Background: Haemorrhagic shock is the leading cause of early preventable death in severe trauma. Delayed treatment is a recognized prognostic factor that can be prevented by efficient organization of care. This study aimed to develop and validate Red Flag, a binary alert identifying blunt trauma patients with high risk of severe haemorrhage (SH), to be used by the pre-hospital trauma team in order to trigger an adequate intra-hospital standardized haemorrhage control response: massive transfusion protocol and/or immediate haemostatic procedures. Methods: A multicentre retrospective study of prospectively collected data from a trauma registry (Traumabase®) was performed. SH was defined as: packed red blood cell (RBC) transfusion in the trauma room, or transfusion ≥ 4RBC in the first 6 h, or lactate ≥ 5 mmol/L, or immediate haemostatic surgery, or interventional radiology and/or death of haemorrhagic shock. Pre-hospital characteristics were selected using a multiple logistic regression model in a derivation cohort to develop a Red Flag binary alert whose performances were confirmed in a validation cohort. Results: Among the 3675 patients of the derivation cohort, 672 (18%) had SH. The final prediction model included five pre-hospital variables: Shock Index ≥ 1, mean arterial blood pressure ≤ 70 mmHg, point of care haemoglobin ≤ 13 g/dl, unstable pelvis and pre-hospital intubation. The Red Flag alert was triggered by the presence of any combination of at least two criteria. Its predictive performances were sensitivity 75% (72–79%), specificity 79% (77–80%) and area under the receiver operating characteristic curve 0.83 (0.81–0.84) in the derivation cohort, and were not significantly different in the independent validation cohort of 2999 patients. Conclusion: The Red Flag alert developed and validated in this study has high performance to accurately predict or exclude SH. Keywords: Severe trauma, Severe haemorrhage, Protocol, Organization, Anticipation * Correspondence: email@example.com Alexandra Rouquette and Jacques Duranteau contributed equally to this work. Université Paris Sud, Department of Anesthesiology and Critical Care, AP-HP, Bicêtre Hôpitaux Universitaires Paris Sud, 78 rue du Général Leclerc, F-94275 Le Kremlin Bicêtre, France CESP, INSERM, Université Paris-Sud, UVSQ, Université Paris-Saclay, Paris, CESP, INSERM, Maison de Solenn, 97 boulevard de Port-Royal, 75014 Paris, France Full list of author information is available at the end of the article © The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Hamada et al. Critical Care (2018) 22:113 Page 2 of 12 Background Methods Haemorrhage remains the leading cause of early prevent- Trauma centres and registry able death in severe trauma [1, 2]. A multidisciplinary ana- This multicentre observational study used the data col- lysis showed that approximately 2.5% of the deaths in a lected prospectively from a trauma registry (Trauma- trauma centre are preventable or potentially preventable. base®, traumabase.eu) shared between the six trauma Among the main causes were haemorrhage (39%) and mul- centres of the Paris area in France. These six centres tiple organ failure (28%), often a consequence of haemor- progressively joined the registry between January 2011 rhagic shock. The main reasons for preventable death due and June 2015. Since then, data collection is exhaustive to haemorrhage were delayed recognition and management and covers the whole administrative area around Paris, . Organizational optimization is essential to control Ile-de-France. The Traumabase® obtained approval from bleeding as quickly as possible and to reduce patient mor- the Advisory Committee for Information Processing in tality [4–6]. It is therefore crucial to identify during the pre- Health Research (CCTIRS, 11.305bis) and from the Na- hospital phase those patients at high risk of severe haemor- tional Commission on Informatics and Liberties (CNIL, rhage (SH) to quickly activate a specific intra-hospital stan- 911461) and meets the requirements of the local and na- dardized haemorrhage control response, connecting the tional ethics committee (Comité de Protection des Per- multispecialty trauma team, blood bank, transfusion proto- sonnes, Paris VI). The structure of the database integrates cols, interventional radiology and surgery . algorithms for consistency and coherence. Data monitoring Thus, to address efficiently the challenge of SH and is performed by a central administrator. Sociodemographic, shape the response, the design of a haemorrhage specific clinical, biological and therapeutic data (from the pre- alert is necessary. Standard triage algorithms are designed hospital phase to discharge from the intensive care unit) to guide severe trauma patients to appropriate trauma are systematically collected for all trauma patients. A de- centres [8, 9] and trigger trauma team activation . The scription of the Emergency Medical System (EMS) and the pre-hospital MGAP score (mechanism, Glasgow coma Trauma System and the characteristics of the Ile-de- scale, age and arterial pressure)  was developed to pre- France can be found in Hamada et al. . In France, dict mortality but showed a proper ability to predict SH a 24/7 available dispatching physician located in a (area under the curve 0.7, 95% CI 0.66–0.73) . Estab- centralized call centre decides which emergency vec- lished haemorrhage scores predict the need for massive tor, either a paramedic-staffed ambulance or a transfusion [13–15]. The TASH score is probably one of physician-staffed mobile intensive care unit, is to be the most widely cited scores to predict massive transfu- deployed on the basis of the trauma bystanders’ call. sion . However, massive transfusion only applies to a Patients brought to dedicated trauma rooms in minority of patients, whereas a timely integrative haemo- trauma centres are suspected to be major trauma by static strategy could decrease overall transfusion require- the pre-hospital team and are necessarily transported ments. The aforementioned massive transfusion scores by a physician-staffed ambulance. All patients trans- are only validated with intra-hospital data, which renders ported to the trauma rooms of the participating cen- their application questionable during the pre-hospital tres were included in the registry (Traumabase®). phase. A “Code Red” policy has been implemented in Our methodology used a three-step approach: devel- trauma centres across the UK [17, 18] with the pre-arrival oping a model, deriving a score and transforming the organization seen as an integral part of the severe haemor- score into a binary alert by choosing the cut-off value. rhage pathway . This activation code consists of three criteria (suspicion or evidence of active haemorrhage, sys- Patient selection tolic arterial blood pressure < 90 mmHg, failure to re- Every trauma patient registered in the Traumabase® since spond to a fluid bolus) but its predictive accuracy has not January 2011 was included in the study. Patients were ex- yet been evaluated. On a pragmatic standpoint, this type cluded if they were admitted after secondary transfer, after of alert is of utmost importance for trauma centres since penetrating trauma or after pre-hospital traumatic cardiac emergency care may compete with elective care because arrest, or if no pre-hospital data were available. Two sub- of common facilities and workforces. samples were constituted: the derivation cohort which in- The aim of our study was to develop and validate an cluded all patients admitted from January 2011 to May easy-to-use pre-hospital prediction tool for SH in blunt 2015, and the validation cohort which included all patients trauma patients derived from a prediction model. This admitted from June 2015 to November 2016 (Fig. 1). tool is meant to be used as a binary Red Flag alert to acti- vate a specific intra-hospital severe haemorrhage response. Definition of severe haemorrhage The Transparent Reporting of a multivariable prediction Patients were retrospectively considered to present SH model for Individual Prognosis Or Diagnosis (TRIPOD) on admission to the trauma centre if any of the follow- statement was followed to report its results . ing criteria was present: need for any packed red blood Hamada et al. Critical Care (2018) 22:113 Page 3 of 12 Fig. 1 Flowchart of the study. SH severe haemorrhage cell (RBC) transfusion upon arrival in the resuscitation chosen as a reflection of haemorrhage intensity, and for room, transfusion of 4 packed RBCs or more within the further anticipation of the need of haemostatic resource first 6 h [22, 23], blood lactate concentration ≥ 5 mmol/L mobilization. The resuscitation room transfusion criter- upon arrival , need for immediate haemostatic sur- ion and death secondary to haemorrhagic shock were gery or interventional radiology before complete injury chosen to select the most severe and actively bleeding pa- assessment by whole-body CT scan or death from haem- tients. The transfusion-related criteria represented the orrhagic shock . These criteria were chosen to reflect dynamic and evolving character of bleeding and need for the heterogeneity and complexity of SH clinical presenta- transfusion over the first hours. Massive transfusion is tion as there is no consensual definition in the literature. usually defined as 10 packed RBCs in the first 24 h  The need for an immediate haemostatic intervention was but this definition is currently questioned [23, 27] and Hamada et al. Critical Care (2018) 22:113 Page 4 of 12 other criteria (transfusion requirements in the first 6 h, validation studies) and therefore is expected to provide transfusion of at least 3 packed RBCs in 1 h or of 5 robust estimates . packed RBCs in 4 h) have been shown to be better correl- ate with mortality [22, 27]. The transfusion-related cri- Statistical analysis teria were chosen in this study to be a good balance Continuous data were described as mean ± standard de- between these different definitions. Finally, the criterion viation or median (quartiles 1–3) according to their dis- concerning blood lactate level at admission quantified the tribution, and categorical variables as count magnitude of tissue hypoperfusion related to haemor- (percentages). The derivation cohort was used to train rhage. Apart from metabolic sources for elevated blood the prediction model. Univariate analyses were per- lactate level (e.g. alcohol, fast, ethylene glycol), a level formed to evaluate crude associations between pre- higher than 5 mmol/L during the first 24 h is a risk factor hospital data and the presence of SH using chi-square for mortality or multiple organ failure [28–30]. For each and Student’s t tests (or the Mann–Whitney test when patient, the presence or absence of SH at admission was necessary) depending on the variable type. Each variable adjudicated by SRH, initially blinded to the pre-hospital with p < 0.2 was retained as a candidate variable and variables. Spearman correlation coefficients were computed to evaluate collinearity (r > 0.8). Candidate continuous vari- Potential pre-hospital predictors ables were dichotomized using receiver operating char- Thirteen predictors of SH were selected or computed acteristic (ROC) curves and Youden’s index to using exclusively pre-hospital data from management on identify the cut-off value, except for SpO for which the scene and during transport. These criteria were selected clinically relevant cut-off value of ≤ 90% was chosen. based on their clinical significance and ease of use in the Then, candidate binary variables were entered altogether pre-hospital time-critical setting: age, sex, minimal sys- into a multivariate logistic regression model and se- tolic, diastolic and mean arterial blood pressure (SBP, lection was performed using a backward stepwise pro- DBP and MBP), maximal heart rate (HR), minimal oxy- cedure to optimize the Akaike criterion. All variables gen saturation (SpO ), minimal Glasgow Coma Scale were tested for pairwise univariate interactions. Mul- (GCS), clinically unstable pelvis, early on-scene point-of- tiple imputation via a chained equation was used to care haemoglobin concentration , tracheal intubation handle missing data (R PACkage “mice” V2.3 ). and vasopressor administration. The maximal Shock A maximum of 5% missing data was observed for the Index was calculated according to the formula: SI = imputed variables. Model calibration was assessed maximal HR / minimal SBP . using the Hosmer and Lemeshow statistic . Model discrimination was assessed using the ROC curve Other measured variables (areaunder theROC curve(AUC))and abootstrap Demographic data, trauma characteristics and outcome methodology (1000 samples)  was used to quan- were also recorded. For the Simplified Acute Physiology tify any optimism (averaged difference between the Score (SAPS), the worst value of each variable over the apparent AUC of the model developed on each boot- first 24 h was taken into account. The Abbreviated In- strap sample and its AUC on the original sample) in jury Scale (AIS) version 2005 and the Injury Severity the final prediction model. Score (ISS) were calculated when the whole-body injury To derive the Red Flag binary alert from this model, a assessment was completed. The expected probability of score was computed for each patient of the derivation survival was calculated using the Trauma and Injury Se- cohort using various combinations of the number of verity Score (TRISS) with the most recent coefficients points attributed to each variable that remained in the [33, 34]. Due to missing data, the TRISS was computed final prediction model (one point for each variable or by giving a respiratory rate of 20/min in all patients . two points for variables with a higher model coefficient The MGAP score was computed as a comparison basis than the others). Their predictive accuracy was evaluated for SH prediction . using the AUC and the Youden’s index was used to iden- tify the cut-off value for the Red Flag binary alert with Sample size the best balance between simplicity of use and predictive We used the entire large cohort (7945 patients) as there performance assessed using sensitivity, specificity, posi- is no generally accepted approach to estimate the sample tive and negative predictive values (PPV and NPV) and size requirements for derivation and validation studies of positive likelihood ratio (+LR). Contingency mosaics risk prediction models. The number of events in our were drafted for the values around the determined cut- sample far exceeds the required number established off value. using previously published rules (10 events per candidate Finally, the predictive accuracy of the final prediction variable for derivation studies and at least 100 events for model and the predictive performance of the resulting Hamada et al. Critical Care (2018) 22:113 Page 5 of 12 Red Flag binary alert were independently assessed in the Discrimination was explored using the AUC ROC. A validation cohort. Model calibration was graphically bootstrap methodology was used to compute the confi- checked using a calibration plot representing the agree- dence interval (CI) of the AUCs (R package “pROC” V1. ment between risks of SH predicted by the model and 10 ). The AUC of Red Flag was compared to the observed proportions in the validation cohort. AUC of the MGAP score. All tests that were two-sided Table 1 Derivation and validation cohort characteristics Derivation Validation Comparison SH No SH SH No SH SH No SH (n = 672) (n = 3003) (n = 415) (n = 2584) p p Demography and outcome Age (years) 42 ± 19 37 ± 16 44 ± 19 38 ± 17 ns ns Male (%) 465 (69%) 2340 (78%) 312 (75%) 2017 (78%) 0.040 ns BMI (kg/m ) 25.3 ± 5 24.7 ± 4.3 25.2 ± 4.4 24.8 ± 4.7 ns ns SAPS II 45 ± 22 21 ± 15 46 ± 23 21 ± 15 ns ns ICU LOS 18 ± 23 8 ± 15 14 ± 19 7 ± 15 0.001 0.001 Hospital mortality 163 (25%) 137 (5%) 93 (23%) 110 (4%) ns ns Predicted mortality by TRISS (%) 27 6 27 6 Mechanism of injury (all blunt) MVA 133 (20%) 765 (26%) 97 (23%) 617 (24%) Motorbike 151 (23%) 905 (30%) 95 (23%) 861 (33%) Pedestrian and bicycle 109 (16%) 381 (13%) 47 (11%) 343 (13%) ns ns Fall 243 (36%) 792 (26%) 144 (35%) 621 (24%) Miscellaneous 36 (5%) 150 (5%) 32 (8%) 142 (6%) Severity of injuries ISS 30 (18–38) 12 (5–20) 27 (14–41) 12 (5–21) ns ns Head and neck AIS 2 (0–4) 0 (0–3) 1 (0–3) 0 (0–2) ns ns Thorax AIS 3 (0–3) 0 (0–3) 3 (0–4) 0 (0–3) ns ns Abdomen AIS 2 (0–3) 0 (0–2) 2 (0–3) 0 (0–2) ns ns Extremities pelvis AIS 3 (2–3) 2 (0–2) 3 (1–4) 0 (0–3) ns 0.001 At admission Total pre-hospital time (min) 85 ± 39 80 ± 37 80 ± 37 77 ± 34 0.010 ns SBP (mmHg) 102 ± 34 129 ± 23 107 ± 35 130 ± 24 ns ns DBP (mmHg) 62 ± 23 76 ± 16 65 ± 24 79 ± 17 ns 0.001 Haemoglobin (g/dl) 10.2 ± 2.6 13.4 ± 1.7 10.7 ± 2.7 13.5 ± 1.7 0.010 ns Lactate (mmol/L) 4.8 ± 3.4 1.9 ± 0.9 4.9 ± 3.4 2 ± 0.9 ns 0.002 Prothrombin time (%) 57 ± 22 83 ± 15 61 ± 23 84 ± 15 0.010 ns Surgery day 1 562 (84%) 1867 (62%) 305 (74%) 1235 (48%) 0.001 0.001 Angio-embolization day 1 111 (17%) 67 (2%) 88 (21%) 86 (3%) 0.001 0.001 SH characteristics Immediate surgery 123 (16%) – 88 (21%) – ns – Transfusion in trauma room 425 (63%) – 245 (59%) – ns – Lactates > 5 mmol/L 323 (51%) – 194 (52%) – ns – ≥ 4 RBCs in first 6 h 385 (57%) – 204 (49%) – 0.010 – Results expressed as mean ± standard deviation, n (%) or median (1st quartile–3rd quartile) SH severe haemorrhage, ns not significant, BMI body mass index, SAPS Simplified Acute Physiology Index, ICU LOS intensive care unit length of stay, TRISS Trauma Injury Severity Score, MVA motor vehicle accident, ISS Injury Severity Score, AIS Abbreviated Injury Scale, SBP systolic blood pressure, DBP diastolic blood pressure RBC red blood cell TRISS computed by giving a respiratory rate of 20/min in all patients  Hamada et al. Critical Care (2018) 22:113 Page 6 of 12 at p ≤ 0.05 were considered significant. R 3.3.3 software (R Score development Foundation for Statistical Computing, Vienna, Austria) Table 2 presents the results of the univariate analysis was used for analysis. performed in the derivation cohort (Additional file 2). All of the pre-hospital variables were significantly associ- ated with SH. Collinear variables were HR, SBP, Shock Results Index (Shock Index was kept), and SBP, DBP and MBP; The flowchart of both the derivation (n = 4339) and val- the latter was kept as not collinear to the Shock Index. idation (n = 3606) cohorts is presented in Fig. 1. Patients As reported in the lower part of Table 2, when there was presenting with initial cardiac arrest, penetrating trauma no clinically relevant cut-off value, Youden’s index was or no pre-hospital data available were excluded, leaving used to dichotomize continuous variables (resulting 3675 (85%) patients for analysis in the derivation cohort thresholds are presented in Additional file 3). and 2999 (83%) in the validation cohort. The distribu- Ten variables were thus included in the multivariate tion of patient across centres is described and illustrated model and seven variables were selected for the final in Additional file 1. prediction model (i.e. all but gender, GCS and vasopres- The main characteristics of both cohorts are presented sor administration) (Table 3). The model goodness-of-fit in Table 1. The observed mortality was lower in both was good according to the Hosmer–Lemeshow statistic groups than the expected mortality according to the (p = 0.60). The discrimination as evaluated by the AUC TRISS. The two cohorts significantly differed mainly in was 0.84 (95% CI 0.82–0.86). As internal validation, opti- terms of haemostatic strategies with less surgery (58 vs mism was evaluated at 0.001 using bootstrap method- 51%), more angio-embolization (3% vs 6%) and a shorter ology, so the optimism-corrected AUC was 0.84 (95% CI intensive care unit length of stay in the validation cohort. 0.82–0.85). Table 2 Univariate analysis of pre-hospital variables in the derivation cohort SH No SH (n = 3003) Missing values, p (n = 672) n (%) Male 465 (69%) 2258 (78%) 7 (0%) < 0.001 Age (years) 42 ± 19 37 ± 16 7 (0%) < 0.001 SBP min (mmHg) 93 ± 30 118 ± 22 50 (1%) < 0.001 DBP min (mmHg) 55 ± 18 70 ± 15 65 (2%) < 0.001 MBP min (mmHg) 68 ± 21 86 ± 16 50 (1%) < 0.001 HR max (/min) 108 ± 27 93 ± 20 73 (2%) < 0.001 Shock Index (HR/SBP) 1.3 ± 0.8 0.8 ± 0.4 77 (2%) < 0.001 Capillary haemoglobin (g/dl) 12.8 ± 2.2 14.2 ± 1.7 183 (5%) < 0.001 SpO min (%) 97 (92–100) 98 (96–100) 99 (3%) < 0.001 Glasgow Coma Scale 14 (7–15) 15 (14–15) 17 (0%) < 0.001 Pelvic trauma 115 (18%) 106 (4%) 141 (5%) < 0.001 Vasopressor 216 (33%) 140 (5%) 37 (1%) < 0.001 Pre-hospital intubation 385 (57%) 692 (23%) 6 (0%) < 0.001 Binarized variables (Youden’s Index) SBP min ≤ 100 421 (64%) 569 (19%) < 0.001 MBP ≤ 70 mmHg 382 (58%) 448 (15%) < 0.001 HR max ≥100 418 (64%) 1050 (36%) < 0.001 Shock Index (HR/SBP) ≥ 1 394 (60%) 419 (14%) < 0.001 Capillary haemoglobin ≤ 13 382 (59%) 812 (29%) < 0.001 SpO min ≤ 90% 142 (22%) 189 (6%) < 0.001 Glasgow Coma Scale ≤13 321 (48%) 712 (24%) < 0.001 Results expressed as mean ± standard deviation, n (%) or median (1st quartile–3rd quartile) SH severe haemorrhage, SBP systolic blood pressure, DBP diastolic blood pressure, MBP mean blood pressure, HR heart rate, SpO peripheral oxygen saturation, min minimal, max maximal Cut-off value not binarized with receiver operating characteristic curves Hamada et al. Critical Care (2018) 22:113 Page 7 of 12 Table 3 Results of multivariate stepwise analysis or equal to 2 points. Its predictive performances are pre- sented in Table 4. The three-variable combination (thresh- Pre-hospital criteria Coefficient OR 95% CI P old 1 point, Table 4) offers less performance (Delong test Shock Index > 1 1.32 3.76 2.96–4.78 < 0.001 for ROC curves of paired data, p<0.001), but seems to Pelvic trauma 1.32 3.76 2.68–5.28 < 0.001 be an interesting option especially in a non-physician- Pre-hospital intubation 0.98 2.67 2.17–3.28 < 0.001 staffed EMS (less oro-tracheal intubation and no available Capillary haemoglobin ≤ 13 g/dl 0.92 2.51 2.05–3.08 < 0.001 pre-hospital haemoglobin measurement). MBP ≤ 70 mmHg 0.87 2.38 1.88–3.02 < 0.001 Oxygen saturation minimal ≤ 90% 0.59 1.79 1.35–2.39 < 0.001 External validation Figure 2 shows the calibration plot of the final predic- Age > 50 years O.42 1.52 1.21–1.92 < 0.001 tion model in the 2999 patients of the validation cohort. Model intercept was −3.33 (p < 0.0001) OR odds ratio, CI confidence interval, MBP mean blood pressure The agreement between predicted probabilities and ob- served proportions was adequate, except in the group Table 4 presents the predictive performances of several with a predicted risk of SH from 40 to 50% in which the combinations of points attributed to each variable observed proportion of patients with SH was 32%. The remaining in the final prediction model. The combination AUC of the final prediction model in this population was associated with the optimal trade-off between predictive 0.80 (95% CI 0.78–0.83), similar to the AUC computed in accuracy (AUC 0.83 (0.81–0.84)), performance (sensitivity the derivation cohort (p = 0.19) and significantly higher 75% (72–79%), specificity 79% (77–80%)) and ease of use than the AUC of the MGAP score equal to 0.72 (95% CI was the simple sum of the following five criteria: SI ≥ 1, 0.69–0.73) (p < 0.001). The predictive performances of point of care haemoglobin ≤ 13 g/dl, pre-hospital intub- the Red Flag binary alert assessed in this population were: ation, MBP minimum ≤ 70 mmHg and clinical signs of sensitivity = 70% (95% CI 66–75), specificity = 80% (95% unstable pelvic fracture at any time during pre-hospital CI 78–81), PPV = 36% (95% CI 34–38) and NPV = 94% management. The Red Flag binary alert was considered to (95% CI 94–95). The contingency mosaics drafted for a be activated if this score, ranging from 0 to 5, was superior threshold of 2 and 3 points are shown in Fig. 3. Table 4 Predictive properties of the various combinations studied to identify the Red Flag binary alert Shading indicates chosen combination T threshold, Se sensitivity, Sp specificity, PPV positive predictive value, NPV negative predictive value, +LR positive likelihood ratio, –LR negative likelihood ratio, AUC area under the receiver operating characteristic curve, SI Shock Index, Pelvis unstable pelvis, OTI oro-tracheal intubation, Hb point-of-care haemoglobin, MBP mean blood pressure, SpO peripheral oxygen saturation 2 Hamada et al. Critical Care (2018) 22:113 Page 8 of 12 Fig. 2 Calibration plot of the model in the validation cohort: agreement between observed and predicted proportion of severe haemorrhage (SH) by the model Discussion haemorrhage [16, 41]. It is a source of internal bleeding In this study, a Red Flag binary alert derived from an ef- that is difficult to control especially in the case of arterial ficient combination of pre-hospital criteria was identified bleeding (20%) but also in the case of venous bleeding, des- with high predictive performances to detect patients at pite pelvic binding. In the TASH score, unstable pelvic risk of SH. Its high predictive performances were con- fracture accounts for about 20% of the total score (6 points firmed in internal and external validations. To our from 28), asinour study(1from5). TheShockIndexhas knowledge this is the first report of a validated pre- been demonstrated as a useful sign to diagnose acute hypo- hospital triggered haemorrhage pre-alert. In practice, the volemia and as a good marker of severe haemorrhagic presence of any combination of at least two criteria during shock . The threshold used in the Red Flag is 1, while the pre-hospital care phase among patients with SI (HR/ the most frequently suggested SI cut-off value to predict SBP) ≥ 1, unstable pelvic fracture, intubation, point of care massive transfusion is 0.9 in the literature . Also, the haemoglobin ≤ 13 g/dl or MBP ≤ 70 mmHg activates the threshold of haemoglobin concentration identified in our Red Flag and provides a powerful signal to initiate an ad- study was 1 point higher than the threshold used in the equate intra-hospital standardized haemorrhage control TASH score (13 g/dl vs 12 g/dl) . The timing and the response (massive transfusion protocol and/or immediate technique of measurement used in our study may explain haemostatic procedures). the difference. Indeed, in the present study, haemoglobin The criteria identified in this study as associated with SH concentration was assessed with a point-of-care technique share similarities with some used in previously described on scene, thus at a very early stage, whereas the TASH haemorrhage control pathways or massive transfusion score uses the haemoglobin laboratory concentration at scores. Unstable pelvic fracture is part of the TASH score, hospital admission. Blood pressure is also a key variable in and part of numerous existing scores predicting ongoing almost all existing predictive scores for severe haemorrhage Hamada et al. Critical Care (2018) 22:113 Page 9 of 12 Fig. 3 a Red Flag alert. b Contingency mosaic according to threshold of activation. FN false negative, FP false positive, Hb haemoglobin MBP mean arterial blood pressure, OTI Oro-tracheal intubation, SI Shock Index, TN true negative, TP true positive [16, 41]. Nevertheless, only SBP is used, while the oscillo- Those latter scores were validated for an intra-hospital metric sphygmomanometer, used in many EMSs, measures setting and include variables such as ultrasound use or a MBP and extrapolates SBP and DBP via an algorithm blood gas results, variables that are not systematically . For this reason, MBP was chosen in the Red Flag and available in the pre-hospital environment. the information carried by this variable was found inde- The present work is the first to attempt an extensive pendent of that carried by the SI. Intubation by the pre- assessment of the predictive performance of routine pre- hospital team, however, has never been suggested as being hospital data and to include a validation. A code should associated with severe haemorrhage in previous studies. In be easy to remember and the criteria routinely available; our pre-hospital system, physicians are involved in the pre- both apply to the Red Flag. Indeed, any prediction tool hospital setting and this may explain this association as it requires evaluation and validation in the very specific is usually the most severe patients who are intubated dur- setting it will be implemented in. In the case of an in- ing pre-hospital care . appropriate activation, the complete set of the haemor- The major advantage of this Red Flag alert is its sim- rhage control infrastructure may be activated and plicity of use and pragmatism as it is computed with disorganize programmed care for a while. So, any activa- routinely assessed variables and thus directly available tion code requires a delicate balance between sensitivity criteria for the pre-hospital care team. It allows the rapid and specificity; that is, between the risk of not activating identification of patients who require mobilization of im- the haemorrhage control pathway when it is needed portant human and material resources to control haem- (false negative, potentially detrimental to the patient) orrhage (advanced immediate resuscitation and/or and over-activation (increase in false positive). On the haemostatic procedures, early and sustained transfusion, one hand, it is crucial not to miss any haemorrhagic pa- etc.). The predictive performances of the previously de- tients and get activated for their arrival, but, on the scribed “Code Red” have not yet been extensively evalu- other, over-activation can generate waste of precious re- ated [17, 18]. The other existing simple scores are not sources and induces team fatigue leading to further non- based on pre-hospital variables , or were built to pre- compliance. Unjustified activation may even reduce the dict outcome such as mortality [11, 46]. The Red Flag chance of other patients to benefit from the resources could not be compared to the TASH or ABC score. inadequately put on standby. An appropriate number of Hamada et al. Critical Care (2018) 22:113 Page 10 of 12 activations, however, maintains team readiness and Additional file 2: Binarization of continuous variables according to training. The clinical consequences of this alert will have Youden’s Index. SpO binarized according to literature (cut-off value 90%). AUC area under the ROC curve. (DOCX 16 kb) to be assessed (times, process, outcomes, etc.). Additional file 3: Univariate analysis of pre-hospital variables in Our study obviously has some limitations. The first is its derivation cohort. Results expressed as mean ± standard deviation or retrospective design, as it usually precludes the ad hoc *median [1st quartile–3rd quartile]. SBP systolic blood pressure, DBP choice of the data studied. It might have been interesting diastolic blood pressure, MBP mean blood pressure, HR heart rate, SpO peripheral oxygen saturation, Min minimal, Max maximal. #Cut-off value not to investigate the contribution of other criteria which were binarized with ROC curves. (DOCX 19 kb) not collected in our study: pre-hospital ultrasonography, described as a bleeding characterization criterion in the Abbreviations ABC score , or pre-hospital blood lactate dosage de- +LR: Positive likelihood ratio; AIS: Abbreviated Injury Scale; AUC: Area under scribed as a predictor of trauma severity . However, the ROC curve; BMI: Body mass index; DBP: Diastolic arterial blood pressure; EMS: Emergency Medical Service; GCS: Glasgow Coma Scale; Hb: Haemoglobin; ICU this is actually a strength of our study, as it allowed the LOS: Intensive care unit length of stay; ISS: Injury Severity Score; MBP: Mean arterial analysis of pre-hospital data that are routinely collected in blood pressure; MGAP: Mechanism, Glasgow, age and arterial pressure; MVA: Motor practice by the EMS which reinforces the interest in Red vehicle accident; NPV: Negative predictive value; OTI: Oro-tracheal intubation; POC: Point of care; PPV: Positive predictive value; ROC: Receiver operating Flag as a pragmatic, easy-to-use tool. Moreover, beside characteristic; SAPS: Simplified Acute Physiology Index; SBP: Systolic arterial blood this retrospective analysis, data collection was prospective pressure; Se: Sensitivity; SH: Severe haemorrhage; SI: Shock Index; Sp: Specificity; as this study used data from the Traumabase®, and this SpO : Peripheral oxygen saturation; TRISS: Trauma Injury Severity Score has limited data loss and biases inherent to retrospective Acknowledgements data collection . Furthermore, this study is the first to The authors are grateful to Alexander Fordyce for his English editing. They evaluate the question in a physician-staffed EMS, whereas also thank Samuel Degoul for his statistical input on using the R software. *The Traumabase Group: Arie Attias, MD (Department of Anaesthesiology existing work has been generated in a paramedic-staffed and Critical Care, Hôpital Henri Mondor, APHP, Créteil, France); Sylvain Ausset, EMS. Transposition of experiences and data from one sys- MD (Anaesthesiology and Critical Care, Hôpital Interarmées Percy, Clamart, tem to another can be difficult. The external validity of France); Mathieu Boutonnet, MD (Anaesthesiology and Critical Care, Hôpital Interarmées Percy, Clamart, France); Gilles Dhonneur, MD, PhD (Université Paris the study could only be assessed by testing and validating Est and Department of Anaesthesiology and Critical Care, Hôpital Henri it outside the original centres. The characteristics of these Mondor, APHP, Créteil, France); Jacques Duranteau, MD, PhD (Université Paris centres, however, are quite different with regard to equip- Sud and Department of Anaesthesiology and Critical Care, Hôpital Bicêtre, Groupement Hôpitaux Universitaires Paris Sud, AP-HP, Kremlin Bicêtre, France); ment, internal organization and case mix as there is a lot Olivier Langeron, MD, PhD (Sorbonne Universités, UPMC Univ Paris 06 and of contrast in demographic characteristics of the different Department of Anaesthesiology and Critical Care, Groupe Hospitalier area covered by each centre within the region. So, results Pitié-Salpêtrière Charles Foix, AP-HP, Paris, France); Catherine Paugam-Burtz, MD, PhD (Université Denis Diderot and Beaujon University Hospital, Hôpitaux from this study are thus likely to be transposable to other Universitaires Paris Nord-Val-De-Seine, Clichy, AP-HP, France); Romain Pirracchio, centres. Furthermore, demographic and clinical character- MD, PhD (Université Paris Descartes and Department of Anaesthesiology and istics of our cohort, as well as mortality, were similar to Critical Care, Hôpital Européen Georges Pompidou, APHP, Paris, France); Bruno Riou, MD, PhD (Sorbonne Université, UMRS 1166, IHU ICAN and Department of those in the trauma literature [14, 49]. It is noteworthy Emergency Medicine, Groupe Hospitalier Pitié-Salpêtrière-Charles Foix, AP-HP, that this alert does not apply to penetrating trauma and Paris, France); Guillaume de St Maurice, MD (Anaesthesiology and Critical Care, has not been validated for children. Finally, the impact of Hôpital Interarmées Percy, Clamart, France); Bernard Vigué, MD (Department of Anaesthesiology and Critical Care, Hôpital Bicêtre, Groupement Hôpitaux this Red Flag alert on patient care has not been evaluated Universitaires Paris Sud, AP-HP, Kremlin Bicêtre, France). and requires a separate prospective study. Funding The Traumabase® has been sponsored by the Regional Health Agency of Ile de France for 2014–2017. Conclusion We have constructed and validated a simple Red Flag Availability of data and materials The datasets generated and analysed during the current study are not alert for identifying severe blunt trauma patients during publicly available due the restrictions around health data demanded by the the pre-hospital care phase and activating a specific im- Advisory Committee for Information Processing in Health Research and the mediate intra-hospital haemorrhage control response National Commission on Informatics and Liberties. An extract can be available from the corresponding author on reasonable request and with permission of prior to arrival. The impact of its use on severe trauma the aforementioned committees. patient care and on resource utilization remains to be determined. Authors’ contributions SRH contributed to the study design, data collection, data analysis and interpretation, literature search, and writing. ARos contributed to the data collection, data analysis and interpretation, literature search, and writing. TG contributed to study design, data collection, literature search, and writing. Additional files J-PD, MR, AH, AF, FC, and MB contributed to data collection and critical revision. JD contributed to study design, critical revision and writing. ARou contributed to the data analysis and interpretation, literature search, and writing. The Additional file 1: Distribution of origin of patients for derivation and Traumabase Group contributed to data collection and critical revision. The validation cohorts. (TIFF 14113 kb) Traumabase registry has provided all data used. All of the authors and the Hamada et al. Critical Care (2018) 22:113 Page 11 of 12 Traumabase Group are responsible for the scientific content. The final version 8. van der Wulp I, van Baar ME, Schrijvers AJP. Reliability and validity of the of the manuscript has been validated by the all of the authors and the Manchester Triage System in a general emergency department patient Traumabase Group. population in the Netherlands: results of a simulation study. Emerg Med J. 2008;25:431–4. 9. Riou B. Le traumatisé grave. Actualités en réanimation préhospitalière. Ethics approval and consent to participate Journées scientifiques des SAMU de France. In: SFEM; 2002. pp. 115-8. The data come from the Traumabase®. This registry obtained approval from 10. Brown JB, Lerner EB, Sperry JL, Billiar TR, Peitzman AB, Guyette FX. the Advisory Committee for Information Processing in Health Research Prehospital lactate improves accuracy of prehospital criteria for designating (CCTIRS, 11.305bis) and from the National Commission on Informatics and trauma activation level. J Trauma Acute Care Surg. 2016;81:445–52. Liberties (CNIL, 911461), and the analysis meets the requirement of the local 11. Sartorius D, Le Manach Y, David J-S, Rancurel E, Smail N, Thicoïpé M, et al. and national ethics committee (Comité de Protection des Personnes, Paris VI). Mechanism, glasgow coma scale, age, and arterial pressure (MGAP): a new simple prehospital triage score to predict mortality in trauma patients. Crit Competing interests Care Med. 2010;38:831–7. The authors declare that they have no competing interests. 12. Raux M, Sartorius D, Le Manach Y, David J-S, Riou B, Vivien B. What do prehospital trauma scores predict besides mortality? J Trauma Acute Care Surg. 2011;71:754–9. Publisher’sNote 13. Yücel N, Lefering R, Maegele M, Vorweg M, Tjardes T, Ruchholtz S, et al. Springer Nature remains neutral with regard to jurisdictional claims in Trauma Associated Severe Hemorrhage (TASH)-Score: probability of mass published maps and institutional affiliations. transfusion as surrogate for life threatening hemorrhage after multiple trauma. J Trauma. 2006;60:1228–36. Author details 14. McLaughlin DF, Niles SE, Salinas J, Perkins JG, Cox ED, Wade CE, et al. A Université Paris Sud, Department of Anesthesiology and Critical Care, AP-HP, predictive model for massive transfusion in combat casualty patients. J Bicêtre Hôpitaux Universitaires Paris Sud, 78 rue du Général Leclerc, F-94275 Trauma. 2008;64:S57–63. discussion S63 Le Kremlin Bicêtre, France. CESP, INSERM, Université Paris-Sud, UVSQ, 15. Nunez TC, Voskresensky IV, Dossett LA, Shinall R, Dutton WD, Cotton BA. Université Paris-Saclay, Paris, CESP, INSERM, Maison de Solenn, 97 boulevard Early prediction of massive transfusion in trauma: simple as ABC de Port-Royal, 75014 Paris, France. Université Paris Sud, Department of (assessment of blood consumption)? J Trauma. 2009;66:346–52. Anesthesiology and Critical Care, Assistance Publique-Hôpitaux de Paris 16. Brockamp T, Nienaber U, Mutschler M, Wafaisade A, Peiniger S, Lefering R, (AP-HP), Béclère Hôpitaux Universitaires Paris Sud, 157 rue de la porte de et al. Predicting on-going hemorrhage and transfusion requirement after Trivaux, 92140 Clamart, France. Hôpitaux Universitaires Paris Nord Val de severe trauma: a validation of six scoring systems and algorithms on the Seine, Department of Anesthesiology and Critical Care, AP-HP, Beaujon, 100 TraumaRegister DGU. Crit Care. 2012;16:R129. avenue du Général Leclerc, 92110 Clichy, France. Centre Hospitalier 17. Weaver AE, Hunter-Dunn C, Lyon RM, Lockey D, Krogh CL. The effectiveness Sud-Francilien, SAMU 91 Emergency Department, 40 avenue Serge Dassault, of a “Code Red” transfusion request policy initiated by pre-hospital 91100 Corbeil-Essonnes, France. Sorbonne Université and Department of physicians. Injury. 2016;47:3–6. Anesthesiology and Critical Care, AP-HP, Hôpitaux Universitaires 18. Reed MJ, Glover A, Byrne L, Donald M, McMahon N, Hughes N, et al. Pitié-Salpêtrière, 47–83 Boulevard de l’Hôpital, 75013 Paris, France. Université Experience of implementing a national pre-hospital Code Red bleeding Paris Descartes, Department of Anesthesiology and Critical Care, AP-HP, protocol in Scotland. Injury. 2017;48(1):41–6. Hôpital Européen Georges Pompidou—Hôpitaux Universitaires Paris Ouest, 19. Khan S, Allard S, Weaver A, Barber C, Davenport R, Brohi K. A major 20 rue Leblanc, 75015 Paris, France. Université Paris Est, Department of haemorrhage protocol improves the delivery of blood component therapy Anesthesiology and Critical Care, APHP, Hôpital Henri Mondor, 51 avenue du and reduces waste in trauma massive transfusion. Injury. 2013;44:587–92. Marechal de Lattre de Tassigny, 94010 Créteil, France. Department of 20. Moons KGM, Altman DG, Reitsma JB, Ioannidis JPA, Macaskill P, Steyerberg Anesthesiology and Critical Care, Percy Military Teaching Hospital, 101 EW, et al. Transparent Reporting of a multivariable prediction model for avenue Henri-Barbusse, 92140 Clamart, France. AP-HP, Bicêtre Hôpitaux Individual Prognosis Or Diagnosis (TRIPOD): explanation and elaboration. Universitaires Paris Sud, Public Health and Epidemiology Department, AP-HP, Ann Intern Med. 2015;162:W1. Bicêtre Hôpitaux Universitaires Paris Sud, 78 rue du Général Leclerc, F-94275 Le Kremlin Bicêtre, France. 21. Hamada SR, Gauss T, Duchateau F-X, Truchot J, Harrois A, Raux M, et al. Evaluation of the performance of French physician-staffed emergency Received: 5 December 2017 Accepted: 3 April 2018 medical service in the triage of major trauma patients. J Trauma Acute Care Surg. 2014;76:1476–83. 22. Mitra B, Cameron PA, Gruen RL, Mori A, Fitzgerald M, Street A. The definition of massive transfusion in trauma: a critical variable in examining References evidence for resuscitation. Eur J Emerg Med. 2011;18:137–42. 1. Kauvar DS, Wade CE. The epidemiology and modern management of 23. Savage SA, Zarzaur BL, Croce MA, Fabian TC. Redefining massive traumatic hemorrhage: US and international perspectives. Crit Care. 2005; transfusion when every second counts. J Trauma Acute Care Surg. 2013; 9(Suppl 5):S1–9. 74:396-400-402. 2. Dutton RP, Stansbury LG, Leone S, Kramer E, Hess JR, Scalea TM. Trauma 24. Callaway DW, Shapiro NI, Donnino MW, Baker C, Rosen CL. Serum lactate mortality in mature trauma systems: are we doing better? An analysis of and base deficit as predictors of mortality in normotensive elderly blunt trauma mortality patterns, 1997-2008. J Trauma. 2010;69:620–6. trauma patients. J Trauma. 2009;66:1040–4. 3. Teixeira PGR, Inaba K, Hadjizacharia P, Brown C, Salim A, Rhee P, et al. 25. Clarke JR, Trooskin SZ, Doshi PJ, Greenwald L, Mode CJ. Time to laparotomy Preventable or potentially preventable mortality at a mature trauma center. for intra-abdominal bleeding from trauma does affect survival for delays up J Trauma. 2007;63:1338–46. to 90 minutes. J Trauma. 2002;52:420–5. 4. Kauvar DS, Lefering R, Wade CE. Impact of hemorrhage on trauma 26. Como JJ, Dutton RP, Scalea TM, Edelman BB, Hess JR. Blood transfusion outcome: an overview of epidemiology, clinical presentations, and rates in the care of acute trauma. Transfusion. 2004;44:809–13. therapeutic considerations. J Trauma. 2006;60:S3–11. 27. Savage SA, Sumislawski JJ, Bell TM, Zarzaur BL. Utilizing group-based trajectory 5. Gerardo CJ, Glickman SW, Vaslef SN, Chandra A, Pietrobon R, Cairns CB. The modeling to understand patterns of hemorrhage and resuscitation. Ann Surg. rapid impact on mortality rates of a dedicated care team including trauma 2016;264:1135–41. and emergency physicians at an academic medical center. J Emerg Med. 2011;40:586–91. 28. Montmany Vioque S, Navarro Soto S, Rebasa Cladera P, Luna Aufroy A, 6. Cotton BA, Gunter OL, Isbell J, Au BK, Robertson AM, Morris JA, et al. Damage Gómez Díaz C, Llaquet BH. Measurement of lactic acid in multiple injury control hematology: the impact of a trauma exsanguination protocol on patients and its usefulness as a predictor of multiorgan failure and mortality. survival and blood product utilization. J Trauma. 2008;64:1177-1182-1183. Cir Esp. 2012;90:107–13. 7. Glen J, Constanti M, Brohi K, Guideline Development Group. Assessment 29. Régnier M-A, Raux M, Le Manach Y, Asencio Y, Gaillard J, Devilliers C, et al. and initial management of major trauma: summary of NICE guidance. BMJ. Prognostic significance of blood lactate and lactate clearance in trauma 2016;353:i3051. patients. Anesthesiology. 2012;117:1276–88. Hamada et al. Critical Care (2018) 22:113 Page 12 of 12 30. Raux M, Le Manach Y, Gauss T, Baumgarten R, Hamada S, Harrois A, et al. Comparison of the prognostic significance of initial blood lactate and base deficit in trauma patients. Anesthesiology. 2017;126:522–33. 31. Sanchis-Gomar F, Cortell-Ballester J, Pareja-Galeano H, Banfi G, Lippi G. Hemoglobin point-of-care testing: the HemoCue system. J Lab Autom. 2013;18:198–205. 32. Mutschler M, Nienaber U, Münzberg M, Wölfl C, Schoechl H, Paffrath T, et al. The Shock Index revisited—a fast guide to transfusion requirement? A retrospective analysis on 21,853 patients derived from the TraumaRegister DGU®. Crit Care. 2013;17:R172. 33. Boyd CR, Tolson MA, Copes WS. Evaluating trauma care: the TRISS method. Trauma Score and the Injury Severity Score. J Trauma. 1987;27:370–8. 34. Champion HR, Sacco WJ, Copes WS. Injury severity scoring again. J Trauma. 1995;38:94–5. 35. Raux M, Thicoïpé M, Wiel E, Rancurel E, Savary D, David J-S, et al. Comparison of respiratory rate and peripheral oxygen saturation to assess severity in trauma patients. Intensive Care Med. 2006;32:405–12. 36. Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3:32–5. 37. Buuren S, Groothuis-Oudshoorn K. mice: Multivariate imputation by chained equations in R. J Stat Softw. 2011 45. Available from: http://doc.utwente.nl/ 78938/. [cited 2017 Jun 26]. 38. Hosmer Jr DW, Lemeshow S, Sturdivant RX. Applied logistic regression. Wiley; 2013. Available from: http://books.google.com/books?hl=en&lr=&id= 64JYAwAAQBAJ&oi=fnd&pg=PA313&dq=info:QYB4u56LccYJ:scholar.google. com&ots=DrkS603smK&sig=hKtWAl9-_-bIkvzkyftnCdfc-cI [cited 2017 Jun 26]. 39. Steyerberg EW, Harrell FE, Borsboom GJ, Eijkemans MJC, Vergouwe Y, Habbema JDF. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol. 2001;54:774–81. 40. Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez J-C, et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC bioinformatics. 2011;12:77. 41. Tonglet ML. Early prediction of ongoing hemorrhage in severe trauma: presentation of the existing scoring systems. Arch Trauma Res. 2016;5(4): e33377. 42. Moffat B, Vogt KN, Inaba K. The Shock Index: is it ready for primetime? Crit Care. 2013;17:196. 43. Olaussen A, Blackburn T, Mitra B, Fitzgerald M. Review article: shock index for prediction of critical bleeding post-trauma: a systematic review. Emerg Med Australas. 2014;26:223–8. 44. Meldrum SJ. The principles underlying Dinamap—a microprocessor based instrument for the automatic determination of mean arterial pressure. J Med Eng Technol. 1978;2:243–6. 45. Yeguiayan J-M, Garrigue D, Binquet C, Jacquot C, Duranteau J, Martin C, et al. Medical pre-hospital management reduces mortality in severe blunt trauma: a prospective epidemiological study. Crit Care. 2011;15:R34. 46. Champion HR, Sacco WJ, Copes WS, Gann DS, Gennarelli TA, Flanagan ME. A revision of the Trauma Score. J Trauma. 1989;29:623–9. 47. Jansen TC, van Bommel J, Mulder PG, Rommes JH, Schieveld SJM, Bakker J. The prognostic value of blood lactate levels relative to that of vital signs in the pre-hospital setting: a pilot study. Crit Care. 2008;12:R160. 48. Euser AM, Zoccali C, Jager KJ, Dekker FW. Cohort studies: prospective versus retrospective. Nephron Clin Pract. 2009;113:c214–7. 49. MacKenzie EJ, Rivara FP, Jurkovich GJ, Nathens AB, Frey KP, Egleston BL, et al. A national evaluation of the effect of trauma-center care on mortality. N Engl J Med. 2006;354:366–78.
Critical Care – Springer Journals
Published: May 5, 2018
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
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera