Identifying Trigger Cues for Hospital Blood Transfusions Based on Ensemble Learning Methods

Identifying Trigger Cues for Hospital Blood Transfusions Based on Ensemble Learning Methods | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Identifying Trigger Cues for Hospital Blood Transfusions Based on Ensemble Learning Methods Eva V. Zadorozny, Tyler Weigel, Samuel M. Galvagno, Joshua B. Brown, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3944131/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 6 You are reading this latest preprint version Abstract Background Traumatic shock is the leading cause of preventable death with most patients dying within the first 6 hours. This underscores the importance of prehospital interventions, and growing evidence suggests prehospital transfusion improves survival. Optimizing transfusion triggers in the prehospital setting is key to improving outcomes for patients in hemorrhagic shock. Our objective was to identify factors associated with early in-hospital transfusion requirements available to prehospital clinicians in the field to develop a simple algorithm for prehospital transfusion, particularly for patients with occult shock. Methods We included trauma patients transported by a single critical care transport service to a level I trauma center between 2012 and 2019. We used logistic regression, Fast and Frugal Trees (FFTs), and Bayesian analysis to identify factors associated with early in-hospital blood transfusion as a potential trigger for prehospital transfusion. Results We included 2,157 patients transported from the scene or emergency department (ED) of whom 207 (9.60%) required blood transfusion within 4 hours of admission. The mean age was 47 (IQR = 28–62) and 1,480 (68.6%) patients were male. From 13 clinically relevant factors for early hospital transfusions, four were incorporated into the FFT in following order: 1) SBP, 2) prehospital lactate concentration, 3) Shock Index, 4) AIS of chest (sensitivity = 0.81, specificity = 0.71). The chosen thresholds were similar to conventional ones. Using conventional thresholds resulted in lower model sensitivity. Consistently, prehospital lactate was among most decisive factors of hospital transfusions identified by Bayesian analysis (OR = 2.31; 95% CI 1.55–3.37). Conclusions Using an ensemble of frequentist statistics, Bayesian analysis and machine learning, we developed a simple, clinically relevant, prehospital algorithm to help identify patients requiring transfusion within 4 hours of hospital arrival. prehospital transfusion early hospital transfusion hemorrhagic shock prehospital lactate concentration Fast Frugal Trees Bayesian analysis decision support models Figures Figure 1 Figure 2 Figure 3 Introduction Hemorrhagic shock is the leading cause of preventable death among injured patients. 1 Shock occurs in a continuum with progressive end-organ damage and leads to death if inadequately treated. Aggressive resuscitation according to damage control principles reduces the risk of death from hemorrhagic shock. 2 Damage control resuscitation with prehospital blood products lowers the risk of death, although the role for prehospital blood remains unclear. 3,4 Early resuscitation prevents the consequences of hemorrhagic shock and poor outcomes but is difficult to achieve in the prehospital environment with constrained diagnostic and therapeutic capabilities. Current field triage guidelines use vital signs and level of consciousness to determine the need for expedient transport to a trauma center, but these guidelines may overlook many patients with unrecognized or compensated shock who may benefit from early blood administration. 5 Indications for prehospital blood transfusion after injury vary considerably and rely on arbitrary vital sign thresholds and obvious symptoms of hemorrhagic shock. 6 Prior work shows that elevated serum lactate levels in trauma patients may indicate sepsis and multiorgan dysfunction, increasing the chance of mortality. 7,8 Prehospital clinicians can measure serum lactate levels using rapid, relatively inexpensive point of care tests to guide current triage decisions in the case of serious injury. In our previous work, we found that increased prehospital lactate levels were associated with higher odds of 24-hour hospital transfusion, even among patients without hypotension. 5 Prehospital lactate may be a useful prompt for prehospital transfusion. To mitigate significant physiologic derangement, prehospital professionals need a reliable but simple approach to rapidly and accurately identify patients who are most likely to benefit from prehospital blood. Our objective was to develop a parsimonious clinically relevant algorithm to identify patients requiring early hospital transfusion using data available in the prehospital setting. This algorithm may be a guide for prehospital blood product administration. We hypothesized that using state of the art statistical techniques to control for known confounders, we would identify a subset of factors highly predictive of transfusion need after injury, thereby creating a simple in-field operational model for identifying patients who need blood during trauma resuscitation. We aimed to compare the accuracy of data-driven methods with conventional triage criteria thresholds to determine variables with the optimal sensitivity and specificity for identifying trauma patients who require a blood transfusion. We also aimed to develop proof of concept decision models with components that could be adapted to different prehospital services such as rural versus urban settings. Methods We performed a retrospective analysis of prehospital factors that predict the need for emergent blood administration (within 4 hours) in adult (age > 16 years) trauma patients. The hours were calculated as number of minutes between ED arrival and discharge dates divided by 60. These dates are electronic timestamps. We included trauma patients with recorded venous lactate who were transported by a regional critical care transport service between 2012 and 2019. We excluded subjects with isolated traumatic brain injury (TBI) (18.6%), those that died in the emergency department (0.4%), and those with missing data ( 2) and no other severe injuries (AIS face, neck, chest, spine, arms, abdomen, legs, external > 2) as these patients are not likely to require transfusion. The University Human Research Protections Office approved this study. The data was from a regional critical care transport service that has 18 helicopter and 2 ground bases across four states. Blood is available at all bases; 2 units of PRBCs is taken by helicopters on all missions. Crews complete 13,000 missions annually and include a minimum of a critical care nurse and paramedic. They are trained to perform point of care testing for blood gases and lactate concentration (iSTAT One, CG4+, Abbott Laboratories Princeton, NJ). They use these data to inform resuscitation and titrate mechanical ventilation. To build an operational in-field model to identify the need for blood use, we used an ensemble of methodologic approaches. Our first approach was to construct Fast and Frugal Trees (FFTs) using prehospital factors associated with hospital blood administration, previously identified using logistic regression as influencing hospital blood decisions (Table 1). 5 Factors associated with hospital blood administration were used to find data-driven thresholds. The algorithm that builds FFTs compares FFT receiver operating characteristics to those of other common model-building approaches: CART, logistic regression. Random Forest (RF) and Support Vector Machine (SVM) methods (see Appendix). 9 We implemented FFT and Bayesian approaches as independent yet complimentary methods that validate each other’s findings. A heuristic (rule of thumb) FFT approach minimizes variance but is more prone to bias, 9 whereas a Bayesian approach is less biased and more prone to higher variance. 10 Using both FFT and Bayesian approaches minimizes the overall error from both bias and variance. FFTs are decision trees that differ from conventional decision trees in three ways: 1) they contain a minimal number of variables/cues needed to decide, 2) they make a decision after every node, and 3) they can only have two branches per node. 9 These trees are salient (we know how the machine arrived at the decision), robust against overfitting and good at identifying new cases of the outcome variable. This makes FFTs ideal to guide fast decisions in dynamic and dangerous environments. 9 We split the analysis data set 50/50 into training and testing datasets (a common starting point for evaluating machine learning algorithms) 11 and applied the FTT algorithm. For more information about the FFT algorithm, please see the Appendix/Supplemental Methods section. Our second approach was a Bayesian analysis of factors predicting in-hospital transfusion to confirm or supplement our prior approaches. Our goal was to identify a parsimonious model to predict transfusion within 4 hours of hospital admission. A Bayesian approach was employed for several reasons. First, prior information from our group and others may be used to provide updated knowledge about variables most strongly associated with the probability that a trauma patient requires a blood transfusion. Second, a hazard with frequentist statistics is that P values and confidence intervals may be difficult to interpret; highly significant P values may not be clinically meaningful or intuitively comprehensible. Third, Bayesian methods yield the probability of a specific outcome given the data. 10 Finally, we synthesized the results of our approaches to create a proposed clinical algorithm of indications for prehospital blood transfusion. This work adheres to STROBE guidelines of reporting in observational studies (Appendix Table 1). Data analysis was performed using R® version 4.1.2 (Vienna, Austria), SAS® version 9.4 (Carry, NC), and Stata® version 17 (College Station, TX). Results Of the patients transported over the 7-year study period, we identified 2,157 trauma patients with a prehospital lactate value (Fig. 1 ) obtained according to the Blood Administration protocol (Supplemental Table 1 and Appendix 2). STROBE guidelines are shown in grey rectangles. Subjects with trauma were received in a trauma or burn unit and/or had the following mechanisms of injury: assault, animal bite, burn, electrocution (non-lightning), gunshot wound, stabbing/cutting, machinery accident; pedestrian, bicycle, motor vehicle, all-terrain vehicle, motorcycle, water transport, or aircraft accident, crash or collision. Among the cohort, 1,480 (68.6%) patients were male, mean age was 47 (IQR = 28–62), and 207 (9.60%) patients had the primary outcome of requiring a blood transfusion within 4 hours of admission to the Emergency Department (Table 1). Table 1. Cohort characteristics * n (%) shown for categorical variables, median (IQR) shown for continuous variables ₀ within 24 hours of hospital admission + the rest of population was transported from scene • the rest of population had blunt injuries ♦ means for categorical variables were compared using Fisher’s exact test, for continuous variables – using Mann-Whitney U test Variable All Subjects (n = 2,157) 4h hospital ED blood products Yes (n = 207; 10%) No (n = 1,950; 90%) P value ♦ Prehospital venous lactate (mmol/L) 2.71 (1.40–3.15) 4.85 (2.30–5.80) 2.48 (1.30–2.98) < 0.01 Age (years) 47 (28–62) 49 (28– 65) 47 (29–62) 0.26 Sex (male) 1,480 (69) 141 (68.0) 1,339 (68.7) 0.88 CCI 0.67 (0–1) 0.78 (0–1) 0.66 (0–1) 0.62 ISS 11 ( 4 – 14 ) 20 (10–29) 9.55 ( 4 – 13 ) < 0.01 Lowest SI 0.59 (0.47–0.68) 0.74 (0.56–0.89) 0.58 (0.47–0.67) < 0.01 SI range 0 (-0.08–0.08) -0.02 (-0.15–0.14) 0 (-0.07–0.07) 0.68 Lowest SBP 115 (100–132) 85 (67–102) 118 (103–133) < 0.01 SBP < 90 mmHg * 345 ( 16 ) 129 (62) 216 ( 11 ) 120 bpm * 442 ( 20 ) 79 (38) 363 ( 19 ) < 0.01 Blood prior to EMS (ml) 37 (0–0) 110 (0–0) 29 (0–0) < 0.01 Blood by EMS (ml) 27 (0–0) 173 (0–300) 12 (0–0) < 0.01 Crystalloids prior to EMS (ml) 418 (0–500) 793 (100–1,000) 378 (0–500) < 0.01 Crystalloids by EMS (ml) 201 (50–200) 173 (0–300) 165 (50–150) < 0.01 Transfer * + 931 (43) 90 (43) 841 (43) 0.94 Penetrating * • 206 ( 10 ) 31 ( 14 ) 175 ( 9 ) 2 * 199 ( 9 ) 51 (25) 148 ( 8 ) 2 * 550 (26) 95 (46) 455 (23) 2 * 148 ( 7 ) 15 ( 7 ) 133 ( 7 ) 0.77 AIS abdomen > 2 * 28 ( 1 ) 4 ( 2 ) 24 ( 1 ) 0.34 AIS legs > 2 * 311 ( 14 ) 62 (30) 249 ( 13 ) < 0.01 Volume of hospital blood (ml) ₀ 497 (0–0) 3,610 (600–4,600) 167 (0–0) < 0.01 Mortality * ₀ 34 ( 2 ) 19 ( 10 ) 15 ( 1 ) < 0.01 Laparotomy * ₀ 319 ( 15 ) 125 (60) 194 ( 10 ) < 0.01 Thoracotomy * ₀ 274 ( 13 ) 81 (39) 193 ( 10 ) < 0.01 Craniotomy * ₀ 34 ( 2 ) 13 ( 7 ) 21 ( 1 ) < 0.01 Interventional radiology * ₀ 125 ( 6 ) 26 ( 13 ) 99 ( 5 ) < 0.01 Pelvic fixation * ₀ 9 (0.4) 4 ( 2 ) 5 (0.3) < 0.01 Vascular repair * ₀ 38 ( 2 ) 12 ( 6 ) 26 ( 1 ) < 0.01 The median prehospital lactate concentration was 4.85 mmol/L for the subjects who received blood products (IQR = 2.30–5.80), and 2.48 mmol/L for the subjects who did not require hospital blood products within 4 hours of arrival (IQR = 1.30–2.98). Of the subjects who received hospital blood products, 19 (10%) died within 24 hours of admission. Only 1% of the subjects who did not require hospital blood died within 24 hours of admission (n = 15). Consistently, a greater percentage of subjects who received hospital blood products needed other hospital life-saving interventions (LSIs) (Table 1). We excluded information about prehospital blood and crystalloids given by the prehospital care service and prior to arrival from the decision process because of significant collinearity (i.e., relationship between model predictors) related to in-hospital blood administration. We provided the FFT algorithm with 13 variables to choose from based on clinical value and availability to the prehospital clinicians. 12 Among them were AIS scores provided as a surrogate for injury condition that is visible to prehospital clinician, which we also previously found to associate with hospital transfusion. While we acknowledge the AIS value would not be available in the prehospital setting; however, we use them here as a proxy for clinically recognizable anatomic injury patterns that are used in the field by EMS clinicians for trauma triage purposes. Five of the thirteen variables were not selected by the algorithm as they were not associated with need for blood transfusion: 1) critical high heart > 120 bpm, 2) AIS abdomen > 2, 3) AIS spine > 2, 4) injury type (blunt or penetrating), and 5) shock index (SI) range (i.e., difference between highest and lowest SI). The algorithm generated four variables highly associated with hospital blood transfusions within 4 hours of arrival (Fig. 2 ). The variable chosen by the algorithm were evaluated in the following sequence: 1) minimum SBP (continuous), 2) prehospital venous lactate (continuous), 3) minimal SI (continuous), and 4) AIS chest > 2 (categorical). The predictors that were not selected by the FFT algorithm were 1) age, 2) mission type (scene or interfacility transfer), 3) AIS head > 2, and 4) AIS lower extremities > 2. The sensitivity for this FFT was 0.81 and specificity 0.71 based on data-driven variable sequence and thresholds. A pilot FFT was obtained using training and testing datasets (the testing dataset N = 1,121) and selected from a “fan” of possible trees as having the best balance between sensitivity and specificity. A default sensitivity weight of 0.5 resulted in a “zig-zag” shape with alternating decisions. The ROC panel shows a comparison of parameters for the resulting FFT and other common model-building approaches: CART (C, red), Logistic Regression (LR, blue), Random Forest (RF, purple) and Support Vector Machine (SVM, yellow). We applied the FFT definitions from the pilot experiment with rounded thresholds to the entire study population and got similar performance (Supplemental Fig. 1A, sensitivity = 0.84, specificity = 0.70). Next, we maximized the sensitivity parameter with an aim to administer hospital blood to the greatest number of eligible patients while minimizing erroneous administrations. Setting the weighting parameter to any value in 0.7-1 range resulted in a “positive-rake” FFT that made positive blood decisions after every node (Supplemental Fig. 1B, sensitivity = 0.93, specificity = 0.39). Also, from Supplemental Fig. 1B, we can see that the Positive Predictive Value (PPV) for our model is 14.0% (192 / 1,373), while the Negative Predictive Value is 98.1% (769 / 784), confirming that our model rarely mis-identifies a patient needing 4-hour hospital transfusion. The resulting FFT out-performed other model-building approaches (e.g., CART and logistic regression (LR)) by creating a decision support model for early hospital blood administration with higher sensitivity and specificity (Supplemental Fig. 1B). Finally, we altered the tree definitions with conventional thresholds used in current field triage guidelines and the literature to simplify for potential use in the prehospital environment. 13 The FFT algorithm found variable thresholds that were different from conventional ones (Fig. 2 ). We explored thresholds already in common use (i.e., SBP threshold of 90 mmHg and prehospital lactate of 4 mmol/L) or based on ease of calculation for the prehospital provider (SI > 1 = HR > BP). 14 Applying conventional thresholds (Supplemental Fig. 1C) instead data-driven ones (Supplemental Fig. 1B) to the dataset greatly reduces the sensitivity but increases the specificity parameter. We tested (a) how altering the FFT definition with conventional thresholds would influence the sensitivity and specificity parameters (Supplemental Fig. 1C, Table 2 , first blue row) and (b) if a balance between specificity and sensitivity can be reached by using a combination of conventional and newly found thresholds (Table 2 , yellow rows). The trees were created the same way as in Supplemental Fig. 1B (Table 2 , first row) differing only by the threshold values (thresholds and parameters of FFT from Supplemental Fig. 1B are highlighted orange in Table 2 ). Table 2 illustrates how varying the threshold for SBP, lactate, and shock index alters the sensitivity, specificity, and overall performance based on Youden's As expected, using a higher SBP, lower lactate, or lower SI threshold increases sensitivity but decreases specificity. Table 2. Effect of using deduced, conventional, or mixed thresholds on FFT parameters # - indicates the FFT model number; models with number ‘2.1’ or higher did not include lactate as a variable; Sens. – sensitivity, Spec. – specificity Youden’s J statistic = sensitivity + specificity – 1 summarizes the performance of each model. Tree from row 1.1 is depicted in Supplemental Figure 1B. Tree from row 1.8 is depicted in Supplemental Figure 1C. We performed sensitivity analyses by removing the lactate term from the models and using FFT-derived vs. conventional thresholds for SBP and SI (Table 2 , rows 2.1–2.4), recognizing that prehospital lactate may not be widely available. The sensitivity was often higher for the models containing the lactate term (compare rows 2.1 and 1.1/3, 2.2 and 1.2/4, 2.3 and 1.5/7, 2.4 and 1.6/8), but the specificity and Youden’s J index were lower. We also assessed current practice of prehospital blood transfusion by the critical care service and the need for early in-hospital transfusion. Table 3 shows a cross-tabulation of actual prehospital blood administration by early hospital transfusions. Of 207 subjects who required early hospital transfusions, 79 (38.2%) subjects also received blood before arriving to the hospital (Table 3 , upper left quadrant). The majority (73) of these 79 subjects had SBP < 90 mmHg and received prehospital blood according to the prehospital care service protocol for blood transfusions. Among 60 patients who received prehospital transfusions but did not require hospital blood (Table 3 , upper right quadrant), 33 (55.0%) patients had SBP 90 mmHg, either received the product on the order of the physician or in deviation from the protocol. Table 3 Cross-tabulation of prehospital transfusions by 24-hour hospital transfusions 4h hospital ED blood = YES 4h hospital ED blood = NO Prehospital blood = YES 79 (3.66%) 60 (2.78%) Prehospital blood = NO 128 (5.93%) 1,890 (87.6%) In our Bayesian analysis the most predictive model demonstrated statistically significant associations with tachycardia (OR = 1.74; 95% CI 1.12–2.55), elevated prehospital lactate (OR = 2.31; 95% CI 1.55–3.37), and hypotension (OR = 11.59; 95% CI 7.70-16.98) for early in-hospital transfusion. In the Bayesian subgroup analysis of patients with SBP > 90mmHg (N = 1,901; 87.6%), the most predictive model included minimum shock index (OR = 25.6; 95% CI 2.54–113.2), elevated lactate (OR = 2.17; 95% CI 1.11–3.77), and tachycardia (OR = 1.59; 95% CI 0.72–2.94). Based on the 95% credible intervals, in the hypotensive cohort, lactate and minimum shock index were significantly associated with a higher posterior probability of early in-hospital transfusion. Synthesizing and operationalizing the results from our approaches for potential field use, we developed an algorithm for prehospital blood transfusion that incorporates prehospital SBP, prehospital lactate, shock index, and severe abdominal injuries (Fig. 3 ). This algorithm allows for different threshold values that may be tailored according to system resources and time considerations. We also applied the FFT definitions from Supplemental Fig. 1B but excluding the node for severe chest injuries (Supplemental Fig. 2). The resulting sensitivity and specificity parameters were slightly lower than those of the four-factor model (Supplemental Fig. 1B, Supplemental Fig. 2). Discussion Using advanced statistical methods to control for confounders and to maximize the information provided by a large cohort of adult trauma patients with granular prehospital data, we identified four variables that predict early in-hospital transfusions. These variables, which are accessible by prehospital clinicians, were selected by an FFT algorithm to facilitate the decision to administer prehospital blood quickly with a parsimonious (small) set of data. We confirmed these findings using Bayesian analysis to identify strong predictors of early in-hospital transfusion. Prehospital lactate emerged as a strong predictor for transfusion need from both the FFT and Bayesian approaches among patients who were not hypotensive. This is consistent with recent study by Griggs et al. who also predicted in hospital transfusion using prehospital lactate concentration. 15 Administration of prehospital blood products to patients in hemorrhagic shock reduces mortality. 4 A systematic review and meta-analysis by Rijnhout et al. describes the administration of prehospital blood products as feasible and safe but describes the evidence as low quality and difficult to compare because there are no standard indication for transfusion. 16 While tools have been developed to identify patient at risk of Trauma Associated Severe Hemorrhage (TASH) and for massive transfusion (ABC score), they rely on data not readily available in the prehospital environment (hemoglobin and ultrasound) and neither was developed for the prehospital environment. 17,18 To find the simplest decision model to identify people who need prehospital blood transfusions, we are faced with two competing considerations: 1) correctly identifying the greatest number of people who need blood (i.e., maximizing the sensitivity of the model), and 2) conserving limited resources of blood. Using these considerations, an EMS Medical Director may conclude that the model with a lactate concentration threshold of 2.5 mmol/L (Sensitivity = 0.89, Specificity = 0.48, Table 2 row 1.6) is more appropriate for use in a rural setting with delayed access to a trauma center and subsequent damage control resuscitation, while a model with a 4 mmol/L threshold (Sensitivity = 0.82, Specificity = 0.62, Table 2 row 1.8) could be more suitable for urban settings with short prehospital times. Similar trade-offs can be made with the thresholds for SBP and shock index. We adjusted the model thresholds to create simple rules for quick reference in the field (Table 2 ). The results depicted in Table 2 have broad implications for prehospital clinicians, ranging from urban and rural EMS systems to austere military environments that might require prolonged field care. Using the four variables derived from our models, prehospital system leadership can decide what thresholds are appropriate for transfusions in their respective systems, based on existing resources and trauma center access. Previous studies associate prehospital lactate with mortality and morbidity in trauma patients. 19,20 Subsequent work demonstrated the association between lactate and need for life saving interventions. 21 Recent work by Fukuma et al. and Galvagno et al. established that prehospital lactate threshold of > 4 mmol/L is associated with the need life-saving interventions for hemorrhage control. 14,22 This threshold is more conservative than the one found by the FFT algorithm (2.5 mmol/L). The last cue identified to trigger potential transfusion is severe chest injury. In the data we used AIS > 2; however, recognizing this is not available in the field setting as an objective number, this would rely on clinical exam evidence, much in the way the anatomic triage criteria for the national field triage guidelines are identified. We suggest operationalizing this as flail chest, unstable chest fractures, or need for needle decompression (Fig. 3 ). Local medical directors certainly would have discretion to operationalize this cue in an alternative way given the personnel, resources, and trauma population seen by his or her EMS agency. We do show comparable accuracy if this cue is omitted, allowing further adaptation to local circumstances given it is the most subjective cue in operational form. A key limitation of our study is that decision to transfuse blood is not always synonymous with the need to transfuse blood. Also, our analyses are retrospective and derived from a single EMS agency serving a regional trauma system. The dataset was limited to patients who had lactate sampled which imparts bias among patients with hemorrhagic shock. Selection bias may result when treatment priorities preclude sampling of lactate in the sickest patients. EMS data is rarely inputted into the record contemporaneously with care and is subject to recall and reporting bias. We import data electronically (vital signs, times and point of care labs) into the prehospital health record which mitigates these biases. There is likely a selection and sensitivity bias as our critical care organization is called for patients with more severe injuries or those who are geographically distant from trauma care. Conclusion We developed a parsimonious, clinically relevant algorithm to identify patients who may require prehospital transfusion. This algorithm accounts for prehospital lactate concentration which is useful for identifying patients with occult shock not meeting the conventional threshold for hypotension. Thresholds of decision factors should be adjusted to meet the needs and resources of a given prehospital trauma system. Further work is necessary to externally validate this algorithm for prehospital blood transfusion. We are including the Appendix describing the FFT algorithm, Blood Administration protocol, and the study checklist for adhering to STROBE guidelines as Supplemental Digital Content. Abbreviations AIS – Abbreviated Injury Scale ; bpm – beats per minute ; ED – Emergency Department; EHR – Electronic Health Record; EMS – Emergency Medical Services; ISS – Injury Severity Score; SBP – Systolic Blood Pressure; SI – Shock Index; TC – Trauma Center; ROC - Receiver Operating Characteristics; FFT – Fast Frugal Tree Declarations Sources of Funding: Our research is supported by the NIH through grant 5K23NS097629. Author contributions: FXG, JBB and SMG guided the research. SMG and TW worked on Bayesian analysis. FXG, JBB and EVZ worked on FFTs. All authors contributed to writing and editing the manuscript. Ethical approval We certify that the study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. University of Pittsburgh Internal Review Board (IRB) PRO15020269 approved this study. Consent to participate IRB granted a waiver of consent for this study. Consent for publication Not applicable Availability of data and materials The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. Competing interests The authors declare that they have no competing interests. References Drake SA, Holcomb JB, Yang Y, et al. Establishing a regional trauma preventable/potentially preventable death rate. Ann Surg . Published online 2020. Holcomb JB, Tilley BC, Baraniuk S, et al. Transfusion of plasma, platelets, and red blood cells in a 1:1:1 vs a 1:1:2 ratio and mortality in patients with severe trauma: The PROPPR randomized clinical trial. JAMA - Journal of the American Medical Association . Published online 2015. Brown JB, Sperry JL, Fombona A, Billiar TR, Peitzman AB, Guyette FX. Pre-trauma center red blood cell transfusion is associated with improved early outcomes in air medical trauma patients. J Am Coll Surg . Published online 2015. Sperry JL, Guyette FX, Brown JB, et al. Prehospital plasma during air medical transport in trauma patients at risk for hemorrhagic shock. New England Journal of Medicine . 2018;379(4):315-326. Zadorozny E v., Weigel T, Stone A, et al. Prehospital Lactate is Associated with the Need for Blood in Trauma. Prehospital Emergency Care . Published online 2021. Pokorny DM, Braverman MA, Edmundson PM, et al. The use of prehospital blood products in the resuscitation of trauma patients: a review of prehospital transfusion practices and a description of our regional whole blood program in San Antonio, TX . ISBT Sci Ser . 2019;14(3):332-342. Tobias AZ, Guyette FX, Seymour CW, et al. Pre-resuscitation lactate and hospital mortality in prehospital patients. Prehospital Emergency Care . Published online 2014. Mikkelsen ME, Miltiades AN, Gaieski DF, et al. Serum lactate is associated with mortality in severe sepsis independent of organ failure and shock. Crit Care Med . Published online 2009. Phillips ND, Neth H, Woike JK, Gaissmaier W. FFTrees: A toolbox to create, visualize, and evaluate fast-and-frugal decision trees. Judgm Decis Mak . Published online 2017. Jansen JO, Pallmann P, Maclennan G, Campbell MK. Bayesian clinical trial designs: Another option for trauma trials? Journal of Trauma and Acute Care Surgery . 2017;83(4):736-741. Richert Willi, Coelho LPedro. Building Machine Learning Systems with Python - Willi Richert - Google Bøker. Packit Publishing . 2013;1:290. Zadorozny E v., Weigel T, Stone A, et al. Prehospital Lactate is Associated with the Need for Blood in Trauma. Prehospital Emergency Care . Published online 2021. Sasser S, Hunt R, Faul M, et al. Guidelines for Field Triage of Injured Patients. MMWR Recommendations and Reports . 2012;61(1):1-21. Galvagno SM, Sikorski RA, Floccare DJ, et al. Prehospital Point of Care Testing for the Early Detection of Shock and Prediction of Lifesaving Interventions. Shock . 2020;54(6):710-716. Griggs JE, Lyon RM, Sherriff M, Barrett JW, Wareham G, ter Avest E. Predictive clinical utility of pre-hospital point of care lactate for transfusion of blood product in patients with suspected traumatic haemorrhage: derivation of a decision-support tool. Scand J Trauma Resusc Emerg Med . 2022;30(1):1-9. Rijnhout TWH, Wever KE, Marinus RHAR, Hoogerwerf N, Geeraedts LMG, Tan ECTH. Is prehospital blood transfusion effective and safe in haemorrhagic trauma patients? A systematic review and meta-analysis. Injury . Published online 2019. Yücel N, Lefering R, Maegele M, et al. Trauma Associated Severe Hemorrhage (TASH)-score: Probability of mass transfusion as surrogate for life threatening hemorrhage after multiple trauma. Journal of Trauma - Injury, Infection and Critical Care . Published online 2006. Nunez TC, Voskresensky I V., Dossett LA, Shinall R, Dutton WD, Cotton BA. Early prediction of massive transfusion in trauma: Simple as ABC (Assessment of Blood Consumption)? Journal of Trauma - Injury, Infection and Critical Care . Published online 2009. Guyette F, Suffoletto B, Castillo JL, Quintero J, Callaway C, Puyana JC. Prehospital serum lactate as a predictor of outcomes in trauma patients: A retrospective observational study. Journal of Trauma - Injury, Infection and Critical Care . Published online 2011. St John AE, McCoy AM, Moyes AG, Guyette FX, Bulger EM, Sayre MR. Prehospital lactate predicts need for resuscitative care in non-hypotensive trauma patients. Western Journal of Emergency Medicine . Published online 2018. Guyette FX, Meier EN, Newgard C, et al. A comparison of prehospital lactate and systolic blood pressure for predicting the need for resuscitative care in trauma transported by ground. Journal of Trauma and Acute Care Surgery . Published online 2015. Fukuma H, Nakada T aki, Shimada T, et al. Prehospital lactate improves prediction of the need for immediate interventions for hemorrhage after trauma. Sci Rep . Published online 2019. Additional Declarations No competing interests reported. Supplementary Files SDCAppendix.docx SDCAppendix2.docx SupplementalTable1.docx SupplementalFigure.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 24 Apr, 2024 Reviewers agreed at journal 13 Apr, 2024 Reviewers invited by journal 08 Apr, 2024 Editor assigned by journal 18 Feb, 2024 Submission checks completed at journal 18 Feb, 2024 First submitted to journal 09 Feb, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3944131","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":273602150,"identity":"017eb54a-1e7b-4d92-9d2d-3ce69ad677dd","order_by":0,"name":"Eva V. Zadorozny","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8ElEQVRIiWNgGAWjYHACAyR2BRAzMzeQouUMSAsjKVoY28Akfi3y7Yc3Pq6oqMszOMB8TOLnvNpo/naglh8V23BqYexJKzY8c4at2OAAW5pk77bjuTMOMzYw9py5jVMLM0OOmWRjG0/ihgM8xga8247lNgC1MDO24dbCxv8GpEUCqIX/s+HfOcdy5xPSwiMBtsUAZAvjY96GmtwNhLRISDwrNmw4k5A48zCb4WOZYwdyNwK1HMTnF/n+5I0PGyrqEvuONz84+KamLnfe+cMHH/yowK0FDhQOgykIeYCwepB1DWCqjijFo2AUjIJRMLIAAGkyW1jArJKyAAAAAElFTkSuQmCC","orcid":"","institution":"University of Pittsburgh","correspondingAuthor":true,"prefix":"","firstName":"Eva","middleName":"V.","lastName":"Zadorozny","suffix":""},{"id":273602151,"identity":"f0e62948-e983-4a94-b01f-39aeb5a14331","order_by":1,"name":"Tyler Weigel","email":"","orcid":"","institution":"University of Pittsburgh School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Tyler","middleName":"","lastName":"Weigel","suffix":""},{"id":273602152,"identity":"06282ca6-7500-4a66-82b1-91d48edaaa9b","order_by":2,"name":"Samuel M. Galvagno","email":"","orcid":"","institution":"University of Maryland Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Samuel","middleName":"M.","lastName":"Galvagno","suffix":""},{"id":273602153,"identity":"87343f06-28dd-4b92-9018-aaf61efd30d8","order_by":3,"name":"Joshua B. Brown","email":"","orcid":"","institution":"University of Pittsburgh School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Joshua","middleName":"B.","lastName":"Brown","suffix":""},{"id":273602154,"identity":"84604827-bc38-4fbe-9b65-09b8249a8bb7","order_by":4,"name":"Francis X. Guyette","email":"","orcid":"","institution":"University of Pittsburgh School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Francis","middleName":"X.","lastName":"Guyette","suffix":""}],"badges":[],"createdAt":"2024-02-09 20:44:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3944131/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3944131/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":51396582,"identity":"6b8b950d-d99b-498e-9b55-4b9e1903fbb0","added_by":"auto","created_at":"2024-02-20 20:18:55","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":61101,"visible":true,"origin":"","legend":"\u003cp\u003eFlow diagram illustrating cohort selection\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/9b6d057c634ff4e9772d2c57.png"},{"id":51396875,"identity":"df331175-130e-409d-bc77-ca68403db053","added_by":"auto","created_at":"2024-02-20 20:26:55","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":32852,"visible":true,"origin":"","legend":"\u003cp\u003ePilot FFT chosen by the algorithm\u003c/p\u003e\n\u003cp\u003eThe top panels show the number of observations and outcome (4-hour hospital blood administration).\u003c/p\u003e\n\u003cp\u003eSBP_min – minimal SBP (mmHg), nlacven – prehospital lactate concentration (mmol/L), si_min – minimum SI (bpm/mmHg), ais_ab – AIS for abdomen (0/1, equal to 1 if the AIS is greater than 2).\u003c/p\u003e\n\u003cp\u003e“Hits” (green triangles) refer to correct blood administrations; “misses” (red triangles) – to incorrect rejections. Sensitivity (triangles) = Hits / (Hits + Misses)\u003c/p\u003e\n\u003cp\u003eCorrect rejections (green circles) refer to correct decisions to \u003cem\u003enot\u003c/em\u003e give blood, false alarms (red circles) – to false positives, or incorrect decisions to give blood. Specificity (circles) = Correct Rejections / (Correct Rejections + False Alarms)\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/4a36555f414c3cec83f93f79.png"},{"id":51396584,"identity":"631e1f05-1873-458a-8593-8a1c22b81270","added_by":"auto","created_at":"2024-02-20 20:18:55","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":37893,"visible":true,"origin":"","legend":"\u003cp\u003eSummary of obtained decision rules and how they may guide prehospital transfusions\u003c/p\u003e\n\u003cp\u003eThe rules were obtained based on the need for 4-hour in-hospital transfusions.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/2f76ee12a8aacb95d5736d73.png"},{"id":51397329,"identity":"9bff6a67-6bc5-4551-bf71-f8444bec382b","added_by":"auto","created_at":"2024-02-20 20:34:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":393865,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/c86c675c-a008-452c-be8d-d0bdcc55cae6.pdf"},{"id":51396583,"identity":"441ee8c5-8360-4cee-a5c5-6f3e55079f1e","added_by":"auto","created_at":"2024-02-20 20:18:55","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":21144,"visible":true,"origin":"","legend":"","description":"","filename":"SDCAppendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/25bb9b16f15299ca39c98890.docx"},{"id":51396586,"identity":"99bdda14-127d-4aad-944d-93f759b590e1","added_by":"auto","created_at":"2024-02-20 20:18:55","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":725689,"visible":true,"origin":"","legend":"","description":"","filename":"SDCAppendix2.docx","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/8e9061fd25b48e182d318ab2.docx"},{"id":51396587,"identity":"1ded736a-af6d-4d40-b2b3-035601e99f15","added_by":"auto","created_at":"2024-02-20 20:18:55","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":14498,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalTable1.docx","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/353c2fa25264135f3ad24aab.docx"},{"id":51396588,"identity":"4fabea66-fe9b-4bce-aca4-7ab6fe511c8b","added_by":"auto","created_at":"2024-02-20 20:18:55","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":502211,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure.docx","url":"https://assets-eu.researchsquare.com/files/rs-3944131/v1/d563a2f3bbae8721fe9e5b4f.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Identifying Trigger Cues for Hospital Blood Transfusions Based on Ensemble Learning Methods","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHemorrhagic shock is the leading cause of preventable death among injured patients.\u003csup\u003e1\u003c/sup\u003e Shock occurs in a continuum with progressive end-organ damage and leads to death if inadequately treated. Aggressive resuscitation according to damage control principles reduces the risk of death from hemorrhagic shock.\u003csup\u003e2\u003c/sup\u003e Damage control resuscitation with prehospital blood products lowers the risk of death, although the role for prehospital blood remains unclear.\u003csup\u003e3,4\u003c/sup\u003e Early resuscitation prevents the consequences of hemorrhagic shock and poor outcomes but is difficult to achieve in the prehospital environment with constrained diagnostic and therapeutic capabilities. Current field triage guidelines use vital signs and level of consciousness to determine the need for expedient transport to a trauma center, but these guidelines may overlook many patients with unrecognized or compensated shock who may benefit from early blood administration.\u003csup\u003e5\u003c/sup\u003e Indications for prehospital blood transfusion after injury vary considerably and rely on arbitrary vital sign thresholds and obvious symptoms of hemorrhagic shock.\u003csup\u003e6\u003c/sup\u003e\u003c/p\u003e \u003cp\u003ePrior work shows that elevated serum lactate levels in trauma patients may indicate sepsis and multiorgan dysfunction, increasing the chance of mortality.\u003csup\u003e7,8\u003c/sup\u003e Prehospital clinicians can measure serum lactate levels using rapid, relatively inexpensive point of care tests to guide current triage decisions in the case of serious injury. In our previous work, we found that increased prehospital lactate levels were associated with higher odds of 24-hour hospital transfusion, even among patients without hypotension.\u003csup\u003e5\u003c/sup\u003e Prehospital lactate may be a useful prompt for prehospital transfusion. To mitigate significant physiologic derangement, prehospital professionals need a reliable but simple approach to rapidly and accurately identify patients who are most likely to benefit from prehospital blood. Our objective was to develop a parsimonious clinically relevant algorithm to identify patients requiring early hospital transfusion using data available in the prehospital setting. This algorithm may be a guide for prehospital blood product administration.\u003c/p\u003e \u003cp\u003eWe hypothesized that using state of the art statistical techniques to control for known confounders, we would identify a subset of factors highly predictive of transfusion need after injury, thereby creating a simple in-field operational model for identifying patients who need blood during trauma resuscitation. We aimed to compare the accuracy of data-driven methods with conventional triage criteria thresholds to determine variables with the optimal sensitivity and specificity for identifying trauma patients who require a blood transfusion. We also aimed to develop proof of concept decision models with components that could be adapted to different prehospital services such as rural versus urban settings.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eWe performed a retrospective analysis of prehospital factors that predict the need for emergent blood administration (within 4 hours) in adult (age\u0026thinsp;\u0026gt;\u0026thinsp;16 years) trauma patients. The hours were calculated as number of minutes between ED arrival and discharge dates divided by 60. These dates are electronic timestamps. We included trauma patients with recorded venous lactate who were transported by a regional critical care transport service between 2012 and 2019. We excluded subjects with isolated traumatic brain injury (TBI) (18.6%), those that died in the emergency department (0.4%), and those with missing data (\u0026lt;\u0026thinsp;4%). Isolated TBI was defined as head abbreviated injury scale (AIS\u0026thinsp;\u0026gt;\u0026thinsp;2) and no other severe injuries (AIS face, neck, chest, spine, arms, abdomen, legs, external\u0026thinsp;\u0026gt;\u0026thinsp;2) as these patients are not likely to require transfusion. The University Human Research Protections Office approved this study.\u003c/p\u003e \u003cp\u003eThe data was from a regional critical care transport service that has 18 helicopter and 2 ground bases across four states. Blood is available at all bases; 2 units of PRBCs is taken by helicopters on all missions. Crews complete 13,000 missions annually and include a minimum of a critical care nurse and paramedic. They are trained to perform point of care testing for blood gases and lactate concentration (iSTAT One, CG4+, Abbott Laboratories Princeton, NJ). They use these data to inform resuscitation and titrate mechanical ventilation.\u003c/p\u003e \u003cp\u003eTo build an operational in-field model to identify the need for blood use, we used an ensemble of methodologic approaches. Our first approach was to construct Fast and Frugal Trees (FFTs) using prehospital factors associated with hospital blood administration, previously identified using logistic regression as influencing hospital blood decisions (Table\u0026nbsp;1).\u003csup\u003e5\u003c/sup\u003e Factors associated with hospital blood administration were used to find data-driven thresholds. The algorithm that builds FFTs compares FFT receiver operating characteristics to those of other common model-building approaches: CART, logistic regression. Random Forest (RF) and Support Vector Machine (SVM) methods (see Appendix).\u003csup\u003e9\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eWe implemented FFT and Bayesian approaches as independent yet complimentary methods that validate each other\u0026rsquo;s findings. A heuristic (rule of thumb) FFT approach minimizes variance but is more prone to bias,\u003csup\u003e9\u003c/sup\u003e whereas a Bayesian approach is less biased and more prone to higher variance.\u003csup\u003e10\u003c/sup\u003e Using both FFT and Bayesian approaches minimizes the overall error from both bias and variance.\u003c/p\u003e \u003cp\u003eFFTs are decision trees that differ from conventional decision trees in three ways: 1) they contain a minimal number of variables/cues needed to decide, 2) they make a decision after every node, and 3) they can only have two branches per node.\u003csup\u003e9\u003c/sup\u003e These trees are salient (we know how the machine arrived at the decision), robust against overfitting and good at identifying new cases of the outcome variable. This makes FFTs ideal to guide fast decisions in dynamic and dangerous environments.\u003csup\u003e9\u003c/sup\u003e We split the analysis data set 50/50 into training and testing datasets (a common starting point for evaluating machine learning algorithms)\u003csup\u003e11\u003c/sup\u003e and applied the FTT algorithm. For more information about the FFT algorithm, please see the Appendix/Supplemental Methods section.\u003c/p\u003e \u003cp\u003eOur second approach was a Bayesian analysis of factors predicting in-hospital transfusion to confirm or supplement our prior approaches. Our goal was to identify a parsimonious model to predict transfusion within 4 hours of hospital admission. A Bayesian approach was employed for several reasons. First, prior information from our group and others may be used to provide updated knowledge about variables most strongly associated with the probability that a trauma patient requires a blood transfusion. Second, a hazard with frequentist statistics is that P values and confidence intervals may be difficult to interpret; highly significant P values may not be clinically meaningful or intuitively comprehensible. Third, Bayesian methods yield the probability of a specific outcome given the data.\u003csup\u003e10\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eFinally, we synthesized the results of our approaches to create a proposed clinical algorithm of indications for prehospital blood transfusion.\u003c/p\u003e \u003cp\u003e This work adheres to STROBE guidelines of reporting in observational studies (Appendix Table\u0026nbsp;1). Data analysis was performed using R\u0026reg; version 4.1.2 (Vienna, Austria), SAS\u0026reg; version 9.4 (Carry, NC), and Stata\u0026reg; version 17 (College Station, TX).\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eOf the patients transported over the 7-year study period, we identified 2,157 trauma patients with a prehospital lactate value (Fig. \u003cspan\u003e1\u003c/span\u003e) obtained according to the Blood Administration protocol (Supplemental Table\u0026nbsp;1 and Appendix 2).\u003c/p\u003e\n\u003cp\u003eSTROBE guidelines are shown in grey rectangles.\u003c/p\u003e\n\u003cp\u003eSubjects with trauma were received in a trauma or burn unit and/or had the following mechanisms of injury: assault, animal bite, burn, electrocution (non-lightning), gunshot wound, stabbing/cutting, machinery accident; pedestrian, bicycle, motor vehicle, all-terrain vehicle, motorcycle, water transport, or aircraft accident, crash or collision.\u003c/p\u003e\n\u003cp\u003eAmong the cohort, 1,480 (68.6%) patients were male, mean age was 47 (IQR\u0026thinsp;=\u0026thinsp;28\u0026ndash;62), and 207 (9.60%) patients had the primary outcome of requiring a blood transfusion within 4 hours of admission to the Emergency Department (Table\u0026nbsp;1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable\u0026nbsp;1.\u003c/strong\u003e Cohort characteristics\u003c/p\u003e\n\u003cp\u003e* n (%) shown for categorical variables, median (IQR) shown for continuous variables\u003c/p\u003e\n\u003cp\u003e₀ within 24 hours of hospital admission\u003c/p\u003e\n\u003cp\u003e+ the rest of population was transported from scene\u003c/p\u003e\n\u003cp\u003e\u0026bull; the rest of population had blunt injuries\u003c/p\u003e\n\u003cp\u003e\u0026diams; means for categorical variables were compared using Fisher\u0026rsquo;s exact test, for continuous variables \u0026ndash; using Mann-Whitney U test\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Taba\" border=\"1\"\u003e\n \u003ccolgroup cols=\"5\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eAll Subjects (n\u0026thinsp;=\u0026thinsp;2,157)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003e4h hospital ED blood products\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eYes (n\u0026thinsp;=\u0026thinsp;207; 10%)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNo (n\u0026thinsp;=\u0026thinsp;1,950; 90%)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eP value\u003csup\u003e\u0026diams;\u003c/sup\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePrehospital venous lactate (mmol/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.71 (1.40\u0026ndash;3.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.85 (2.30\u0026ndash;5.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.48 (1.30\u0026ndash;2.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e47 (28\u0026ndash;62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e49 (28\u0026ndash; 65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e47 (29\u0026ndash;62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.26\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSex (male)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,480 (69)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e141 (68.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,339 (68.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.67 (0\u0026ndash;1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.78 (0\u0026ndash;1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.66 (0\u0026ndash;1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eISS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (\u003cspan\u003e4\u003c/span\u003e\u0026ndash;\u003cspan\u003e14\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e20 (10\u0026ndash;29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9.55 (\u003cspan\u003e4\u003c/span\u003e\u0026ndash;\u003cspan\u003e13\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLowest SI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.59 (0.47\u0026ndash;0.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.74 (0.56\u0026ndash;0.89)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.58 (0.47\u0026ndash;0.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSI range\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0 (-0.08\u0026ndash;0.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-0.02 (-0.15\u0026ndash;0.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0 (-0.07\u0026ndash;0.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLowest SBP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e115 (100\u0026ndash;132)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e85 (67\u0026ndash;102)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e118 (103\u0026ndash;133)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSBP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg *\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e345 (\u003cspan\u003e16\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e129 (62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e216 (\u003cspan\u003e11\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHeart rate\u0026thinsp;\u0026gt;\u0026thinsp;120 bpm *\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e442 (\u003cspan\u003e20\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e79 (38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e363 (\u003cspan\u003e19\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlood prior to EMS (ml)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e37 (0\u0026ndash;0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e110 (0\u0026ndash;0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e29 (0\u0026ndash;0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlood by EMS (ml)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27 (0\u0026ndash;0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e173 (0\u0026ndash;300)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12 (0\u0026ndash;0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCrystalloids prior to EMS (ml)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e418 (0\u0026ndash;500)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e793 (100\u0026ndash;1,000)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e378 (0\u0026ndash;500)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCrystalloids by EMS (ml)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e201 (50\u0026ndash;200)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e173 (0\u0026ndash;300)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e165 (50\u0026ndash;150)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTransfer * \u003csup\u003e+\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e931 (43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e90 (43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e841 (43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePenetrating * \u003csup\u003e\u0026bull;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e206 (\u003cspan\u003e10\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e31 (\u003cspan\u003e14\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e175 (\u003cspan\u003e9\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAIS head\u0026thinsp;\u0026gt;\u0026thinsp;2 *\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e199 (\u003cspan\u003e9\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e51 (25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e148 (\u003cspan\u003e8\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAIS chest\u0026thinsp;\u0026gt;\u0026thinsp;2 *\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e550 (26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e95 (46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e455 (23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAIS spine\u0026thinsp;\u0026gt;\u0026thinsp;2 *\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e148 (\u003cspan\u003e7\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15 (\u003cspan\u003e7\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e133 (\u003cspan\u003e7\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAIS abdomen\u0026thinsp;\u0026gt;\u0026thinsp;2 *\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (\u003cspan\u003e1\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4 (\u003cspan\u003e2\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24 (\u003cspan\u003e1\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.34\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAIS legs\u0026thinsp;\u0026gt;\u0026thinsp;2 *\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e311 (\u003cspan\u003e14\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e62 (30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e249 (\u003cspan\u003e13\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVolume of hospital blood (ml)\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e497 (0\u0026ndash;0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3,610 (600\u0026ndash;4,600)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e167 (0\u0026ndash;0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMortality *\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34 (\u003cspan\u003e2\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e19 (\u003cspan\u003e10\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15 (\u003cspan\u003e1\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLaparotomy *\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e319 (\u003cspan\u003e15\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e125 (60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e194 (\u003cspan\u003e10\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eThoracotomy *\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e274 (\u003cspan\u003e13\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81 (39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e193 (\u003cspan\u003e10\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCraniotomy *\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34 (\u003cspan\u003e2\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 (\u003cspan\u003e7\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e21 (\u003cspan\u003e1\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInterventional radiology *\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e125 (\u003cspan\u003e6\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26 (\u003cspan\u003e13\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e99 (\u003cspan\u003e5\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePelvic fixation *\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9 (0.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4 (\u003cspan\u003e2\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5 (0.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVascular repair *\u003csup\u003e₀\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e38 (\u003cspan\u003e2\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12 (\u003cspan\u003e6\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26 (\u003cspan\u003e1\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe median prehospital lactate concentration was 4.85 mmol/L for the subjects who received blood products (IQR\u0026thinsp;=\u0026thinsp;2.30\u0026ndash;5.80), and 2.48 mmol/L for the subjects who did not require hospital blood products within 4 hours of arrival (IQR\u0026thinsp;=\u0026thinsp;1.30\u0026ndash;2.98). Of the subjects who received hospital blood products, 19 (10%) died within 24 hours of admission. Only 1% of the subjects who did not require hospital blood died within 24 hours of admission (n\u0026thinsp;=\u0026thinsp;15). Consistently, a greater percentage of subjects who received hospital blood products needed other hospital life-saving interventions (LSIs) (Table\u0026nbsp;1).\u003c/p\u003e\n\u003cp\u003eWe excluded information about prehospital blood and crystalloids given by the prehospital care service and prior to arrival from the decision process because of significant collinearity (i.e., relationship between model predictors) related to in-hospital blood administration. We provided the FFT algorithm with 13 variables to choose from based on clinical value and availability to the prehospital clinicians.\u003csup\u003e12\u003c/sup\u003e Among them were AIS scores provided as a surrogate for injury condition that is visible to prehospital clinician, which we also previously found to associate with hospital transfusion. While we acknowledge the AIS value would not be available in the prehospital setting; however, we use them here as a proxy for clinically recognizable anatomic injury patterns that are used in the field by EMS clinicians for trauma triage purposes. Five of the thirteen variables were not selected by the algorithm as they were not associated with need for blood transfusion: 1) critical high heart\u0026thinsp;\u0026gt;\u0026thinsp;120 bpm, 2) AIS abdomen\u0026thinsp;\u0026gt;\u0026thinsp;2, 3) AIS spine\u0026thinsp;\u0026gt;\u0026thinsp;2, 4) injury type (blunt or penetrating), and 5) shock index (SI) range (i.e., difference between highest and lowest SI).\u003c/p\u003e\n\u003cp\u003eThe algorithm generated four variables highly associated with hospital blood transfusions within 4 hours of arrival (Fig. \u003cspan\u003e2\u003c/span\u003e). The variable chosen by the algorithm were evaluated in the following sequence: 1) minimum SBP (continuous), 2) prehospital venous lactate (continuous), 3) minimal SI (continuous), and 4) AIS chest\u0026thinsp;\u0026gt;\u0026thinsp;2 (categorical). The predictors that were not selected by the \u003cem\u003eFFT\u003c/em\u003e algorithm were 1) age, 2) mission type (scene or interfacility transfer), 3) AIS head\u0026thinsp;\u0026gt;\u0026thinsp;2, and 4) AIS lower extremities\u0026thinsp;\u0026gt;\u0026thinsp;2. The sensitivity for this FFT was 0.81 and specificity 0.71 based on data-driven variable sequence and thresholds.\u003c/p\u003e\n\u003cp\u003eA pilot FFT was obtained using training and testing datasets (the testing dataset N\u0026thinsp;=\u0026thinsp;1,121) and selected from a \u0026ldquo;fan\u0026rdquo; of possible trees as having the best balance between sensitivity and specificity. A default sensitivity weight of 0.5 resulted in a \u0026ldquo;zig-zag\u0026rdquo; shape with alternating decisions. The ROC panel shows a comparison of parameters for the resulting FFT and other common model-building approaches: CART (C, red), Logistic Regression (LR, blue), Random Forest (RF, purple) and Support Vector Machine (SVM, yellow).\u003c/p\u003e\n\u003cp\u003eWe applied the FFT definitions from the pilot experiment with rounded thresholds to the entire study population and got similar performance (Supplemental Fig.\u0026nbsp;1A, sensitivity\u0026thinsp;=\u0026thinsp;0.84, specificity\u0026thinsp;=\u0026thinsp;0.70). Next, we maximized the sensitivity parameter with an aim to administer hospital blood to the greatest number of eligible patients while minimizing erroneous administrations. Setting the weighting parameter to any value in 0.7-1 range resulted in a \u0026ldquo;positive-rake\u0026rdquo; FFT that made positive blood decisions after every node (Supplemental Fig.\u0026nbsp;1B, sensitivity\u0026thinsp;=\u0026thinsp;0.93, specificity\u0026thinsp;=\u0026thinsp;0.39). Also, from Supplemental Fig.\u0026nbsp;1B, we can see that the Positive Predictive Value (PPV) for our model is 14.0% (192 / 1,373), while the Negative Predictive Value is 98.1% (769 / 784), confirming that our model rarely mis-identifies a patient needing 4-hour hospital transfusion.\u003c/p\u003e\n\u003cp\u003eThe resulting FFT out-performed other model-building approaches (e.g., CART and logistic regression (LR)) by creating a decision support model for early hospital blood administration with higher sensitivity and specificity (Supplemental Fig.\u0026nbsp;1B). Finally, we altered the tree definitions with conventional thresholds used in current field triage guidelines and the literature to simplify for potential use in the prehospital environment.\u003csup\u003e13\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eThe FFT algorithm found variable thresholds that were different from conventional ones (Fig. \u003cspan\u003e2\u003c/span\u003e). We explored thresholds already in common use (i.e., SBP threshold of 90 mmHg and prehospital lactate of 4 mmol/L) or based on ease of calculation for the prehospital provider (SI\u0026thinsp;\u0026gt;\u0026thinsp;1\u0026thinsp;=\u0026thinsp;HR\u0026thinsp;\u0026gt;\u0026thinsp;BP).\u003csup\u003e14\u003c/sup\u003e Applying conventional thresholds (Supplemental Fig. 1C) instead data-driven ones (Supplemental Fig. 1B) to the dataset greatly reduces the sensitivity but increases the specificity parameter. We tested (a) how altering the FFT definition with conventional thresholds would influence the sensitivity and specificity parameters (Supplemental Fig. 1C, Table \u003cspan\u003e2\u003c/span\u003e, first blue row) and (b) if a balance between specificity and sensitivity can be reached by using a combination of conventional and newly found thresholds (Table \u003cspan\u003e2\u003c/span\u003e, yellow rows). The trees were created the same way as in Supplemental Fig. 1B (Table \u003cspan\u003e2\u003c/span\u003e, first row) differing only by the threshold values (thresholds and parameters of FFT from Supplemental Fig. 1B are highlighted orange in Table \u003cspan\u003e2\u003c/span\u003e). Table \u003cspan\u003e2\u003c/span\u003e illustrates how varying the threshold for SBP, lactate, and shock index alters the sensitivity, specificity, and overall performance based on Youden\u0026apos;s As expected, using a higher SBP, lower lactate, or lower SI threshold increases sensitivity but decreases specificity. \u0026nbsp; \u0026nbsp; \u0026nbsp;\u003c/p\u003e\n\u003cdiv\u003e\u003cbr\u003e\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u0026nbsp;\u003c/strong\u003eEffect of using deduced, conventional, or mixed thresholds on FFT parameters\u003c/p\u003e\n\u003cp\u003e# - indicates the FFT model number; models with number \u0026lsquo;2.1\u0026rsquo; or higher did not include lactate as a variable; Sens. \u0026ndash; sensitivity, Spec. \u0026ndash; specificity\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eYouden\u0026rsquo;s J statistic = sensitivity + specificity \u0026ndash; 1 summarizes the performance of each model.\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"https://myfiles.space/user_files/122228_c8a1650c59388082/122228_custom_files/img1708424996.png\"\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cdiv\u003e\n \u003cp\u003eTree from row 1.1 is depicted in Supplemental Figure 1B.\u003c/p\u003e\n \u003cp\u003eTree from row 1.8 is depicted in Supplemental Figure 1C.\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003eWe performed sensitivity analyses by removing the lactate term from the models and using FFT-derived vs. conventional thresholds for SBP and SI (Table \u003cspan\u003e2\u003c/span\u003e, rows 2.1\u0026ndash;2.4), recognizing that prehospital lactate may not be widely available. The sensitivity was often higher for the models containing the lactate term (compare rows 2.1 and 1.1/3, 2.2 and 1.2/4, 2.3 and 1.5/7, 2.4 and 1.6/8), but the specificity and Youden\u0026rsquo;s J index were lower.\u003c/p\u003e\n\u003cp\u003eWe also assessed current practice of prehospital blood transfusion by the critical care service and the need for early in-hospital transfusion. Table \u003cspan\u003e3\u003c/span\u003e shows a cross-tabulation of actual prehospital blood administration by early hospital transfusions. Of 207 subjects who required early hospital transfusions, 79 (38.2%) subjects also received blood before arriving to the hospital (Table \u003cspan\u003e3\u003c/span\u003e, upper left quadrant). The majority (73) of these 79 subjects had SBP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg and received prehospital blood according to the prehospital care service protocol for blood transfusions. Among 60 patients who received prehospital transfusions but did not require hospital blood (Table \u003cspan\u003e3\u003c/span\u003e, upper right quadrant), 33 (55.0%) patients had SBP\u0026thinsp;\u0026lt;\u0026thinsp;90 mmHg. Patients who received blood with systolic blood pressures\u0026thinsp;\u0026gt;\u0026thinsp;90 mmHg, either received the product on the order of the physician or in deviation from the protocol.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 3\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eCross-tabulation of prehospital transfusions by 24-hour hospital transfusions\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"3\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e4h hospital ED blood\u0026thinsp;=\u0026thinsp;\u003cem\u003eYES\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e4h hospital ED blood\u0026thinsp;=\u0026thinsp;\u003cem\u003eNO\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrehospital blood\u0026thinsp;=\u003c/strong\u003e\u0026thinsp;\u003cstrong\u003eYES\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e79 (3.66%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e60 (2.78%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrehospital blood\u0026thinsp;=\u003c/strong\u003e\u0026thinsp;\u003cstrong\u003eNO\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128 (5.93%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1,890 (87.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eIn our Bayesian analysis the most predictive model demonstrated statistically significant associations with tachycardia (OR\u0026thinsp;=\u0026thinsp;1.74; 95% CI 1.12\u0026ndash;2.55), elevated prehospital lactate (OR\u0026thinsp;=\u0026thinsp;2.31; 95% CI 1.55\u0026ndash;3.37), and hypotension (OR\u0026thinsp;=\u0026thinsp;11.59; 95% CI 7.70-16.98) for early in-hospital transfusion. In the Bayesian subgroup analysis of patients with SBP\u0026thinsp;\u0026gt;\u0026thinsp;90mmHg (N\u0026thinsp;=\u0026thinsp;1,901; 87.6%), the most predictive model included minimum shock index (OR\u0026thinsp;=\u0026thinsp;25.6; 95% CI 2.54\u0026ndash;113.2), elevated lactate (OR\u0026thinsp;=\u0026thinsp;2.17; 95% CI 1.11\u0026ndash;3.77), and tachycardia (OR\u0026thinsp;=\u0026thinsp;1.59; 95% CI 0.72\u0026ndash;2.94). Based on the 95% credible intervals, in the hypotensive cohort, lactate and minimum shock index were significantly associated with a higher posterior probability of early in-hospital transfusion.\u003c/p\u003e\n\u003cp\u003eSynthesizing and operationalizing the results from our approaches for potential field use, we developed an algorithm for prehospital blood transfusion that incorporates prehospital SBP, prehospital lactate, shock index, and severe abdominal injuries (Fig. \u003cspan\u003e3\u003c/span\u003e). This algorithm allows for different threshold values that may be tailored according to system resources and time considerations.\u003c/p\u003e\n\u003cp\u003eWe also applied the FFT definitions from Supplemental Fig.\u0026nbsp;1B but excluding the node for severe chest injuries (Supplemental Fig.\u0026nbsp;2). The resulting sensitivity and specificity parameters were slightly lower than those of the four-factor model (Supplemental Fig.\u0026nbsp;1B, Supplemental Fig.\u0026nbsp;2).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eUsing advanced statistical methods to control for confounders and to maximize the information provided by a large cohort of adult trauma patients with granular prehospital data, we identified four variables that predict early in-hospital transfusions. These variables, which are accessible by prehospital clinicians, were selected by an FFT algorithm to facilitate the decision to administer prehospital blood quickly with a parsimonious (small) set of data. We confirmed these findings using Bayesian analysis to identify strong predictors of early in-hospital transfusion. Prehospital lactate emerged as a strong predictor for transfusion need from both the FFT and Bayesian approaches among patients who were not hypotensive. This is consistent with recent study by \u003cem\u003eGriggs et al.\u003c/em\u003e who also predicted in hospital transfusion using prehospital lactate concentration.\u003csup\u003e15\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eAdministration of prehospital blood products to patients in hemorrhagic shock reduces mortality.\u003csup\u003e4\u003c/sup\u003e A systematic review and meta-analysis by Rijnhout et al. describes the administration of prehospital blood products as feasible and safe but describes the evidence as low quality and difficult to compare because there are no standard indication for transfusion.\u003csup\u003e16\u003c/sup\u003e While tools have been developed to identify patient at risk of Trauma Associated Severe Hemorrhage (TASH) and for massive transfusion (ABC score), they rely on data not readily available in the prehospital environment (hemoglobin and ultrasound) and neither was developed for the prehospital environment.\u003csup\u003e17,18\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eTo find the simplest decision model to identify people who need prehospital blood transfusions, we are faced with two competing considerations: 1) correctly identifying the greatest number of people who need blood (i.e., maximizing the sensitivity of the model), and 2) conserving limited resources of blood. Using these considerations, an EMS Medical Director may conclude that the model with a lactate concentration threshold of 2.5 mmol/L (Sensitivity\u0026thinsp;=\u0026thinsp;0.89, Specificity\u0026thinsp;=\u0026thinsp;0.48, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e row 1.6) is more appropriate for use in a rural setting with delayed access to a trauma center and subsequent damage control resuscitation, while a model with a 4 mmol/L threshold (Sensitivity\u0026thinsp;=\u0026thinsp;0.82, Specificity\u0026thinsp;=\u0026thinsp;0.62, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e row 1.8) could be more suitable for urban settings with short prehospital times. Similar trade-offs can be made with the thresholds for SBP and shock index.\u003c/p\u003e \u003cp\u003eWe adjusted the model thresholds to create simple rules for quick reference in the field (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The results depicted in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e have broad implications for prehospital clinicians, ranging from urban and rural EMS systems to austere military environments that might require prolonged field care. Using the four variables derived from our models, prehospital system leadership can decide what thresholds are appropriate for transfusions in their respective systems, based on existing resources and trauma center access.\u003c/p\u003e \u003cp\u003ePrevious studies associate prehospital lactate with mortality and morbidity in trauma patients.\u003csup\u003e19,20\u003c/sup\u003eSubsequent work demonstrated the association between lactate and need for life saving interventions.\u003csup\u003e21\u003c/sup\u003e Recent work by Fukuma et al. and Galvagno et al. established that prehospital lactate threshold of \u0026gt;\u0026thinsp;4 mmol/L is associated with the need life-saving interventions for hemorrhage control.\u003csup\u003e14,22\u003c/sup\u003e This threshold is more conservative than the one found by the FFT algorithm (2.5 mmol/L).\u003c/p\u003e \u003cp\u003eThe last cue identified to trigger potential transfusion is severe chest injury. In the data we used AIS\u0026thinsp;\u0026gt;\u0026thinsp;2; however, recognizing this is not available in the field setting as an objective number, this would rely on clinical exam evidence, much in the way the anatomic triage criteria for the national field triage guidelines are identified. We suggest operationalizing this as flail chest, unstable chest fractures, or need for needle decompression (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Local medical directors certainly would have discretion to operationalize this cue in an alternative way given the personnel, resources, and trauma population seen by his or her EMS agency. We do show comparable accuracy if this cue is omitted, allowing further adaptation to local circumstances given it is the most subjective cue in operational form.\u003c/p\u003e \u003cp\u003eA key limitation of our study is that decision to transfuse blood is not always synonymous with the need to transfuse blood. Also, our analyses are retrospective and derived from a single EMS agency serving a regional trauma system. The dataset was limited to patients who had lactate sampled which imparts bias among patients with hemorrhagic shock. Selection bias may result when treatment priorities preclude sampling of lactate in the sickest patients. EMS data is rarely inputted into the record contemporaneously with care and is subject to recall and reporting bias. We import data electronically (vital signs, times and point of care labs) into the prehospital health record which mitigates these biases. There is likely a selection and sensitivity bias as our critical care organization is called for patients with more severe injuries or those who are geographically distant from trauma care.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eWe developed a parsimonious, clinically relevant algorithm to identify patients who may require prehospital transfusion. This algorithm accounts for prehospital lactate concentration which is useful for identifying patients with occult shock not meeting the conventional threshold for hypotension. Thresholds of decision factors should be adjusted to meet the needs and resources of a given prehospital trauma system. Further work is necessary to externally validate this algorithm for prehospital blood transfusion.\u003c/p\u003e \u003cp\u003e We are including the Appendix describing the FFT algorithm, Blood Administration protocol, and the study checklist for adhering to STROBE guidelines as Supplemental Digital Content.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003e\u003cem\u003eAIS \u0026ndash;\u003c/em\u003e Abbreviated Injury Scale\u003cstrong\u003e;\u0026nbsp;\u003c/strong\u003e\u003cem\u003ebpm \u0026ndash;\u003c/em\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003ebeats per minute\u003cstrong\u003e;\u0026nbsp;\u003c/strong\u003e\u003cem\u003eED \u0026ndash;\u003c/em\u003e Emergency Department; \u003cem\u003eEHR \u0026ndash;\u003c/em\u003e Electronic Health Record; \u003cem\u003eEMS \u0026ndash;\u0026nbsp;\u003c/em\u003eEmergency Medical Services; \u003cem\u003eISS \u0026ndash;\u003c/em\u003e Injury Severity Score; \u003cem\u003eSBP \u0026ndash;\u003c/em\u003e Systolic Blood Pressure; \u003cem\u003eSI \u0026ndash;\u003c/em\u003e Shock Index; \u003cem\u003eTC \u0026ndash;\u003c/em\u003e Trauma Center; \u0026nbsp;\u003cem\u003eROC\u0026nbsp;\u003c/em\u003e - Receiver Operating Characteristics; \u003cem\u003eFFT\u003c/em\u003e \u0026ndash; Fast Frugal Tree\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eSources of Funding:\u003c/strong\u003e Our research is supported by the NIH through grant 5K23NS097629.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eAuthor contributions:\u0026nbsp;\u003c/strong\u003eFXG, JBB and SMG guided the research. SMG and TW worked on Bayesian analysis. FXG, JBB and EVZ worked on FFTs. All authors contributed to writing and editing the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEthical approval\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; We certify that the study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. University of Pittsburgh Internal Review Board (IRB) PRO15020269 approved this study.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eConsent to participate\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; IRB granted a waiver of consent for this study.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eConsent for publication\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAvailability of data and materials\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCompeting interests\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/em\u003eThe authors declare that they have no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDrake SA, Holcomb JB, Yang Y, et al. Establishing a regional trauma preventable/potentially preventable death rate. \u003cem\u003eAnn Surg\u003c/em\u003e. Published online 2020.\u003c/li\u003e\n\u003cli\u003eHolcomb JB, Tilley BC, Baraniuk S, et al. Transfusion of plasma, platelets, and red blood cells in a 1:1:1 vs a 1:1:2 ratio and mortality in patients with severe trauma: The PROPPR randomized clinical trial. \u003cem\u003eJAMA - Journal of the American Medical Association\u003c/em\u003e. Published online 2015.\u003c/li\u003e\n\u003cli\u003eBrown JB, Sperry JL, Fombona A, Billiar TR, Peitzman AB, Guyette FX. Pre-trauma center red blood cell transfusion is associated with improved early outcomes in air medical trauma patients. \u003cem\u003eJ Am Coll Surg\u003c/em\u003e. Published online 2015.\u003c/li\u003e\n\u003cli\u003eSperry JL, Guyette FX, Brown JB, et al. Prehospital plasma during air medical transport in trauma patients at risk for hemorrhagic shock. \u003cem\u003eNew England Journal of Medicine\u003c/em\u003e. 2018;379(4):315-326.\u003c/li\u003e\n\u003cli\u003eZadorozny E v., Weigel T, Stone A, et al. Prehospital Lactate is Associated with the Need for Blood in Trauma. \u003cem\u003ePrehospital Emergency Care\u003c/em\u003e. Published online 2021.\u003c/li\u003e\n\u003cli\u003ePokorny DM, Braverman MA, Edmundson PM, et al. The use of prehospital blood products in the resuscitation of trauma patients: a review of prehospital transfusion practices and a description of our regional whole blood program in San Antonio, TX . \u003cem\u003eISBT Sci Ser\u003c/em\u003e. 2019;14(3):332-342.\u003c/li\u003e\n\u003cli\u003eTobias AZ, Guyette FX, Seymour CW, et al. Pre-resuscitation lactate and hospital mortality in prehospital patients. \u003cem\u003ePrehospital Emergency Care\u003c/em\u003e. Published online 2014.\u003c/li\u003e\n\u003cli\u003eMikkelsen ME, Miltiades AN, Gaieski DF, et al. Serum lactate is associated with mortality in severe sepsis independent of organ failure and shock. \u003cem\u003eCrit Care Med\u003c/em\u003e. Published online 2009.\u003c/li\u003e\n\u003cli\u003ePhillips ND, Neth H, Woike JK, Gaissmaier W. FFTrees: A toolbox to create, visualize, and evaluate fast-and-frugal decision trees. \u003cem\u003eJudgm Decis Mak\u003c/em\u003e. Published online 2017.\u003c/li\u003e\n\u003cli\u003eJansen JO, Pallmann P, Maclennan G, Campbell MK. Bayesian clinical trial designs: Another option for trauma trials? \u003cem\u003eJournal of Trauma and Acute Care Surgery\u003c/em\u003e. 2017;83(4):736-741.\u003c/li\u003e\n\u003cli\u003eRichert Willi, Coelho LPedro. Building Machine Learning Systems with Python - Willi Richert - Google B\u0026oslash;ker. \u003cem\u003ePackit Publishing\u003c/em\u003e. 2013;1:290.\u003c/li\u003e\n\u003cli\u003eZadorozny E v., Weigel T, Stone A, et al. Prehospital Lactate is Associated with the Need for Blood in Trauma. \u003cem\u003ePrehospital Emergency Care\u003c/em\u003e. Published online 2021.\u003c/li\u003e\n\u003cli\u003eSasser S, Hunt R, Faul M, et al. Guidelines for Field Triage of Injured Patients. \u003cem\u003eMMWR Recommendations and Reports\u003c/em\u003e. 2012;61(1):1-21.\u003c/li\u003e\n\u003cli\u003eGalvagno SM, Sikorski RA, Floccare DJ, et al. Prehospital Point of Care Testing for the Early Detection of Shock and Prediction of Lifesaving Interventions. \u003cem\u003eShock\u003c/em\u003e. 2020;54(6):710-716.\u003c/li\u003e\n\u003cli\u003eGriggs JE, Lyon RM, Sherriff M, Barrett JW, Wareham G, ter Avest E. Predictive clinical utility of pre-hospital point of care lactate for transfusion of blood product in patients with suspected traumatic haemorrhage: derivation of a decision-support tool. \u003cem\u003eScand J Trauma Resusc Emerg Med\u003c/em\u003e. 2022;30(1):1-9.\u003c/li\u003e\n\u003cli\u003eRijnhout TWH, Wever KE, Marinus RHAR, Hoogerwerf N, Geeraedts LMG, Tan ECTH. Is prehospital blood transfusion effective and safe in haemorrhagic trauma patients? A systematic review and meta-analysis. \u003cem\u003eInjury\u003c/em\u003e. Published online 2019.\u003c/li\u003e\n\u003cli\u003eY\u0026uuml;cel N, Lefering R, Maegele M, et al. Trauma Associated Severe Hemorrhage (TASH)-score: Probability of mass transfusion as surrogate for life threatening hemorrhage after multiple trauma. \u003cem\u003eJournal of Trauma - Injury, Infection and Critical Care\u003c/em\u003e. Published online 2006.\u003c/li\u003e\n\u003cli\u003eNunez TC, Voskresensky I V., Dossett LA, Shinall R, Dutton WD, Cotton BA. Early prediction of massive transfusion in trauma: Simple as ABC (Assessment of Blood Consumption)? \u003cem\u003eJournal of Trauma - Injury, Infection and Critical Care\u003c/em\u003e. Published online 2009.\u003c/li\u003e\n\u003cli\u003eGuyette F, Suffoletto B, Castillo JL, Quintero J, Callaway C, Puyana JC. Prehospital serum lactate as a predictor of outcomes in trauma patients: A retrospective observational study. \u003cem\u003eJournal of Trauma - Injury, Infection and Critical Care\u003c/em\u003e. Published online 2011.\u003c/li\u003e\n\u003cli\u003eSt John AE, McCoy AM, Moyes AG, Guyette FX, Bulger EM, Sayre MR. Prehospital lactate predicts need for resuscitative care in non-hypotensive trauma patients. \u003cem\u003eWestern Journal of Emergency Medicine\u003c/em\u003e. Published online 2018.\u003c/li\u003e\n\u003cli\u003eGuyette FX, Meier EN, Newgard C, et al. A comparison of prehospital lactate and systolic blood pressure for predicting the need for resuscitative care in trauma transported by ground. \u003cem\u003eJournal of Trauma and Acute Care Surgery\u003c/em\u003e. Published online 2015.\u003c/li\u003e\n\u003cli\u003eFukuma H, Nakada T aki, Shimada T, et al. Prehospital lactate improves prediction of the need for immediate interventions for hemorrhage after trauma. \u003cem\u003eSci Rep\u003c/em\u003e. Published online 2019.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-emergency-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ijem","sideBox":"Learn more about [International Journal of Emergency Medicine](https://intjem.biomedcentral.com/)","snPcode":"12245","submissionUrl":"https://submission.nature.com/new-submission/12245/3","title":"International Journal of Emergency Medicine","twitterHandle":"@IntJEmergMed","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"prehospital transfusion, early hospital transfusion, hemorrhagic shock, prehospital lactate concentration, Fast Frugal Trees, Bayesian analysis, decision support models","lastPublishedDoi":"10.21203/rs.3.rs-3944131/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3944131/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eTraumatic shock is the leading cause of preventable death with most patients dying within the first 6 hours. This underscores the importance of prehospital interventions, and growing evidence suggests prehospital transfusion improves survival. Optimizing transfusion triggers in the prehospital setting is key to improving outcomes for patients in hemorrhagic shock. Our objective was to identify factors associated with early in-hospital transfusion requirements available to prehospital clinicians in the field to develop a simple algorithm for prehospital transfusion, particularly for patients with occult shock.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe included trauma patients transported by a single critical care transport service to a level I trauma center between 2012 and 2019. We used logistic regression, Fast and Frugal Trees (FFTs), and Bayesian analysis to identify factors associated with early in-hospital blood transfusion as a potential trigger for prehospital transfusion.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eWe included 2,157 patients transported from the scene or emergency department (ED) of whom 207 (9.60%) required blood transfusion within 4 hours of admission. The mean age was 47 (IQR\u0026thinsp;=\u0026thinsp;28\u0026ndash;62) and 1,480 (68.6%) patients were male. From 13 clinically relevant factors for early hospital transfusions, four were incorporated into the FFT in following order: 1) SBP, 2) prehospital lactate concentration, 3) Shock Index, 4) AIS of chest (sensitivity\u0026thinsp;=\u0026thinsp;0.81, specificity\u0026thinsp;=\u0026thinsp;0.71). The chosen thresholds were similar to conventional ones. Using conventional thresholds resulted in lower model sensitivity. Consistently, prehospital lactate was among most decisive factors of hospital transfusions identified by Bayesian analysis (OR\u0026thinsp;=\u0026thinsp;2.31; 95% CI 1.55\u0026ndash;3.37).\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eUsing an ensemble of frequentist statistics, Bayesian analysis and machine learning, we developed a simple, clinically relevant, prehospital algorithm to help identify patients requiring transfusion within 4 hours of hospital arrival.\u003c/p\u003e","manuscriptTitle":"Identifying Trigger Cues for Hospital Blood Transfusions Based on Ensemble Learning Methods","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-02-20 20:18:50","doi":"10.21203/rs.3.rs-3944131/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2024-04-24T23:39:49+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"0bbcbd24-3fc7-4f50-8b39-2f8e25e03c92","date":"2024-04-13T14:01:37+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-04-08T06:34:27+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-02-19T04:51:29+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-02-19T04:25:29+00:00","index":"","fulltext":""},{"type":"submitted","content":"International Journal of Emergency Medicine","date":"2024-02-09T20:39:58+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-emergency-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ijem","sideBox":"Learn more about [International Journal of Emergency Medicine](https://intjem.biomedcentral.com/)","snPcode":"12245","submissionUrl":"https://submission.nature.com/new-submission/12245/3","title":"International Journal of Emergency Medicine","twitterHandle":"@IntJEmergMed","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"73245e20-bc13-4009-abc1-ec574cba5b36","owner":[],"postedDate":"February 20th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2024-05-23T04:23:48+00:00","versionOfRecord":[],"versionCreatedAt":"2024-02-20 20:18:50","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3944131","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3944131","identity":"rs-3944131","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}