraints on human behaviour, it is then possible to formulate hypotheses about the behaviour of past populations under the same or very closely similar conditions. Wylie [39] notes that this relates to a very narrow set of circumstances, “in those rare limiting cases where the reconstructed behaviour is, by nature, a direct and exclusive consequence of impinging ecological or material conditions”, but she suggests that in these cases the inferences may be raised “to the level of deductive security”. Gould too agrees that such instances are rare, such as “where a particularly restrictive natural environment limits the options for survival”. We argue that the choice of seasonal encampments is, and was in the Bronze Age, governed by very specific ecological and environmental constraints which limit the influence of cultural differences to a degree that modern ethnographic data may be used to construct hypotheses about Bronze Age seasonal patterns of movement. 4. Methodology and results The ethnographic study clearly shows that in understanding transhumance patterns in western Xinjiang, elevation alone is too simplistic as a proxy for seasonal patterning. Selection of seasonal locations is defined by the accessibility, quality and availability of grazing, but not necessarily all together. Winter accessibility is intricately linked to snow cover which sees major variations throughout the years. Pasture quality is related to the available water supply for a specific range land. Considering these as the main environmental parameters which define the ideal usage pattern of a given range land, we are developing a model to determine pastoralists’ choices for seasonal movements. Like every model this is a radical simplification of a complex human-environmental interaction, but we are trying to capture the essence of what makes a specific location preferable to others with regard to a localized subsistence economy. We analyze the dynamic parameters of pasture quality and snow cover throughout the year. It is clear that this approach omits for example interactions between different human groups, such as access limitations, traditionalized land tenure, or special treatment of parts of the landscape connected to grazing bans during certain times of the year. The limitations and simplicity, however, have the advantage that we can validate the model with the ethnographic accounts provided above and identify mismatches and problems for which additional explanations are necessary. We want to test whether snow cover and pasture quality are factors that allow us to determine the ideal use of a particular range land and validate the outcome based on the ethnographic data. 4.1 Study area spatial limits and boundary definitions The analysis was conducted covering the entire Bortala Valley. In order to distinguish meaningful grazing ground entities, we applied the designations made by local pastoral nomads living in the area. Using raster data covering the entire valley, we would be able to define the suitability of a grazing area on a pixel by pixel basis (depending on the resolution of the data this is 500m x 500m for MOD10A2 and 10m x 10m for Sentinel2 red and NIR bands). However, we choose to base our evaluation on the coarser, but more meaningful ethnographic usage patterns. This allows for better delineation of the grazing grounds and a direct comparability as to whether the applied approach is able to reflect the current land use patterns in the area. The analyses were conducted for the following pasture areas: Adunqiaolu winter pasture, the core area of Adunqiaolu winter pasture, the Kazan winter pasture, the Wenquan spring/autumn pasture, the Husta spring/summer/autumn pasture, the core area of the Husta pasture, the eastern and western part of the Harnur summer pasture, the north-western Adunqiaolu summer pasture, the core area of the north-western Adunqiaolu summer pasture, and the Sarimu Lake summer pasture (Fig 10). The delineation of these pasture areas is based on the ethnographic survey described above. Local herders mentioned these areas of usage to P. Jia, who then mapped the extent of the range lands by the key points derived from the handheld GPS. We analysed three areas which were designated as winter pastures through the ethnographic survey: The Adunqiaolu winter pasture, its core area, and the Kazan winter pasture. Note that the core area of the Adunqiaolu winter pasture is fully included in the larger Adunqiaolu winter pasture and therefore highly correlated with the latter. Three areas which were considered to be fit for grazing in spring and autumn were analysed: the Wenquan spring/autumn pasture, the Husta spring/summer/autumn pasture, and its core area. Five designated summer pastures were analysed: Harnur East summer pasture, Harnur West summer pasture, Adunqiaolu Northwest summer pasture and its core area, and the Sarimu Lake summer pasture. Again, Adunqiaolu and its core area are ethnographically distinguished, and the former encompasses the latter which leads to a high degree of correlation. Harnur East and Harnur West are connected. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 10. Location and area of seasonal pastureland in Wenquan: 1. Kazan Winter pasture 2, Adunqiaolu northwest summer pasture, 3. Adunqiaolu northwest core area of summer pasture, 4. Adunqiaolu winter pasture, 5. Core area Adunqiaolu winter pasture, 6. Wenquan spring-autumn pasture, 7. Entire Husta Grassland -summer-spring–autumn, 8. The Centre of Husta-summer-spring-autumn, 9. Harnur summer 1, 10. Harnur summer pasture 2, 11. Xiaowenquan Summer pasture, 12. Sarimu Lake summer pasture. Image created using ArcGIS Pro 2.5.0. Single use licence issued to Meng Qi by ESRI Australia Pty Ltd (Environmental Systems Research Institute). We acknowledge the use of Esri base maps as noted in the image. https://doi.org/10.1371/journal.pone.0240739.g010 4.2 NDVI processing The Normalized Difference Vegetation Index (NDVI) is one of the most important methods for vegetation monitoring since Rouse developed it in 1972 [43]. This method is mainly dependent on the difference between healthy green vegetation and other land surface reflectance (including unhealthy/dry vegetation) in both red and infrared wavelengths [44]. The underlying principle of the NDVI is that healthy green vegetation reflects more infrared radiation and absorbs more energy in the red wavelength when compared with damaged vegetation or non-vegetated surfaces [44]. The NDVI value is defined by the ratio of: (1) where NIR is the near-infrared wavelength reflectance, and RED is the reflectance in the red wavelength. The resulting values are ranging from −1 to 1, where −1 indicates no presence of vegetation at all, and 1 indicates dense levels of healthy vegetation [44]. In this model, 161 Sentinel-2 images acquired between December 2016 and February 2018 [45], were utilized to calculate NDVI rasters for the Bortala Valley, Xinjiang, China. 96 images were acquired by Sentinel-2A (S2-A) sensor between December 2016 and February 2018, and sixty-five images were acquired by Sentinel-2B (S2-B) between July 2017 and February 2018. After each image is radiometrically calibrated, the NDVI is processed. Then, a stack of NDVI results is created for each month between December 2016 and February 2018 as presented in (Table 1). We created a mosaic for each month based on the highest NDVI value among all stack images. The selection of the maximum value was chosen in order to overcome influences of cloud cover and haze problems in data acquisition. This resulted in monthly maps displaying the highest NDVI value for each pixel throughout the month. By overlaying seasonal pasture areas, the average of NDVI values within each pasture was retrieved for further analysis. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Sentinel2 and MODIS data inventory. https://doi.org/10.1371/journal.pone.0240739.t001 4.3 Snow cover analysis This study used Moderate Resolution Imaging Spectroradiometer (MODIS/Terra) Snow Cover data (MOD10A2) [46] for snow cover analysis during the study period. These data provide the maximum extent of snow and ice coverage within eight days [47] and use a snow-conservative approach for snow detection based on reflectance features. Data are then screened for false snow detections [48]. The MODIS snow cover data are mainly based on the Normalized Difference Snow Index (NDSI) snow-mapping algorithm. NDSI is based on the high reflectivity of snow in the visible part of the spectrum (VIS) and low reflectance in the shortwave infrared spectrum (SWIR). We can create a normalized difference index for snow similar to the NDVI [49], based on the ratio in equation number 2: (2) In MOD10A2 data, snow cover was mapped by select MODIS radiance data of band 4 for VIS, and band 6 for SWIR. Snow presents a value of >0.0 in the NDSI. However, not all surfaces within these characteristics are snow which can result in a snow commission error. Especially within cloud fringes similar values can be encountered. Hence, these data must go through a process of snow commission errors alleviation. Several data screens were applied based on the spectral features of snow and other characteristics to flag uncertain snow cover detection and reverse it [48]. The MOD10A2 consists of two products; the first displays the mapped maximum snow extent over a period of eight days. The second displays the chronology of snow occurrence observations during the data acquisition period [47]. Here, we used the second product that displays snow continuity during an eight-day period as a chronology of observed snow cover. In this product, cloud cover was only included if there was persistent cloud cover on all eight days. If this was the case, then cloud cover was reported for a grid cell and the cell was not included in the analysis [48]. Table 1 presents the number of MOD10A2 data products used for each month during the study period. We classified monthly snow cover in each pasture to thin, moderate, and thick based on the average snow cover. If an area was covered in all eight-day datasets for a specific month it was considered thick coverage. When it was covered in only one out of three or four datasets, it was considered thin coverage. If the value fell between the two, we assigned moderate coverage. Finally, the percentage of each category including no snow cover was calculated to describe relative snow coverage over pastures surfaces for each month. 4.4 Results Below we go through each area designated by local pastoralists as having a specific usage, briefly elaborating on the vegetation growth cycles and the snow fall patterns as well as any issues with our simplified modelling approach. We group these areas into the seasonal movement stages they belong to in order to see whether they do show commonalities or disparities. 4.4.1 Winter grazing grounds. As expected, the main requirement for a winter pasture is limited snow cover during the cold months of the year, so animals can access food. In all three areas we see little snow in the months of December through March. For Adunqiaolu there is some thicker snow cover in February 2017, but then the next year sees almost no snow in the same month. Kazan shows snow in December of 2016 but is completely free of snow in December 2017. Crucially, there is no month during the year which sees complete thick or moderate snow cover. Some parts of the pasture always remain open or are at most temporarily covered with a thin layer of snow. Clearly the significance of observations made during two consecutive years is limited, but neither of the years was considered extreme by the locals. Neither black nor white disasters were recorded for the time period the underlying data of the model date to (December 2016 –February 2018). Hence, the time period covered can be assumed to lie within the range of normal years where no shocks to the pastoralist system and therefore no changes in herding patterns are observed. The limited snow cover during the winter months is connected to the low quality of the pasture in the following months. Even during the peak of the vegetation cycle in July the average NDVI value barely exceeds 0.3 for the Adunqiaolu winter pasture and its core area and 0.25 for the Kazan winter pasture. The grazing grounds dry off quickly in summer and the NDVI values decline. It is important to note that the NDVI values for barren areas, rock, and snow lie around 0.1 to -0.1. Considerable snow cover therefore impacts the average NDVI values and these averages are not directly connected to the quality of the grazing during the winter months. Slightly higher grazing qualities can be expected since snow cover biases the NDVI of a pasture area towards 0. Average NDVI values for winter pastures during the months of usage seem to hover around a value of 0.1. 4.4.2 Spring/autumn grazing grounds. Spring/autumn pastures are more difficult to identify, since they serve a transitional function and are usually only used during six to eight weeks from late March to early May and from late September to early December. In terms of snow cover the data showed a still considerable snow cover in March. With the beginning of April, the vegetation saw rapid development which is consistent with the first movement of pastoralists during the year. However, both Husta, including its core area, as well as the Wenquan spring/autumn pasture showed characteristics that need to be further elaborated upon since they demonstrate the limits of a pasture classification approach based on NDVI averages and snow cover. Wenquan, whereas consistent with the melting period and the first move of the year, showed very low NDVI averages over the entire year, even trailing the overgrazed and dried off winter pastures. The topographic location of the Wenquan spring/autumn pasture offered explanations beyond the simplistic assumptions of the model. The Wenquan pasture seemed to only provide meagre grazing throughout the entire year. The area lies to both sides of a river. Water has a very low NDVI value approaching -1 and could considerably influence an average. However, even when excluding the river and the denuded surfaces of the riverbed, the NDVI average stayed low. Therefore, other explanations needed to be considered. According to ethnographic data, the most important reason for the usage of Wenquan as a spring pasture was the early temperature increase which allows for a shortening of the winter camping time. These range lands provided a convenient location within the landscape as a transition area for heading up to the summer pastures as soon as snow started to melt there. It also had small areas of lush vegetation next to the riverbank which were too small in area size to influence the overall NDVI average much but provided ideal grazing for a limited amount of time. The abundant availability of water next to the river was convenient for herders and even allowed for limited agricultural activity including fodder crops and sometimes vegetables. Beyond environmental factors, Wenquan served as a place for exchange of animals and goods and stacking supplies for winter/summer periods. The Husta grazing lands are not only used in spring and autumn but also in summer. The reason for this lies in the favourable topographic location of the pasture. Its lower elevation starts around 1400 m a.s.l. and the highest areas lie at 2200 m a.s.l. This allows for slow movements inside the pasture area itself, depending on where the grazing is best. To the south, the Husta pasture is framed by a small ridge of around 1600–1900 m a.s.l. The north side is backed by the mountain range which defines the Bortala valley [38]. This creates a topographic situation where seasonal springs and small creeks provide water for the vegetation throughout the growing season from spring to autumn and create one of the best grazing grounds in the entire valley. 4.4.3 Summer grazing grounds. The summer pastures are characterized by late snow melt in the spring, usually in the month of April, with a good chance of considerable quantities of snow by October. Harnur East and West, as well as the summer pasture west of the Sarimu Lake, clearly fit this pattern. Harnur East, one of the highest lying summer pastures, shows consistent snow cover of between 70% and 100% from October through April, with most areas displaying thick or moderate snow cover. The Adunqiaolu Northwest summer pasture and its core area can experience full snow cover during the winter months (December 2016), but otherwise see relatively little snow during the study period. We think that the chance of a full moderate or thick snow cover excludes this area from safe usage during the winter, but there may be sociocultural and practical factors at play as well. As discussed earlier, less snow cover is not the only condition to be considered in defining a suitable winter pastureland. Winter camping grounds also require specific landforms such as a small hill or depression for shelter as well as easy external access. At an elevation of 2900 m a.s.l. the Adunqiaolu Northwest summer pasture is extremely cold in winter and the landforms offer no places for good shelter from the wind. The track leading up to the pastures is difficult, particularly in winter when it would be necessary to climb or descend steep, potentially snow-covered slopes. All summer pastures show a late but accelerated start to vegetation growth in the month of May. This fits with the second movement of the year in early May. The herds arrive on the summer pastures around the time when vegetation growth is spiking up from its low level in April. The average NDVI values for the summer pastures peaks in July with values around 0.7. These average values are extremely high for grassland and speak for the excellent quality of the grazing on these summer pastures. NDVI has been shown to correlate with productivity [50, 51]. The interrelations are certainly more complex, but as an approximation the summer pastures can be seen as roughly twice as productive as the winter pastures in our case. This has clear implications for the calorie intake of the animals over the course of the year and demonstrates why these seasonal movements are worthwhile. 4.1 Study area spatial limits and boundary definitions The analysis was conducted covering the entire Bortala Valley. In order to distinguish meaningful grazing ground entities, we applied the designations made by local pastoral nomads living in the area. Using raster data covering the entire valley, we would be able to define the suitability of a grazing area on a pixel by pixel basis (depending on the resolution of the data this is 500m x 500m for MOD10A2 and 10m x 10m for Sentinel2 red and NIR bands). However, we choose to base our evaluation on the coarser, but more meaningful ethnographic usage patterns. This allows for better delineation of the grazing grounds and a direct comparability as to whether the applied approach is able to reflect the current land use patterns in the area. The analyses were conducted for the following pasture areas: Adunqiaolu winter pasture, the core area of Adunqiaolu winter pasture, the Kazan winter pasture, the Wenquan spring/autumn pasture, the Husta spring/summer/autumn pasture, the core area of the Husta pasture, the eastern and western part of the Harnur summer pasture, the north-western Adunqiaolu summer pasture, the core area of the north-western Adunqiaolu summer pasture, and the Sarimu Lake summer pasture (Fig 10). The delineation of these pasture areas is based on the ethnographic survey described above. Local herders mentioned these areas of usage to P. Jia, who then mapped the extent of the range lands by the key points derived from the handheld GPS. We analysed three areas which were designated as winter pastures through the ethnographic survey: The Adunqiaolu winter pasture, its core area, and the Kazan winter pasture. Note that the core area of the Adunqiaolu winter pasture is fully included in the larger Adunqiaolu winter pasture and therefore highly correlated with the latter. Three areas which were considered to be fit for grazing in spring and autumn were analysed: the Wenquan spring/autumn pasture, the Husta spring/summer/autumn pasture, and its core area. Five designated summer pastures were analysed: Harnur East summer pasture, Harnur West summer pasture, Adunqiaolu Northwest summer pasture and its core area, and the Sarimu Lake summer pasture. Again, Adunqiaolu and its core area are ethnographically distinguished, and the former encompasses the latter which leads to a high degree of correlation. Harnur East and Harnur West are connected. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 10. Location and area of seasonal pastureland in Wenquan: 1. Kazan Winter pasture 2, Adunqiaolu northwest summer pasture, 3. Adunqiaolu northwest core area of summer pasture, 4. Adunqiaolu winter pasture, 5. Core area Adunqiaolu winter pasture, 6. Wenquan spring-autumn pasture, 7. Entire Husta Grassland -summer-spring–autumn, 8. The Centre of Husta-summer-spring-autumn, 9. Harnur summer 1, 10. Harnur summer pasture 2, 11. Xiaowenquan Summer pasture, 12. Sarimu Lake summer pasture. Image created using ArcGIS Pro 2.5.0. Single use licence issued to Meng Qi by ESRI Australia Pty Ltd (Environmental Systems Research Institute). We acknowledge the use of Esri base maps as noted in the image. https://doi.org/10.1371/journal.pone.0240739.g010 4.2 NDVI processing The Normalized Difference Vegetation Index (NDVI) is one of the most important methods for vegetation monitoring since Rouse developed it in 1972 [43]. This method is mainly dependent on the difference between healthy green vegetation and other land surface reflectance (including unhealthy/dry vegetation) in both red and infrared wavelengths [44]. The underlying principle of the NDVI is that healthy green vegetation reflects more infrared radiation and absorbs more energy in the red wavelength when compared with damaged vegetation or non-vegetated surfaces [44]. The NDVI value is defined by the ratio of: (1) where NIR is the near-infrared wavelength reflectance, and RED is the reflectance in the red wavelength. The resulting values are ranging from −1 to 1, where −1 indicates no presence of vegetation at all, and 1 indicates dense levels of healthy vegetation [44]. In this model, 161 Sentinel-2 images acquired between December 2016 and February 2018 [45], were utilized to calculate NDVI rasters for the Bortala Valley, Xinjiang, China. 96 images were acquired by Sentinel-2A (S2-A) sensor between December 2016 and February 2018, and sixty-five images were acquired by Sentinel-2B (S2-B) between July 2017 and February 2018. After each image is radiometrically calibrated, the NDVI is processed. Then, a stack of NDVI results is created for each month between December 2016 and February 2018 as presented in (Table 1). We created a mosaic for each month based on the highest NDVI value among all stack images. The selection of the maximum value was chosen in order to overcome influences of cloud cover and haze problems in data acquisition. This resulted in monthly maps displaying the highest NDVI value for each pixel throughout the month. By overlaying seasonal pasture areas, the average of NDVI values within each pasture was retrieved for further analysis. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Sentinel2 and MODIS data inventory. https://doi.org/10.1371/journal.pone.0240739.t001 4.3 Snow cover analysis This study used Moderate Resolution Imaging Spectroradiometer (MODIS/Terra) Snow Cover data (MOD10A2) [46] for snow cover analysis during the study period. These data provide the maximum extent of snow and ice coverage within eight days [47] and use a snow-conservative approach for snow detection based on reflectance features. Data are then screened for false snow detections [48]. The MODIS snow cover data are mainly based on the Normalized Difference Snow Index (NDSI) snow-mapping algorithm. NDSI is based on the high reflectivity of snow in the visible part of the spectrum (VIS) and low reflectance in the shortwave infrared spectrum (SWIR). We can create a normalized difference index for snow similar to the NDVI [49], based on the ratio in equation number 2: (2) In MOD10A2 data, snow cover was mapped by select MODIS radiance data of band 4 for VIS, and band 6 for SWIR. Snow presents a value of >0.0 in the NDSI. However, not all surfaces within these characteristics are snow which can result in a snow commission error. Especially within cloud fringes similar values can be encountered. Hence, these data must go through a process of snow commission errors alleviation. Several data screens were applied based on the spectral features of snow and other characteristics to flag uncertain snow cover detection and reverse it [48]. The MOD10A2 consists of two products; the first displays the mapped maximum snow extent over a period of eight days. The second displays the chronology of snow occurrence observations during the data acquisition period [47]. Here, we used the second product that displays snow continuity during an eight-day period as a chronology of observed snow cover. In this product, cloud cover was only included if there was persistent cloud cover on all eight days. If this was the case, then cloud cover was reported for a grid cell and the cell was not included in the analysis [48]. Table 1 presents the number of MOD10A2 data products used for each month during the study period. We classified monthly snow cover in each pasture to thin, moderate, and thick based on the average snow cover. If an area was covered in all eight-day datasets for a specific month it was considered thick coverage. When it was covered in only one out of three or four datasets, it was considered thin coverage. If the value fell between the two, we assigned moderate coverage. Finally, the percentage of each category including no snow cover was calculated to describe relative snow coverage over pastures surfaces for each month. 4.4 Results Below we go through each area designated by local pastoralists as having a specific usage, briefly elaborating on the vegetation growth cycles and the snow fall patterns as well as any issues with our simplified modelling approach. We group these areas into the seasonal movement stages they belong to in order to see whether they do show commonalities or disparities. 4.4.1 Winter grazing grounds. As expected, the main requirement for a winter pasture is limited snow cover during the cold months of the year, so animals can access food. In all three areas we see little snow in the months of December through March. For Adunqiaolu there is some thicker snow cover in February 2017, but then the next year sees almost no snow in the same month. Kazan shows snow in December of 2016 but is completely free of snow in December 2017. Crucially, there is no month during the year which sees complete thick or moderate snow cover. Some parts of the pasture always remain open or are at most temporarily covered with a thin layer of snow. Clearly the significance of observations made during two consecutive years is limited, but neither of the years was considered extreme by the locals. Neither black nor white disasters were recorded for the time period the underlying data of the model date to (December 2016 –February 2018). Hence, the time period covered can be assumed to lie within the range of normal years where no shocks to the pastoralist system and therefore no changes in herding patterns are observed. The limited snow cover during the winter months is connected to the low quality of the pasture in the following months. Even during the peak of the vegetation cycle in July the average NDVI value barely exceeds 0.3 for the Adunqiaolu winter pasture and its core area and 0.25 for the Kazan winter pasture. The grazing grounds dry off quickly in summer and the NDVI values decline. It is important to note that the NDVI values for barren areas, rock, and snow lie around 0.1 to -0.1. Considerable snow cover therefore impacts the average NDVI values and these averages are not directly connected to the quality of the grazing during the winter months. Slightly higher grazing qualities can be expected since snow cover biases the NDVI of a pasture area towards 0. Average NDVI values for winter pastures during the months of usage seem to hover around a value of 0.1. 4.4.2 Spring/autumn grazing grounds. Spring/autumn pastures are more difficult to identify, since they serve a transitional function and are usually only used during six to eight weeks from late March to early May and from late September to early December. In terms of snow cover the data showed a still considerable snow cover in March. With the beginning of April, the vegetation saw rapid development which is consistent with the first movement of pastoralists during the year. However, both Husta, including its core area, as well as the Wenquan spring/autumn pasture showed characteristics that need to be further elaborated upon since they demonstrate the limits of a pasture classification approach based on NDVI averages and snow cover. Wenquan, whereas consistent with the melting period and the first move of the year, showed very low NDVI averages over the entire year, even trailing the overgrazed and dried off winter pastures. The topographic location of the Wenquan spring/autumn pasture offered explanations beyond the simplistic assumptions of the model. The Wenquan pasture seemed to only provide meagre grazing throughout the entire year. The area lies to both sides of a river. Water has a very low NDVI value approaching -1 and could considerably influence an average. However, even when excluding the river and the denuded surfaces of the riverbed, the NDVI average stayed low. Therefore, other explanations needed to be considered. According to ethnographic data, the most important reason for the usage of Wenquan as a spring pasture was the early temperature increase which allows for a shortening of the winter camping time. These range lands provided a convenient location within the landscape as a transition area for heading up to the summer pastures as soon as snow started to melt there. It also had small areas of lush vegetation next to the riverbank which were too small in area size to influence the overall NDVI average much but provided ideal grazing for a limited amount of time. The abundant availability of water next to the river was convenient for herders and even allowed for limited agricultural activity including fodder crops and sometimes vegetables. Beyond environmental factors, Wenquan served as a place for exchange of animals and goods and stacking supplies for winter/summer periods. The Husta grazing lands are not only used in spring and autumn but also in summer. The reason for this lies in the favourable topographic location of the pasture. Its lower elevation starts around 1400 m a.s.l. and the highest areas lie at 2200 m a.s.l. This allows for slow movements inside the pasture area itself, depending on where the grazing is best. To the south, the Husta pasture is framed by a small ridge of around 1600–1900 m a.s.l. The north side is backed by the mountain range which defines the Bortala valley [38]. This creates a topographic situation where seasonal springs and small creeks provide water for the vegetation throughout the growing season from spring to autumn and create one of the best grazing grounds in the entire valley. 4.4.3 Summer grazing grounds. The summer pastures are characterized by late snow melt in the spring, usually in the month of April, with a good chance of considerable quantities of snow by October. Harnur East and West, as well as the summer pasture west of the Sarimu Lake, clearly fit this pattern. Harnur East, one of the highest lying summer pastures, shows consistent snow cover of between 70% and 100% from October through April, with most areas displaying thick or moderate snow cover. The Adunqiaolu Northwest summer pasture and its core area can experience full snow cover during the winter months (December 2016), but otherwise see relatively little snow during the study period. We think that the chance of a full moderate or thick snow cover excludes this area from safe usage during the winter, but there may be sociocultural and practical factors at play as well. As discussed earlier, less snow cover is not the only condition to be considered in defining a suitable winter pastureland. Winter camping grounds also require specific landforms such as a small hill or depression for shelter as well as easy external access. At an elevation of 2900 m a.s.l. the Adunqiaolu Northwest summer pasture is extremely cold in winter and the landforms offer no places for good shelter from the wind. The track leading up to the pastures is difficult, particularly in winter when it would be necessary to climb or descend steep, potentially snow-covered slopes. All summer pastures show a late but accelerated start to vegetation growth in the month of May. This fits with the second movement of the year in early May. The herds arrive on the summer pastures around the time when vegetation growth is spiking up from its low level in April. The average NDVI values for the summer pastures peaks in July with values around 0.7. These average values are extremely high for grassland and speak for the excellent quality of the grazing on these summer pastures. NDVI has been shown to correlate with productivity [50, 51]. The interrelations are certainly more complex, but as an approximation the summer pastures can be seen as roughly twice as productive as the winter pastures in our case. This has clear implications for the calorie intake of the animals over the course of the year and demonstrates why these seasonal movements are worthwhile. 4.4.1 Winter grazing grounds. As expected, the main requirement for a winter pasture is limited snow cover during the cold months of the year, so animals can access food. In all three areas we see little snow in the months of December through March. For Adunqiaolu there is some thicker snow cover in February 2017, but then the next year sees almost no snow in the same month. Kazan shows snow in December of 2016 but is completely free of snow in December 2017. Crucially, there is no month during the year which sees complete thick or moderate snow cover. Some parts of the pasture always remain open or are at most temporarily covered with a thin layer of snow. Clearly the significance of observations made during two consecutive years is limited, but neither of the years was considered extreme by the locals. Neither black nor white disasters were recorded for the time period the underlying data of the model date to (December 2016 –February 2018). Hence, the time period covered can be assumed to lie within the range of normal years where no shocks to the pastoralist system and therefore no changes in herding patterns are observed. The limited snow cover during the winter months is connected to the low quality of the pasture in the following months. Even during the peak of the vegetation cycle in July the average NDVI value barely exceeds 0.3 for the Adunqiaolu winter pasture and its core area and 0.25 for the Kazan winter pasture. The grazing grounds dry off quickly in summer and the NDVI values decline. It is important to note that the NDVI values for barren areas, rock, and snow lie around 0.1 to -0.1. Considerable snow cover therefore impacts the average NDVI values and these averages are not directly connected to the quality of the grazing during the winter months. Slightly higher grazing qualities can be expected since snow cover biases the NDVI of a pasture area towards 0. Average NDVI values for winter pastures during the months of usage seem to hover around a value of 0.1. 4.4.2 Spring/autumn grazing grounds. Spring/autumn pastures are more difficult to identify, since they serve a transitional function and are usually only used during six to eight weeks from late March to early May and from late September to early December. In terms of snow cover the data showed a still considerable snow cover in March. With the beginning of April, the vegetation saw rapid development which is consistent with the first movement of pastoralists during the year. However, both Husta, including its core area, as well as the Wenquan spring/autumn pasture showed characteristics that need to be further elaborated upon since they demonstrate the limits of a pasture classification approach based on NDVI averages and snow cover. Wenquan, whereas consistent with the melting period and the first move of the year, showed very low NDVI averages over the entire year, even trailing the overgrazed and dried off winter pastures. The topographic location of the Wenquan spring/autumn pasture offered explanations beyond the simplistic assumptions of the model. The Wenquan pasture seemed to only provide meagre grazing throughout the entire year. The area lies to both sides of a river. Water has a very low NDVI value approaching -1 and could considerably influence an average. However, even when excluding the river and the denuded surfaces of the riverbed, the NDVI average stayed low. Therefore, other explanations needed to be considered. According to ethnographic data, the most important reason for the usage of Wenquan as a spring pasture was the early temperature increase which allows for a shortening of the winter camping time. These range lands provided a convenient location within the landscape as a transition area for heading up to the summer pastures as soon as snow started to melt there. It also had small areas of lush vegetation next to the riverbank which were too small in area size to influence the overall NDVI average much but provided ideal grazing for a limited amount of time. The abundant availability of water next to the river was convenient for herders and even allowed for limited agricultural activity including fodder crops and sometimes vegetables. Beyond environmental factors, Wenquan served as a place for exchange of animals and goods and stacking supplies for winter/summer periods. The Husta grazing lands are not only used in spring and autumn but also in summer. The reason for this lies in the favourable topographic location of the pasture. Its lower elevation starts around 1400 m a.s.l. and the highest areas lie at 2200 m a.s.l. This allows for slow movements inside the pasture area itself, depending on where the grazing is best. To the south, the Husta pasture is framed by a small ridge of around 1600–1900 m a.s.l. The north side is backed by the mountain range which defines the Bortala valley [38]. This creates a topographic situation where seasonal springs and small creeks provide water for the vegetation throughout the growing season from spring to autumn and create one of the best grazing grounds in the entire valley. 4.4.3 Summer grazing grounds. The summer pastures are characterized by late snow melt in the spring, usually in the month of April, with a good chance of considerable quantities of snow by October. Harnur East and West, as well as the summer pasture west of the Sarimu Lake, clearly fit this pattern. Harnur East, one of the highest lying summer pastures, shows consistent snow cover of between 70% and 100% from October through April, with most areas displaying thick or moderate snow cover. The Adunqiaolu Northwest summer pasture and its core area can experience full snow cover during the winter months (December 2016), but otherwise see relatively little snow during the study period. We think that the chance of a full moderate or thick snow cover excludes this area from safe usage during the winter, but there may be sociocultural and practical factors at play as well. As discussed earlier, less snow cover is not the only condition to be considered in defining a suitable winter pastureland. Winter camping grounds also require specific landforms such as a small hill or depression for shelter as well as easy external access. At an elevation of 2900 m a.s.l. the Adunqiaolu Northwest summer pasture is extremely cold in winter and the landforms offer no places for good shelter from the wind. The track leading up to the pastures is difficult, particularly in winter when it would be necessary to climb or descend steep, potentially snow-covered slopes. All summer pastures show a late but accelerated start to vegetation growth in the month of May. This fits with the second movement of the year in early May. The herds arrive on the summer pastures around the time when vegetation growth is spiking up from its low level in April. The average NDVI values for the summer pastures peaks in July with values around 0.7. These average values are extremely high for grassland and speak for the excellent quality of the grazing on these summer pastures. NDVI has been shown to correlate with productivity [50, 51]. The interrelations are certainly more complex, but as an approximation the summer pastures can be seen as roughly twice as productive as the winter pastures in our case. This has clear implications for the calorie intake of the animals over the course of the year and demonstrates why these seasonal movements are worthwhile. 4.5 Summary The general model does provide insights into potential usage patterns of the landscape (Fig 11). The combination of snow cover and grazing quality seems to show relatively clearly what a particular area is best used for and the results match up with the ethnographically assigned usage. The Kazan winter pasture, for example, has relatively low pasture quality, but it is almost free of snow in the harshest winter months of January and February, thus guaranteeing access of animals to food. On the Wenquan spring pasture, the growth phase starts early and when the first seasonal movements are made by pastoralists, the pasture is already relatively lush, seeing additional growth throughout the months of April and May. The quality of the grass then stays fairly consistent throughout the summer. The herds, however, are moved to the summer pastures like, for example, Sarimu. These have been under thick snow all winter, but as soon as it melts in late April / early May, the intensive growth phase starts. These pastures provide the best grazing throughout the summer into early fall. Before the first snow starts covering the ground in October, the animals are moved back to the autumn pasture where they can graze on the grass that was left untouched throughout the summer for a few more weeks before eventually the move to the winter pasture completes the year. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 11. Idealised pastoralist movements based on ethnographic and remote sensing data. Left y-axis: snow cover in % displayed as a bar diagram distinguishing thin/moderate/thick snow cover. Right y-axis: average NDVI value. https://doi.org/10.1371/journal.pone.0240739.g011 5. Discussion Studying patterns of modern seasonal herding movement, reinforced by analysis of pasture quality and snow cover, makes it possible to infer specific seasonal occupation periods for Bronze Age house sites. The archaeological evidence for Bronze Age occupation sites is limited but sufficient to begin to address this question in a meaningful way. 5.1 Archaeological site distribution as compared to model-identified pastures 5.1.1 Winter pastures. Terrain analyses were conducted for the wider Adunqiaolu winter pasture as well as the core area, and also the Kazan winter pasture. Adunqiaolu, as the area studied most intensively, has the strongest evidence for seasonality of use. Archaeological field survey and excavation has recorded more than a hundred campsites around Adunqiaolu so it is clear that the location was a favoured one in the Bronze Age. Several factors suggest that House F1 at Adunqiaolu [18] was a winter encampment. A modern winter camp is located immediately adjacent to F1, both taking advantage of the shelter provided by a low hillock just behind the houses (Fig 6). Local informants testify that this pasture is ideal for, and in fact specifically reserved for, winter camps, while the analysis of snow cover confirms these ethnographic accounts. Botanical studies provide additional supporting evidence. Pollen analysis of sheep dung from the Bronze Age house exhibits a much more limited variety of species than modern sheep dung collected near there in summer (sheep belonging to the pasture guardian protecting the lands from out of season use), while the phytoliths present in the modern summer sheep dung come from less mature plants than those in the Bronze Age dung, suggesting that the animals were very likely kept in the house during autumn or winter [52]. 5.1.2 Summer pastures. Terrain analyses were conducted for Husta spring/summer/autumn pasture, the core area of the Husta pasture, the eastern and western part of the Harnur summer pasture, the north-western Adunqiaolu summer pasture, the core area of the north-western Adunqiaolu summer pasture, and the Sarimu Lake summer pasture. Husta is a particularly favoured pasture, available from spring through summer into autumn. There is archaeological evidence to suggest that it was so significant in the Bronze Age that it was necessary to defend it with walled lookout posts [29]. More than a hundred Bronze Age campsites have been identified there. Harnur is a smaller pasture and the season of possible use is much shorter than at Husta. Harnur East is one of the highest lying summer pastures. Fewer than ten campsites have been recorded there across both Harnur East and West, but this is still sufficient to demonstrate its use in the Bronze Age. There are also high mountain summer pastures at Adunqiaolu where more than twenty camps have been identified. No survey has been conducted in the Sarimu Lake summer pasture. 5.1.3 Spring/autumn. Terrain analyses were conducted for the Wenquan spring/autumn pasture and the Husta spring/summer/autumn pasture. The low-lying land around Wenquan has been subject to considerable modern disturbance and it has proved difficult to undertake survey there, so the presence of Bronze Age sites cannot yet be confirmed. However, the terrain analysis matches the ethnographic accounts, showing that the early appearance of grazing and its late persistence is more important in spring and autumn than the quality, as it helps to protect use of the winter pastures, the most critical of all throughout the seasons. 5.2 Paleoclimate For the period of maximum Bronze Age occupation in the Bortala Valley from c. 3900–3400 cal yr BP [18] available palaeobotanical data indicate steppic vegetation broadly similar to that of the present day. However, there are shifts within the paleoclimate that may have impacted Bronze Age land use. Holocene climate records for western Xinjiang and surrounding regions indicate localised variation over time, but they do exhibit a general pattern [53–55]. The start of the Holocene saw generally arid vegetation in low lying areas, and steppe and forest at high elevations. This changed with a somewhat cooler, wetter period starting around 6400 to 5500 cal yr BP, lasting until c. 4000 cal yr BP when there is a warming shift again. Cooler and wetter conditions return in the very late Holocene. Within the Bortala valley itself, a recent study of sediment from the Wenquan wetlands showed a shift from an arid sub-desertic landscape to steppic conditions around 7500 cal. yr BP, with a more humid steppic environment from c. 4200 cal. yr BP (J. Dodson, pers. comm.). The Andronovo culture in Xinjiang is dated between the 18th and the 14th centuries BCE, coinciding with the post 4000 BP warming shift. This may have created more favorable conditions for summer grazing and may also have ameliorated the severity of winter snowfall. Generalized as this evidence is, it indicates suitable conditions for Bronze Age transhumant pastoralism and does not contradict the results of the combined archaeological, ethnographic and terrain analyses discussed above. As with the ethnographic data, for considerations of the influence of climate on site patterning, some constants such as elevation, aspect and access still remain. The presence of residential structures in the Adunqiaolu winter pastures, with botanical evidence supporting their use in winter, suggests that snow cover was also slight there in the Bronze Age. The large number of residential structures at Husta [29] indicate that this was also a very much favored summer pasture in the Bronze Age, although it is not clear as to how early and how late in the season it was used. 5.3 Pastoral mobility Despite the apparent domestication of the horse at Early Bronze Age Botai in northern Kazakhstan [56], horses did not form a significant component of Middle and Late Bronze Age pastoral economies in the eastern Eurasian steppe which were dependent on cattle, sheep and goat, in varying proportions determined by the micro-niches they exploited. Studies at Begash in the western Dzhungar Mountains show a focus on sheep and goat over cattle as early as the mid-3rd millennium BCE [25]. This may have helped to encourage the development of the vertical pastoralism that facilitated the spread of Andronovo groups in and across much of Xinjiang during the 2nd millennium BCE. This pattern of a predominance of sheep and goat, with fewer cattle, and horses exploited mainly for transport, is also present today in western Xinjiang. However, from the Iron Age up to Late Medieval times, the steppes were dominated by mounted tribal confederacies whose wide-ranging movements were facilitated by their reliance on the horse for food, transport and warfare. This shift to new economic practices and subsequent increasing social complexity can be seen by the Late Bronze Age with the appearance of the Deer-Stone Khirigsuur culture [57, 58] and the emergence of highly mobile nomadic pastoralism associated with the earliest “Scythian” horizon [59–62]. In western Xinjiang, traces of the Andronovo fade away towards the end of the 2nd millennium BCE, to be replaced by transitional Bronze/Iron Age groups who may have exploited the landscape in quite different ways. Over the past few centuries, with increasing state control over the steppes, and over Xinjiang, tribal confederacies gradually lost their power. Together with wider access to market economies, lifestyles began to differentiate again according to local conditions, permitting us to draw close parallels between the Bronze Age and the present day. 5.1 Archaeological site distribution as compared to model-identified pastures 5.1.1 Winter pastures. Terrain analyses were conducted for the wider Adunqiaolu winter pasture as well as the core area, and also the Kazan winter pasture. Adunqiaolu, as the area studied most intensively, has the strongest evidence for seasonality of use. Archaeological field survey and excavation has recorded more than a hundred campsites around Adunqiaolu so it is clear that the location was a favoured one in the Bronze Age. Several factors suggest that House F1 at Adunqiaolu [18] was a winter encampment. A modern winter camp is located immediately adjacent to F1, both taking advantage of the shelter provided by a low hillock just behind the houses (Fig 6). Local informants testify that this pasture is ideal for, and in fact specifically reserved for, winter camps, while the analysis of snow cover confirms these ethnographic accounts. Botanical studies provide additional supporting evidence. Pollen analysis of sheep dung from the Bronze Age house exhibits a much more limited variety of species than modern sheep dung collected near there in summer (sheep belonging to the pasture guardian protecting the lands from out of season use), while the phytoliths present in the modern summer sheep dung come from less mature plants than those in the Bronze Age dung, suggesting that the animals were very likely kept in the house during autumn or winter [52]. 5.1.2 Summer pastures. Terrain analyses were conducted for Husta spring/summer/autumn pasture, the core area of the Husta pasture, the eastern and western part of the Harnur summer pasture, the north-western Adunqiaolu summer pasture, the core area of the north-western Adunqiaolu summer pasture, and the Sarimu Lake summer pasture. Husta is a particularly favoured pasture, available from spring through summer into autumn. There is archaeological evidence to suggest that it was so significant in the Bronze Age that it was necessary to defend it with walled lookout posts [29]. More than a hundred Bronze Age campsites have been identified there. Harnur is a smaller pasture and the season of possible use is much shorter than at Husta. Harnur East is one of the highest lying summer pastures. Fewer than ten campsites have been recorded there across both Harnur East and West, but this is still sufficient to demonstrate its use in the Bronze Age. There are also high mountain summer pastures at Adunqiaolu where more than twenty camps have been identified. No survey has been conducted in the Sarimu Lake summer pasture. 5.1.3 Spring/autumn. Terrain analyses were conducted for the Wenquan spring/autumn pasture and the Husta spring/summer/autumn pasture. The low-lying land around Wenquan has been subject to considerable modern disturbance and it has proved difficult to undertake survey there, so the presence of Bronze Age sites cannot yet be confirmed. However, the terrain analysis matches the ethnographic accounts, showing that the early appearance of grazing and its late persistence is more important in spring and autumn than the quality, as it helps to protect use of the winter pastures, the most critical of all throughout the seasons. 5.1.1 Winter pastures. Terrain analyses were conducted for the wider Adunqiaolu winter pasture as well as the core area, and also the Kazan winter pasture. Adunqiaolu, as the area studied most intensively, has the strongest evidence for seasonality of use. Archaeological field survey and excavation has recorded more than a hundred campsites around Adunqiaolu so it is clear that the location was a favoured one in the Bronze Age. Several factors suggest that House F1 at Adunqiaolu [18] was a winter encampment. A modern winter camp is located immediately adjacent to F1, both taking advantage of the shelter provided by a low hillock just behind the houses (Fig 6). Local informants testify that this pasture is ideal for, and in fact specifically reserved for, winter camps, while the analysis of snow cover confirms these ethnographic accounts. Botanical studies provide additional supporting evidence. Pollen analysis of sheep dung from the Bronze Age house exhibits a much more limited variety of species than modern sheep dung collected near there in summer (sheep belonging to the pasture guardian protecting the lands from out of season use), while the phytoliths present in the modern summer sheep dung come from less mature plants than those in the Bronze Age dung, suggesting that the animals were very likely kept in the house during autumn or winter [52]. 5.1.2 Summer pastures. Terrain analyses were conducted for Husta spring/summer/autumn pasture, the core area of the Husta pasture, the eastern and western part of the Harnur summer pasture, the north-western Adunqiaolu summer pasture, the core area of the north-western Adunqiaolu summer pasture, and the Sarimu Lake summer pasture. Husta is a particularly favoured pasture, available from spring through summer into autumn. There is archaeological evidence to suggest that it was so significant in the Bronze Age that it was necessary to defend it with walled lookout posts [29]. More than a hundred Bronze Age campsites have been identified there. Harnur is a smaller pasture and the season of possible use is much shorter than at Husta. Harnur East is one of the highest lying summer pastures. Fewer than ten campsites have been recorded there across both Harnur East and West, but this is still sufficient to demonstrate its use in the Bronze Age. There are also high mountain summer pastures at Adunqiaolu where more than twenty camps have been identified. No survey has been conducted in the Sarimu Lake summer pasture. 5.1.3 Spring/autumn. Terrain analyses were conducted for the Wenquan spring/autumn pasture and the Husta spring/summer/autumn pasture. The low-lying land around Wenquan has been subject to considerable modern disturbance and it has proved difficult to undertake survey there, so the presence of Bronze Age sites cannot yet be confirmed. However, the terrain analysis matches the ethnographic accounts, showing that the early appearance of grazing and its late persistence is more important in spring and autumn than the quality, as it helps to protect use of the winter pastures, the most critical of all throughout the seasons. 5.2 Paleoclimate For the period of maximum Bronze Age occupation in the Bortala Valley from c. 3900–3400 cal yr BP [18] available palaeobotanical data indicate steppic vegetation broadly similar to that of the present day. However, there are shifts within the paleoclimate that may have impacted Bronze Age land use. Holocene climate records for western Xinjiang and surrounding regions indicate localised variation over time, but they do exhibit a general pattern [53–55]. The start of the Holocene saw generally arid vegetation in low lying areas, and steppe and forest at high elevations. This changed with a somewhat cooler, wetter period starting around 6400 to 5500 cal yr BP, lasting until c. 4000 cal yr BP when there is a warming shift again. Cooler and wetter conditions return in the very late Holocene. Within the Bortala valley itself, a recent study of sediment from the Wenquan wetlands showed a shift from an arid sub-desertic landscape to steppic conditions around 7500 cal. yr BP, with a more humid steppic environment from c. 4200 cal. yr BP (J. Dodson, pers. comm.). The Andronovo culture in Xinjiang is dated between the 18th and the 14th centuries BCE, coinciding with the post 4000 BP warming shift. This may have created more favorable conditions for summer grazing and may also have ameliorated the severity of winter snowfall. Generalized as this evidence is, it indicates suitable conditions for Bronze Age transhumant pastoralism and does not contradict the results of the combined archaeological, ethnographic and terrain analyses discussed above. As with the ethnographic data, for considerations of the influence of climate on site patterning, some constants such as elevation, aspect and access still remain. The presence of residential structures in the Adunqiaolu winter pastures, with botanical evidence supporting their use in winter, suggests that snow cover was also slight there in the Bronze Age. The large number of residential structures at Husta [29] indicate that this was also a very much favored summer pasture in the Bronze Age, although it is not clear as to how early and how late in the season it was used. 5.3 Pastoral mobility Despite the apparent domestication of the horse at Early Bronze Age Botai in northern Kazakhstan [56], horses did not form a significant component of Middle and Late Bronze Age pastoral economies in the eastern Eurasian steppe which were dependent on cattle, sheep and goat, in varying proportions determined by the micro-niches they exploited. Studies at Begash in the western Dzhungar Mountains show a focus on sheep and goat over cattle as early as the mid-3rd millennium BCE [25]. This may have helped to encourage the development of the vertical pastoralism that facilitated the spread of Andronovo groups in and across much of Xinjiang during the 2nd millennium BCE. This pattern of a predominance of sheep and goat, with fewer cattle, and horses exploited mainly for transport, is also present today in western Xinjiang. However, from the Iron Age up to Late Medieval times, the steppes were dominated by mounted tribal confederacies whose wide-ranging movements were facilitated by their reliance on the horse for food, transport and warfare. This shift to new economic practices and subsequent increasing social complexity can be seen by the Late Bronze Age with the appearance of the Deer-Stone Khirigsuur culture [57, 58] and the emergence of highly mobile nomadic pastoralism associated with the earliest “Scythian” horizon [59–62]. In western Xinjiang, traces of the Andronovo fade away towards the end of the 2nd millennium BCE, to be replaced by transitional Bronze/Iron Age groups who may have exploited the landscape in quite different ways. Over the past few centuries, with increasing state control over the steppes, and over Xinjiang, tribal confederacies gradually lost their power. Together with wider access to market economies, lifestyles began to differentiate again according to local conditions, permitting us to draw close parallels between the Bronze Age and the present day. 6. Conclusion While it has been generally accepted that the Eurasian Andronovo peoples were mobile pastoralists, details of their specific adaptation to transhumant pastoralism in the Inner Asian Mountain Corridor have not been widely explored archaeologically. This study, using detailed ethnographic fieldwork and analysis of modern snow and grass cover, has shown that there is a strong correlation between modern patterns of seasonal movement and those apparently practiced in the Bronze Age. Investment in built structures in both periods further reinforces the close relationship between the two. Botanical data obtained from House F1 in the Adunqiaolu winter pastures provides a range of evidence strongly supporting the hypothesis that the structure was used as a winter camp while the widespread distribution of Bronze Age sites in modern winter and summer pastures emphasises similarities in patterns of seasonal movement between the two periods. Acknowledgments Fieldwork in the Bortala Valley is a collaboration between the Institute of Archaeology, Chinese Academy of Social Sciences (Beijing), University of Sydney and Monash University. We are thankful for the additional support from the China Studies Centre and the University of Sydney, and the Centre for Classical and Near Eastern Studies, University of Sydney.

Verot, Elise;Denois, Véronique Regnier;Macron, Corinne;Chauvin, Franck

doi: 10.1371/journal.pone.0242693pmid: 33232348

Objective Five French oncology institutions had participated in a funded study aiming at implementing an Evidence-Based Practice tool (PAM-13), which allowed nurses to measure the level of activation of the patient to support his or her own empowerment in the cancer care pathway. The purpose of this ancillary study is to (i) describe the caregivers’ perceptions of addictions and their management concurrently with cancer treatment, (ii) explore the role that Motivational Interviewing techniques can play. Methods 15 individual semi-structured interviews of caregivers were performed, using the Theoretical Domains Framework for the interview guide. Qualitative data were analyzed inductively, in a thematic analysis. The COREQ guided the reporting of this qualitative study. Results Views on addictions influence the way caregivers manage patients suffering from unresolved issues of addiction. Care is mainly focused on the pathology (cancer-centered) and strictly curative. When practiced, Motivational Interviewing is patient-centered, fostering the patient’s empowerment on the cancer care pathway. Conclusions The dissemination of Motivational Interviewing techniques in current practices in oncology, both in terms of doctors and nursing teams, would enable improvement to the management of addictions on the cancer care pathway, by deploying a patient-centered approach. This new paradigm of care would support the empowerment of patients enrolled in the cancer care pathway and promote better communication between caregivers and patients. Hence, a paradigm shift is essential. Motivational Interviewing techniques could provide a caring approach that promotes communication between the patient and the caregiver and also supports the former’s empowerment. This research suggests the need to adapt the cancer care pathway in order to integrate the necessary care for patients who concomitantly suffer from unresolved addictive disorders. Trial registration NCT03706937 1 Introduction France has turned its health policy towards a patient-centered care strategy. The National Health Strategy advocates placing the patient back at the heart of care, with a particular interest in patient experience and feedback, to enable the individual to play an active role in his or her own care [1, 2]. This paradigm shift goes against the paternalistic traditions of French care. The engagement of patients with chronic diseases has been a sought-after goal for many years [3]. Specifically, the French Cancer Plan of 2014–2019 recommended systematic support for smokers who attempt to quit while in cancer treatment facilities [4]. Benarous et al. explained that Motivational Interviewing (MI) has proven it supports and helps encourage motivation of patients to manage themselves in terms of alcohol and tobacco addiction, also by giving positive results on therapeutic adherence [5]. Regarding smoking cessation, Perriot et al. showed that a behavioral approach was more efficient (RR = 2,17; IC95%: 1,52–3,11) than a simple advisory one (RR = 1,24; IC95%: 1,16–1,33) or a total lack of advice [4]. They suggested that, as soon the cancer diagnosis was given, caregivers proposed support intervention for unresolved addictive disorders. Motivational interviewing techniques, combined with medication, turned out to ensure prolonged abstinence [4]. With this two-pronged approach, patient engagement in the cancer care pathway could be made easier [4]. McCarley stated that MI provides effective engagement among patient and care providers, in a collaborative relationship centered on patient goal-setting and self-management [6]. Miller and Rollnick define MI as “a style of collaborative conversation that strengthens a person’s own motivation and commitment to change” [7]. The question of addiction management during cancer care through a motivational approach, using Motivational interviewing techniques, is worth raising. In this paper, we considered the dependence on tobacco, alcohol and illicit drugs in terms of addiction. It should be noted that verbatim interviews that focused on patients with opioid addiction did not include patients with a higher opioid tolerance as a result of cancer pain management. According to the French National Health Ministry, the consumption of psychoactive substances (alcohol, tobacco, illicit drugs) was responsible for 100,000 deaths per year, caused either by accident or illness. Of these 100,000 deaths, 40,000 were attributable to cancer. Research also shows that addictive behavior was prevalent in 30% of deaths of persons under age 65 [8]. In the spring of 2018, the “onco-addiction group” was created within the French comprehensive cancer centers network, which is a first in France and internationally [9]. Thus, while conducting research at five French comprehensive centers aimed at implementing an Evidence-Based Practice (EBP) tool in nursing practices (in order to foster patient empowerment in the cancer care pathway) [10], we believed it was relevant to undertake an ancillary research study on the subject, given its novelty in France. Indeed, the orientation of the new French health policy recognized prevention and patient experience as two essential elements in the cancer care pathway. However, the recent results of the Compare Cohort demonstrated the persistent gap between expectations, needs expressed by patients and the representations made by healthcare professionals [11]. We therefore sought to question the continuous stigma that plagues those dealing with addiction problems while being treated by healthcare workers, and the latter’s role-related perceptions in such a context. Moreover, we aimed to demonstrate how motivational interviewing techniques could improve patient care. The aim of this qualitative study is to (i) describe caregivers’ perceptions of addictions and how they manage them concurrently with cancer treatment (ii) explore the role that motivational interviewing techniques can play. 2 Methods A qualitative study was carried out in a French public comprehensive cancer facility situated in a district heavily affected by cancer. The choice to perform a qualitative study was based on the potential that this type of research could potentially contribute to understanding a problem in various dimensions or to study phenomena not yet detected [12]. The Hygée Centre, located in the same area, is a facility of regional resources for cancer prevention, information and education. 2.1 Characteristics of the sample 2.1.1 Source and method of recruitment. After meeting with the Administrative Director of nurses and caregivers at the different facilities, the former provided a list of names of all the nurses’ managers to contact in each care unit (oncology and hematology). Each healthcare manager presented the study to his/her entire team, supported by the study information leaflet at their disposal, which had been validated by the Ethics Committee. Depending on the dates of meetings that we had arranged beforehand, the study was proposed to all the professionals who were scheduled to work in the units on these specific days. Concerning the treatment teams, caregivers were contacted either directly by the representatives, who personally arranged the appointments, or through an e-mail sent by the representative, with a copy of the e-mail to our team, requesting they make an appointment with the interviewers. 2.1.2 Inclusion and exclusion criteria. The inclusion criterion consisted of an agreement by the nurses to participate in collecting data. Consent was given by e-mail or verbally during the appointment arrangement process. The exclusion criterion was a refusal to take part in the study. All those designated agreed to participate and were subsequently interviewed. 2.2 Semi-structured interviews 2.2.1 Using the Theoretical Domains Framework (TDF). The TDF was developed by Michie et al. to understand health professionals’ practices and identify elements of these practices to target for the implementation of new recommendations and evidence-based practice [13, 14]. According to Lou Atkins et al., this framework was primarily used in healthcare settings to explore factors influencing clinical behaviors in order to design implementation interventions; to identify barriers and facilitators to change [15–17]. Alan Glasper and Colin Rees explained that each domain of the TDF represented a behavioral determinant, mediators of behavioral change. [18] The TDF was composed of 12 original theoretical domains: (1) knowledge, (2) skills, (3) social/ professional role and identity, (4) beliefs about capabilities, (5) beliefs about consequences, (6) motivation and goals, (7) memory, attention and decision processes, (8) environmental context and resources, (9) social influences, (10) emotion regulation, (11) behavioral regulation, and (12) nature of the behavior. [14] Employing this tool helped identify elements that either hindered or facilitated the implementation of new modes of care in everyday practice. Michie et al. suggested a prompt containing the wording of a series of follow-up questions related to each theoretical area [14]. We performed a double translation of the prompt. We then conducted a testing phase with five persons of differing profiles to ensure adequate understanding in French and subsequently presented it to four health professionals to confirm its relevance to the target audience. 2.2.2 The interview guide. The form and substance of the interview schedule were approved by a medical anthropologist with a PhD, specialized in the field of cancer research. To begin with, the guide consisted of the following open prompt: “Please tell me about the management of addictions in cancer patient care.” The interview was organized to assess how health professionals defined MI and motivational positioning, addictions and their management in the cancer care pathway. Each interview was conducted in a single session, individually, in a quiet office, with the schedule organized to ensure the professional’s full attention. 2.3 Data collection phase Data was collected from October to November 2018. Interviews were conducted by the first author. She is an MSc nurse, a PhD candidate, trained in qualitative research, with no previous association to the caregivers interviewed. The framework in which the collection took place was explained to each of the participants, ensuring their freedom of speech, data protection, anonymization, and a policy of non-reporting to their hierarchy. All interviews took place in a calm office, face-to-face, and were recorded in their entirety on a digital audio recorder. They were then transcribed in verbatim form on a word processing program by a professional secretary. The objective was to collect data until saturation. The average duration of each interview was 22 minutes. The sample consisted of 15 cancer care providers, mainly oncology nurses. The sample represented a wide variety of ages, seniority and experience. It included nurses working in oncology or hematology in different departments and contexts, ranging from day hospitalization or full conventional hospitalization to interdisciplinary positions (nurse navigators) or the sterile services sector or the hematology transplant department. Although a minority, the sample also included other health professionals such as a tobacco specialist nurse, a social worker and an onco-psychologist, working in both departments (Table 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Participants’ characteristics. https://doi.org/10.1371/journal.pone.0242693.t001 2.4 Qualitative data analysis These data were analyzed in order to generate themes that could subsequently be associated with the Theoretical Domains Framework of behavior change to investigate implementation problems [14–19]. Specifically, we applied the methodology framework proposed by Braun, V. and Clarke, V. [20]: Familiarization with the data. This phase involved reading and re-reading the data to become immersed in and intimately familiar with its content. Generating initial Codes: this phase involved generating succinct codes that identified important features of the data that might be relevant to answering our research question. It involved coding the entire dataset, and then collating all the codes and all relevant data extracts together for later stages of analysis. Nvivo 11 pro software (QSR International) was used to perform the analyses. Searching for themes: the collection of codes was worked on in pairs with the PhD researcher (VRD) to validate a comprehensive interpretation and a grouping of code elements into themes. The resulting themes were contrasted to the TDF categories independently by the two researchers, and then compared and discussed. Items that could not be included in the TDF framework were thematized separately. Finally, a more refined coding was used for each part of the verbatim used to illustrate the themes. Three themes and eight sub-themes were generated. Nvivo 11 pro (QSR International) software was used to analyze the collected data, performed by the first author. Once completed, this analysis was discussed in detail with the second author, a medical anthropologist with a PhD, specialized in the cancer research field. The participants did not provide any feedback on the results. 2.5 Ethical considerations The study protocol was approved by the Ethics Committee of the Saint Etienne University Hospital, France (IORG0007394/ N° IRBN712018/CHUSTE). All participants were given written and oral information about the study and gave informed consent to participate. The COREQ checklist guided the preparation of this manuscript [21]. 2.5.1 Ethics approval. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or the national research committee (Ethics Committee of the Saint Etienne University Hospital, France. NCT03706937) and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. 2.5.2 Consent to participate. Informed consent was obtained from all individual participants included in the study. 2.5.3 Consent for publication. Informed consent was obtained from all individual participants included in the study. 3. Results The three main themes emerging from the qualitative analysis were (i) representations about addictions-guided care (ii) a difference in practice that depended on the role in care (iii) Motivational Interviewing and motivational positioning driving patient-centered care. An overview of the different major themes and sub-themes are presented in Table 2. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Overview of the major themes and sub-themes. https://doi.org/10.1371/journal.pone.0242693.t002 3.1 Representations about addictions-guided care 3.1.1 More complex patients. For the majority of nurses, patients with alcohol addiction were described as "somewhat uncommon patients,” “Always more complicated than other patients!" or "those kinds of patients.” They were perceived as apart. Most nurses believed these patients required more time and that an effort had to be made to avoid confrontation, language hostilities or upsetting the patient. "For us, it is complicated to treat patients who have an ear, nose or throat cancer and who still have addiction issues; it is difficult for the treatment I give them (…) they are somewhat uncommon… they are non-communicative, they do not talk to us or they are verbally a bit aggressive (…) It is not always easy to give care; they refuse treatment." (Nurse) For the onco-psychologist, anchored in a psychoanalytical approach like his or her colleagues, patients with an unresolved addictive disorder were viewed as more vulnerable than patients already weakened by cancer. They described complex care management without behavioral changes in the patients in question. "Addiction, which is added on top of cancer (…) (is an) exacerbating factor or a vulnerability." "They represent two major entities: addiction and cancer. Together, they make for a particularly complex situation.” (Onco-Psychologist) 3.1.2 Cancer seen as a burden. Cancer remained central to concerns. The patient was seen as stupefied by the announcement of a cancer diagnosis. Caregivers perceived it as a unique disease compared to other care pathways. "Cancer is a very heavy disease.” “Patients suffering from cancer are in a state of stupefaction (…) they have just been told that they have cancer, which defines the specificity of care." (Nurse) Nurses described the diagnosis of cancer as representing a drastic change in the patient's life. They explained that as soon as the disease was announced, patients became psychologically affected, due to the intensity of emotions and upheaval it caused. They also mentioned that the entire daily life of the patient required a reorganization; that these sudden changes were transformations for the body, but also for the mind of the patient. "So, in fact, the specificity is that you have to give them time too! They have so much information, that their brain cannot do everything at the same time! This is the difference, this stupefaction!" (Nurse) Patients are often seen as very tired, psychologically and physically. "There are also patients who are not doing well, they have cardiac issues, so this must be mentioned too! There are patients with advanced, already metastatic diseases, with a full range of symptoms.” (Nurse) For patients with an advanced stage of the disease, the concept of project and future projection was modified. Nurses struggled to cope with unsolved addictive disorders associated with cancer, since the cancer was an advanced state and the patient’s life span uncertain. Nurses felt the patient was accumulating burdens. "I think it's the cancer, on top of the addiction, that makes the problem harder, it adds an additional burden, anyway! And afterwards, the prognosis is often short-term, often… for conditions that are a bit particular, anyway!" (Nurse) Managing a cancer patient’s addiction presented a challenge for nurses, who felt like they were overloading the patient already dealing with cancer. “We are going to ask people to make an extra effort, when they are already facing cancer. So, it’s true that it can actually be complicated. (…) Cancer is a whirlwind that turns everything upside down, and yet we are asking that of them too! It’s the part about adding something more again!” (Nurse) 3.1.3 Lack of awareness of smoking cessation. The tobacco specialist nurse emphasized the overused image that nurses and doctors had of their role in the care. Caregivers appeared to pose a major obstacle to specific support for patients with unresolved addictive disorders. “We have several barriers to seeing all the patients; the barrier of the caregivers, in any case!” (Tobacco specialist nurse) For many nurses, managing tobacco addiction was tantamount to total abstinence, which explained why they found it hard to understand the modalities of management based on a reduction in consumption associated with a prescription for nicotine replacement (in this case, patches). “It won't change anything if they stop (…) Once they have met the tobacco specialist nurse, there is not always a change in the patient—some of them will still go outside and smoke their cigarette, like they did before meeting her. They tell us that they have an agreement with the tobacco specialist nurse and that they can smoke, even with the patches, despite things like that, yes!” (Nurse) For the other staff members, the opposite was true. They held the conviction that the total abstinence approach of the tobacco specialist nurse caused conflict and hurt the patient, forcing him/her to stop smoking when cancer had just been diagnosed. The tobacco specialist nurse insisted that it was vital to treating physical dependence in the cancer care pathway. “Addiction is not only about wearing the patch (…) Sometimes, we can be sure of temporary abstinence; a patient who has cancer and who is confined to his/her bed—it is not normal that the person suffer from abstinence.” (Tobacco specialist nurse) The care management strategies of this specialist opposed the representations of his peers, not trained in the specific management of addictions. "We have the specificity regarding the staff, where we have quite a few obstacles to overcome, in any case. Some experience our intervention as something ‘violent’ regarding the patient who has just been diagnosed with cancer. For some, it’s not obvious." (Tobacco specialist nurse) 3.2 A difference in practice depending on the role in care 3.2.1 A taboo on alcohol and drug addiction. For the majority of the sample participants, tobacco was the addiction addressed the most often. Alcohol came in second place but was seldom discussed in a meaningful way. Drug addiction was only very briefly brought up by the three professionals (n = 3: 1 psychologist and 2 caregivers). "There are drugs. People who take drugs (…) I don’t know if it’s not too taboo because we don’t talk about it much.” “It can apply to alcohol as well and opiates can be even more taboo; we don’t even talk about it. Ultra-taboo." (Onco-psychologist) For sterile containment units, addictions seemed inexistant, with the exception of tobacco, which remained questionable for nurses, who lacked a rational explanation. "Frankly no, it's very strange (…) no, curiously, we don't have people, I mean, I have come across very few in our department. Whereas, when you go to other departments, you will find some! But not in our department!" (Nurse) Nurses pointed out that the consideration of dependence was directly related to the spontaneous reporting of the patient. For example, in the case of a young opioid-dependent patient (as stated in the medical record), nothing had been implemented, since she had not declared or requested anything. “There was nothing specific about opiates. There weren't any things implemented at that time. She wasn't describing a dependence, but there too, they have to be able to say they're dependent! (Nurse) However, the nurse was obliged to collect information from the patient to fill out the medical file. The majority of nurses were uncomfortable with how they approached the addiction subject with the patient. When the nurse broached the topic, it was in a mode that emphasized thoroughness. Otherwise, as seen above, the nurse waited for the patient to declare unresolved issues of addiction on his/her own. “Because, it’s always taken badly when I ask this question, because it means I put the word ‘addiction’ on it (…) It has a pejorative side, I find (…) I don’t know how to ask the question in another way.” (Nurse) The interview was described as containing leading questions, essentially akin to an interrogation, with the computer as an intermediary, employing closed questions that required binary answers of yes or no. The patient therefore answered these questions of the nurse, without the possibility of speaking about other areas, problems or individual priorities. “To know the patient’s priorities, I will ask them the question. But it remains a pointed question; it is an interrogation.” (Nurse) Thus, the directive nursing interview, employing closed questions, like an inventory checklist, seemed to protect the nurse from having to raise questions on sensitive subjects such as the consumption of or addiction to alcohol or other illicit substances. "For me, it's easier to ask (…) like a quote-unquote ‘inventory checklist,’ because falling into judging (them) I find difficult (…) To avoid being judgmental.” (Nurse) 3.2.2 The curative approach: Cancer-centered. For nurses, the fact that the patient continued to consume tobacco while managing cancer did not change the standards of care. The caregiver did not invest heavily in addiction. "Afterwards, I'm not going to make any specific adjustment in treatment, because there is no real point (…) I realize that we haven’t done anything concrete about it." (Nurse) However, an exception could be made for the sterile hematology sector. Indeed, nurses explained that these patients in sterile isolation rooms had no choice but to avoid the tobacco or other drugs. Abstinence appeared to be an easy “action,” since the patient could not take any other action, given the hospitalization conditions: complete hospitalization over several weeks without the possibility of leaving the sterile room. “Afterwards, I’d say that sterile hospitalization is easy, because, all of a sudden, they don’t have a choice!” (Nurse) According to the nurses, the medical oncologists seemed to focus only on the treatment aspect of cancer. Other ailments, such as pain or fatigue, did not appear to have a place in starting care. It was therefore not uncommon for the patient, experiencing a certain level of pain, not to go to his or her treatment appointment. The medical approach remained pathological-centered and left no room for a holistic approach of the person. “For example, the doctor, what is incredible is that he will only be interested in chemo! We have the impression that chemo interests him more than the patient! And it’s a bit complicated! (…) Later on, we sometimes realize there is a problem with pain, and that’s why the patient didn’t come; but the problem wasn’t handled on the first day because it wasn’t the question of the day! For the doctor, the question of the day was to know whether the patient had chemo or not!” (Nurse) Addictions were added as an additional burden (although they often existed prior to the cancer), with the cancer remaining the central concern, directing management toward a disease-centered approach rather than a patient-centered approach. In most cases, nurses were helpless in the face of cancer-treated patients with unresolved issues of addiction. Cancer care was the priority of the institution and the healthcare teams working there. Therefore, unresolved issues of addiction were not high on the nurses’ list of responsibilities, even if it was sometimes one of the patient's priorities. "We are treating a cancerous disease; we do not treat addiction. Addiction is not the priority; I’d say that, even for me, it is not my priority, whereas sometimes it is the priority for the patient!” (Nurse) This is probably the reason why nurses pointed out the absence of smoking prevention in the hospital. “It is true that the team doesn’t do prevention (…) in terms of the establishment… there is not a whole lot of prevention, not in the hallways either.” (Nurse) 3.2.3 Motivational positioning at the margin of current practices. We noted one exception: a nurse with experience in treating alcohol and tobacco dependence stood out from his/her peers. This nurse suffered from the lack of consideration and specific management of addictions by his/her colleagues. This nurse managed to identify and create a rapport with the patient to encourage the latter to talk and built trust with patients during their cancer care. This nurse supported patients in their daily lives, while trying to inspire them to change their behavior. This nurse practiced motivational positioning, but the action remained strictly isolated, with no connection or possibility of sharing with other colleagues. For the tobacco specialist nurse, who was affiliated with another institution, his/her intervention seemed to fall entirely outside the care pathway, again, focused 100% on cancer. This specialist nurse explained that there was no link to overall care, which, in his/her view, influenced the effectiveness of care. “The problem of addictology in the cancer treatment facility—I have the impression that we are a pawn among the rest of it (…) Our care is very, very, very ‘psy’ (psychological). And the info that we do not have here, it’s kind of a pity!” (Tobacco specialist nurse) The tobacco specialist nurse practiced MI on a daily basis in care. This specialist nurse, trained in this process, saw the benefits of care in terms of results with the patient. The current nursing practice was based on all the fundamentals of MI, utilized to care for patients undergoing cancer treatment and suffering from unresolved issues of addiction. This specialist nurse explained that alcohol was often addressed by the tobacco person, because patients suffering from alcohol addiction found it difficult to admit. In this case, the nurse adopted motivational positioning with regard to alcohol and subsequently directed the patient to the addiction specialist team, who intended to follow up. "People have a little trouble confessing their alcohol consumption, so through tobacco, we're going to bring up the other products that exist." (Tobacco specialist nurse) The Social worker informed us of training received in conducting social interviews in the course of his/her studies. Mastering the social interview technique was a requisite for obtaining the degree. The foundations of MI provided the framework that guided the social interview: an interview for and with the patient, to bring out the patient’s motivations for change. The Social worker stressed the need to adopt an empathetic posture, rely on reflective listening and use reformulation and enhancement. Respect for the patient’s rhythm was central to the professional’s approach, not acting in the patient’s place. The formulation of hypotheses and the use of the decision-making balance were an integral part of the practice. “It is this relational technique that we are asked to use, which combines empathy, a patient-centered approach, valorization, reformulation, active listening.” (Social worker) The Social worker related how, when there was a medical injunction to arrange directive home care, against the will of the patient, it ended in failure, with the patient doing an about-face upon returning home. “The person tells me, ‘No! I don’t want it!’ But I still arrange everything urgently, because the patient will return home, which comes with a medical protocol, an obligation, and we are responsible for their discharge (…) The home aides called and said that he didn’t want it, you know? (…) When we don’t follow the timing of the patient, it does (results in) this!” (Social worker) The Social worker highlighted the difference that this directive stance created with the patient, compared to patient-centered care, where the patient felt considered. “The patient feels much more considered in terms of his care, that’s it, he feels he is receiving care! It makes a difference! I see this difference.” (Social worker) Almost a majority of nurses had never heard of MI. A third of the participants of the sample attempted to explain representations of MI but did not corroborate its definition and modalities. They always positioned themselves first, with their filters and their care and management priorities, which was in contrast to the posture taken in motivational positioning. Some claimed it was due to their training and educational experience. They mentioned not knowing how to work differently: the nurse is in a perpetual state of motion with the patient; the nurse must always provide a solution or meet a measurable objective nearly instantaneously. “People who listen fully- that is not an objective for me! I'm not going to play the pop psychologist!” (Nurse) The tobacco specialist nurse tried to show other nurses that listening to the patient is the key to managing unresolved issues of addiction. In motivational positioning, listening is a treatment, it is considered care. "But it does not prevent us from fighting against this; talking is a treatment, but it's not the same as administering an injection. There are degrees of ‘nobility.’” (Tobacco specialist nurse) At times when the nurse wanted to help the patient move towards weaning, the nurse quickly found himself/herself in an impasse. Clearly, the posture adopted by the nurse, combined with the way of connecting with the patient and any addictive problems, was met with a resistance to change in the patient. Furthermore, the discourse increased this resistance to modify his or her habits. "After a quarter of an hour, quickly, we were stuck! I thought something was missing and I couldn’t see how to get out of this situation! I didn't find an opening or anything, I didn’t know what to do! (…) It was so painstaking for fifteen minutes! There was an opening that told me that he was worried about it, but I felt blocked and I didn’t see how! Then he asked me if I had a lighter! I told him no, there’s a limit to everything! It was too much for me!” (Nurse) Clinical care remained under the influence of the nurses’ corrective reflex, referring to the tendency to advise patients about the “right” path to follow for good health. Thus, the nurse advised and warned the patient about the dangers of addiction. The nurse only intended to convince the patient that his/her health would improve, and treatments would prove more efficient after the changes, to provide support. Every fact and gesture of the patient was known to each member of the care team. Caregiver treatment was directive or even injunctive. “This is what we are trying to do, but it just doesn’t work.” (Make the patient’s own motivation emerge and depart from the common discourse). “I think we are not very good at highlighting the positive, we know how to criticize. We have a hard time moving forward at the patient's pace as a caregiver.” (Nurse) For hospitalized patients, nurses claimed they insisted on education. They reiterated that they were obliged to follow up on what the oncologist did because the latter “made a big deal out of it.” The nurses explained to the patient that smoking increased risks that affected the curability of the cancer. They concluded with the patient, “If there are other addictive elements that come into play, we will not be able to cure them!” Many nurses disclosed that they used the technique of fear to stimulate the weaning process necessary for the hospital environment (complete hospitalization, sterile post-transplant area). "My technique is fear (…) It is my way of being in contact with this specific domain." (Nurse) 3.3 Motivational Interviewing and motivational positioning driving patient-centered care 3.3.1 The contributions of motivational positioning to cancer care. Those who practice MI and motivational positioning explained that it was the best way to encourage the patient to change. The decision-making scale involved a technique widely used to enable patients to integrate the notion of management of current addictions while undergoing cancer treatment. “We are not necessarily going to talk about completely quitting tobacco because often, people are scared, thinking we are going to forbid them to smoke (…) So we’re going to cite all the pros of quitting smoking and all the cons of cigarettes (…) Then we are going to show them a rather empathetic attitude, we are not going to impose things on them, we are going to gently help them come around to quitting, which is different (…) We evoke the efforts they have made, highlight what they have already achieved, and in no case are we going to make them feel guilty." (Tobacco specialist nurse) Thanks to this motivational support technique, patients were often satisfied and comforted. Non-directive care, with full listening, without guilt or judgement, reassured patients. They could thus move forward at their own pace and take an active part in changing their behavior and habits, if they so desired. The healthcare professional trained in this approach was there to respect their timing and support them in this process. “Actively listening to the patient, (being) empathetic (…) many people have an idea of the care regarding tobacco that is completely different; often people do not want to see us (….) (but) at the end of the appointment, they are happy because, and this is the point, they did not expect our particular approach.” (Tobacco specialist nurse) 3.3.2 A need for training for a patient-centered approach emerging among nurses. For nurses, a need for training was raised—training that would enable them to relate in a relevant and effective manner to patients undergoing cancer treatment and suffering from an unresolved addictive disorder. “For starters, I don’t have the skills because I have no specific training.” “In my opinion, we have to come to a realization (…) (that) I don’t have the skills; I don’t have the training.” (Nurse) Nurses justifiably questioned the mirror game between themselves and the patients. The former seemed to be regularly challenged when facing these patients, amassing counter-attitudes or results against productivity, even though they had fervently invested in care. “I think if we had training to approach this type of patient, it wouldn’t be so bad! (…) I wonder if we are showing them the appropriate attitude.” (Nurse) As the nurses had previously mentioned, they were the product of a training system with a medico-centered approach. It was the nursing posture anchored in the paternalistic and prescriptive attitude, tinged with the corrective reflex described here. “I tell myself that there might be a way to approach it in a different manner.” (Nurse) 2.1 Characteristics of the sample 2.1.1 Source and method of recruitment. After meeting with the Administrative Director of nurses and caregivers at the different facilities, the former provided a list of names of all the nurses’ managers to contact in each care unit (oncology and hematology). Each healthcare manager presented the study to his/her entire team, supported by the study information leaflet at their disposal, which had been validated by the Ethics Committee. Depending on the dates of meetings that we had arranged beforehand, the study was proposed to all the professionals who were scheduled to work in the units on these specific days. Concerning the treatment teams, caregivers were contacted either directly by the representatives, who personally arranged the appointments, or through an e-mail sent by the representative, with a copy of the e-mail to our team, requesting they make an appointment with the interviewers. 2.1.2 Inclusion and exclusion criteria. The inclusion criterion consisted of an agreement by the nurses to participate in collecting data. Consent was given by e-mail or verbally during the appointment arrangement process. The exclusion criterion was a refusal to take part in the study. All those designated agreed to participate and were subsequently interviewed. 2.1.1 Source and method of recruitment. After meeting with the Administrative Director of nurses and caregivers at the different facilities, the former provided a list of names of all the nurses’ managers to contact in each care unit (oncology and hematology). Each healthcare manager presented the study to his/her entire team, supported by the study information leaflet at their disposal, which had been validated by the Ethics Committee. Depending on the dates of meetings that we had arranged beforehand, the study was proposed to all the professionals who were scheduled to work in the units on these specific days. Concerning the treatment teams, caregivers were contacted either directly by the representatives, who personally arranged the appointments, or through an e-mail sent by the representative, with a copy of the e-mail to our team, requesting they make an appointment with the interviewers. 2.1.2 Inclusion and exclusion criteria. The inclusion criterion consisted of an agreement by the nurses to participate in collecting data. Consent was given by e-mail or verbally during the appointment arrangement process. The exclusion criterion was a refusal to take part in the study. All those designated agreed to participate and were subsequently interviewed. 2.2 Semi-structured interviews 2.2.1 Using the Theoretical Domains Framework (TDF). The TDF was developed by Michie et al. to understand health professionals’ practices and identify elements of these practices to target for the implementation of new recommendations and evidence-based practice [13, 14]. According to Lou Atkins et al., this framework was primarily used in healthcare settings to explore factors influencing clinical behaviors in order to design implementation interventions; to identify barriers and facilitators to change [15–17]. Alan Glasper and Colin Rees explained that each domain of the TDF represented a behavioral determinant, mediators of behavioral change. [18] The TDF was composed of 12 original theoretical domains: (1) knowledge, (2) skills, (3) social/ professional role and identity, (4) beliefs about capabilities, (5) beliefs about consequences, (6) motivation and goals, (7) memory, attention and decision processes, (8) environmental context and resources, (9) social influences, (10) emotion regulation, (11) behavioral regulation, and (12) nature of the behavior. [14] Employing this tool helped identify elements that either hindered or facilitated the implementation of new modes of care in everyday practice. Michie et al. suggested a prompt containing the wording of a series of follow-up questions related to each theoretical area [14]. We performed a double translation of the prompt. We then conducted a testing phase with five persons of differing profiles to ensure adequate understanding in French and subsequently presented it to four health professionals to confirm its relevance to the target audience. 2.2.2 The interview guide. The form and substance of the interview schedule were approved by a medical anthropologist with a PhD, specialized in the field of cancer research. To begin with, the guide consisted of the following open prompt: “Please tell me about the management of addictions in cancer patient care.” The interview was organized to assess how health professionals defined MI and motivational positioning, addictions and their management in the cancer care pathway. Each interview was conducted in a single session, individually, in a quiet office, with the schedule organized to ensure the professional’s full attention. 2.2.1 Using the Theoretical Domains Framework (TDF). The TDF was developed by Michie et al. to understand health professionals’ practices and identify elements of these practices to target for the implementation of new recommendations and evidence-based practice [13, 14]. According to Lou Atkins et al., this framework was primarily used in healthcare settings to explore factors influencing clinical behaviors in order to design implementation interventions; to identify barriers and facilitators to change [15–17]. Alan Glasper and Colin Rees explained that each domain of the TDF represented a behavioral determinant, mediators of behavioral change. [18] The TDF was composed of 12 original theoretical domains: (1) knowledge, (2) skills, (3) social/ professional role and identity, (4) beliefs about capabilities, (5) beliefs about consequences, (6) motivation and goals, (7) memory, attention and decision processes, (8) environmental context and resources, (9) social influences, (10) emotion regulation, (11) behavioral regulation, and (12) nature of the behavior. [14] Employing this tool helped identify elements that either hindered or facilitated the implementation of new modes of care in everyday practice. Michie et al. suggested a prompt containing the wording of a series of follow-up questions related to each theoretical area [14]. We performed a double translation of the prompt. We then conducted a testing phase with five persons of differing profiles to ensure adequate understanding in French and subsequently presented it to four health professionals to confirm its relevance to the target audience. 2.2.2 The interview guide. The form and substance of the interview schedule were approved by a medical anthropologist with a PhD, specialized in the field of cancer research. To begin with, the guide consisted of the following open prompt: “Please tell me about the management of addictions in cancer patient care.” The interview was organized to assess how health professionals defined MI and motivational positioning, addictions and their management in the cancer care pathway. Each interview was conducted in a single session, individually, in a quiet office, with the schedule organized to ensure the professional’s full attention. 2.3 Data collection phase Data was collected from October to November 2018. Interviews were conducted by the first author. She is an MSc nurse, a PhD candidate, trained in qualitative research, with no previous association to the caregivers interviewed. The framework in which the collection took place was explained to each of the participants, ensuring their freedom of speech, data protection, anonymization, and a policy of non-reporting to their hierarchy. All interviews took place in a calm office, face-to-face, and were recorded in their entirety on a digital audio recorder. They were then transcribed in verbatim form on a word processing program by a professional secretary. The objective was to collect data until saturation. The average duration of each interview was 22 minutes. The sample consisted of 15 cancer care providers, mainly oncology nurses. The sample represented a wide variety of ages, seniority and experience. It included nurses working in oncology or hematology in different departments and contexts, ranging from day hospitalization or full conventional hospitalization to interdisciplinary positions (nurse navigators) or the sterile services sector or the hematology transplant department. Although a minority, the sample also included other health professionals such as a tobacco specialist nurse, a social worker and an onco-psychologist, working in both departments (Table 1). Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Participants’ characteristics. https://doi.org/10.1371/journal.pone.0242693.t001 2.4 Qualitative data analysis These data were analyzed in order to generate themes that could subsequently be associated with the Theoretical Domains Framework of behavior change to investigate implementation problems [14–19]. Specifically, we applied the methodology framework proposed by Braun, V. and Clarke, V. [20]: Familiarization with the data. This phase involved reading and re-reading the data to become immersed in and intimately familiar with its content. Generating initial Codes: this phase involved generating succinct codes that identified important features of the data that might be relevant to answering our research question. It involved coding the entire dataset, and then collating all the codes and all relevant data extracts together for later stages of analysis. Nvivo 11 pro software (QSR International) was used to perform the analyses. Searching for themes: the collection of codes was worked on in pairs with the PhD researcher (VRD) to validate a comprehensive interpretation and a grouping of code elements into themes. The resulting themes were contrasted to the TDF categories independently by the two researchers, and then compared and discussed. Items that could not be included in the TDF framework were thematized separately. Finally, a more refined coding was used for each part of the verbatim used to illustrate the themes. Three themes and eight sub-themes were generated. Nvivo 11 pro (QSR International) software was used to analyze the collected data, performed by the first author. Once completed, this analysis was discussed in detail with the second author, a medical anthropologist with a PhD, specialized in the cancer research field. The participants did not provide any feedback on the results. 2.5 Ethical considerations The study protocol was approved by the Ethics Committee of the Saint Etienne University Hospital, France (IORG0007394/ N° IRBN712018/CHUSTE). All participants were given written and oral information about the study and gave informed consent to participate. The COREQ checklist guided the preparation of this manuscript [21]. 2.5.1 Ethics approval. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or the national research committee (Ethics Committee of the Saint Etienne University Hospital, France. NCT03706937) and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. 2.5.2 Consent to participate. Informed consent was obtained from all individual participants included in the study. 2.5.3 Consent for publication. Informed consent was obtained from all individual participants included in the study. 2.5.1 Ethics approval. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or the national research committee (Ethics Committee of the Saint Etienne University Hospital, France. NCT03706937) and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. 2.5.2 Consent to participate. Informed consent was obtained from all individual participants included in the study. 2.5.3 Consent for publication. Informed consent was obtained from all individual participants included in the study. 3. Results The three main themes emerging from the qualitative analysis were (i) representations about addictions-guided care (ii) a difference in practice that depended on the role in care (iii) Motivational Interviewing and motivational positioning driving patient-centered care. An overview of the different major themes and sub-themes are presented in Table 2. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Overview of the major themes and sub-themes. https://doi.org/10.1371/journal.pone.0242693.t002 3.1 Representations about addictions-guided care 3.1.1 More complex patients. For the majority of nurses, patients with alcohol addiction were described as "somewhat uncommon patients,” “Always more complicated than other patients!" or "those kinds of patients.” They were perceived as apart. Most nurses believed these patients required more time and that an effort had to be made to avoid confrontation, language hostilities or upsetting the patient. "For us, it is complicated to treat patients who have an ear, nose or throat cancer and who still have addiction issues; it is difficult for the treatment I give them (…) they are somewhat uncommon… they are non-communicative, they do not talk to us or they are verbally a bit aggressive (…) It is not always easy to give care; they refuse treatment." (Nurse) For the onco-psychologist, anchored in a psychoanalytical approach like his or her colleagues, patients with an unresolved addictive disorder were viewed as more vulnerable than patients already weakened by cancer. They described complex care management without behavioral changes in the patients in question. "Addiction, which is added on top of cancer (…) (is an) exacerbating factor or a vulnerability." "They represent two major entities: addiction and cancer. Together, they make for a particularly complex situation.” (Onco-Psychologist) 3.1.2 Cancer seen as a burden. Cancer remained central to concerns. The patient was seen as stupefied by the announcement of a cancer diagnosis. Caregivers perceived it as a unique disease compared to other care pathways. "Cancer is a very heavy disease.” “Patients suffering from cancer are in a state of stupefaction (…) they have just been told that they have cancer, which defines the specificity of care." (Nurse) Nurses described the diagnosis of cancer as representing a drastic change in the patient's life. They explained that as soon as the disease was announced, patients became psychologically affected, due to the intensity of emotions and upheaval it caused. They also mentioned that the entire daily life of the patient required a reorganization; that these sudden changes were transformations for the body, but also for the mind of the patient. "So, in fact, the specificity is that you have to give them time too! They have so much information, that their brain cannot do everything at the same time! This is the difference, this stupefaction!" (Nurse) Patients are often seen as very tired, psychologically and physically. "There are also patients who are not doing well, they have cardiac issues, so this must be mentioned too! There are patients with advanced, already metastatic diseases, with a full range of symptoms.” (Nurse) For patients with an advanced stage of the disease, the concept of project and future projection was modified. Nurses struggled to cope with unsolved addictive disorders associated with cancer, since the cancer was an advanced state and the patient’s life span uncertain. Nurses felt the patient was accumulating burdens. "I think it's the cancer, on top of the addiction, that makes the problem harder, it adds an additional burden, anyway! And afterwards, the prognosis is often short-term, often… for conditions that are a bit particular, anyway!" (Nurse) Managing a cancer patient’s addiction presented a challenge for nurses, who felt like they were overloading the patient already dealing with cancer. “We are going to ask people to make an extra effort, when they are already facing cancer. So, it’s true that it can actually be complicated. (…) Cancer is a whirlwind that turns everything upside down, and yet we are asking that of them too! It’s the part about adding something more again!” (Nurse) 3.1.3 Lack of awareness of smoking cessation. The tobacco specialist nurse emphasized the overused image that nurses and doctors had of their role in the care. Caregivers appeared to pose a major obstacle to specific support for patients with unresolved addictive disorders. “We have several barriers to seeing all the patients; the barrier of the caregivers, in any case!” (Tobacco specialist nurse) For many nurses, managing tobacco addiction was tantamount to total abstinence, which explained why they found it hard to understand the modalities of management based on a reduction in consumption associated with a prescription for nicotine replacement (in this case, patches). “It won't change anything if they stop (…) Once they have met the tobacco specialist nurse, there is not always a change in the patient—some of them will still go outside and smoke their cigarette, like they did before meeting her. They tell us that they have an agreement with the tobacco specialist nurse and that they can smoke, even with the patches, despite things like that, yes!” (Nurse) For the other staff members, the opposite was true. They held the conviction that the total abstinence approach of the tobacco specialist nurse caused conflict and hurt the patient, forcing him/her to stop smoking when cancer had just been diagnosed. The tobacco specialist nurse insisted that it was vital to treating physical dependence in the cancer care pathway. “Addiction is not only about wearing the patch (…) Sometimes, we can be sure of temporary abstinence; a patient who has cancer and who is confined to his/her bed—it is not normal that the person suffer from abstinence.” (Tobacco specialist nurse) The care management strategies of this specialist opposed the representations of his peers, not trained in the specific management of addictions. "We have the specificity regarding the staff, where we have quite a few obstacles to overcome, in any case. Some experience our intervention as something ‘violent’ regarding the patient who has just been diagnosed with cancer. For some, it’s not obvious." (Tobacco specialist nurse) 3.1.1 More complex patients. For the majority of nurses, patients with alcohol addiction were described as "somewhat uncommon patients,” “Always more complicated than other patients!" or "those kinds of patients.” They were perceived as apart. Most nurses believed these patients required more time and that an effort had to be made to avoid confrontation, language hostilities or upsetting the patient. "For us, it is complicated to treat patients who have an ear, nose or throat cancer and who still have addiction issues; it is difficult for the treatment I give them (…) they are somewhat uncommon… they are non-communicative, they do not talk to us or they are verbally a bit aggressive (…) It is not always easy to give care; they refuse treatment." (Nurse) For the onco-psychologist, anchored in a psychoanalytical approach like his or her colleagues, patients with an unresolved addictive disorder were viewed as more vulnerable than patients already weakened by cancer. They described complex care management without behavioral changes in the patients in question. "Addiction, which is added on top of cancer (…) (is an) exacerbating factor or a vulnerability." "They represent two major entities: addiction and cancer. Together, they make for a particularly complex situation.” (Onco-Psychologist) 3.1.2 Cancer seen as a burden. Cancer remained central to concerns. The patient was seen as stupefied by the announcement of a cancer diagnosis. Caregivers perceived it as a unique disease compared to other care pathways. "Cancer is a very heavy disease.” “Patients suffering from cancer are in a state of stupefaction (…) they have just been told that they have cancer, which defines the specificity of care." (Nurse) Nurses described the diagnosis of cancer as representing a drastic change in the patient's life. They explained that as soon as the disease was announced, patients became psychologically affected, due to the intensity of emotions and upheaval it caused. They also mentioned that the entire daily life of the patient required a reorganization; that these sudden changes were transformations for the body, but also for the mind of the patient. "So, in fact, the specificity is that you have to give them time too! They have so much information, that their brain cannot do everything at the same time! This is the difference, this stupefaction!" (Nurse) Patients are often seen as very tired, psychologically and physically. "There are also patients who are not doing well, they have cardiac issues, so this must be mentioned too! There are patients with advanced, already metastatic diseases, with a full range of symptoms.” (Nurse) For patients with an advanced stage of the disease, the concept of project and future projection was modified. Nurses struggled to cope with unsolved addictive disorders associated with cancer, since the cancer was an advanced state and the patient’s life span uncertain. Nurses felt the patient was accumulating burdens. "I think it's the cancer, on top of the addiction, that makes the problem harder, it adds an additional burden, anyway! And afterwards, the prognosis is often short-term, often… for conditions that are a bit particular, anyway!" (Nurse) Managing a cancer patient’s addiction presented a challenge for nurses, who felt like they were overloading the patient already dealing with cancer. “We are going to ask people to make an extra effort, when they are already facing cancer. So, it’s true that it can actually be complicated. (…) Cancer is a whirlwind that turns everything upside down, and yet we are asking that of them too! It’s the part about adding something more again!” (Nurse) 3.1.3 Lack of awareness of smoking cessation. The tobacco specialist nurse emphasized the overused image that nurses and doctors had of their role in the care. Caregivers appeared to pose a major obstacle to specific support for patients with unresolved addictive disorders. “We have several barriers to seeing all the patients; the barrier of the caregivers, in any case!” (Tobacco specialist nurse) For many nurses, managing tobacco addiction was tantamount to total abstinence, which explained why they found it hard to understand the modalities of management based on a reduction in consumption associated with a prescription for nicotine replacement (in this case, patches). “It won't change anything if they stop (…) Once they have met the tobacco specialist nurse, there is not always a change in the patient—some of them will still go outside and smoke their cigarette, like they did before meeting her. They tell us that they have an agreement with the tobacco specialist nurse and that they can smoke, even with the patches, despite things like that, yes!” (Nurse) For the other staff members, the opposite was true. They held the conviction that the total abstinence approach of the tobacco specialist nurse caused conflict and hurt the patient, forcing him/her to stop smoking when cancer had just been diagnosed. The tobacco specialist nurse insisted that it was vital to treating physical dependence in the cancer care pathway. “Addiction is not only about wearing the patch (…) Sometimes, we can be sure of temporary abstinence; a patient who has cancer and who is confined to his/her bed—it is not normal that the person suffer from abstinence.” (Tobacco specialist nurse) The care management strategies of this specialist opposed the representations of his peers, not trained in the specific management of addictions. "We have the specificity regarding the staff, where we have quite a few obstacles to overcome, in any case. Some experience our intervention as something ‘violent’ regarding the patient who has just been diagnosed with cancer. For some, it’s not obvious." (Tobacco specialist nurse) 3.2 A difference in practice depending on the role in care 3.2.1 A taboo on alcohol and drug addiction. For the majority of the sample participants, tobacco was the addiction addressed the most often. Alcohol came in second place but was seldom discussed in a meaningful way. Drug addiction was only very briefly brought up by the three professionals (n = 3: 1 psychologist and 2 caregivers). "There are drugs. People who take drugs (…) I don’t know if it’s not too taboo because we don’t talk about it much.” “It can apply to alcohol as well and opiates can be even more taboo; we don’t even talk about it. Ultra-taboo." (Onco-psychologist) For sterile containment units, addictions seemed inexistant, with the exception of tobacco, which remained questionable for nurses, who lacked a rational explanation. "Frankly no, it's very strange (…) no, curiously, we don't have people, I mean, I have come across very few in our department. Whereas, when you go to other departments, you will find some! But not in our department!" (Nurse) Nurses pointed out that the consideration of dependence was directly related to the spontaneous reporting of the patient. For example, in the case of a young opioid-dependent patient (as stated in the medical record), nothing had been implemented, since she had not declared or requested anything. “There was nothing specific about opiates. There weren't any things implemented at that time. She wasn't describing a dependence, but there too, they have to be able to say they're dependent! (Nurse) However, the nurse was obliged to collect information from the patient to fill out the medical file. The majority of nurses were uncomfortable with how they approached the addiction subject with the patient. When the nurse broached the topic, it was in a mode that emphasized thoroughness. Otherwise, as seen above, the nurse waited for the patient to declare unresolved issues of addiction on his/her own. “Because, it’s always taken badly when I ask this question, because it means I put the word ‘addiction’ on it (…) It has a pejorative side, I find (…) I don’t know how to ask the question in another way.” (Nurse) The interview was described as containing leading questions, essentially akin to an interrogation, with the computer as an intermediary, employing closed questions that required binary answers of yes or no. The patient therefore answered these questions of the nurse, without the possibility of speaking about other areas, problems or individual priorities. “To know the patient’s priorities, I will ask them the question. But it remains a pointed question; it is an interrogation.” (Nurse) Thus, the directive nursing interview, employing closed questions, like an inventory checklist, seemed to protect the nurse from having to raise questions on sensitive subjects such as the consumption of or addiction to alcohol or other illicit substances. "For me, it's easier to ask (…) like a quote-unquote ‘inventory checklist,’ because falling into judging (them) I find difficult (…) To avoid being judgmental.” (Nurse) 3.2.2 The curative approach: Cancer-centered. For nurses, the fact that the patient continued to consume tobacco while managing cancer did not change the standards of care. The caregiver did not invest heavily in addiction. "Afterwards, I'm not going to make any specific adjustment in treatment, because there is no real point (…) I realize that we haven’t done anything concrete about it." (Nurse) However, an exception could be made for the sterile hematology sector. Indeed, nurses explained that these patients in sterile isolation rooms had no choice but to avoid the tobacco or other drugs. Abstinence appeared to be an easy “action,” since the patient could not take any other action, given the hospitalization conditions: complete hospitalization over several weeks without the possibility of leaving the sterile room. “Afterwards, I’d say that sterile hospitalization is easy, because, all of a sudden, they don’t have a choice!” (Nurse) According to the nurses, the medical oncologists seemed to focus only on the treatment aspect of cancer. Other ailments, such as pain or fatigue, did not appear to have a place in starting care. It was therefore not uncommon for the patient, experiencing a certain level of pain, not to go to his or her treatment appointment. The medical approach remained pathological-centered and left no room for a holistic approach of the person. “For example, the doctor, what is incredible is that he will only be interested in chemo! We have the impression that chemo interests him more than the patient! And it’s a bit complicated! (…) Later on, we sometimes realize there is a problem with pain, and that’s why the patient didn’t come; but the problem wasn’t handled on the first day because it wasn’t the question of the day! For the doctor, the question of the day was to know whether the patient had chemo or not!” (Nurse) Addictions were added as an additional burden (although they often existed prior to the cancer), with the cancer remaining the central concern, directing management toward a disease-centered approach rather than a patient-centered approach. In most cases, nurses were helpless in the face of cancer-treated patients with unresolved issues of addiction. Cancer care was the priority of the institution and the healthcare teams working there. Therefore, unresolved issues of addiction were not high on the nurses’ list of responsibilities, even if it was sometimes one of the patient's priorities. "We are treating a cancerous disease; we do not treat addiction. Addiction is not the priority; I’d say that, even for me, it is not my priority, whereas sometimes it is the priority for the patient!” (Nurse) This is probably the reason why nurses pointed out the absence of smoking prevention in the hospital. “It is true that the team doesn’t do prevention (…) in terms of the establishment… there is not a whole lot of prevention, not in the hallways either.” (Nurse) 3.2.3 Motivational positioning at the margin of current practices. We noted one exception: a nurse with experience in treating alcohol and tobacco dependence stood out from his/her peers. This nurse suffered from the lack of consideration and specific management of addictions by his/her colleagues. This nurse managed to identify and create a rapport with the patient to encourage the latter to talk and built trust with patients during their cancer care. This nurse supported patients in their daily lives, while trying to inspire them to change their behavior. This nurse practiced motivational positioning, but the action remained strictly isolated, with no connection or possibility of sharing with other colleagues. For the tobacco specialist nurse, who was affiliated with another institution, his/her intervention seemed to fall entirely outside the care pathway, again, focused 100% on cancer. This specialist nurse explained that there was no link to overall care, which, in his/her view, influenced the effectiveness of care. “The problem of addictology in the cancer treatment facility—I have the impression that we are a pawn among the rest of it (…) Our care is very, very, very ‘psy’ (psychological). And the info that we do not have here, it’s kind of a pity!” (Tobacco specialist nurse) The tobacco specialist nurse practiced MI on a daily basis in care. This specialist nurse, trained in this process, saw the benefits of care in terms of results with the patient. The current nursing practice was based on all the fundamentals of MI, utilized to care for patients undergoing cancer treatment and suffering from unresolved issues of addiction. This specialist nurse explained that alcohol was often addressed by the tobacco person, because patients suffering from alcohol addiction found it difficult to admit. In this case, the nurse adopted motivational positioning with regard to alcohol and subsequently directed the patient to the addiction specialist team, who intended to follow up. "People have a little trouble confessing their alcohol consumption, so through tobacco, we're going to bring up the other products that exist." (Tobacco specialist nurse) The Social worker informed us of training received in conducting social interviews in the course of his/her studies. Mastering the social interview technique was a requisite for obtaining the degree. The foundations of MI provided the framework that guided the social interview: an interview for and with the patient, to bring out the patient’s motivations for change. The Social worker stressed the need to adopt an empathetic posture, rely on reflective listening and use reformulation and enhancement. Respect for the patient’s rhythm was central to the professional’s approach, not acting in the patient’s place. The formulation of hypotheses and the use of the decision-making balance were an integral part of the practice. “It is this relational technique that we are asked to use, which combines empathy, a patient-centered approach, valorization, reformulation, active listening.” (Social worker) The Social worker related how, when there was a medical injunction to arrange directive home care, against the will of the patient, it ended in failure, with the patient doing an about-face upon returning home. “The person tells me, ‘No! I don’t want it!’ But I still arrange everything urgently, because the patient will return home, which comes with a medical protocol, an obligation, and we are responsible for their discharge (…) The home aides called and said that he didn’t want it, you know? (…) When we don’t follow the timing of the patient, it does (results in) this!” (Social worker) The Social worker highlighted the difference that this directive stance created with the patient, compared to patient-centered care, where the patient felt considered. “The patient feels much more considered in terms of his care, that’s it, he feels he is receiving care! It makes a difference! I see this difference.” (Social worker) Almost a majority of nurses had never heard of MI. A third of the participants of the sample attempted to explain representations of MI but did not corroborate its definition and modalities. They always positioned themselves first, with their filters and their care and management priorities, which was in contrast to the posture taken in motivational positioning. Some claimed it was due to their training and educational experience. They mentioned not knowing how to work differently: the nurse is in a perpetual state of motion with the patient; the nurse must always provide a solution or meet a measurable objective nearly instantaneously. “People who listen fully- that is not an objective for me! I'm not going to play the pop psychologist!” (Nurse) The tobacco specialist nurse tried to show other nurses that listening to the patient is the key to managing unresolved issues of addiction. In motivational positioning, listening is a treatment, it is considered care. "But it does not prevent us from fighting against this; talking is a treatment, but it's not the same as administering an injection. There are degrees of ‘nobility.’” (Tobacco specialist nurse) At times when the nurse wanted to help the patient move towards weaning, the nurse quickly found himself/herself in an impasse. Clearly, the posture adopted by the nurse, combined with the way of connecting with the patient and any addictive problems, was met with a resistance to change in the patient. Furthermore, the discourse increased this resistance to modify his or her habits. "After a quarter of an hour, quickly, we were stuck! I thought something was missing and I couldn’t see how to get out of this situation! I didn't find an opening or anything, I didn’t know what to do! (…) It was so painstaking for fifteen minutes! There was an opening that told me that he was worried about it, but I felt blocked and I didn’t see how! Then he asked me if I had a lighter! I told him no, there’s a limit to everything! It was too much for me!” (Nurse) Clinical care remained under the influence of the nurses’ corrective reflex, referring to the tendency to advise patients about the “right” path to follow for good health. Thus, the nurse advised and warned the patient about the dangers of addiction. The nurse only intended to convince the patient that his/her health would improve, and treatments would prove more efficient after the changes, to provide support. Every fact and gesture of the patient was known to each member of the care team. Caregiver treatment was directive or even injunctive. “This is what we are trying to do, but it just doesn’t work.” (Make the patient’s own motivation emerge and depart from the common discourse). “I think we are not very good at highlighting the positive, we know how to criticize. We have a hard time moving forward at the patient's pace as a caregiver.” (Nurse) For hospitalized patients, nurses claimed they insisted on education. They reiterated that they were obliged to follow up on what the oncologist did because the latter “made a big deal out of it.” The nurses explained to the patient that smoking increased risks that affected the curability of the cancer. They concluded with the patient, “If there are other addictive elements that come into play, we will not be able to cure them!” Many nurses disclosed that they used the technique of fear to stimulate the weaning process necessary for the hospital environment (complete hospitalization, sterile post-transplant area). "My technique is fear (…) It is my way of being in contact with this specific domain." (Nurse) 3.2.1 A taboo on alcohol and drug addiction. For the majority of the sample participants, tobacco was the addiction addressed the most often. Alcohol came in second place but was seldom discussed in a meaningful way. Drug addiction was only very briefly brought up by the three professionals (n = 3: 1 psychologist and 2 caregivers). "There are drugs. People who take drugs (…) I don’t know if it’s not too taboo because we don’t talk about it much.” “It can apply to alcohol as well and opiates can be even more taboo; we don’t even talk about it. Ultra-taboo." (Onco-psychologist) For sterile containment units, addictions seemed inexistant, with the exception of tobacco, which remained questionable for nurses, who lacked a rational explanation. "Frankly no, it's very strange (…) no, curiously, we don't have people, I mean, I have come across very few in our department. Whereas, when you go to other departments, you will find some! But not in our department!" (Nurse) Nurses pointed out that the consideration of dependence was directly related to the spontaneous reporting of the patient. For example, in the case of a young opioid-dependent patient (as stated in the medical record), nothing had been implemented, since she had not declared or requested anything. “There was nothing specific about opiates. There weren't any things implemented at that time. She wasn't describing a dependence, but there too, they have to be able to say they're dependent! (Nurse) However, the nurse was obliged to collect information from the patient to fill out the medical file. The majority of nurses were uncomfortable with how they approached the addiction subject with the patient. When the nurse broached the topic, it was in a mode that emphasized thoroughness. Otherwise, as seen above, the nurse waited for the patient to declare unresolved issues of addiction on his/her own. “Because, it’s always taken badly when I ask this question, because it means I put the word ‘addiction’ on it (…) It has a pejorative side, I find (…) I don’t know how to ask the question in another way.” (Nurse) The interview was described as containing leading questions, essentially akin to an interrogation, with the computer as an intermediary, employing closed questions that required binary answers of yes or no. The patient therefore answered these questions of the nurse, without the possibility of speaking about other areas, problems or individual priorities. “To know the patient’s priorities, I will ask them the question. But it remains a pointed question; it is an interrogation.” (Nurse) Thus, the directive nursing interview, employing closed questions, like an inventory checklist, seemed to protect the nurse from having to raise questions on sensitive subjects such as the consumption of or addiction to alcohol or other illicit substances. "For me, it's easier to ask (…) like a quote-unquote ‘inventory checklist,’ because falling into judging (them) I find difficult (…) To avoid being judgmental.” (Nurse) 3.2.2 The curative approach: Cancer-centered. For nurses, the fact that the patient continued to consume tobacco while managing cancer did not change the standards of care. The caregiver did not invest heavily in addiction. "Afterwards, I'm not going to make any specific adjustment in treatment, because there is no real point (…) I realize that we haven’t done anything concrete about it." (Nurse) However, an exception could be made for the sterile hematology sector. Indeed, nurses explained that these patients in sterile isolation rooms had no choice but to avoid the tobacco or other drugs. Abstinence appeared to be an easy “action,” since the patient could not take any other action, given the hospitalization conditions: complete hospitalization over several weeks without the possibility of leaving the sterile room. “Afterwards, I’d say that sterile hospitalization is easy, because, all of a sudden, they don’t have a choice!” (Nurse) According to the nurses, the medical oncologists seemed to focus only on the treatment aspect of cancer. Other ailments, such as pain or fatigue, did not appear to have a place in starting care. It was therefore not uncommon for the patient, experiencing a certain level of pain, not to go to his or her treatment appointment. The medical approach remained pathological-centered and left no room for a holistic approach of the person. “For example, the doctor, what is incredible is that he will only be interested in chemo! We have the impression that chemo interests him more than the patient! And it’s a bit complicated! (…) Later on, we sometimes realize there is a problem with pain, and that’s why the patient didn’t come; but the problem wasn’t handled on the first day because it wasn’t the question of the day! For the doctor, the question of the day was to know whether the patient had chemo or not!” (Nurse) Addictions were added as an additional burden (although they often existed prior to the cancer), with the cancer remaining the central concern, directing management toward a disease-centered approach rather than a patient-centered approach. In most cases, nurses were helpless in the face of cancer-treated patients with unresolved issues of addiction. Cancer care was the priority of the institution and the healthcare teams working there. Therefore, unresolved issues of addiction were not high on the nurses’ list of responsibilities, even if it was sometimes one of the patient's priorities. "We are treating a cancerous disease; we do not treat addiction. Addiction is not the priority; I’d say that, even for me, it is not my priority, whereas sometimes it is the priority for the patient!” (Nurse) This is probably the reason why nurses pointed out the absence of smoking prevention in the hospital. “It is true that the team doesn’t do prevention (…) in terms of the establishment… there is not a whole lot of prevention, not in the hallways either.” (Nurse) 3.2.3 Motivational positioning at the margin of current practices. We noted one exception: a nurse with experience in treating alcohol and tobacco dependence stood out from his/her peers. This nurse suffered from the lack of consideration and specific management of addictions by his/her colleagues. This nurse managed to identify and create a rapport with the patient to encourage the latter to talk and built trust with patients during their cancer care. This nurse supported patients in their daily lives, while trying to inspire them to change their behavior. This nurse practiced motivational positioning, but the action remained strictly isolated, with no connection or possibility of sharing with other colleagues. For the tobacco specialist nurse, who was affiliated with another institution, his/her intervention seemed to fall entirely outside the care pathway, again, focused 100% on cancer. This specialist nurse explained that there was no link to overall care, which, in his/her view, influenced the effectiveness of care. “The problem of addictology in the cancer treatment facility—I have the impression that we are a pawn among the rest of it (…) Our care is very, very, very ‘psy’ (psychological). And the info that we do not have here, it’s kind of a pity!” (Tobacco specialist nurse) The tobacco specialist nurse practiced MI on a daily basis in care. This specialist nurse, trained in this process, saw the benefits of care in terms of results with the patient. The current nursing practice was based on all the fundamentals of MI, utilized to care for patients undergoing cancer treatment and suffering from unresolved issues of addiction. This specialist nurse explained that alcohol was often addressed by the tobacco person, because patients suffering from alcohol addiction found it difficult to admit. In this case, the nurse adopted motivational positioning with regard to alcohol and subsequently directed the patient to the addiction specialist team, who intended to follow up. "People have a little trouble confessing their alcohol consumption, so through tobacco, we're going to bring up the other products that exist." (Tobacco specialist nurse) The Social worker informed us of training received in conducting social interviews in the course of his/her studies. Mastering the social interview technique was a requisite for obtaining the degree. The foundations of MI provided the framework that guided the social interview: an interview for and with the patient, to bring out the patient’s motivations for change. The Social worker stressed the need to adopt an empathetic posture, rely on reflective listening and use reformulation and enhancement. Respect for the patient’s rhythm was central to the professional’s approach, not acting in the patient’s place. The formulation of hypotheses and the use of the decision-making balance were an integral part of the practice. “It is this relational technique that we are asked to use, which combines empathy, a patient-centered approach, valorization, reformulation, active listening.” (Social worker) The Social worker related how, when there was a medical injunction to arrange directive home care, against the will of the patient, it ended in failure, with the patient doing an about-face upon returning home. “The person tells me, ‘No! I don’t want it!’ But I still arrange everything urgently, because the patient will return home, which comes with a medical protocol, an obligation, and we are responsible for their discharge (…) The home aides called and said that he didn’t want it, you know? (…) When we don’t follow the timing of the patient, it does (results in) this!” (Social worker) The Social worker highlighted the difference that this directive stance created with the patient, compared to patient-centered care, where the patient felt considered. “The patient feels much more considered in terms of his care, that’s it, he feels he is receiving care! It makes a difference! I see this difference.” (Social worker) Almost a majority of nurses had never heard of MI. A third of the participants of the sample attempted to explain representations of MI but did not corroborate its definition and modalities. They always positioned themselves first, with their filters and their care and management priorities, which was in contrast to the posture taken in motivational positioning. Some claimed it was due to their training and educational experience. They mentioned not knowing how to work differently: the nurse is in a perpetual state of motion with the patient; the nurse must always provide a solution or meet a measurable objective nearly instantaneously. “People who listen fully- that is not an objective for me! I'm not going to play the pop psychologist!” (Nurse) The tobacco specialist nurse tried to show other nurses that listening to the patient is the key to managing unresolved issues of addiction. In motivational positioning, listening is a treatment, it is considered care. "But it does not prevent us from fighting against this; talking is a treatment, but it's not the same as administering an injection. There are degrees of ‘nobility.’” (Tobacco specialist nurse) At times when the nurse wanted to help the patient move towards weaning, the nurse quickly found himself/herself in an impasse. Clearly, the posture adopted by the nurse, combined with the way of connecting with the patient and any addictive problems, was met with a resistance to change in the patient. Furthermore, the discourse increased this resistance to modify his or her habits. "After a quarter of an hour, quickly, we were stuck! I thought something was missing and I couldn’t see how to get out of this situation! I didn't find an opening or anything, I didn’t know what to do! (…) It was so painstaking for fifteen minutes! There was an opening that told me that he was worried about it, but I felt blocked and I didn’t see how! Then he asked me if I had a lighter! I told him no, there’s a limit to everything! It was too much for me!” (Nurse) Clinical care remained under the influence of the nurses’ corrective reflex, referring to the tendency to advise patients about the “right” path to follow for good health. Thus, the nurse advised and warned the patient about the dangers of addiction. The nurse only intended to convince the patient that his/her health would improve, and treatments would prove more efficient after the changes, to provide support. Every fact and gesture of the patient was known to each member of the care team. Caregiver treatment was directive or even injunctive. “This is what we are trying to do, but it just doesn’t work.” (Make the patient’s own motivation emerge and depart from the common discourse). “I think we are not very good at highlighting the positive, we know how to criticize. We have a hard time moving forward at the patient's pace as a caregiver.” (Nurse) For hospitalized patients, nurses claimed they insisted on education. They reiterated that they were obliged to follow up on what the oncologist did because the latter “made a big deal out of it.” The nurses explained to the patient that smoking increased risks that affected the curability of the cancer. They concluded with the patient, “If there are other addictive elements that come into play, we will not be able to cure them!” Many nurses disclosed that they used the technique of fear to stimulate the weaning process necessary for the hospital environment (complete hospitalization, sterile post-transplant area). "My technique is fear (…) It is my way of being in contact with this specific domain." (Nurse) 3.3 Motivational Interviewing and motivational positioning driving patient-centered care 3.3.1 The contributions of motivational positioning to cancer care. Those who practice MI and motivational positioning explained that it was the best way to encourage the patient to change. The decision-making scale involved a technique widely used to enable patients to integrate the notion of management of current addictions while undergoing cancer treatment. “We are not necessarily going to talk about completely quitting tobacco because often, people are scared, thinking we are going to forbid them to smoke (…) So we’re going to cite all the pros of quitting smoking and all the cons of cigarettes (…) Then we are going to show them a rather empathetic attitude, we are not going to impose things on them, we are going to gently help them come around to quitting, which is different (…) We evoke the efforts they have made, highlight what they have already achieved, and in no case are we going to make them feel guilty." (Tobacco specialist nurse) Thanks to this motivational support technique, patients were often satisfied and comforted. Non-directive care, with full listening, without guilt or judgement, reassured patients. They could thus move forward at their own pace and take an active part in changing their behavior and habits, if they so desired. The healthcare professional trained in this approach was there to respect their timing and support them in this process. “Actively listening to the patient, (being) empathetic (…) many people have an idea of the care regarding tobacco that is completely different; often people do not want to see us (….) (but) at the end of the appointment, they are happy because, and this is the point, they did not expect our particular approach.” (Tobacco specialist nurse) 3.3.2 A need for training for a patient-centered approach emerging among nurses. For nurses, a need for training was raised—training that would enable them to relate in a relevant and effective manner to patients undergoing cancer treatment and suffering from an unresolved addictive disorder. “For starters, I don’t have the skills because I have no specific training.” “In my opinion, we have to come to a realization (…) (that) I don’t have the skills; I don’t have the training.” (Nurse) Nurses justifiably questioned the mirror game between themselves and the patients. The former seemed to be regularly challenged when facing these patients, amassing counter-attitudes or results against productivity, even though they had fervently invested in care. “I think if we had training to approach this type of patient, it wouldn’t be so bad! (…) I wonder if we are showing them the appropriate attitude.” (Nurse) As the nurses had previously mentioned, they were the product of a training system with a medico-centered approach. It was the nursing posture anchored in the paternalistic and prescriptive attitude, tinged with the corrective reflex described here. “I tell myself that there might be a way to approach it in a different manner.” (Nurse) 3.3.1 The contributions of motivational positioning to cancer care. Those who practice MI and motivational positioning explained that it was the best way to encourage the patient to change. The decision-making scale involved a technique widely used to enable patients to integrate the notion of management of current addictions while undergoing cancer treatment. “We are not necessarily going to talk about completely quitting tobacco because often, people are scared, thinking we are going to forbid them to smoke (…) So we’re going to cite all the pros of quitting smoking and all the cons of cigarettes (…) Then we are going to show them a rather empathetic attitude, we are not going to impose things on them, we are going to gently help them come around to quitting, which is different (…) We evoke the efforts they have made, highlight what they have already achieved, and in no case are we going to make them feel guilty." (Tobacco specialist nurse) Thanks to this motivational support technique, patients were often satisfied and comforted. Non-directive care, with full listening, without guilt or judgement, reassured patients. They could thus move forward at their own pace and take an active part in changing their behavior and habits, if they so desired. The healthcare professional trained in this approach was there to respect their timing and support them in this process. “Actively listening to the patient, (being) empathetic (…) many people have an idea of the care regarding tobacco that is completely different; often people do not want to see us (….) (but) at the end of the appointment, they are happy because, and this is the point, they did not expect our particular approach.” (Tobacco specialist nurse) 3.3.2 A need for training for a patient-centered approach emerging among nurses. For nurses, a need for training was raised—training that would enable them to relate in a relevant and effective manner to patients undergoing cancer treatment and suffering from an unresolved addictive disorder. “For starters, I don’t have the skills because I have no specific training.” “In my opinion, we have to come to a realization (…) (that) I don’t have the skills; I don’t have the training.” (Nurse) Nurses justifiably questioned the mirror game between themselves and the patients. The former seemed to be regularly challenged when facing these patients, amassing counter-attitudes or results against productivity, even though they had fervently invested in care. “I think if we had training to approach this type of patient, it wouldn’t be so bad! (…) I wonder if we are showing them the appropriate attitude.” (Nurse) As the nurses had previously mentioned, they were the product of a training system with a medico-centered approach. It was the nursing posture anchored in the paternalistic and prescriptive attitude, tinged with the corrective reflex described here. “I tell myself that there might be a way to approach it in a different manner.” (Nurse) 4 Discussion 4.1 Care centered on cancer at the expense of addiction management Hamant et al. conducted one of the few French studies of dealing with alcohol and tobacco addiction in patients undergoing treatment for ear, nose or throat cancer [22]. Less than 1% of patients treated for cancer reported that they were informed about appointments to help with tobacco dependence or possible support in the management of alcohol consumption. Healthcare professionals involved in the cancer care pathways did not seem invested in the opportunity to support patients with this type of care, complementary to the treatment of their cancer [22]. The professionals interviewed in this qualitative study corroborated these facts. Caregiver management was central to cancer and did not include the possibility that the patient could also be suffering from unresolved addictive disorders. However, the Oncology Nursing Society (ONS) has shown that the management of tobacco addiction has been essential in oncology nursing. It has also established that the role of the nurse in the cancer care pathway is primordial in supporting patients who are experiencing withdrawal or who must wean themselves from their tobacco addiction [23]. However, smoking cessation programs have only recently been integrated as an essential part of cancer management. Active smokers treated for cancer have had poorer outcomes (reduced survival rates and weaker responses to cancer treatments) than those who were able to quit [24]. Results from this ancillary study showed that the oncology nurses often felt helpless to manage and support patients with unresolved addictive disorders. Recent studies have highlighted the need to integrate comprehensive patient care into cancer care pathways. Dauchy and Curé explained that this involves the systematic detection of the patient's vulnerability criteria, to include him/her in a risk group. This would strengthen the care process. One of the criteria was the search for persistent risk behaviors during cancer disease [25]. For this, as evidenced by the findings of this study, a change in the representations of caregivers on the patient treated for a cancer would be necessary. Nurses would also need to change the way they connect with the patient to be able to manage them more holistically, in a patient-centered manner of care, throughout the cancer care pathway. 4.2 The need to change practitioners’ attitudes to enable them to provide appropriate care to patients In France, the paternalistic cultural model of care was the predominant approach until recently. The sensitive relationship often remained characterized by an asymmetrical relationship. This aspect of the nursing position was clearly highlighted in this study. Nurses were often challenged because of their relationship with the patient. This relationship was not one of equality. Conversely, the patient-centered approach aimed to foster patient engagement by placing the caregiver and the patient in a more symmetrical relationship that encouraged the patient to be able to express himself/herself freely, without fear of judgement, on subjects of importance to him/her. Nurses tended to be locked in a pattern of their traditional role and found it decidedly difficult to connect with the patient differently [26]. The results of this research have supported what has been highlighted in the literature. It is well known that the traditional mode of communication focused on a directive approach to delivering information and advice is becoming less effective in maintaining change and significantly changing behavior [5]. Studies have shown that when doctors provided their patients with useful health information, while responding to their emotions, patients experienced a sense of self-control and greater hope. This has had a directly positive impact on their quality of life as well as their survival [27]. The method of entering into a relationship with the patient has therefore proved decisive, with the possibility that the caregiver could let the patient engage and be involved in his/her own care continuum. Barello et al. explained that nursing was the key to patient engagement. This commitment was associated with patients’ perceptions of the positive attitudes of healthcare professionals on their self-care behaviors. This focused on the essential role of nurses in charge [28]. It also appeared that patients tended to no longer actively be actors in their care when they did not feel they were in a symmetrical relationship with the caregiver, due to a lack of information about their state of health, or as part of a paternalistic attitude from the care system [28]. Our results have demonstrated that nurses who were not trained in motivational positioning struggled to avoid judging patients. They were also expected to obtain concrete results in the management of withdrawal symptoms or to connect with patients affected by addictive behaviors. Some postures adopted by nurses created a "relationship gap" in the caregiver-patient relationship. The majority of nurses interviewed reported that it was difficult to care for patients suffering from addictions vis-à-vis the traditional way nurses treat patients undergoing cancer care. Certain people questioned this issue of communication with patients but were unable to find a response. Michiels related that this relationship gap was the result of attitudes of care such as directional postures, admonitory speech, warnings, reasoned persuasive attitudes, administering advice or even using moralizing words [29]. Situations of a relationship impasse (such as those described in a verbatim where the nurse could not find a suitable angle to help the patient out of ambivalence to discussing smoking) were extremely stressful for the caregivers. They experienced feelings of helplessness and powerlessness in their efforts to pursue care goals. Motivational positioning and its “philosophy” would be a relevant tool to enable nurses to approach the care of patients who suffer from unresolved issues of addiction from another perspective. 4.3 The advantages of motivational techniques in the cancer care process for patients suffering from unresolved addiction problems The motivational techniques employed by two of the health professionals in the sample helped support the patient’s self-esteem (for example, a feeling of self-efficacy) and elicited, among other things, his/her active participation. As Benarous et al. pointed out, a professional caregiver trained in MI did not experience an increase in his/her appointment time or frequency [5]. However, this approach is still not well known in France outside the sphere of addictology, which corresponds to the results of this research. According to Ong et al, MI has a prime role to play in these intervention programs in order to enable patients to solve their issues of ambivalence and identify barriers to smoking cessation, among others. It is recommended that the first interview be held as close as possible to the announcement of the cancer diagnosis; the moment of diagnosis is favorable to the promotion of smoking cessation. The second most suitable time is during the patient's hospitalization. In both periods, patients have increased interaction with healthcare professionals. Nurses can support them in the quest to quit smoking [24]. Furthermore, as our results have illustrated, most of the nurses we met believed that the patient needed to be protected during this period. They held such a belief almost certainly because of their representations on the effects of the cancer announcement on the psychological sphere of the patient. They described a patient who is stupefied and profoundly stricken emotionally. This perception did not engage them in a patient-support approach to addressing any addictive disorders in progress. They felt they were accompanying the patient, which put them at risk in their own nursing role. On the other hand, for professionals trained in motivational positioning, the approach greatly differed. Their care was adapted to the fragility generated by the announcement of the cancer diagnosis. Still, according to Ong et al., the majority of patients undergoing cancer treatment who are still active smokers generally fail to treat it themselves. Many would like to quit, but the majority still smoke. A combination of MI and substitution treatments, initiated as close as possible to diagnosis, is more likely to make withdrawal permanent [24]. This demonstrates the amount of work that remains to be done with medical and nursing teams managing patients treated for cancer and suffering from addictive disorders, in order to change their representations in terms of times conducive to intervention (of smoking in particular). With regard to tobacco, overall care must be organized at all levels of the care pathway, including actions to prevent and manage smoking in cancer care facilities [25]. These recommendations should be incorporated into good cancer practice, in accordance with the latest guidance from several organizations such as the American Society of Clinical Oncology (ASCO) and the Association for Cancer Research (AACR) [30]. Developing patient-centered care on the care pathway in oncology is necessary to be able to take account of unresolved addictive disorders, where they exist. A paradigm shift is needed to improve communication between healthcare professionals and patients. Motivational positioning is a posture of care that promotes this communication, in addition to supporting his/her empowerment. This research suggests the need to adapt the cancer care pathway in order to integrate the necessary care for patients who concomitantly suffer from unresolved addictive disorders. 4.3.1 Limitations of the study. The results of our research should be compared with similar studies at other cancer treatment facilities. The sample did not include ‘advanced practice nurses’ because, as a new profession, French establishments generally do not employ them. We suggest that research should be pursued on a wider scale, in several comprehensive cancer centers, when ‘advanced practice nursing’ becomes more prevalent in cancer treatment facilities to collect information for further findings. The small size of the sample cannot be used to extract representative quantitative statistical data due to lack of power. However, the sample size is compatible with reliable qualitative analysis [31]. 4.1 Care centered on cancer at the expense of addiction management Hamant et al. conducted one of the few French studies of dealing with alcohol and tobacco addiction in patients undergoing treatment for ear, nose or throat cancer [22]. Less than 1% of patients treated for cancer reported that they were informed about appointments to help with tobacco dependence or possible support in the management of alcohol consumption. Healthcare professionals involved in the cancer care pathways did not seem invested in the opportunity to support patients with this type of care, complementary to the treatment of their cancer [22]. The professionals interviewed in this qualitative study corroborated these facts. Caregiver management was central to cancer and did not include the possibility that the patient could also be suffering from unresolved addictive disorders. However, the Oncology Nursing Society (ONS) has shown that the management of tobacco addiction has been essential in oncology nursing. It has also established that the role of the nurse in the cancer care pathway is primordial in supporting patients who are experiencing withdrawal or who must wean themselves from their tobacco addiction [23]. However, smoking cessation programs have only recently been integrated as an essential part of cancer management. Active smokers treated for cancer have had poorer outcomes (reduced survival rates and weaker responses to cancer treatments) than those who were able to quit [24]. Results from this ancillary study showed that the oncology nurses often felt helpless to manage and support patients with unresolved addictive disorders. Recent studies have highlighted the need to integrate comprehensive patient care into cancer care pathways. Dauchy and Curé explained that this involves the systematic detection of the patient's vulnerability criteria, to include him/her in a risk group. This would strengthen the care process. One of the criteria was the search for persistent risk behaviors during cancer disease [25]. For this, as evidenced by the findings of this study, a change in the representations of caregivers on the patient treated for a cancer would be necessary. Nurses would also need to change the way they connect with the patient to be able to manage them more holistically, in a patient-centered manner of care, throughout the cancer care pathway. 4.2 The need to change practitioners’ attitudes to enable them to provide appropriate care to patients In France, the paternalistic cultural model of care was the predominant approach until recently. The sensitive relationship often remained characterized by an asymmetrical relationship. This aspect of the nursing position was clearly highlighted in this study. Nurses were often challenged because of their relationship with the patient. This relationship was not one of equality. Conversely, the patient-centered approach aimed to foster patient engagement by placing the caregiver and the patient in a more symmetrical relationship that encouraged the patient to be able to express himself/herself freely, without fear of judgement, on subjects of importance to him/her. Nurses tended to be locked in a pattern of their traditional role and found it decidedly difficult to connect with the patient differently [26]. The results of this research have supported what has been highlighted in the literature. It is well known that the traditional mode of communication focused on a directive approach to delivering information and advice is becoming less effective in maintaining change and significantly changing behavior [5]. Studies have shown that when doctors provided their patients with useful health information, while responding to their emotions, patients experienced a sense of self-control and greater hope. This has had a directly positive impact on their quality of life as well as their survival [27]. The method of entering into a relationship with the patient has therefore proved decisive, with the possibility that the caregiver could let the patient engage and be involved in his/her own care continuum. Barello et al. explained that nursing was the key to patient engagement. This commitment was associated with patients’ perceptions of the positive attitudes of healthcare professionals on their self-care behaviors. This focused on the essential role of nurses in charge [28]. It also appeared that patients tended to no longer actively be actors in their care when they did not feel they were in a symmetrical relationship with the caregiver, due to a lack of information about their state of health, or as part of a paternalistic attitude from the care system [28]. Our results have demonstrated that nurses who were not trained in motivational positioning struggled to avoid judging patients. They were also expected to obtain concrete results in the management of withdrawal symptoms or to connect with patients affected by addictive behaviors. Some postures adopted by nurses created a "relationship gap" in the caregiver-patient relationship. The majority of nurses interviewed reported that it was difficult to care for patients suffering from addictions vis-à-vis the traditional way nurses treat patients undergoing cancer care. Certain people questioned this issue of communication with patients but were unable to find a response. Michiels related that this relationship gap was the result of attitudes of care such as directional postures, admonitory speech, warnings, reasoned persuasive attitudes, administering advice or even using moralizing words [29]. Situations of a relationship impasse (such as those described in a verbatim where the nurse could not find a suitable angle to help the patient out of ambivalence to discussing smoking) were extremely stressful for the caregivers. They experienced feelings of helplessness and powerlessness in their efforts to pursue care goals. Motivational positioning and its “philosophy” would be a relevant tool to enable nurses to approach the care of patients who suffer from unresolved issues of addiction from another perspective. 4.3 The advantages of motivational techniques in the cancer care process for patients suffering from unresolved addiction problems The motivational techniques employed by two of the health professionals in the sample helped support the patient’s self-esteem (for example, a feeling of self-efficacy) and elicited, among other things, his/her active participation. As Benarous et al. pointed out, a professional caregiver trained in MI did not experience an increase in his/her appointment time or frequency [5]. However, this approach is still not well known in France outside the sphere of addictology, which corresponds to the results of this research. According to Ong et al, MI has a prime role to play in these intervention programs in order to enable patients to solve their issues of ambivalence and identify barriers to smoking cessation, among others. It is recommended that the first interview be held as close as possible to the announcement of the cancer diagnosis; the moment of diagnosis is favorable to the promotion of smoking cessation. The second most suitable time is during the patient's hospitalization. In both periods, patients have increased interaction with healthcare professionals. Nurses can support them in the quest to quit smoking [24]. Furthermore, as our results have illustrated, most of the nurses we met believed that the patient needed to be protected during this period. They held such a belief almost certainly because of their representations on the effects of the cancer announcement on the psychological sphere of the patient. They described a patient who is stupefied and profoundly stricken emotionally. This perception did not engage them in a patient-support approach to addressing any addictive disorders in progress. They felt they were accompanying the patient, which put them at risk in their own nursing role. On the other hand, for professionals trained in motivational positioning, the approach greatly differed. Their care was adapted to the fragility generated by the announcement of the cancer diagnosis. Still, according to Ong et al., the majority of patients undergoing cancer treatment who are still active smokers generally fail to treat it themselves. Many would like to quit, but the majority still smoke. A combination of MI and substitution treatments, initiated as close as possible to diagnosis, is more likely to make withdrawal permanent [24]. This demonstrates the amount of work that remains to be done with medical and nursing teams managing patients treated for cancer and suffering from addictive disorders, in order to change their representations in terms of times conducive to intervention (of smoking in particular). With regard to tobacco, overall care must be organized at all levels of the care pathway, including actions to prevent and manage smoking in cancer care facilities [25]. These recommendations should be incorporated into good cancer practice, in accordance with the latest guidance from several organizations such as the American Society of Clinical Oncology (ASCO) and the Association for Cancer Research (AACR) [30]. Developing patient-centered care on the care pathway in oncology is necessary to be able to take account of unresolved addictive disorders, where they exist. A paradigm shift is needed to improve communication between healthcare professionals and patients. Motivational positioning is a posture of care that promotes this communication, in addition to supporting his/her empowerment. This research suggests the need to adapt the cancer care pathway in order to integrate the necessary care for patients who concomitantly suffer from unresolved addictive disorders. 4.3.1 Limitations of the study. The results of our research should be compared with similar studies at other cancer treatment facilities. The sample did not include ‘advanced practice nurses’ because, as a new profession, French establishments generally do not employ them. We suggest that research should be pursued on a wider scale, in several comprehensive cancer centers, when ‘advanced practice nursing’ becomes more prevalent in cancer treatment facilities to collect information for further findings. The small size of the sample cannot be used to extract representative quantitative statistical data due to lack of power. However, the sample size is compatible with reliable qualitative analysis [31]. 4.3.1 Limitations of the study. The results of our research should be compared with similar studies at other cancer treatment facilities. The sample did not include ‘advanced practice nurses’ because, as a new profession, French establishments generally do not employ them. We suggest that research should be pursued on a wider scale, in several comprehensive cancer centers, when ‘advanced practice nursing’ becomes more prevalent in cancer treatment facilities to collect information for further findings. The small size of the sample cannot be used to extract representative quantitative statistical data due to lack of power. However, the sample size is compatible with reliable qualitative analysis [31]. 5 Conclusions It is becoming essential in France that nurses change their practices to support patient engagement in the management of unresolved addictive disorders along the oncology care pathway. This must involve strengthening their training in this domain, specifically the dissemination of MI techniques to support patient motivation, precisely considering the traditional positioning of nurses who have acquired counterproductive effects. The results of this ancillary study illustrate that such improvements are not yet integrated into French standard practices in oncology. The generalization of the MI and its tools, in current oncology practice, both in terms of physicians and nursing teams, would make it possible to improve the management of addictions in a patient-centered approach, support the empowerment of patients in the cancer care pathway and promote positive communication between caregivers and patients. Supporting information S1 Checklist. COREQ (COnsolidated criteria for Reporting Qualitative research) checklist. https://doi.org/10.1371/journal.pone.0242693.s001 (PDF) Acknowledgments The authors would like to thank Olivia Miyataki-Pellarin (Aloha Language Consulting) for translation and editing services.

Aguiar, Délia Cristina Figueira;Pereira, Washington Luiz Assunção;de Cássia Bastos de Matos, Gyselly;Silva, Klena Sarges Marruaz da;de Loiola, Rosane do Socorro Pompeu;Corvelo, Tereza Cristina Oliveira

doi: 10.1371/journal.pone.0241487pmid: 33206689

Introduction Antigens of the A, B and O blood groups are carbohydrate structures of glycoproteins and glycolipids. Several polymorphic genes are involved in regulating the synthesis of these glycoconjugates. The steps for their biosynthesis and chemical structure are explained by the relation between the ABO, H and Se systems [1]. ABO and H antigens may be formed from at least four types of precursor chains. In most tissues, the stages of biosynthesis of ABH structures are correlated with the embryological development of the tissue and its cellular differentiation. The expression of those varies from cell to cell and organ and organ [2], and the detection and location of ABH structures in various normal tissues in humans has been described in many studies [3]. The presence of ABH antigens was described initially on the surface of erythrocytes, although those are primarily histo-antigens. The distribution of those structures is extremely varied; from an evolutionary point of view they occur in tissues and fluids of more primitive mammalian species and appear in red blood cells only among the great apes and humans [4, 5]. Three α-2-fucosyltransferase genes (FUT1, FUT2, Sec1) has been characterized in primates [6]. They shared a high degree of DNA sequence identity, suggesting that they were generated by successive gene duplications and divergent evolution. In the course of evolution, the duplication event at the origin of H and Se genes occurred before the great mammalian radiation [7]. Besides, it has been proposed that an ancestral Se gene has been duplicated in two related genes, Se and Sec 1 [8]. Then FUT1 regulates the expression of H antigens on red cells and vascular epithelium. The secretor (Se) FUT2 regulates the expression of H antigen in the endoderm derived epithelium. There exists a third α-2-fucosyltransferase gene called Sec1, which is inactivation. In human Sec1 is a pseudogene but others nonsense inactivating mutation might have occurred in the Sec1 gene families, which may be responsible for the changes present in different species of New World Monkeys, suggesting the non-functionality as found for Callithrix, Aotus, and Pithecia whereas Sapajus and Saimiri have an active state for Sec 1 [9]. In this context, Old World Monkeys (OWM) and New World Monkeys (NWM) express ABH substances in secretions and body fluids, whereas in their red blood cells have only a factor related to the human B factor known as “like-B” [10]. In the specific case of ABH antigens derived from type 1 and 2 precursor chains, the type 1 chain is expressed mainly in tissues originating endodermally, under control of the Se gene, and found in fluids and bodily secretions [11]. The type 2 chain is expressed mainly in tissues of mesodermal and ectodermal origin, under control of gene H [12]. Regarding the distribution of these molecules in non-human primates, the studies are restricted to a few tissues and a few species [13–16]. The purpose of the present study was to clarify the distribution of ABH antigen expressions in several tissues of different New World Monkeys species of the Amazon region. Material and methods Animals Tissues from various organs of 19 New World primates of the species Aotus infulatus (n = 3), Callithrix jacchus (n = 6), Sapajus apella (n = 6) and Saimiri sciureus (n = 4) were obtained from the Centro Nacional de Primatas (Pará, Brazil) for this study. In all CENP non-human primate colonies, adequate environmental and nutritional conditions are maintained for the well-being of the animals and reproducing them, either to use them in scientific research or with the conservation of species. To achieve these objectives, there are reproductive, health and nutritional management in animals. Also, laboratory tests, routine clinical and surgical care, diagnostic imaging tests, microbiological control of conservation and reproduction areas and post mortem tests are carried out to investigate the cause of death of animals found dead in cages, as well as those who were already being treated at the clinic. The species studied were kept in indoor cages and received different types of environmental enrichment. All species received feeding enrichment daily with different food items such as larvae of mealworm (Tenebrio molitor), varied seasonal fruits and seeds, offered on alternate days. In the case of C. jachus and A. infulatus species, environmental enrichment with furniture and bedding material was also offered, such as nest boxes and pine bedding. Non-human primates received specific industrial animal feed (MegaZoo®) and were fed according to the management of each species adopted by the institution: different types of fruits, vegetables, roots, tubers, milk, eggs, vitamins, and mineral supplements into the water ad libitum. No animals in this study were euthanized. Most of the primates used were found dead on the floor of the cages due to natural death or a fight, while the rest were being treated at the clinic for different causes, such as abortion, cage fights and also due to the callitrichid slimming syndrome. The ones found dead in the morning were placed in refrigeration for the autopsy procedure by the pathologist responsible for investigating the cause of death. It is important to mention that both for samples of primates belonging to the collection of paraffinized tissues and for those obtained from animals that died during this research, none of them were participating in other scientific experiments, which was an exclusion criterion for due research. Through the analysis of the animals’ clinical files, necropsies, and the result of the histopathological analysis of the organs, Table 1 indicates the cause of death of each animal. Besides, another point that must be informed is that another work of this research has already been published, by the same research group, relating the expression of ABH antigens in the stomach and the association of infection by the bacterium Helicobacter sp. [12]. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Species used in the study, their medical records, identifications used in CENP and the causes of death of the animals. https://doi.org/10.1371/journal.pone.0241487.t001 The number of organs of the animals available among the species for use in the study varied, since some were obtained from the collection of tissues paraffinized at o CENP in 2004, while the remainder were collected during post-mortem autopsies on the animals. The samples collected during the autopsies were fixed in 10% buffered formaldehyde and processed in paraffin. Later, 5μm slices were made and placed on histological slides. Only tissues without any pathology were used in this study. Methods The content of the manuscript is original and it has not been published or accepted for publication, either in whole or in part, in any form. No part of the manuscript is currently under consideration for publication elsewhere. Ethics statement. This research was approved by the Environmental Agency (IBAMA N° 086/2004) and by the Research Ethics Committee of the Instituto Evandro Chagas (N° 069–2005). Immunohistochemistry. To determine the expression of A, B, and H antigens in primate organs in CENP we used the modified immunoperoxidase technique of Pedal et al., 1989 [17]. Anti-A, anti-B (Fresenius) monoclonal antibodies were used at a 1:10 dilution, while for H antigen, we used the Ulex europaeus lectin (Sigma—UEA I) linked to the peroxidase at a (1:50) dilution. The tissue slices that had been processed beforehand were then deparaffinized in xylol and treated with methanol containing 0.3% of H2O2. Next, the sections were washed in phosphate buffer (pH 7.6) and incubated with a blocking solution (phosphate buffer and bovine albumin). This was followed by incubation with the monoclonal antibodies and the U. europaeus lectin for one hour at room temperature, followed by washing with the phosphate buffer. The slides incubated with the anti-A and anti-B antibodies were treated with a blocking solution followed by a second incubation with anti-mouse IgG linked to peroxidase for one hour. After washing in phosphate buffer, the reaction was developed for all of the antigens (Tris buffer, diaminobenzidine, and hydrogen peroxide). The slides were stained with hematoxylin, dried, and mounted with Entellan® (Sigma). Slides were evaluated under a light microscope with results expressed as a score based on the percentage of the total field staining positively based of the following semi-quantitative scale: (+++) > 50% of the field showing positive staining; (++) < 50% of the field showing heterogeneous positive staining. Characterization of the animals’ blood group phenotype. The ABH phenotype of each animal was determined by the expression of these antigens in organs that produce fluids and secretions (stomach, intestines, bladder, and salivary glands) since all primates are secretors of ABH substances [6]. This type of determination has already been used in another study [15], where the saliva or other secretions from the animal were not available. Statistical analysis. The intra- and inter-specificities of NWM were estimated based on the variable ABH expression of antigens by the semi-quantitative scale (score). The means, standard deviations, and Shannon-Wiener Index were calculated using Bioestat 5.0 software [18]. Animals Tissues from various organs of 19 New World primates of the species Aotus infulatus (n = 3), Callithrix jacchus (n = 6), Sapajus apella (n = 6) and Saimiri sciureus (n = 4) were obtained from the Centro Nacional de Primatas (Pará, Brazil) for this study. In all CENP non-human primate colonies, adequate environmental and nutritional conditions are maintained for the well-being of the animals and reproducing them, either to use them in scientific research or with the conservation of species. To achieve these objectives, there are reproductive, health and nutritional management in animals. Also, laboratory tests, routine clinical and surgical care, diagnostic imaging tests, microbiological control of conservation and reproduction areas and post mortem tests are carried out to investigate the cause of death of animals found dead in cages, as well as those who were already being treated at the clinic. The species studied were kept in indoor cages and received different types of environmental enrichment. All species received feeding enrichment daily with different food items such as larvae of mealworm (Tenebrio molitor), varied seasonal fruits and seeds, offered on alternate days. In the case of C. jachus and A. infulatus species, environmental enrichment with furniture and bedding material was also offered, such as nest boxes and pine bedding. Non-human primates received specific industrial animal feed (MegaZoo®) and were fed according to the management of each species adopted by the institution: different types of fruits, vegetables, roots, tubers, milk, eggs, vitamins, and mineral supplements into the water ad libitum. No animals in this study were euthanized. Most of the primates used were found dead on the floor of the cages due to natural death or a fight, while the rest were being treated at the clinic for different causes, such as abortion, cage fights and also due to the callitrichid slimming syndrome. The ones found dead in the morning were placed in refrigeration for the autopsy procedure by the pathologist responsible for investigating the cause of death. It is important to mention that both for samples of primates belonging to the collection of paraffinized tissues and for those obtained from animals that died during this research, none of them were participating in other scientific experiments, which was an exclusion criterion for due research. Through the analysis of the animals’ clinical files, necropsies, and the result of the histopathological analysis of the organs, Table 1 indicates the cause of death of each animal. Besides, another point that must be informed is that another work of this research has already been published, by the same research group, relating the expression of ABH antigens in the stomach and the association of infection by the bacterium Helicobacter sp. [12]. Download: PPT PowerPoint slide PNG larger image TIFF original image Table 1. Species used in the study, their medical records, identifications used in CENP and the causes of death of the animals. https://doi.org/10.1371/journal.pone.0241487.t001 The number of organs of the animals available among the species for use in the study varied, since some were obtained from the collection of tissues paraffinized at o CENP in 2004, while the remainder were collected during post-mortem autopsies on the animals. The samples collected during the autopsies were fixed in 10% buffered formaldehyde and processed in paraffin. Later, 5μm slices were made and placed on histological slides. Only tissues without any pathology were used in this study. Methods The content of the manuscript is original and it has not been published or accepted for publication, either in whole or in part, in any form. No part of the manuscript is currently under consideration for publication elsewhere. Ethics statement. This research was approved by the Environmental Agency (IBAMA N° 086/2004) and by the Research Ethics Committee of the Instituto Evandro Chagas (N° 069–2005). Immunohistochemistry. To determine the expression of A, B, and H antigens in primate organs in CENP we used the modified immunoperoxidase technique of Pedal et al., 1989 [17]. Anti-A, anti-B (Fresenius) monoclonal antibodies were used at a 1:10 dilution, while for H antigen, we used the Ulex europaeus lectin (Sigma—UEA I) linked to the peroxidase at a (1:50) dilution. The tissue slices that had been processed beforehand were then deparaffinized in xylol and treated with methanol containing 0.3% of H2O2. Next, the sections were washed in phosphate buffer (pH 7.6) and incubated with a blocking solution (phosphate buffer and bovine albumin). This was followed by incubation with the monoclonal antibodies and the U. europaeus lectin for one hour at room temperature, followed by washing with the phosphate buffer. The slides incubated with the anti-A and anti-B antibodies were treated with a blocking solution followed by a second incubation with anti-mouse IgG linked to peroxidase for one hour. After washing in phosphate buffer, the reaction was developed for all of the antigens (Tris buffer, diaminobenzidine, and hydrogen peroxide). The slides were stained with hematoxylin, dried, and mounted with Entellan® (Sigma). Slides were evaluated under a light microscope with results expressed as a score based on the percentage of the total field staining positively based of the following semi-quantitative scale: (+++) > 50% of the field showing positive staining; (++) < 50% of the field showing heterogeneous positive staining. Characterization of the animals’ blood group phenotype. The ABH phenotype of each animal was determined by the expression of these antigens in organs that produce fluids and secretions (stomach, intestines, bladder, and salivary glands) since all primates are secretors of ABH substances [6]. This type of determination has already been used in another study [15], where the saliva or other secretions from the animal were not available. Statistical analysis. The intra- and inter-specificities of NWM were estimated based on the variable ABH expression of antigens by the semi-quantitative scale (score). The means, standard deviations, and Shannon-Wiener Index were calculated using Bioestat 5.0 software [18]. Ethics statement. This research was approved by the Environmental Agency (IBAMA N° 086/2004) and by the Research Ethics Committee of the Instituto Evandro Chagas (N° 069–2005). Immunohistochemistry. To determine the expression of A, B, and H antigens in primate organs in CENP we used the modified immunoperoxidase technique of Pedal et al., 1989 [17]. Anti-A, anti-B (Fresenius) monoclonal antibodies were used at a 1:10 dilution, while for H antigen, we used the Ulex europaeus lectin (Sigma—UEA I) linked to the peroxidase at a (1:50) dilution. The tissue slices that had been processed beforehand were then deparaffinized in xylol and treated with methanol containing 0.3% of H2O2. Next, the sections were washed in phosphate buffer (pH 7.6) and incubated with a blocking solution (phosphate buffer and bovine albumin). This was followed by incubation with the monoclonal antibodies and the U. europaeus lectin for one hour at room temperature, followed by washing with the phosphate buffer. The slides incubated with the anti-A and anti-B antibodies were treated with a blocking solution followed by a second incubation with anti-mouse IgG linked to peroxidase for one hour. After washing in phosphate buffer, the reaction was developed for all of the antigens (Tris buffer, diaminobenzidine, and hydrogen peroxide). The slides were stained with hematoxylin, dried, and mounted with Entellan® (Sigma). Slides were evaluated under a light microscope with results expressed as a score based on the percentage of the total field staining positively based of the following semi-quantitative scale: (+++) > 50% of the field showing positive staining; (++) < 50% of the field showing heterogeneous positive staining. Characterization of the animals’ blood group phenotype. The ABH phenotype of each animal was determined by the expression of these antigens in organs that produce fluids and secretions (stomach, intestines, bladder, and salivary glands) since all primates are secretors of ABH substances [6]. This type of determination has already been used in another study [15], where the saliva or other secretions from the animal were not available. Statistical analysis. The intra- and inter-specificities of NWM were estimated based on the variable ABH expression of antigens by the semi-quantitative scale (score). The means, standard deviations, and Shannon-Wiener Index were calculated using Bioestat 5.0 software [18]. Results Immunohistochemistry of the substances in the human-type ABO blood groups of NWM All of the animals of the species C. jacchus and A. infulatus belonged to blood groups A and B, respectively. In the species S. apella, four animals had the phenotype for blood group A and two for phenotype B. In the four individuals for the species S. sciureus tested, two had phenotype A and the rest were phenotype AB. In all of the organs tested the ABH expressions were concordant amongst themselves, in relation to the ABO blood group phenotype. Immunohistochemical distribution of antigens of human-type ABH blood groups in different species of NWM Only two of the organs studied (adrenal and heart) did not have expression of antigens from ABO blood groups. The expression found in the organs and tissues of each species analyzed are distributed according to the embryological origin of the tissue and detailed in Tables 2 and 3. Two patterns of antigen expression were observed in histological analyses. One pattern was homogeneous, with expression of the antigen in the entire area of tissue analyzed, and the other was heterogeneous, which areas completely marked as well as areas where the antigen did not occur. The antigen expressions in some of the tissue areas analyzed in the species studied are illustrated in Figs 1 and 2. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Expression of ABH antigens in A. infulatus and C. Jacchus. (a) anti-B in the basal layer (arrows) of the esophagus; (b) anti-H at the superficial (S) and intermediate (I) layers of the gastric mucosa; (c) anti-H in goblet cells of the small intestine; (d) anti-H in the bladder epithelium; (e) anti-A in the renal loop of Henle (arrows) - 400x; (g) anti-A epithelium in the prostatic glands. 100x; (h) anti-A in the epithelium (brown arrows) of the seminal vesicle and spermatozoids (black arrow). 1000x; (h) anti-H in the major and minor hepatic ducts (arrows) - 400x. https://doi.org/10.1371/journal.pone.0241487.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. ABH antigens in S. apella and S. sciureus in organs with an endodermal origin. Pharynx (a) anti-H, including the basal (arrows), 400x; (b) anti-H in ducts (D), serous acini (SA), mucous (MA) of the salivary gland, 400x; (c) anti-A in the prostatic epithelium, 100x; (d) anti-B in goblet cells of the large intestine, 400x; (e) anti-H in the vaginal epithelium, 400x; (f) anti-H in the basal layer of the bladder (arrow), 400x; (g) Anti-B in the greater ducts (brown arrow) and lesser ducts (black of arrow) of the liver in S. apella, 400x; (h) anti-B in the gastric mucosa, 400x. https://doi.org/10.1371/journal.pone.0241487.g002 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Expression of ABH antigens in organs with an endodermal origin in species of New World Monkeys. https://doi.org/10.1371/journal.pone.0241487.t002 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Expression of ABH antigens in organs of mesodermal origin in species of New World Monkeys. https://doi.org/10.1371/journal.pone.0241487.t003 Aotus infulatus. All of the animals are monomorphic belonging to the human B type group. One study [19] found that A. infulatus expressed only B phenotype. Table 1 lists the pattern for expression of antigens B and H depending on the type of cell or tissue layer. For example, only B antigen was detected in the basal layer of the esophagus (Fig 1a), while substance H was also expressed in the gastric mucosa (Fig 1b), in goblet cells of the small intestine (Fig 1c), and in intermediary and basal cells of the epithelium in the bladder (Fig 1d). Callithrix jacchus. In this species, only animals of group A human blood type were found. In C. jacchus the ABO system was monomorphic, where all individuals were classified as A phenotype, as reported by Rocha et al. (1990) [19] and Schneider et al. (1985) [20]. The A antigen was expressed in some areas such as the loop of Henle in renal tissue (Fig 1e), in the prostate (Fig 1f), in the seminal vesicles and spermatozoids (Fig 1g). H antigen was also expressed in the hepatic tissue (Fig 1h). Sapajus paella. Animals were found with phenotypes of human blood groups type A and type B. But, in this species, all four phenotypes (A, B, AB, O) have been found as described in Schneider et al. (1985) [20] and Corvelo et al. (2002) [21]. The H antigen was in basal cells of the pharyngeal epithelium (Fig 2a) and in the ducts, mucosal and serous acini of the salivary gland (Fig 2b). The expression of the A antigen in the prostatic epithelium is demonstrated in Fig 2c while the expression of B antigen is demonstrated in Fig 2d in the small intestine, especially in goblet cells. Saimiri sciureus. In this species, human-type A and AB types were found. However, the squirrel monkeys were observed to have all four phenotypes [22, 23]. The expression of H antigen is identified in the basal layer of the vaginal epithelium (Fig 2e) and bladder epithelium (Fig 2f). The B antigen may be found in the major and minor hepatic ducts in the phenotype AB animal (Fig 2g) and in all layers of the gastric mucosa (Fig 2h). Only the B antigen may be detected in the white pulp of the spleen in an individual of phenotype AB. Intra-and interspecific diversity of the ABH expression in NWM. As can be seen in Table 4, compatible ABH expression was verified in all animals examined. Based on score ABH from histological expression distributions in both origin tissues (endo-/mesodermal), the occurrence of maximum diversity revealed more heterogeneity of H antigens than A/B antigens, which seemed to show homogeneity (Fig 3). Moreover, in some of the different cell/tissue layers of endo- and mesodermal organs were found more diversity to B antigen expression in A. infulatus than S. sciureus and S. apella. Nevertheless, in selection to A antigen, the higher estimated diversity indexes divide the analyzed species in C. jacchus from S. sciureus and S. apella. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Schematic drawing of comparison of ABH histoblood group expression in several tissues from New World Monkey and human. https://doi.org/10.1371/journal.pone.0241487.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 4. Histological expression score of ABH antigens in organs of endo-and mesodermal origin in New World Monkey species. https://doi.org/10.1371/journal.pone.0241487.t004 In this way, the statistical analysis by (Shannon-Wiener Index) showed that the diversity is confirmed by a scenario, in which the history of the species is result of differences in selection pressures among these closely related species so that the variation at ABO system reflects a multiallelic balanced polymorphism. Immunohistochemistry of the substances in the human-type ABO blood groups of NWM All of the animals of the species C. jacchus and A. infulatus belonged to blood groups A and B, respectively. In the species S. apella, four animals had the phenotype for blood group A and two for phenotype B. In the four individuals for the species S. sciureus tested, two had phenotype A and the rest were phenotype AB. In all of the organs tested the ABH expressions were concordant amongst themselves, in relation to the ABO blood group phenotype. Immunohistochemical distribution of antigens of human-type ABH blood groups in different species of NWM Only two of the organs studied (adrenal and heart) did not have expression of antigens from ABO blood groups. The expression found in the organs and tissues of each species analyzed are distributed according to the embryological origin of the tissue and detailed in Tables 2 and 3. Two patterns of antigen expression were observed in histological analyses. One pattern was homogeneous, with expression of the antigen in the entire area of tissue analyzed, and the other was heterogeneous, which areas completely marked as well as areas where the antigen did not occur. The antigen expressions in some of the tissue areas analyzed in the species studied are illustrated in Figs 1 and 2. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 1. Expression of ABH antigens in A. infulatus and C. Jacchus. (a) anti-B in the basal layer (arrows) of the esophagus; (b) anti-H at the superficial (S) and intermediate (I) layers of the gastric mucosa; (c) anti-H in goblet cells of the small intestine; (d) anti-H in the bladder epithelium; (e) anti-A in the renal loop of Henle (arrows) - 400x; (g) anti-A epithelium in the prostatic glands. 100x; (h) anti-A in the epithelium (brown arrows) of the seminal vesicle and spermatozoids (black arrow). 1000x; (h) anti-H in the major and minor hepatic ducts (arrows) - 400x. https://doi.org/10.1371/journal.pone.0241487.g001 Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 2. ABH antigens in S. apella and S. sciureus in organs with an endodermal origin. Pharynx (a) anti-H, including the basal (arrows), 400x; (b) anti-H in ducts (D), serous acini (SA), mucous (MA) of the salivary gland, 400x; (c) anti-A in the prostatic epithelium, 100x; (d) anti-B in goblet cells of the large intestine, 400x; (e) anti-H in the vaginal epithelium, 400x; (f) anti-H in the basal layer of the bladder (arrow), 400x; (g) Anti-B in the greater ducts (brown arrow) and lesser ducts (black of arrow) of the liver in S. apella, 400x; (h) anti-B in the gastric mucosa, 400x. https://doi.org/10.1371/journal.pone.0241487.g002 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 2. Expression of ABH antigens in organs with an endodermal origin in species of New World Monkeys. https://doi.org/10.1371/journal.pone.0241487.t002 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 3. Expression of ABH antigens in organs of mesodermal origin in species of New World Monkeys. https://doi.org/10.1371/journal.pone.0241487.t003 Aotus infulatus. All of the animals are monomorphic belonging to the human B type group. One study [19] found that A. infulatus expressed only B phenotype. Table 1 lists the pattern for expression of antigens B and H depending on the type of cell or tissue layer. For example, only B antigen was detected in the basal layer of the esophagus (Fig 1a), while substance H was also expressed in the gastric mucosa (Fig 1b), in goblet cells of the small intestine (Fig 1c), and in intermediary and basal cells of the epithelium in the bladder (Fig 1d). Callithrix jacchus. In this species, only animals of group A human blood type were found. In C. jacchus the ABO system was monomorphic, where all individuals were classified as A phenotype, as reported by Rocha et al. (1990) [19] and Schneider et al. (1985) [20]. The A antigen was expressed in some areas such as the loop of Henle in renal tissue (Fig 1e), in the prostate (Fig 1f), in the seminal vesicles and spermatozoids (Fig 1g). H antigen was also expressed in the hepatic tissue (Fig 1h). Sapajus paella. Animals were found with phenotypes of human blood groups type A and type B. But, in this species, all four phenotypes (A, B, AB, O) have been found as described in Schneider et al. (1985) [20] and Corvelo et al. (2002) [21]. The H antigen was in basal cells of the pharyngeal epithelium (Fig 2a) and in the ducts, mucosal and serous acini of the salivary gland (Fig 2b). The expression of the A antigen in the prostatic epithelium is demonstrated in Fig 2c while the expression of B antigen is demonstrated in Fig 2d in the small intestine, especially in goblet cells. Saimiri sciureus. In this species, human-type A and AB types were found. However, the squirrel monkeys were observed to have all four phenotypes [22, 23]. The expression of H antigen is identified in the basal layer of the vaginal epithelium (Fig 2e) and bladder epithelium (Fig 2f). The B antigen may be found in the major and minor hepatic ducts in the phenotype AB animal (Fig 2g) and in all layers of the gastric mucosa (Fig 2h). Only the B antigen may be detected in the white pulp of the spleen in an individual of phenotype AB. Intra-and interspecific diversity of the ABH expression in NWM. As can be seen in Table 4, compatible ABH expression was verified in all animals examined. Based on score ABH from histological expression distributions in both origin tissues (endo-/mesodermal), the occurrence of maximum diversity revealed more heterogeneity of H antigens than A/B antigens, which seemed to show homogeneity (Fig 3). Moreover, in some of the different cell/tissue layers of endo- and mesodermal organs were found more diversity to B antigen expression in A. infulatus than S. sciureus and S. apella. Nevertheless, in selection to A antigen, the higher estimated diversity indexes divide the analyzed species in C. jacchus from S. sciureus and S. apella. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Schematic drawing of comparison of ABH histoblood group expression in several tissues from New World Monkey and human. https://doi.org/10.1371/journal.pone.0241487.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 4. Histological expression score of ABH antigens in organs of endo-and mesodermal origin in New World Monkey species. https://doi.org/10.1371/journal.pone.0241487.t004 In this way, the statistical analysis by (Shannon-Wiener Index) showed that the diversity is confirmed by a scenario, in which the history of the species is result of differences in selection pressures among these closely related species so that the variation at ABO system reflects a multiallelic balanced polymorphism. Aotus infulatus. All of the animals are monomorphic belonging to the human B type group. One study [19] found that A. infulatus expressed only B phenotype. Table 1 lists the pattern for expression of antigens B and H depending on the type of cell or tissue layer. For example, only B antigen was detected in the basal layer of the esophagus (Fig 1a), while substance H was also expressed in the gastric mucosa (Fig 1b), in goblet cells of the small intestine (Fig 1c), and in intermediary and basal cells of the epithelium in the bladder (Fig 1d). Callithrix jacchus. In this species, only animals of group A human blood type were found. In C. jacchus the ABO system was monomorphic, where all individuals were classified as A phenotype, as reported by Rocha et al. (1990) [19] and Schneider et al. (1985) [20]. The A antigen was expressed in some areas such as the loop of Henle in renal tissue (Fig 1e), in the prostate (Fig 1f), in the seminal vesicles and spermatozoids (Fig 1g). H antigen was also expressed in the hepatic tissue (Fig 1h). Sapajus paella. Animals were found with phenotypes of human blood groups type A and type B. But, in this species, all four phenotypes (A, B, AB, O) have been found as described in Schneider et al. (1985) [20] and Corvelo et al. (2002) [21]. The H antigen was in basal cells of the pharyngeal epithelium (Fig 2a) and in the ducts, mucosal and serous acini of the salivary gland (Fig 2b). The expression of the A antigen in the prostatic epithelium is demonstrated in Fig 2c while the expression of B antigen is demonstrated in Fig 2d in the small intestine, especially in goblet cells. Saimiri sciureus. In this species, human-type A and AB types were found. However, the squirrel monkeys were observed to have all four phenotypes [22, 23]. The expression of H antigen is identified in the basal layer of the vaginal epithelium (Fig 2e) and bladder epithelium (Fig 2f). The B antigen may be found in the major and minor hepatic ducts in the phenotype AB animal (Fig 2g) and in all layers of the gastric mucosa (Fig 2h). Only the B antigen may be detected in the white pulp of the spleen in an individual of phenotype AB. Intra-and interspecific diversity of the ABH expression in NWM. As can be seen in Table 4, compatible ABH expression was verified in all animals examined. Based on score ABH from histological expression distributions in both origin tissues (endo-/mesodermal), the occurrence of maximum diversity revealed more heterogeneity of H antigens than A/B antigens, which seemed to show homogeneity (Fig 3). Moreover, in some of the different cell/tissue layers of endo- and mesodermal organs were found more diversity to B antigen expression in A. infulatus than S. sciureus and S. apella. Nevertheless, in selection to A antigen, the higher estimated diversity indexes divide the analyzed species in C. jacchus from S. sciureus and S. apella. Download: PPT PowerPoint slide PNG larger image TIFF original image Fig 3. Schematic drawing of comparison of ABH histoblood group expression in several tissues from New World Monkey and human. https://doi.org/10.1371/journal.pone.0241487.g003 Download: PPT PowerPoint slide PNG larger image TIFF original image Table 4. Histological expression score of ABH antigens in organs of endo-and mesodermal origin in New World Monkey species. https://doi.org/10.1371/journal.pone.0241487.t004 In this way, the statistical analysis by (Shannon-Wiener Index) showed that the diversity is confirmed by a scenario, in which the history of the species is result of differences in selection pressures among these closely related species so that the variation at ABO system reflects a multiallelic balanced polymorphism. Discussion In this study, the tissue distribution pattern of human ABH type in NWP was similar to patterns of expression in humans. In that regard, the results analyzed related to the expression of this histo-antigens in tissue structures of endodermal and mesodermal origin in the species S. sciureus, A. infulatus, S. apella and C. jacchus, revealed some findings regarding the reactivity of these ABH antigens, defining distinct patterns of antigen expression because of this intra and interspecies diversity. In a comparison of ABO status and ABH immunohistological expression antigens of NWM species and human were observed that these distributions seemed to be dependent on the cell/tissue type, where these ABH antigens are formed by the sequential addition of monosaccharides to the growing chain in relations to cellular differentiation as proposed by Ravn and Dabelsteen (2000) [3]. Our findings are also consistent with their hypothesis. However, some changes into the ABH patterns of the expression antigens were found in various cell/tissue layers of the NWM species. A likely mechanism for these differences may be due to the regulation of the types of precursor carbohydrate chains and the activity of glycosyltransferase enzymes (FUT1; FUT2; Sec1; A/B genes) involved in the synthesis of A/B/H antigens (Fig 3). In the overall framework of this study, for each ABH antigen, there seems to be a pattern of tissue expression, conserved and restricted to species, probably because of differentiation in the tissue organization. This results from gene rearrangements and polymorphisms with antigenic activity during morphogenesis. Because of this biochemical diversity, one may view the ABH antigens preferentially as immunological macromolecules, which determine an active role at the cell surface, and thus participate in cell migration and mutual recognition, which are indispensable for biological processes. In this study, comparison of the results was done with the assumption that the genetic pattern of the expression of these ABH type human carbohydrate antigens. These antigens would exhibit the same type of genetic control found in humans, marked by preferential use of the type 1 chain by the human-type FUT2 enzyme (Se), in tissues with endodermal origin, and by action of the human-type FUT1 enzyme (H) in the type 2 chain, in tissues with an ectodermal and mesodermal origin [15]. Furthermore, in this interpretation, one should note the data from analyses of the sequences in Sec1 in primates [16], which supposedly also codify for a potentially functional α-2-fucosiltransferase in OWM and NWM, although it is a pseudogene in humans. Those assumptions were adopted to inform these first studies and identify the probable divergences between these non-human primates and humans. From there, one may better understand how these differences in ABH expression depend upon an evolutionary pattern of gene regulation. Investigation of the expression of ABH antigens in primate tissues is restricted to a handful of studies. These demonstrate expression in epithelial tissues, endothelium and secretions in new and OWP, using antibodies and lectins [13–15], without, however, a detailed description of reactive cell types in the tissue of each of the species studied, as has been done exhaustively with humans [3]. Expression of these antigens in tissues is related to the embryonic origin, cell differentiation and gene control in the H and Se systems of the tissues involved [11, 12]. In this study, with regard to the wide distribution of these antigens, the expression of the H antigen showed great diversity, particularly of the intra-specific type. This expressed variation is because this H antigen serves as a precursor for production of the A and B antigens. Furthermore, one cannot ignore that the distinct expression patterns of substance H are the result of the type of gene regulation, which may be dependent on the synthesis of three distinct α-2-fucosiltransferases that are the product of genes H, Se and Sec1 described in NWM [6]. Another aspect that must be emphasized are the discrepancies observed in comparative patterns between humans and NWM, in different tissue structures with distinct embryonic origins. The stratified epitheliums of the esophagus and pharynx demonstrated a heterogeneous expression of at least one of the ABH antigens in the basal layer. This observation is different from what has been reported humans [24, 25], where only the expression of the H precursor was found in a few parabasal cells [26]. Considering the constant cell renewal in those epitheliums, the basal layer, with a view to cellular differentiation, accordingly modulates the synthesis of new antigenic ABH structures on the cell surface, which reinforces this differentiation in the antigenic intra and interspecies expression. In the gastric mucosa, it was found that the difference in the expression of the H antigen, when compared with data from humans, was related to antigen distribution, with more homogeneity in tissue expression found in humans [27, 28]. Although Ravn and Dabelsteen (2000) [3] did not indicate the presence of the antigen in cells small intestine cells in humans, Shimamoto et al. in 1987 [29] revealed reduced and rare expression patterns for the H antigen. In this study, the results found for the H antigen were of a more intense marking in goblet cells than that reported for humans by Shimamoto et al. (1987) [29]. Studies performed in human large intestines describe ABH antigens being found in absorbent and goblet cells, with a tendency towards higher expression in goblet cells similar to what was observed in the species analyzed in this study [24, 25, 30, 31]. In the bladder epithelium, the detection of ABH antigens differed between the primate species studied, with some species having a reduction and/or absence of expression. In terms of humans, expression of these antigens at the surface layer also showed variation with loss or reduction in antigen expression [24, 25, 32, 33] as described above for the monkeys investigated. This variation observed in surface layer cells in the bladder may be a sign of a process of cell aging (desquamation) that leads to the concurrent disappearance of antigens. In all the species studied, expression of the H antigen was in most cases detected in the mucous glands of the salivary glands. While in the ducts and serous glands, this pattern varied considerably between the species. It may be due to the chemical nature of the ABH substances produced by the different glands, where mucous glands have a production rich in glycoproteins, while in serous glands their concentration is more diluted, making detection of those antigens more difficult [34]. In the animals’ livers antigen marking was found in the major and minor ducts. In humans, studies indicate that marking is not found in the minor ducts [34]. Therefore, those found positive ABH marking in both ducts apparently are reproducing a distribution pattern found in the fetal and post-natal phase of development observed in humans, which in the case of the primates seems to have persisted in both structures in the organs until the adult phase [35]. In the follicular cells of the thyroid gland of the species analyzed, there was no expression of ABH antigens, as also described for humans [3, 24, 25]. As for the vaginal epithelium, the ABH expression pattern is different from humans, where the expression occur in the upper layer with a tendency towards reduction of expression in the intermediate layer and absence in the basal layer [24, 25]. Among the studied monkeys, many basal cells in the vaginal epithelium are capable of expressing ABH already at that level of differentiation. These findings thus seem to demonstrate that this profile of expression in the basal layer results from the ontogenetic process, as described for hepatic ducts. In the prostatic epithelium, there was no expression of the H antigen in the studied monkey species. In humans, descriptions of the presence of H antigen in this epithelium differ among some of the studies. One study [36] did not find expression of this antigen, unlike what was cited in the review by Ravn and Dabelsteen, 2000 [3]. It is thus very likely that non-detection of the H antigen may be a common event, given that this substance is a precursor for antigens in the ABO locus. In the human uterine endometrium and uterine tube, hormonal influences control the growth, differentiation and expression of ABH antigens. The increase in expression of these antigens occurs mainly in the late and secretory proliferative phases [37]. In the study, the female monkeys were in intermediate and late stages of the proliferative phase in the reproductive cycle. The endometrium and uterine tube expressed the ABH antigens in a very similar manner among the species studied, and in humans [3]. Welshinger et al. in 1996 [38] note that in humans, ABH antigens are normally not found in the ovarian epithelium. On the other hand, when there is hormonal activation in this epithelium, a change in the cellular morphology is induced with synthesis of new proteins, including ABH antigens, as found in this analysis. In males, the testicles and related structure examined in Callithrix and Sapajus showed differences in detection of ABH antigens between the species. Thus, a larger sampling number was necessary in order to establish a pattern for ABH expression. However, C. jacchus was the species presenting the greatest similarity in relation to the marking profile described for these humans [39]. In human spleen cells, there is no presence of ABH antigens [24, 25]. This pattern was also reproduced in the majority of the monkeys tested. However, two animals species were found Sapajus and Saimiri, with reactivity for the A and BH antigens, respectively in the white pulp, and for A, B and H in the red pulp Saimiri. This shows a distinct cell-specific pattern type in those areas. Although one may point out that, no histological modification was observed that justify this change in the pattern for histological expression. The appearance of these molecules may be the result of a positive cellular regulation for antigenic activity and the persistence of this phylogenetic level [40]. Human renal tissue shows a permanent expression of ABH antigens at the level of collecting and distal ducts and tubules, but not in the loop of Henle [41, 42]. In the primates studied, the detection of these antigens was also observed in this structure in 98% of the animals. That characteristic indicates a differentiated antigenic expression in this structure between members of the families studied and humans. It was found that differences in the expression of ABH antigens between New World monkey species and humans. This occur mainly in the structures of organs that have undergone ontogenetic, morphological and functional differences. Thus, constant antigenic reactivity was detected on many epithelial cells of all the tissues and organs, which have secretory functions. Thus, they are under apparent control of the like-human FUT-2 (Se) enzyme. On the other hand, a tendency towards the absence or reduction of A/B antigens, and particularly of the H precursor substance was a typical finding in organs of mesodermal origin that apparently are under the control of the like-human FUT-1 (H) enzyme. It is thus plausible that specific predictions regarding the functions of these ABH blood group antigens may be derived from the results of analyses of consistent species-specific variations in the tissue distribution of these antigens. This tissue distribution with reactivity patterns that are similar and conserved between the different species, reinforce the evidence of the evolutionary impact of ABH antigens. The ABO glcosyltransferase and α-2-fucosyltransferase families’ genes are polymorphic at least in humans. Although in some primates species these genes are monomorphic or show low-frequency polymorphism (i.g. C. jacchus and A. infulatus). In this case, is expected that the genic product act at the cellular level where they are indispensable in cell-cell interactions, occurring during differentiation, organogenesis and in the inflammatory process, then these genes are essential and so conserved among the species. In contrast, the multiantigenic variations (or intraspecies diversity) are mainly present on epithelial cells in contact with the external environment. The biological meaning is that various microorganism bind to the epithelial cells via glicoconjugate, expressing either precursor of ABH antigens (non-secretor) or the ABH antigens themselves (secretor). These polymorphisms might also split the population between sensitive and resistant individuals to diseases, due to difference in selection pressures among the species. Furthermore, this study help to understand that the ABH antigens are the result of an ancient polymorphism, in which some alleles are shared among related species (such as S. apella and S. sciureus) and maintained under balancing selection.