Development of a clinical risk score for pain and function following total knee arthroplasty: results from the TRIO study

Development of a clinical risk score for pain and function following total knee arthroplasty:... Objectives To develop and validate a simple clinical prediction model, based on easily collected preoperative information, to identify patients at high risk of pain and functional disability 6 months after total knee arthroplasty (TKA). Methods This was a multi-centre cohort study of patients from 9 centres across the UK, who were undergoing a primary TKA for osteoarthritis. Information on socio-demographic, psychosocial, clinical, and quality of life measures were collected at recruitment. The primary outcome measure for this analysis was Oxford Knee Score, measured 6 months postoperatively by postal questionnaire. Multivariable logistic regression was used to develop the model. Model performance (discrimination and calibration) and internal validity was assessed, and a simple clinical risk score developed. Results 721 participants (mean age 68.3 years; 53% female) provided data for the current analysis and 14% had a poor outcome at 6 months. Key predictors were poor clinical status, widespread body pain, high expectation of postoperative pain, and lack of active coping. The developed model based on these variables demonstrated good discrimination. At the optimal cut-off, the final model had a sensitivity of 83%, specificity of 61%, and positive likelihood ratio of 2.11. Excellent agreement was found between observed and predicted outcomes, and there was no evidence of overfitting in the model. Conclusion We have developed and validated a clinical prediction model that can be used to identify patients at high risk of a poor outcome after TKA. This clinical risk score may be an aid to shared decision-making between patient and clinician. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 4 Keywords: Knee pain, Osteoarthritis, Total knee arthroplasty, Prediction Modelling, Clinical risk score, Model calibration, Model discrimination Key messages: • Predictors of poor outcome following total knee arthroplasty included illness attitudes and behaviours and clinical factors. • A model based on easily measurable variables demonstrates good performance. • The prediction tool developed can be an aid to shared decision-making between patient and clinician. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 5 Introduction Total knee arthroplasty (TKA) is one of the most common and effective treatments for severe knee osteoarthritis with over 100,000 knee replacements performed in the UK annually [1, 2]. Despite success in reduction of pain after knee replacements, approximately 20% to 30% of patients continue to experience pain and limited function after their TKA, which cannot be entirely explained by biomedical factors [3-5]. Clinical determinants of outcomes after TKA which have been consistently shown to be related to outcome across reviews include preoperative pain and function, pain at other sites, and aspects of surgery (longer duration of surgery, lengthy wait times) [6-12]. For other factors, the evidence is not consistent and may be related to the outcome studied. For example, one systematic review focusing on patients’ characteristics found that younger age and being male were related to risk of revision, older age was associated with increased risk of mortality and poorer function after TKA, but age and sex did not influence postoperative pain [13]. The importance of psychosocial and individual factors, as predictors of musculoskeletal outcomes has also been increasingly recognised [14-16]. Adverse psychological factors such as anxiety and depression may have an effect on pain perception and mediate the development of chronic pain and disability [17, 18]. The relationship between psychosocial factors and TKA outcomes has been examined in several systematic reviews, which have consistently indicated poor preoperative mental health and pain catastrophizing to be strongly associated with greater postoperative pain and functional disability [7, 10, 19]. Limited or conflicting evidence was found for other psychological factors. It is clear from the reviews that there is a lack of consensus on the most important clinical and psychological risk factors for poor outcomes after TKA. Although the decision to operate is primarily based on radiographic evidence of osteoarthritis and the patient’s report of symptoms, variation in the use of surgery reflects the different beliefs among patients and surgeons as to the risks and benefits of surgery. In a US-based study, Riddle et al. (2014) Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 6 reported that a third of cases reviewed that underwent knee replacement surgery were ‘inappropriate’ and as a group, these patients demonstrated worse outcomes [20]. The fact that surgery might not be successful for certain patients still highlights the need for robust predictive models to inform the clinical decision-making process. Therefore, our study aimed to firstly predict the impact of pain and functional disability 6 months after TKA using routinely-collected patient preoperative information and secondly, incorporate this information into a clinical prediction tool. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 7 Methods The Targeted Rehabilitation to Improve Outcome- preoperative predictors of unfavourable outcome following knee arthroplasty (TRIO-POPULAR) study was a multi-centre cohort study to investigate potential preoperative predictors of poor outcome following TKA. The study recruited from 9 participating centres across the UK between December 2013 and July 2016. The study was conducted alongside a randomised controlled trial of targeted rehabilitation to improve outcome after TKA [21]. Adults aged 16 years or over, undergoing primary TKA for osteoarthritis were invited to take part in the study either by letter or in person at a clinic visit prior to surgery. Participants were excluded if they: were undergoing a revision TKA or fully constrained knee arthroplasty; had a TKA for a diagnosis other than osteoarthritis; or had existing medical conditions such as stroke, or other musculoskeletal conditions that cause a limitation of function. Participants completed a questionnaire at the time of recruitment and consent was obtained for access to medical records for research purposes. Follow-up questionnaires were mailed to participants 6 weeks, 3, and 6 months after surgery. Ethical approval was granted by the office for Research Ethics Committees Northern Ireland (ORECNI) (13/NI/0101). The preoperative questionnaire included the following items: - Sociodemographic factors: Age, gender, marital status, socio-economic status (highest education level achieved), and employment status were measured. - Clinical factors: Clinical factors measured included duration of knee pain, baseline pain and function using the Oxford Knee Score (OKS) [22], and the Chronic Pain Grade (CPG) [23]. The CPG contains 7-items which allow respondents to be classified into five categories: Grade 0 (no pain), Grade I (low disability/low intensity), Grade II (low disability/high intensity), Grade III (high disability/moderately limiting intensity), and Grade IV (high disability and highly limiting disability). Body manikins were used to determine whether participants meet the Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 8 definition of ‘chronic widespread pain’ used in the American College of Rheumatology criteria for fibromyalgia [24]. The Sleep Problem Scale consists of four questions, rated on a 6-point frequency rating scale, ranging from 0 (not at all) to 5 ( 22 to 31 days/month) [25]. Sleep disturbance was defined as a mean score ≥ 4, corresponding to at least 15 troubled nights per month [25]. Self-reported comorbidities in this cohort were also recorded. - Psychosocial factors: The Illness Attitude Scales (IAS) [26, 27] measures personal attitudes, fears, and beliefs associated with hypochondriasis and abnormal illness behaviour. It consists of nine subscales, each with three items on a 0-4 Likert scale. Scores are summed to give the total IAS score, with a higher score representing greater hypochondriacal fears and beliefs. Among participants who reported they had aches or pains lasting one day or longer in the past month, the Vanderbilt Pain Management Inventory was used to assess chronic pain coping strategies [28]. This questionnaire consists of 18-items, rated on a 5-point frequency Likert scale. From this data, two subscales can be calculated; active coping score and passive coping score. High scores indicate a high use of active and passive coping strategies respectively. Patient expectations of pain, and limitations in everyday activities after TKA were measured using visual analogue scales (VAS); 0 representing ‘not at all painful’ or ‘not limited at all’ and 100 ‘very painful’ or ‘greatly limited’, respectively [29]. - Mental and Physical Health: Mental and physical health was measured by the Hospital Anxiety and Depression Scale (HADS) [30] and the Patient Reported Outcome Measurement System 10 (PROMIS-10) Global Health Questionnaire [31]. The HADS is a 14-item questionnaire; seven items measuring anxiety and seven items measuring depression. Each item is rated on a 0-3 Likert scale with higher scores indicating poorer mental health. The PROMIS-10 Questionnaire has 10-items which allow the Global Physical Health and Global Mental Health sub-scales to be derived. Scores range from 4 to 20 with higher scores indicating better health. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 9 - Quality of life: The EuroQoL-5 dimension (EQ-5D) is a measure of quality of life [32] . It consists of five dimensions; mobility, self-care, usual activities, pain/discomfort, and anxiety/depression, rated on a 3-point scale. Each EQ-5D profile was converted to a single summary index based on the valuation of health states in the UK. A score of 1.0 indicates the best possible health. The outcome for this analysis was the OKS [33], measured 6 months postoperatively by postal questionnaire. The OKS measures the impact of pain and functional disability in patients undergoing knee replacement [34, 35]. Poor outcome was defined by a score of 26 or less (out of a maximum score of 48) according to the modified Kalairajah classification [36]. Statistical analysis The study aimed to recruit 750 participants and if 80% of participants (n=600) provided follow-up data, this would give 80% power to detect an odds ratio of 1.5 for a poor outcome, comparing the highest with the other two tertiles of exposure. Descriptive statistics were carried out to describe the study sample and normality of individual variables were assessed. Categorical variables; the Sleep Problem Scale, CPG, and the HADS anxiety and depression were categorised according to standard cut-offs. In preparation for the modelling, the relationship between continuous predictor variables and the observed log odds of a poor outcome were assessed for linearity. Health scores measured by the EQ5D and the PROMIS-10 questionnaire, measures of active and passive coping strategies determined by the Vanderbilt Pain Management Inventory, patient expectations of outcomes after surgery, and illness attitude scores were analysed as continuous variables. However, a maximum health index of one in the EQ5D results in regression coefficients (expressed as change in outcome per one unit increase in predictor) which are not intuitive to interpret, and values were therefore multiplied by 10 for the purpose of the univariable and multivariable analyses. Logistic regression analysis was used to explore the association between each of the potential preoperative predictor Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 10 variables and the OKS at 6 months. In the univariable analysis, variables showing an association with a significance level of p < 0.2 were candidates for entry into a forward stepwise regression as part of a bootstrap selection process as described below. Entry and removal criteria for the stepwise models were p ≤ 0.1 and p > 0.15, respectively. We used stepwise regression to suggest predictor variables for the model followed by the incorporation of clinical knowledge. Associations were expressed as odds ratios (ORs) with 95% confidence intervals (95% CIs). To aid clinical decision- making, a simplified point-based risk-scoring system was developed using coefficients from the final model [37]. Multiple imputation with chained equations (MICE) was used to impute missing predictor data with the aim to reduce bias and improve efficiency; 20 imputed datasets were generated [38, 39]. Detailed descriptions of the post-estimation procedure can be found in Appendix 1 (electronic supplementary material). Model discrimination was quantified using the area under the Receiver Operating Characteristic (ROC) curve or concordance (c) statistic to estimate predictive accuracy. A c-statistic value of 1 represents perfect discrimination whilst a c-statistic of 0.5 indicates a discriminative value equivalent to chance [40]. A pooled c-statistic of the 20 imputed datasets, was calculated. A shrinkage estimate was also calculated to assess overfitting. A shrinkage estimate of less than 0.8 would reflect a need for shrinkage of the regression coefficients in a prediction model using methods such as Lasso or ridge regression [41]. Model calibration, which refers to the agreement between the observed and predicted probabilities, was also assessed using calibration-in-the-large [42]. This indicates whether the predictions are systematically too low or too high. Overfitting occurs when a model is too strongly tailored to the specifics of the sample population used in development such that it predicts well for patients within the derivative cohort but is not generalisable to other samples [41]. A bootstrap resampling technique was used to test for Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 11 overfitting. Details of the bootstrap approach can be found in Appendix 1. Data were analysed using STATA version 13.0 and Rstudio version 1.0.143. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 12 Results 721 of the 972 (75.7%) participants completed and returned the baseline and 6-month follow-up questionnaire and were eligible for this analysis. The mean age of the participants was 68.6 years, there was an even gender split and approximately half were educated to secondary-school level (Table 1). Most participants were retired (56.5%) but approximately 1 in 4 were still working either full-time or part-time. 99 patients (14.1%) met the definition of poor outcome 6-months post-TKA. Univariable analysis There were several preoperative factors which predicted a poor outcome (see Table 2). Firstly, clinical status: severe chronic pain (CPG grade IV) (OR 11.25, 95% CI 3.92 to 32.30), chronic widespread pain (2.34, 1.30 to 4.19), and a high number of co-morbidities (≥ 4 comorbidities: 3.75, 1.90 to 7.40). In contrast, a better OKS was associated with reduced risk of poor outcome (0.87 per unit increase in score; 0.84 to 0.91). Secondly, psychosocial factors: illness attitudes were strongly related to poor outcome, for every one point increase in illness attitude score (1.03, 1.01 to 1.05), the risk of poor outcome increased. Among participants who had reported aches or pains, the odds of a poor outcome also increased for every unit increase in passive coping score (1.08; 1.05 to 1.12), while poor outcome was less likely for every unit increase in active coping strategies (0.87; 0.83 to 0.92). Expectations were strongly associated with poor outcome, for every one point increase in expected knee pain after recovery (1.01, 1.01 to 1.02) or expected limitations in everyday activities (1.02, 1.01 to 1.02), the risk of poor outcome increased. Thirdly, mental health: severe anxiety (2.58, 1.48 to 4.49) and depression (3.67, 1.88 to 7.15) were associated with poor outcome and for every one unit increase in the PROMIS mental score, the risk of poor outcome decreased (0.93, 0.89 to 0.97). Finally, poor outcome was less likely amongst those with good preoperative physical health (PROMIS- physical health) and quality of life (EQ-5D). There were other factors which were not significantly associated with outcome but which met the criteria for being considered in the multivariable model: severely disturbed sleep and a long duration Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 13 of knee pain. In contrast, age and gender were not related to outcome and were not considered further. Model development and validation Of the factors eligible for inclusion in the multivariable models (p < 0.2), four were entered and retained in the final model predicting poor outcome: low preoperative OKS, chronic widespread pain, high expectations of knee pain after recovery, and lack of active coping strategies (Table 3). The model demonstrated good discrimination between patients at high and low risk of poor outcome following TKA, as indicated by a pooled c-statistic of 0.78 (pooled estimates of the 20 imputations). The final predictive model had a sensitivity of 82.8%, a specificity of 60.7%, and a positive likelihood ratio of 2.11 at the optimal cut-off identified by Youden’s index (J). Excellent agreement was found between observed and predicted probabilities. The estimate obtained with the bootstrap resampling was very close to the original estimate across the 20 imputed datasets. After correcting for optimism, the average c-statistic was 0.77. This suggested a reliable optimism-corrected c-statistic. Calibration-in-the-large showed no evidence of systematic overestimation or underestimation of the predicted probability of outcome. The average calibration- in-the-large was 0.16 (-0.07 to 0.34), which indicated there was no evidence of overfitting in the model. Clinical prediction tool A simple risk-scoring system was developed from the multivariable model, which can be found in Appendix 2 (electronic supplementary material). Scores range from 0 to 19, with higher scores corresponding to higher risk of poor outcome at 6 months post-TKA. Risk estimates are attached to each point total as shown in Figure 1. Two case studies demonstrating the relationship between the estimated risks of the prediction tool and those from the logistic regression model are available in Appendix 3 (electronic supplementary material). Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 14 Discussion Expectations (of poor outcome) and behaviour (lack of active coping) as well as clinical factors (poor preoperative knee status and chronic widespread pain) were key predictors of a poor outcome in persons undergoing TKA. A clinical prediction model based on these factors demonstrated good performance in identifying patients who had poor outcome based on OKS. A strength of our study is the multi-centre nature and large sample size. We have measured a range of patient-reported factors, particularly focussing on those that have been shown to predict outcome for musculoskeletal disorders, and specifically pain. Robust statistical methods such as multiple imputation and bootstrap resampling were employed to strengthen the development of this clinical prediction tool. Multiple imputation encourages statistical efficiency especially when missing data are assumed to be missing at random (MAR), which is plausible in the context of this study [43]. With many variables and rare events, there is a risk of overfitting the model. To test for this, we measured the shrinkage factor, an indicator for reliable estimations, to determine whether there was a need to reduce the regression coefficients using a shrinkage method (e.g. lasso) and overfitting was not indicated (shrinkage factor > 0.8) [41]. Limitations of our study include that only a few clinical factors were measured and some such as joint damage or body mass index (BMI) was not available. Although BMI is often associated with many conditions including osteoarthritis, there is no evidence in the literature to suggest that BMI is a clinically important predictor of postoperative outcome [44, 45]. Though the absolute risk remains small, higher BMI is, however, associated with an increased relative risk of revisions and post- surgical complications, which are important factors to consider in decision-making [46-48]. There were also no intra-operative factors collected, some of which have been related to poor outcome. However, as the purpose was to develop a clinical prediction tool to aid shared decision-making by the clinician and the patient about proceeding to knee replacement surgery, then by de facto this must be based only on factors available at this time. At the optimal cut-off for clinical use, there was Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 15 a sensitivity and specificity of 82.8% and 60.7% respectively, with a positive likelihood ratio of 2.11. Although the likelihood ratio (LR) of the positive test falls below the recommended value for a strong diagnostic test (LR=5), it is comparable to other prediction rules reported in the literature (e.g. Lungu et al. 2014 [49]). Our study predicted a binary outcome, using a recommended cut-off of the OKS. We tested our model using other cut-offs which have been proposed (OKS ≤ 19/ > 19) [50] and also developed a model which predicted the actual score rather than a binary state. Each of these alternative strategies produced very similar predictive models (data not shown). To our knowledge, only two other studies have translated determinants of TKA outcomes into a clinical prediction rule [44, 49]. Lungu et al. (2014) explored an extensive list of potential predictors, and included 5 of the 24 items from the Western Ontario and McMaster Osteoarthritis Index questionnaire in their prediction rule [49]. Four of the questions were specific to preoperative function and the other measured stiffness. Their model, based on a small sample size of 141 patients, demonstrated good overall predictive validity for outcomes 6 month post-surgery: sensitivity 82%, specificity 72%, positive likelihood ratio of 2.9. The second study was an extensive programme of work funded by the National Institute for Health Research (NIHR) [44]. Using data from the Knee Arthroplasty Trial (KAT), Arden et al. (2017) developed the Clinical Outcomes in Arthroplasty Study (COASt) knee model to predict 12-month postoperative OKS. This model included patient characteristics (age, sex, preoperative OKS, BMI, deprivation score, SF-12 mental component summary score) and clinical factors (the American Society of Anesthesiologists grade (a measure of fitness for surgery), comorbidities, previous knee surgery, fixed flexion deformity, valgus or varus deformity and preoperative anterior cruciate ligament state) [44]. Internal validation of the model demonstrated overall good discrimination (R =20%) and calibration but did not perform well in their validation cohort [44]. They attributed this to fundamental differences in patient characteristics, surgical techniques and implants, proportion of missing data, and varying proxy variables between the development and validation cohorts. A further cost-utility analysis did not find the COASt knee model to be cost-effective and therefore, the implementation in practice could not be Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 16 recommended. It is of note that previous models are solely focussed on clinical factors, while the evidence from this study and others [7, 10, 19] demonstrate that outcome is influenced by both clinical factors and psychosocial factors (including patient beliefs and health behaviour). It is likely that any clinical prediction model will need to incorporate both these domains to be optimal in predicting outcomes. Our findings highlight the importance of biopsychosocial assessment in patients undergoing TKA. Alattas et al. (2017) in a systematic review which included 10 studies, found consistent evidence for the role of anxiety and some evidence for the role of depression in predicting poor outcome [51]. We found that people with high expectations of knee pain after recovery also have poorer outcome. Taking into account their condition and their requirements, patients may make a realistic assessment of their outcome. However, pessimism has been linked to long-term poorer physical health, even when controlling for the health status at the time of pessimism [50]. Misplaced adverse beliefs may influence one’s perception of events and affect the way we cope [16]. Studies have found that active coping strategies such as remaining active and positive refocusing are associated with less pain and functional impairment [28, 52] whereas adopting passive coping strategies such as catastrophizing has been related to poorer functional outcomes [16]. The role of psychosocial factors in predicting outcome is important because such factors are potentially modifiable preoperatively and if the relationship is causal, could improve outcome. Cognitive and other behavioural therapies, which can include focussing on behavioural activation, pacing and changes in lifestyle can alter patients expectation and coping style, and indeed have been shown to have positive effects on pain experience and positive coping measures [53]. The purpose of designing a clinical prediction tool is not to determine who should and should not undergo total knee arthroplasty but instead act as an aid to shared decision-making between the patient and clinician in terms of highlighting patients at higher risk of a poor outcome and also establishing realistic expectations of postoperative pain and function. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 17 In conclusion, we have developed a prediction model for outcome after TKA, including both clinical factors and patient attitudes and behaviour in terms of self-management. Future work may investigate the validation of the model in another cohort and its impact on clinical decision-making. The results also offer the possibility that modifying illness beliefs and behaviours may result in better TKA outcomes. Funding Targeted Rehabilitation to Improve Outcome after knee replacement (TRIO) was supported by Arthritis Research UK (Grant No: 20100) (chief investigator TRIO-Physio, Prof. Hamish A Simpson; chief investigator TRIO-POPULAR, Prof. Gary J Macfarlane). Acknowledgements We would like to thank the staff at the following centres who recruited to the study: NHS Lothian, NHS Grampian, Royal Orthopaedic Hospital, Weston Area Health NHS Trust, Barts Health NHS Trust, Warrington and Halton Hospital, NHS Fife, The Dudley Group NHS Foundation Trust, and The Pennine Acute Hospitals NHS Trust. We are also grateful to the study team of TRIO-Physio which was conducted alongside this (study coordinators: Phillip Rayson and Chris Tuck). GJM developed the idea for TRIO-POPULAR and it was conducted in collaboration with DH, HS and the TRIO-Physio investigators. MB was the study co-ordinator. JS undertook the statistical analysis under the supervision of DM and GJM. JS and GJM drafted the manuscript which was reviewed and revised by all authors. Disclosure statement All authors have no conflicts of interest or financial ties to disclose. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 18 References 1 National Joint Registry. National Joint Registry for England, Wales, Northern Ireland and the Isle of th Man 14 annual report. NJR 2017; 14. 2 Scottish Arthroplasty Project. The Scottish Arthroplasty Project Annual Report 2017. SAP 2017. 3 Baker PN, Deehan DJ, Lees D, et al. The effect of surgical factors on early patient-reported outcome measures (PROMS) following total knee replacement. J Bone Joint Surg Br 2012; 94:1058-66 doi:10.1302/0301-620X.94B8.28786. 4 Hamilton D, Henderson GR, Gaston P, et al. Comparative outcomes of total hip and knee arthroplasty: a prospective cohort study. Postgrad Med J 2012; 88:627-31 doi:10.1136/postgradmedj-2011-130715. 5 Beswick AD, Wylde V, Gooberman-Hill R, et al. What proportion of patients report long-term pain after total hip or knee replacement for osteoarthritis? A systematic review of prospective studies in unselected patients. BMJ Open 2012; 2 doi:10.1136/bmjopen-2011-000435. 6 Jones CA, Beaupre LA, Johnston DW, et al. Total joint arthroplasties: current concepts of patient outcomes after surgery. Clin Geriatr Med 2005; 21:527,41, vi doi:S0749-0690(05)00008-X [pii]. 7 Vissers MM, Bussmann JB, Verhaar JA, et al. Psychological factors affecting the outcome of total hip and knee arthroplasty: a systematic review. Semin Arthritis Rheum 2012; 41:576-88 doi:10.1016/j.semarthrit.2011.07.003. 8 Katz J, Seltzer Z. Transition from acute to chronic postsurgical pain: risk factors and protective factors. Expert Rev Neurother 2009; 9:723-44 doi:10.1586/ern.09.20. 9 da Silva RR, Santos AA, de Sampaio Carvalho Junior J, et al. Quality of life after total knee arthroplasty: systematic review. Rev Bras Ortop 2014; 49:520-7 doi:10.1016/j.rboe.2014.09.007. 10 Lewis GN, Rice DA, McNair PJ, et al. Predictors of persistent pain after total knee arthroplasty: a systematic review and meta-analysis. Br J Anaesth 2015; 114:551-61 doi:10.1093/bja/aeu441. 11 Choi Y, Ra HJ. Patient Satisfaction after Total Knee Arthroplasty. Knee Surgery & Related Research 2015; 28:1-15 doi:10.5792/ksrr.2016.28.1.1. 12 Harmelink KEM, Zeegers AVCM, Hullegie W, et al. Are There Prognostic Factors for One-Year Outcome After Total Knee Arthroplasty? A Systematic Review. J Arthroplasty 2017; 32:3840,3853.e1 doi:S0883-5403(17)30594-6 [pii]. 13 Santaguida PL, Hawker GA, Hudak PL, et al. Patient characteristics affecting the prognosis of total hip and knee joint arthroplasty: a systematic review. Canadian Journal of Surgery 2008; 51:428-36. 14 Pincus T, Burton AK, Vogel S, et al. A systematic review of psychological factors as predictors of chronicity/disability in prospective cohorts of low back pain. Spine (Phila Pa 1976) 2002; 27:E109-20. 15 Turk DC, Okifuji A. Psychological factors in chronic pain: evolution and revolution. J Consult Clin Psychol 2002; 70:678-90. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 19 16 Vargas-Prada S, Coggon D. Psychological and psychosocial determinants of musculoskeletal pain and associated disability. Best practice & research.Clinical rheumatology 2015; 29:374-90 doi:10.1016/j.berh.2015.03.003. 17 Melzack R. Evolution of the neuromatrix theory of pain. The Prithvi Raj Lecture: presented at the third World Congress of World Institute of Pain, Barcelona 2004. Pain Pract 2005; 5:85-94 doi:PPR5203 [pii]. 18 Crofford LJ. Psychological aspects of chronic musculoskeletal pain. Best Pract Res Clin Rheumatol 2015; 29:147-55 doi:10.1016/j.berh.2015.04.027. 19 Khatib Y, Madan A, Naylor JM, et al. Do Psychological Factors Predict Poor Outcome in Patients Undergoing TKA? A Systematic Review. Clinical Orthopaedics and Related Research 2015; 473:2630-8 doi:10.1007/s11999-015-4234-9". 20 Riddle DL, Perera RA, Jiranek WA, et al. Using surgical appropriateness criteria to examine outcomes of total knee arthroplasty in a United States sample. Arthritis Care Res (Hoboken) 2015; 67:349-57 doi:10.1002/acr.22428. 21 Simpson AH, Hamilton DF, Beard DJ, et al. Targeted rehabilitation to improve outcome after total knee replacement (TRIO): study protocol for a randomised controlled trial. Trials 2014; 15:44,6215- 15-44 doi:10.1186/1745-6215-15-44. 22 Dawson J, Fitzpatrick R, Murray D, et al. Questionnaire on the perceptions of patients about total knee replacement. J Bone Joint Surg Br 1998; 80:63-9. 23 Von Korff M, Dworkin SF, Le Resche L. Graded chronic pain status: an epidemiologic evaluation. Pain 1990; 40:279-91. 24 Wolfe F, Smythe HA, Yunus MB, et al. The American College of Rheumatology 1990 Criteria for the Classification of Fibromyalgia. Report of the Multicenter Criteria Committee. Arthritis Rheum 1990; 33:160-72. 25 Jenkins CD, Stanton BA, Niemcryk SJ, et al. A scale for the estimation of sleep problems in clinical research. J Clin Epidemiol 1988; 41:313-21 doi:0895-4356(88)90138-2 [pii]. 26 Kellner R, Abbott P, Winslow WW, et al. Fears, beliefs, and attitudes in DSM-III hypochondriasis. J Nerv Ment Dis 1987; 175:20-5. 27 Kellner R. Abridged Manual of the Illness Attitude Scales: University of New Mexico, Department of Psychiatry, School of Medicine 1987. 28 Brown GK, Nicassio PM. Development of a questionnaire for the assessment of active and passive coping strategies in chronic pain patients. Pain 1987; 31:53-64. 29 Mannion AF, Kampfen S, Munzinger U, et al. The role of patient expectations in predicting outcome after total knee arthroplasty. Arthritis Res Ther 2009; 11:R139 doi:10.1186/ar2811. 30 Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand 1983; 67:361-70. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 20 31 Hays RD, Bjorner JB, Revicki DA, et al. Development of physical and mental health summary scores from the patient-reported outcomes measurement information system (PROMIS) global items. Quality of Life Research 2009; 18:873-80 doi:10.1007/s11136-009-9496-9. 32 EuroQol Group. EuroQol--a new facility for the measurement of health-related quality of life. Health Policy 1990; 16:199-208. 33 Davies AP. Rating systems for total knee replacement. Knee 2002; 9:261-6 doi:S0968016002000959 [pii]. 34 Garratt AM, Brealey S, Gillespie WJ, et al. Patient-assessed health instruments for the knee: a structured review. Rheumatology 2004; 43:1414-23 doi:10.1093/rheumatology/keh362. 35 Murray DW, Fitzpatrick R, Rogers K, et al. The use of the Oxford hip and knee scores. J Bone Joint Surg Br 2007; 89:1010-4 doi:89-B/8/1010 [pii]. 36 Rothwell AG, Hooper GJ, Hobbs A, et al. An analysis of the Oxford hip and knee scores and their relationship to early joint revision in the New Zealand Joint Registry. J Bone Joint Surg Br 2010; 92:413-8 doi:10.1302/0301-620X.92B3.22913. 37 Sullivan LM, Massaro JM, D'Agostino RB. Presentation of multivariate data for clinical use: The Framingham Study risk score functions. Stat Med 2004; 23:1631-60 doi:10.1002/sim.1742. 38 Sterne JA, White IR, Carlin JB, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ 2009; 338:b2393 doi:10.1136/bmj.b2393. 39 White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med 2011; 30:377-99 doi:10.1002/sim.4067. 40 Hosmer DW, Lemeshow S, eds. Applied Logistic Regression. New York, NY: John Wiley & Sons 41 Steyerberg E. Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating: Springer New York 2009. 42 Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology 2010; 21:128-38 doi:10.1097/EDE.0b013e3181c30fb2. 43 Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York: Wiley 1987. 44 Arden N, Altman D, Beard D, et al. Lower limb arthroplasty: can we produce a tool to predict outcome and failure, and is it cost-effective? An epidemiological study. Programme Grants for Applied Research 2017;5 doi:10.3310/pgfar05120. 45 Kane RL, Saleh KJ, Wilt TJ, et al. Total knee replacement. Evid Rep Technol Assess 2003:1-8. 46 Culliford D, Maskell J, Judge A, et al. A population-based survival analysis describing the association of body mass index on time to revision for total hip and knee replacements: results from the UK general practice research database. BMJ Open 2013; 3 doi:10.1136/bmjopen-2013-003614. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 21 47 Judge A, Arden NK, Cooper C, et al. Predictors of outcomes of total knee replacement surgery. Rheumatology (Oxford) 2012; 51:1804-13 doi:kes075 [pii]. 48 Fortin PR, Penrod JR, Clarke AE, et al. Timing of total joint replacement affects clinical outcomes among patients with osteoarthritis of the hip or knee. Arthritis Rheum 2002; 46:3327-30 doi:10.1002/art.10631. 49 Lungu E, Vendittoli P, Desmeules F. Preoperative Determinants of Patient-reported Pain and Physical Function Levels Following Total Knee Arthroplasty: A Systematic Review. The Open Orthopaedics Journal 2016; 10:213-31 doi:10.2174/1874325001610010213. 50 Peterson C, Seligman ME, Vaillant GE. Pessimistic explanatory style is a risk factor for physical illness: a thirty-five-year longitudinal study. J Pers Soc Psychol 1988; 55:23-7. 51 Alattas SA, Smith T, Bhatti M, et al. Greater pre-operative anxiety, pain and poorer function predict a worse outcome of a total knee arthroplasty. Knee Surg Sports Traumatol Arthrosc 2017; 25:3403-10 doi:10.1007/s00167-016-4314-8. 52 Jamison RN, Edwards RR. Integrating Pain Management in Clinical Practice. Journal of clinical psychology in medical settings 2012;19:49-64 doi:10.1007/s10880-012-9295-2. 53 Morley S, Eccleston C, Williams A. Systematic review and meta-analysis of randomized controlled trials of cognitive behaviour therapy and behaviour therapy for chronic pain in adults, excluding headache. Pain 1999; 80:1-13. 54 van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis. Stat Med 1999; 18:681-94 doi:10.1002/(SICI)1097-0258(19990330)18:63.0.CO;2- R [pii]. 55 SWBOOT: Stata module to bootstrap stepwise linear or logistic regression models. 56 Austin PC, Tu JV. Bootstrap Methods for Developing Predictive Models. The American Statistician 2004; 58:131-7 doi:10.1198/0003130043277. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 1 Table 1. Characteristics of the study population Predictors N respondents Demographic and socioeconomic characteristics Age (median years, IQR) 68.6 63.3-74.6 721 No. female (n, %) 379 52.6% 721 Marital status (n, %) Single 35 4.9% Married 485 67.5% Widowed 100 13.9% Divorced 67 9.3% Separated 8 1.1% Co-habiting 24 3.4% 719 Education (n, %) Secondary school 356 49.5% Apprenticeship 81 11.3% Further education college 188 26.2% University degree 69 9.6% Further degree 25 3.5% 719 Centres Edinburgh 242 33.6% Aberdeen 118 16.4% Royal Orthopaedic Hospital 146 20.3% Weston General Hospital 45 6.2% Barts Health NHS Trust 17 2.4% Warrington 20 2.8% Fife 67 9.3% Dudley 13 1.8% Pennine Acute 53 7.4% 721 Work Current employment status (n, %) Working full time 117 16.6% Working part-time 68 9.7% Retired 397 56.5% Unable to work because of illness or disability 41 5.8% Student 0 0 Unemployed and looking for work 6 0.9% Not looking for paid employment 74 10.5% 703 Clinical factors Duration of knee pain (median years, IQR) 7.2 2.0-10.0 699 Baseline Oxford Knee Score (mean, IQR) 20.6 15.0-26.0 709 Chronic Pain Grade (n, %) 664 Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 2 No pain – Grade 0 126 19.0% Low disability and low intensity – Grade I 55 8.3% Low disability and high intensity – Grade II 175 26.4% High disability and moderate intensity – Grade III 145 21.8% High disability and high intensity – Grade IV 163 24.6% Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 3 Table 2. Univariable associations between individual preoperative variables and poor outcome Predictors Persons N Odds 95% Confidence with poor total Ratio Interval outcome Sociodemographic factors Age, years 67.8 (9.0) 704 0.99 0.96-1.01 Gender Female 55 (15.1) reference category Male 44 (12.9) 704 0.84 0.54-1.28 Clinical factors Duration of knee pain, years 8.4 (7.6) 682 1.02 0.998-1.05* Baseline Oxford Knee Score; per unit (0-48) 15.2 (6.8) 695 0.87 0.84-0.91* Chronic Pain Grade No pain – Grade 0 4 (3.3) reference category Low disability and low intensity – Grade I 2 (3.8) 1.16 0.21-6.53 Low disability and high intensity – Grade II 12 (7.1) 2.25 0.71-7.17 High disability and moderate intensity – Grade 24 (16.7) 5.90 1.98-17.54* III High disability and high intensity – Grade IV 45(27.6) 651 11.25 3.92-32.30* Chronic Widespread Pain No 78(12.5) reference category Yes 18 (25.0) 697 2.34 1.30-4.19* Sleep Problem Scale Mildly sleep disturbed (≤ 15 nights) 77 (13.0) reference category Severely sleep disturbed (>15 nights) 21 (19.8) 699 1.66 0.97-2.83* Co-morbidities ≤1 comorbidities 14 (8.5) reference category 2-3 comorbidities 53 (12.8) 1.58 0.85-2.93 ≥4 comorbidities 32 (25.8) 704 3.75 1.90-7.40* Psychosocial factors Illness Attitude Score; per unit (0-108) 31.9 (13.4) 655 1.03 1.01-1.05* Active coping; per unit (7-35) 21.1 (4.6) 562 0.87 0.83-0.92* Passive coping; per unit (11-55) 33.6 (7.6) 547 1.08 1.05-1.12* Expectations of pain after recovery; per unit (0-100) 51.4 (29.0) 685 1.01 1.01-1.02* Expectations of limitations after recovery; per unit (0-100) 43.5 (25.4) 685 1.02 1.01-1.02* Mental and Physical Health HADS anxiety For categorical variables, number and percentages of persons with poor outcome are reported. Means and standard deviations of persons with poor outcome are reported for continuous variables Hospital Anxiety and Depression Scale Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 4 Mild to moderate anxiety 84 (12.8) reference category Severe anxiety 15 (34.9) 702 2.58 1.48-4.49* HADS depression Mild to moderate depression 78 (12.5) reference category Severe depression 21 (26.9) 702 3.67 1.88-7.15* PROMIS mental health; per unit (4-20) 42.6 (5.6) 696 0.93 0.89-0.97* PROMIS physical health; per unit (4-20) 35.0 (3.6) 691 0.82 0.77-0.87* Quality of life EQ5D; per tenth of a unit (-0.5-1.0) 2.6 (1.8) 685 0.74 0.65-0.83* *p < 0.2 Patient Reported Outcome Measurement System Patient Reported Outcome Measurement System Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 5 Table 3. Predictors of poor outcome in a multivariable stepwise regression model Predictors Adjusted 95% Confidence Odds ratio Interval Oxford knee score (per unit increase in score) 0.89 0.86-0.93 Expectations of knee pain after recovery (per unit increase 1.01 1.005-1.02 in score) Active coping ( per unit increase in score) 0.91 0.86-0.96 Chronic Widespread Pain 1.65 0.86-3.17 Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 1 Predictors Instrument scale scores (clinical points) Scores Oxford Knee Score 0-4 5-8 9-12 13-16 17-20 21-24 25-28 29-32 33-36 37-40 41-44 45-48 (11 pts) (10 pts) (9 pts) (8 pts) (7 pts) (6 pts) (5 pts) (4 pts) (3 pts) (2 pts) (1 pts) (0 pts) Expectations of knee 0-32 33-65 66-100 pain after recovery (0 pts) (1 pts) (2 pts) Active coping 7-12 13-18 19-24 25-30 31-35 (5 pts) (4 pts) (3 pts) (1 pts) (0 pts) Chronic widespread Yes No pain (1 pts) (0 pts) Total point Total point Risk of OKS ≤ 26 in % Total point Risk of OKS ≤ 26 in % < 4 ≤1 12 26 5 1 13 36 6 2 14 46 7 4 15 57 8 6 16 68 Note: For each instrument scale scores, enter the corresponding clinical points in the box on the right hand side. Add up the points and enter the total. Look 9 8 17 77 for the total point in the lower table and get the percentage risk of a poor 10 12 18 84 outcome post-TKA. 11 18 19 89 Figure 1. Points-based risk-scoring system for estimation of poor outcome (defined as Oxford Knee Score (OKS) ≤ 26) post-TKA. Scores range from 0 to 18 points. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Rheumatology Advances in Practice Oxford University Press

Development of a clinical risk score for pain and function following total knee arthroplasty: results from the TRIO study

Free
26 pages

Loading next page...
 
/lp/ou_press/development-of-a-clinical-risk-score-for-pain-and-function-following-Vd0iQDRAaN
Publisher
British Society for Rheumatology
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the British Society for Rheumatology.
eISSN
2514-1775
D.O.I.
10.1093/rap/rky021
Publisher site
See Article on Publisher Site

Abstract

Objectives To develop and validate a simple clinical prediction model, based on easily collected preoperative information, to identify patients at high risk of pain and functional disability 6 months after total knee arthroplasty (TKA). Methods This was a multi-centre cohort study of patients from 9 centres across the UK, who were undergoing a primary TKA for osteoarthritis. Information on socio-demographic, psychosocial, clinical, and quality of life measures were collected at recruitment. The primary outcome measure for this analysis was Oxford Knee Score, measured 6 months postoperatively by postal questionnaire. Multivariable logistic regression was used to develop the model. Model performance (discrimination and calibration) and internal validity was assessed, and a simple clinical risk score developed. Results 721 participants (mean age 68.3 years; 53% female) provided data for the current analysis and 14% had a poor outcome at 6 months. Key predictors were poor clinical status, widespread body pain, high expectation of postoperative pain, and lack of active coping. The developed model based on these variables demonstrated good discrimination. At the optimal cut-off, the final model had a sensitivity of 83%, specificity of 61%, and positive likelihood ratio of 2.11. Excellent agreement was found between observed and predicted outcomes, and there was no evidence of overfitting in the model. Conclusion We have developed and validated a clinical prediction model that can be used to identify patients at high risk of a poor outcome after TKA. This clinical risk score may be an aid to shared decision-making between patient and clinician. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 4 Keywords: Knee pain, Osteoarthritis, Total knee arthroplasty, Prediction Modelling, Clinical risk score, Model calibration, Model discrimination Key messages: • Predictors of poor outcome following total knee arthroplasty included illness attitudes and behaviours and clinical factors. • A model based on easily measurable variables demonstrates good performance. • The prediction tool developed can be an aid to shared decision-making between patient and clinician. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 5 Introduction Total knee arthroplasty (TKA) is one of the most common and effective treatments for severe knee osteoarthritis with over 100,000 knee replacements performed in the UK annually [1, 2]. Despite success in reduction of pain after knee replacements, approximately 20% to 30% of patients continue to experience pain and limited function after their TKA, which cannot be entirely explained by biomedical factors [3-5]. Clinical determinants of outcomes after TKA which have been consistently shown to be related to outcome across reviews include preoperative pain and function, pain at other sites, and aspects of surgery (longer duration of surgery, lengthy wait times) [6-12]. For other factors, the evidence is not consistent and may be related to the outcome studied. For example, one systematic review focusing on patients’ characteristics found that younger age and being male were related to risk of revision, older age was associated with increased risk of mortality and poorer function after TKA, but age and sex did not influence postoperative pain [13]. The importance of psychosocial and individual factors, as predictors of musculoskeletal outcomes has also been increasingly recognised [14-16]. Adverse psychological factors such as anxiety and depression may have an effect on pain perception and mediate the development of chronic pain and disability [17, 18]. The relationship between psychosocial factors and TKA outcomes has been examined in several systematic reviews, which have consistently indicated poor preoperative mental health and pain catastrophizing to be strongly associated with greater postoperative pain and functional disability [7, 10, 19]. Limited or conflicting evidence was found for other psychological factors. It is clear from the reviews that there is a lack of consensus on the most important clinical and psychological risk factors for poor outcomes after TKA. Although the decision to operate is primarily based on radiographic evidence of osteoarthritis and the patient’s report of symptoms, variation in the use of surgery reflects the different beliefs among patients and surgeons as to the risks and benefits of surgery. In a US-based study, Riddle et al. (2014) Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 6 reported that a third of cases reviewed that underwent knee replacement surgery were ‘inappropriate’ and as a group, these patients demonstrated worse outcomes [20]. The fact that surgery might not be successful for certain patients still highlights the need for robust predictive models to inform the clinical decision-making process. Therefore, our study aimed to firstly predict the impact of pain and functional disability 6 months after TKA using routinely-collected patient preoperative information and secondly, incorporate this information into a clinical prediction tool. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 7 Methods The Targeted Rehabilitation to Improve Outcome- preoperative predictors of unfavourable outcome following knee arthroplasty (TRIO-POPULAR) study was a multi-centre cohort study to investigate potential preoperative predictors of poor outcome following TKA. The study recruited from 9 participating centres across the UK between December 2013 and July 2016. The study was conducted alongside a randomised controlled trial of targeted rehabilitation to improve outcome after TKA [21]. Adults aged 16 years or over, undergoing primary TKA for osteoarthritis were invited to take part in the study either by letter or in person at a clinic visit prior to surgery. Participants were excluded if they: were undergoing a revision TKA or fully constrained knee arthroplasty; had a TKA for a diagnosis other than osteoarthritis; or had existing medical conditions such as stroke, or other musculoskeletal conditions that cause a limitation of function. Participants completed a questionnaire at the time of recruitment and consent was obtained for access to medical records for research purposes. Follow-up questionnaires were mailed to participants 6 weeks, 3, and 6 months after surgery. Ethical approval was granted by the office for Research Ethics Committees Northern Ireland (ORECNI) (13/NI/0101). The preoperative questionnaire included the following items: - Sociodemographic factors: Age, gender, marital status, socio-economic status (highest education level achieved), and employment status were measured. - Clinical factors: Clinical factors measured included duration of knee pain, baseline pain and function using the Oxford Knee Score (OKS) [22], and the Chronic Pain Grade (CPG) [23]. The CPG contains 7-items which allow respondents to be classified into five categories: Grade 0 (no pain), Grade I (low disability/low intensity), Grade II (low disability/high intensity), Grade III (high disability/moderately limiting intensity), and Grade IV (high disability and highly limiting disability). Body manikins were used to determine whether participants meet the Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 8 definition of ‘chronic widespread pain’ used in the American College of Rheumatology criteria for fibromyalgia [24]. The Sleep Problem Scale consists of four questions, rated on a 6-point frequency rating scale, ranging from 0 (not at all) to 5 ( 22 to 31 days/month) [25]. Sleep disturbance was defined as a mean score ≥ 4, corresponding to at least 15 troubled nights per month [25]. Self-reported comorbidities in this cohort were also recorded. - Psychosocial factors: The Illness Attitude Scales (IAS) [26, 27] measures personal attitudes, fears, and beliefs associated with hypochondriasis and abnormal illness behaviour. It consists of nine subscales, each with three items on a 0-4 Likert scale. Scores are summed to give the total IAS score, with a higher score representing greater hypochondriacal fears and beliefs. Among participants who reported they had aches or pains lasting one day or longer in the past month, the Vanderbilt Pain Management Inventory was used to assess chronic pain coping strategies [28]. This questionnaire consists of 18-items, rated on a 5-point frequency Likert scale. From this data, two subscales can be calculated; active coping score and passive coping score. High scores indicate a high use of active and passive coping strategies respectively. Patient expectations of pain, and limitations in everyday activities after TKA were measured using visual analogue scales (VAS); 0 representing ‘not at all painful’ or ‘not limited at all’ and 100 ‘very painful’ or ‘greatly limited’, respectively [29]. - Mental and Physical Health: Mental and physical health was measured by the Hospital Anxiety and Depression Scale (HADS) [30] and the Patient Reported Outcome Measurement System 10 (PROMIS-10) Global Health Questionnaire [31]. The HADS is a 14-item questionnaire; seven items measuring anxiety and seven items measuring depression. Each item is rated on a 0-3 Likert scale with higher scores indicating poorer mental health. The PROMIS-10 Questionnaire has 10-items which allow the Global Physical Health and Global Mental Health sub-scales to be derived. Scores range from 4 to 20 with higher scores indicating better health. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 9 - Quality of life: The EuroQoL-5 dimension (EQ-5D) is a measure of quality of life [32] . It consists of five dimensions; mobility, self-care, usual activities, pain/discomfort, and anxiety/depression, rated on a 3-point scale. Each EQ-5D profile was converted to a single summary index based on the valuation of health states in the UK. A score of 1.0 indicates the best possible health. The outcome for this analysis was the OKS [33], measured 6 months postoperatively by postal questionnaire. The OKS measures the impact of pain and functional disability in patients undergoing knee replacement [34, 35]. Poor outcome was defined by a score of 26 or less (out of a maximum score of 48) according to the modified Kalairajah classification [36]. Statistical analysis The study aimed to recruit 750 participants and if 80% of participants (n=600) provided follow-up data, this would give 80% power to detect an odds ratio of 1.5 for a poor outcome, comparing the highest with the other two tertiles of exposure. Descriptive statistics were carried out to describe the study sample and normality of individual variables were assessed. Categorical variables; the Sleep Problem Scale, CPG, and the HADS anxiety and depression were categorised according to standard cut-offs. In preparation for the modelling, the relationship between continuous predictor variables and the observed log odds of a poor outcome were assessed for linearity. Health scores measured by the EQ5D and the PROMIS-10 questionnaire, measures of active and passive coping strategies determined by the Vanderbilt Pain Management Inventory, patient expectations of outcomes after surgery, and illness attitude scores were analysed as continuous variables. However, a maximum health index of one in the EQ5D results in regression coefficients (expressed as change in outcome per one unit increase in predictor) which are not intuitive to interpret, and values were therefore multiplied by 10 for the purpose of the univariable and multivariable analyses. Logistic regression analysis was used to explore the association between each of the potential preoperative predictor Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 10 variables and the OKS at 6 months. In the univariable analysis, variables showing an association with a significance level of p < 0.2 were candidates for entry into a forward stepwise regression as part of a bootstrap selection process as described below. Entry and removal criteria for the stepwise models were p ≤ 0.1 and p > 0.15, respectively. We used stepwise regression to suggest predictor variables for the model followed by the incorporation of clinical knowledge. Associations were expressed as odds ratios (ORs) with 95% confidence intervals (95% CIs). To aid clinical decision- making, a simplified point-based risk-scoring system was developed using coefficients from the final model [37]. Multiple imputation with chained equations (MICE) was used to impute missing predictor data with the aim to reduce bias and improve efficiency; 20 imputed datasets were generated [38, 39]. Detailed descriptions of the post-estimation procedure can be found in Appendix 1 (electronic supplementary material). Model discrimination was quantified using the area under the Receiver Operating Characteristic (ROC) curve or concordance (c) statistic to estimate predictive accuracy. A c-statistic value of 1 represents perfect discrimination whilst a c-statistic of 0.5 indicates a discriminative value equivalent to chance [40]. A pooled c-statistic of the 20 imputed datasets, was calculated. A shrinkage estimate was also calculated to assess overfitting. A shrinkage estimate of less than 0.8 would reflect a need for shrinkage of the regression coefficients in a prediction model using methods such as Lasso or ridge regression [41]. Model calibration, which refers to the agreement between the observed and predicted probabilities, was also assessed using calibration-in-the-large [42]. This indicates whether the predictions are systematically too low or too high. Overfitting occurs when a model is too strongly tailored to the specifics of the sample population used in development such that it predicts well for patients within the derivative cohort but is not generalisable to other samples [41]. A bootstrap resampling technique was used to test for Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 11 overfitting. Details of the bootstrap approach can be found in Appendix 1. Data were analysed using STATA version 13.0 and Rstudio version 1.0.143. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 12 Results 721 of the 972 (75.7%) participants completed and returned the baseline and 6-month follow-up questionnaire and were eligible for this analysis. The mean age of the participants was 68.6 years, there was an even gender split and approximately half were educated to secondary-school level (Table 1). Most participants were retired (56.5%) but approximately 1 in 4 were still working either full-time or part-time. 99 patients (14.1%) met the definition of poor outcome 6-months post-TKA. Univariable analysis There were several preoperative factors which predicted a poor outcome (see Table 2). Firstly, clinical status: severe chronic pain (CPG grade IV) (OR 11.25, 95% CI 3.92 to 32.30), chronic widespread pain (2.34, 1.30 to 4.19), and a high number of co-morbidities (≥ 4 comorbidities: 3.75, 1.90 to 7.40). In contrast, a better OKS was associated with reduced risk of poor outcome (0.87 per unit increase in score; 0.84 to 0.91). Secondly, psychosocial factors: illness attitudes were strongly related to poor outcome, for every one point increase in illness attitude score (1.03, 1.01 to 1.05), the risk of poor outcome increased. Among participants who had reported aches or pains, the odds of a poor outcome also increased for every unit increase in passive coping score (1.08; 1.05 to 1.12), while poor outcome was less likely for every unit increase in active coping strategies (0.87; 0.83 to 0.92). Expectations were strongly associated with poor outcome, for every one point increase in expected knee pain after recovery (1.01, 1.01 to 1.02) or expected limitations in everyday activities (1.02, 1.01 to 1.02), the risk of poor outcome increased. Thirdly, mental health: severe anxiety (2.58, 1.48 to 4.49) and depression (3.67, 1.88 to 7.15) were associated with poor outcome and for every one unit increase in the PROMIS mental score, the risk of poor outcome decreased (0.93, 0.89 to 0.97). Finally, poor outcome was less likely amongst those with good preoperative physical health (PROMIS- physical health) and quality of life (EQ-5D). There were other factors which were not significantly associated with outcome but which met the criteria for being considered in the multivariable model: severely disturbed sleep and a long duration Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 13 of knee pain. In contrast, age and gender were not related to outcome and were not considered further. Model development and validation Of the factors eligible for inclusion in the multivariable models (p < 0.2), four were entered and retained in the final model predicting poor outcome: low preoperative OKS, chronic widespread pain, high expectations of knee pain after recovery, and lack of active coping strategies (Table 3). The model demonstrated good discrimination between patients at high and low risk of poor outcome following TKA, as indicated by a pooled c-statistic of 0.78 (pooled estimates of the 20 imputations). The final predictive model had a sensitivity of 82.8%, a specificity of 60.7%, and a positive likelihood ratio of 2.11 at the optimal cut-off identified by Youden’s index (J). Excellent agreement was found between observed and predicted probabilities. The estimate obtained with the bootstrap resampling was very close to the original estimate across the 20 imputed datasets. After correcting for optimism, the average c-statistic was 0.77. This suggested a reliable optimism-corrected c-statistic. Calibration-in-the-large showed no evidence of systematic overestimation or underestimation of the predicted probability of outcome. The average calibration- in-the-large was 0.16 (-0.07 to 0.34), which indicated there was no evidence of overfitting in the model. Clinical prediction tool A simple risk-scoring system was developed from the multivariable model, which can be found in Appendix 2 (electronic supplementary material). Scores range from 0 to 19, with higher scores corresponding to higher risk of poor outcome at 6 months post-TKA. Risk estimates are attached to each point total as shown in Figure 1. Two case studies demonstrating the relationship between the estimated risks of the prediction tool and those from the logistic regression model are available in Appendix 3 (electronic supplementary material). Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 14 Discussion Expectations (of poor outcome) and behaviour (lack of active coping) as well as clinical factors (poor preoperative knee status and chronic widespread pain) were key predictors of a poor outcome in persons undergoing TKA. A clinical prediction model based on these factors demonstrated good performance in identifying patients who had poor outcome based on OKS. A strength of our study is the multi-centre nature and large sample size. We have measured a range of patient-reported factors, particularly focussing on those that have been shown to predict outcome for musculoskeletal disorders, and specifically pain. Robust statistical methods such as multiple imputation and bootstrap resampling were employed to strengthen the development of this clinical prediction tool. Multiple imputation encourages statistical efficiency especially when missing data are assumed to be missing at random (MAR), which is plausible in the context of this study [43]. With many variables and rare events, there is a risk of overfitting the model. To test for this, we measured the shrinkage factor, an indicator for reliable estimations, to determine whether there was a need to reduce the regression coefficients using a shrinkage method (e.g. lasso) and overfitting was not indicated (shrinkage factor > 0.8) [41]. Limitations of our study include that only a few clinical factors were measured and some such as joint damage or body mass index (BMI) was not available. Although BMI is often associated with many conditions including osteoarthritis, there is no evidence in the literature to suggest that BMI is a clinically important predictor of postoperative outcome [44, 45]. Though the absolute risk remains small, higher BMI is, however, associated with an increased relative risk of revisions and post- surgical complications, which are important factors to consider in decision-making [46-48]. There were also no intra-operative factors collected, some of which have been related to poor outcome. However, as the purpose was to develop a clinical prediction tool to aid shared decision-making by the clinician and the patient about proceeding to knee replacement surgery, then by de facto this must be based only on factors available at this time. At the optimal cut-off for clinical use, there was Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 15 a sensitivity and specificity of 82.8% and 60.7% respectively, with a positive likelihood ratio of 2.11. Although the likelihood ratio (LR) of the positive test falls below the recommended value for a strong diagnostic test (LR=5), it is comparable to other prediction rules reported in the literature (e.g. Lungu et al. 2014 [49]). Our study predicted a binary outcome, using a recommended cut-off of the OKS. We tested our model using other cut-offs which have been proposed (OKS ≤ 19/ > 19) [50] and also developed a model which predicted the actual score rather than a binary state. Each of these alternative strategies produced very similar predictive models (data not shown). To our knowledge, only two other studies have translated determinants of TKA outcomes into a clinical prediction rule [44, 49]. Lungu et al. (2014) explored an extensive list of potential predictors, and included 5 of the 24 items from the Western Ontario and McMaster Osteoarthritis Index questionnaire in their prediction rule [49]. Four of the questions were specific to preoperative function and the other measured stiffness. Their model, based on a small sample size of 141 patients, demonstrated good overall predictive validity for outcomes 6 month post-surgery: sensitivity 82%, specificity 72%, positive likelihood ratio of 2.9. The second study was an extensive programme of work funded by the National Institute for Health Research (NIHR) [44]. Using data from the Knee Arthroplasty Trial (KAT), Arden et al. (2017) developed the Clinical Outcomes in Arthroplasty Study (COASt) knee model to predict 12-month postoperative OKS. This model included patient characteristics (age, sex, preoperative OKS, BMI, deprivation score, SF-12 mental component summary score) and clinical factors (the American Society of Anesthesiologists grade (a measure of fitness for surgery), comorbidities, previous knee surgery, fixed flexion deformity, valgus or varus deformity and preoperative anterior cruciate ligament state) [44]. Internal validation of the model demonstrated overall good discrimination (R =20%) and calibration but did not perform well in their validation cohort [44]. They attributed this to fundamental differences in patient characteristics, surgical techniques and implants, proportion of missing data, and varying proxy variables between the development and validation cohorts. A further cost-utility analysis did not find the COASt knee model to be cost-effective and therefore, the implementation in practice could not be Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 16 recommended. It is of note that previous models are solely focussed on clinical factors, while the evidence from this study and others [7, 10, 19] demonstrate that outcome is influenced by both clinical factors and psychosocial factors (including patient beliefs and health behaviour). It is likely that any clinical prediction model will need to incorporate both these domains to be optimal in predicting outcomes. Our findings highlight the importance of biopsychosocial assessment in patients undergoing TKA. Alattas et al. (2017) in a systematic review which included 10 studies, found consistent evidence for the role of anxiety and some evidence for the role of depression in predicting poor outcome [51]. We found that people with high expectations of knee pain after recovery also have poorer outcome. Taking into account their condition and their requirements, patients may make a realistic assessment of their outcome. However, pessimism has been linked to long-term poorer physical health, even when controlling for the health status at the time of pessimism [50]. Misplaced adverse beliefs may influence one’s perception of events and affect the way we cope [16]. Studies have found that active coping strategies such as remaining active and positive refocusing are associated with less pain and functional impairment [28, 52] whereas adopting passive coping strategies such as catastrophizing has been related to poorer functional outcomes [16]. The role of psychosocial factors in predicting outcome is important because such factors are potentially modifiable preoperatively and if the relationship is causal, could improve outcome. Cognitive and other behavioural therapies, which can include focussing on behavioural activation, pacing and changes in lifestyle can alter patients expectation and coping style, and indeed have been shown to have positive effects on pain experience and positive coping measures [53]. The purpose of designing a clinical prediction tool is not to determine who should and should not undergo total knee arthroplasty but instead act as an aid to shared decision-making between the patient and clinician in terms of highlighting patients at higher risk of a poor outcome and also establishing realistic expectations of postoperative pain and function. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 17 In conclusion, we have developed a prediction model for outcome after TKA, including both clinical factors and patient attitudes and behaviour in terms of self-management. Future work may investigate the validation of the model in another cohort and its impact on clinical decision-making. The results also offer the possibility that modifying illness beliefs and behaviours may result in better TKA outcomes. Funding Targeted Rehabilitation to Improve Outcome after knee replacement (TRIO) was supported by Arthritis Research UK (Grant No: 20100) (chief investigator TRIO-Physio, Prof. Hamish A Simpson; chief investigator TRIO-POPULAR, Prof. Gary J Macfarlane). Acknowledgements We would like to thank the staff at the following centres who recruited to the study: NHS Lothian, NHS Grampian, Royal Orthopaedic Hospital, Weston Area Health NHS Trust, Barts Health NHS Trust, Warrington and Halton Hospital, NHS Fife, The Dudley Group NHS Foundation Trust, and The Pennine Acute Hospitals NHS Trust. We are also grateful to the study team of TRIO-Physio which was conducted alongside this (study coordinators: Phillip Rayson and Chris Tuck). GJM developed the idea for TRIO-POPULAR and it was conducted in collaboration with DH, HS and the TRIO-Physio investigators. MB was the study co-ordinator. JS undertook the statistical analysis under the supervision of DM and GJM. JS and GJM drafted the manuscript which was reviewed and revised by all authors. Disclosure statement All authors have no conflicts of interest or financial ties to disclose. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 18 References 1 National Joint Registry. National Joint Registry for England, Wales, Northern Ireland and the Isle of th Man 14 annual report. NJR 2017; 14. 2 Scottish Arthroplasty Project. The Scottish Arthroplasty Project Annual Report 2017. SAP 2017. 3 Baker PN, Deehan DJ, Lees D, et al. The effect of surgical factors on early patient-reported outcome measures (PROMS) following total knee replacement. J Bone Joint Surg Br 2012; 94:1058-66 doi:10.1302/0301-620X.94B8.28786. 4 Hamilton D, Henderson GR, Gaston P, et al. Comparative outcomes of total hip and knee arthroplasty: a prospective cohort study. Postgrad Med J 2012; 88:627-31 doi:10.1136/postgradmedj-2011-130715. 5 Beswick AD, Wylde V, Gooberman-Hill R, et al. What proportion of patients report long-term pain after total hip or knee replacement for osteoarthritis? A systematic review of prospective studies in unselected patients. BMJ Open 2012; 2 doi:10.1136/bmjopen-2011-000435. 6 Jones CA, Beaupre LA, Johnston DW, et al. Total joint arthroplasties: current concepts of patient outcomes after surgery. Clin Geriatr Med 2005; 21:527,41, vi doi:S0749-0690(05)00008-X [pii]. 7 Vissers MM, Bussmann JB, Verhaar JA, et al. Psychological factors affecting the outcome of total hip and knee arthroplasty: a systematic review. Semin Arthritis Rheum 2012; 41:576-88 doi:10.1016/j.semarthrit.2011.07.003. 8 Katz J, Seltzer Z. Transition from acute to chronic postsurgical pain: risk factors and protective factors. Expert Rev Neurother 2009; 9:723-44 doi:10.1586/ern.09.20. 9 da Silva RR, Santos AA, de Sampaio Carvalho Junior J, et al. Quality of life after total knee arthroplasty: systematic review. Rev Bras Ortop 2014; 49:520-7 doi:10.1016/j.rboe.2014.09.007. 10 Lewis GN, Rice DA, McNair PJ, et al. Predictors of persistent pain after total knee arthroplasty: a systematic review and meta-analysis. Br J Anaesth 2015; 114:551-61 doi:10.1093/bja/aeu441. 11 Choi Y, Ra HJ. Patient Satisfaction after Total Knee Arthroplasty. Knee Surgery & Related Research 2015; 28:1-15 doi:10.5792/ksrr.2016.28.1.1. 12 Harmelink KEM, Zeegers AVCM, Hullegie W, et al. Are There Prognostic Factors for One-Year Outcome After Total Knee Arthroplasty? A Systematic Review. J Arthroplasty 2017; 32:3840,3853.e1 doi:S0883-5403(17)30594-6 [pii]. 13 Santaguida PL, Hawker GA, Hudak PL, et al. Patient characteristics affecting the prognosis of total hip and knee joint arthroplasty: a systematic review. Canadian Journal of Surgery 2008; 51:428-36. 14 Pincus T, Burton AK, Vogel S, et al. A systematic review of psychological factors as predictors of chronicity/disability in prospective cohorts of low back pain. Spine (Phila Pa 1976) 2002; 27:E109-20. 15 Turk DC, Okifuji A. Psychological factors in chronic pain: evolution and revolution. J Consult Clin Psychol 2002; 70:678-90. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 19 16 Vargas-Prada S, Coggon D. Psychological and psychosocial determinants of musculoskeletal pain and associated disability. Best practice & research.Clinical rheumatology 2015; 29:374-90 doi:10.1016/j.berh.2015.03.003. 17 Melzack R. Evolution of the neuromatrix theory of pain. The Prithvi Raj Lecture: presented at the third World Congress of World Institute of Pain, Barcelona 2004. Pain Pract 2005; 5:85-94 doi:PPR5203 [pii]. 18 Crofford LJ. Psychological aspects of chronic musculoskeletal pain. Best Pract Res Clin Rheumatol 2015; 29:147-55 doi:10.1016/j.berh.2015.04.027. 19 Khatib Y, Madan A, Naylor JM, et al. Do Psychological Factors Predict Poor Outcome in Patients Undergoing TKA? A Systematic Review. Clinical Orthopaedics and Related Research 2015; 473:2630-8 doi:10.1007/s11999-015-4234-9". 20 Riddle DL, Perera RA, Jiranek WA, et al. Using surgical appropriateness criteria to examine outcomes of total knee arthroplasty in a United States sample. Arthritis Care Res (Hoboken) 2015; 67:349-57 doi:10.1002/acr.22428. 21 Simpson AH, Hamilton DF, Beard DJ, et al. Targeted rehabilitation to improve outcome after total knee replacement (TRIO): study protocol for a randomised controlled trial. Trials 2014; 15:44,6215- 15-44 doi:10.1186/1745-6215-15-44. 22 Dawson J, Fitzpatrick R, Murray D, et al. Questionnaire on the perceptions of patients about total knee replacement. J Bone Joint Surg Br 1998; 80:63-9. 23 Von Korff M, Dworkin SF, Le Resche L. Graded chronic pain status: an epidemiologic evaluation. Pain 1990; 40:279-91. 24 Wolfe F, Smythe HA, Yunus MB, et al. The American College of Rheumatology 1990 Criteria for the Classification of Fibromyalgia. Report of the Multicenter Criteria Committee. Arthritis Rheum 1990; 33:160-72. 25 Jenkins CD, Stanton BA, Niemcryk SJ, et al. A scale for the estimation of sleep problems in clinical research. J Clin Epidemiol 1988; 41:313-21 doi:0895-4356(88)90138-2 [pii]. 26 Kellner R, Abbott P, Winslow WW, et al. Fears, beliefs, and attitudes in DSM-III hypochondriasis. J Nerv Ment Dis 1987; 175:20-5. 27 Kellner R. Abridged Manual of the Illness Attitude Scales: University of New Mexico, Department of Psychiatry, School of Medicine 1987. 28 Brown GK, Nicassio PM. Development of a questionnaire for the assessment of active and passive coping strategies in chronic pain patients. Pain 1987; 31:53-64. 29 Mannion AF, Kampfen S, Munzinger U, et al. The role of patient expectations in predicting outcome after total knee arthroplasty. Arthritis Res Ther 2009; 11:R139 doi:10.1186/ar2811. 30 Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand 1983; 67:361-70. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 20 31 Hays RD, Bjorner JB, Revicki DA, et al. Development of physical and mental health summary scores from the patient-reported outcomes measurement information system (PROMIS) global items. Quality of Life Research 2009; 18:873-80 doi:10.1007/s11136-009-9496-9. 32 EuroQol Group. EuroQol--a new facility for the measurement of health-related quality of life. Health Policy 1990; 16:199-208. 33 Davies AP. Rating systems for total knee replacement. Knee 2002; 9:261-6 doi:S0968016002000959 [pii]. 34 Garratt AM, Brealey S, Gillespie WJ, et al. Patient-assessed health instruments for the knee: a structured review. Rheumatology 2004; 43:1414-23 doi:10.1093/rheumatology/keh362. 35 Murray DW, Fitzpatrick R, Rogers K, et al. The use of the Oxford hip and knee scores. J Bone Joint Surg Br 2007; 89:1010-4 doi:89-B/8/1010 [pii]. 36 Rothwell AG, Hooper GJ, Hobbs A, et al. An analysis of the Oxford hip and knee scores and their relationship to early joint revision in the New Zealand Joint Registry. J Bone Joint Surg Br 2010; 92:413-8 doi:10.1302/0301-620X.92B3.22913. 37 Sullivan LM, Massaro JM, D'Agostino RB. Presentation of multivariate data for clinical use: The Framingham Study risk score functions. Stat Med 2004; 23:1631-60 doi:10.1002/sim.1742. 38 Sterne JA, White IR, Carlin JB, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ 2009; 338:b2393 doi:10.1136/bmj.b2393. 39 White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med 2011; 30:377-99 doi:10.1002/sim.4067. 40 Hosmer DW, Lemeshow S, eds. Applied Logistic Regression. New York, NY: John Wiley & Sons 41 Steyerberg E. Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating: Springer New York 2009. 42 Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology 2010; 21:128-38 doi:10.1097/EDE.0b013e3181c30fb2. 43 Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York: Wiley 1987. 44 Arden N, Altman D, Beard D, et al. Lower limb arthroplasty: can we produce a tool to predict outcome and failure, and is it cost-effective? An epidemiological study. Programme Grants for Applied Research 2017;5 doi:10.3310/pgfar05120. 45 Kane RL, Saleh KJ, Wilt TJ, et al. Total knee replacement. Evid Rep Technol Assess 2003:1-8. 46 Culliford D, Maskell J, Judge A, et al. A population-based survival analysis describing the association of body mass index on time to revision for total hip and knee replacements: results from the UK general practice research database. BMJ Open 2013; 3 doi:10.1136/bmjopen-2013-003614. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 21 47 Judge A, Arden NK, Cooper C, et al. Predictors of outcomes of total knee replacement surgery. Rheumatology (Oxford) 2012; 51:1804-13 doi:kes075 [pii]. 48 Fortin PR, Penrod JR, Clarke AE, et al. Timing of total joint replacement affects clinical outcomes among patients with osteoarthritis of the hip or knee. Arthritis Rheum 2002; 46:3327-30 doi:10.1002/art.10631. 49 Lungu E, Vendittoli P, Desmeules F. Preoperative Determinants of Patient-reported Pain and Physical Function Levels Following Total Knee Arthroplasty: A Systematic Review. The Open Orthopaedics Journal 2016; 10:213-31 doi:10.2174/1874325001610010213. 50 Peterson C, Seligman ME, Vaillant GE. Pessimistic explanatory style is a risk factor for physical illness: a thirty-five-year longitudinal study. J Pers Soc Psychol 1988; 55:23-7. 51 Alattas SA, Smith T, Bhatti M, et al. Greater pre-operative anxiety, pain and poorer function predict a worse outcome of a total knee arthroplasty. Knee Surg Sports Traumatol Arthrosc 2017; 25:3403-10 doi:10.1007/s00167-016-4314-8. 52 Jamison RN, Edwards RR. Integrating Pain Management in Clinical Practice. Journal of clinical psychology in medical settings 2012;19:49-64 doi:10.1007/s10880-012-9295-2. 53 Morley S, Eccleston C, Williams A. Systematic review and meta-analysis of randomized controlled trials of cognitive behaviour therapy and behaviour therapy for chronic pain in adults, excluding headache. Pain 1999; 80:1-13. 54 van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis. Stat Med 1999; 18:681-94 doi:10.1002/(SICI)1097-0258(19990330)18:63.0.CO;2- R [pii]. 55 SWBOOT: Stata module to bootstrap stepwise linear or logistic regression models. 56 Austin PC, Tu JV. Bootstrap Methods for Developing Predictive Models. The American Statistician 2004; 58:131-7 doi:10.1198/0003130043277. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 1 Table 1. Characteristics of the study population Predictors N respondents Demographic and socioeconomic characteristics Age (median years, IQR) 68.6 63.3-74.6 721 No. female (n, %) 379 52.6% 721 Marital status (n, %) Single 35 4.9% Married 485 67.5% Widowed 100 13.9% Divorced 67 9.3% Separated 8 1.1% Co-habiting 24 3.4% 719 Education (n, %) Secondary school 356 49.5% Apprenticeship 81 11.3% Further education college 188 26.2% University degree 69 9.6% Further degree 25 3.5% 719 Centres Edinburgh 242 33.6% Aberdeen 118 16.4% Royal Orthopaedic Hospital 146 20.3% Weston General Hospital 45 6.2% Barts Health NHS Trust 17 2.4% Warrington 20 2.8% Fife 67 9.3% Dudley 13 1.8% Pennine Acute 53 7.4% 721 Work Current employment status (n, %) Working full time 117 16.6% Working part-time 68 9.7% Retired 397 56.5% Unable to work because of illness or disability 41 5.8% Student 0 0 Unemployed and looking for work 6 0.9% Not looking for paid employment 74 10.5% 703 Clinical factors Duration of knee pain (median years, IQR) 7.2 2.0-10.0 699 Baseline Oxford Knee Score (mean, IQR) 20.6 15.0-26.0 709 Chronic Pain Grade (n, %) 664 Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 2 No pain – Grade 0 126 19.0% Low disability and low intensity – Grade I 55 8.3% Low disability and high intensity – Grade II 175 26.4% High disability and moderate intensity – Grade III 145 21.8% High disability and high intensity – Grade IV 163 24.6% Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 3 Table 2. Univariable associations between individual preoperative variables and poor outcome Predictors Persons N Odds 95% Confidence with poor total Ratio Interval outcome Sociodemographic factors Age, years 67.8 (9.0) 704 0.99 0.96-1.01 Gender Female 55 (15.1) reference category Male 44 (12.9) 704 0.84 0.54-1.28 Clinical factors Duration of knee pain, years 8.4 (7.6) 682 1.02 0.998-1.05* Baseline Oxford Knee Score; per unit (0-48) 15.2 (6.8) 695 0.87 0.84-0.91* Chronic Pain Grade No pain – Grade 0 4 (3.3) reference category Low disability and low intensity – Grade I 2 (3.8) 1.16 0.21-6.53 Low disability and high intensity – Grade II 12 (7.1) 2.25 0.71-7.17 High disability and moderate intensity – Grade 24 (16.7) 5.90 1.98-17.54* III High disability and high intensity – Grade IV 45(27.6) 651 11.25 3.92-32.30* Chronic Widespread Pain No 78(12.5) reference category Yes 18 (25.0) 697 2.34 1.30-4.19* Sleep Problem Scale Mildly sleep disturbed (≤ 15 nights) 77 (13.0) reference category Severely sleep disturbed (>15 nights) 21 (19.8) 699 1.66 0.97-2.83* Co-morbidities ≤1 comorbidities 14 (8.5) reference category 2-3 comorbidities 53 (12.8) 1.58 0.85-2.93 ≥4 comorbidities 32 (25.8) 704 3.75 1.90-7.40* Psychosocial factors Illness Attitude Score; per unit (0-108) 31.9 (13.4) 655 1.03 1.01-1.05* Active coping; per unit (7-35) 21.1 (4.6) 562 0.87 0.83-0.92* Passive coping; per unit (11-55) 33.6 (7.6) 547 1.08 1.05-1.12* Expectations of pain after recovery; per unit (0-100) 51.4 (29.0) 685 1.01 1.01-1.02* Expectations of limitations after recovery; per unit (0-100) 43.5 (25.4) 685 1.02 1.01-1.02* Mental and Physical Health HADS anxiety For categorical variables, number and percentages of persons with poor outcome are reported. Means and standard deviations of persons with poor outcome are reported for continuous variables Hospital Anxiety and Depression Scale Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 4 Mild to moderate anxiety 84 (12.8) reference category Severe anxiety 15 (34.9) 702 2.58 1.48-4.49* HADS depression Mild to moderate depression 78 (12.5) reference category Severe depression 21 (26.9) 702 3.67 1.88-7.15* PROMIS mental health; per unit (4-20) 42.6 (5.6) 696 0.93 0.89-0.97* PROMIS physical health; per unit (4-20) 35.0 (3.6) 691 0.82 0.77-0.87* Quality of life EQ5D; per tenth of a unit (-0.5-1.0) 2.6 (1.8) 685 0.74 0.65-0.83* *p < 0.2 Patient Reported Outcome Measurement System Patient Reported Outcome Measurement System Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 5 Table 3. Predictors of poor outcome in a multivariable stepwise regression model Predictors Adjusted 95% Confidence Odds ratio Interval Oxford knee score (per unit increase in score) 0.89 0.86-0.93 Expectations of knee pain after recovery (per unit increase 1.01 1.005-1.02 in score) Active coping ( per unit increase in score) 0.91 0.86-0.96 Chronic Widespread Pain 1.65 0.86-3.17 Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018 1 Predictors Instrument scale scores (clinical points) Scores Oxford Knee Score 0-4 5-8 9-12 13-16 17-20 21-24 25-28 29-32 33-36 37-40 41-44 45-48 (11 pts) (10 pts) (9 pts) (8 pts) (7 pts) (6 pts) (5 pts) (4 pts) (3 pts) (2 pts) (1 pts) (0 pts) Expectations of knee 0-32 33-65 66-100 pain after recovery (0 pts) (1 pts) (2 pts) Active coping 7-12 13-18 19-24 25-30 31-35 (5 pts) (4 pts) (3 pts) (1 pts) (0 pts) Chronic widespread Yes No pain (1 pts) (0 pts) Total point Total point Risk of OKS ≤ 26 in % Total point Risk of OKS ≤ 26 in % < 4 ≤1 12 26 5 1 13 36 6 2 14 46 7 4 15 57 8 6 16 68 Note: For each instrument scale scores, enter the corresponding clinical points in the box on the right hand side. Add up the points and enter the total. Look 9 8 17 77 for the total point in the lower table and get the percentage risk of a poor 10 12 18 84 outcome post-TKA. 11 18 19 89 Figure 1. Points-based risk-scoring system for estimation of poor outcome (defined as Oxford Knee Score (OKS) ≤ 26) post-TKA. Scores range from 0 to 18 points. Downloaded from https://academic.oup.com/rheumap/advance-article-abstract/doi/10.1093/rap/rky021/5025119 by Ed 'DeepDyve' Gillespie user on 08 June 2018

Journal

Rheumatology Advances in PracticeOxford University Press

Published: May 29, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off