Predicting Prehospital Blood Transfusion After Motor Vehicle Trauma

Prehospital blood transfusion (PHBT) improves outcomes among patients with traumatic hemorrhagic shock, yet the epidemiology and geographic distribution of patients likely to require PHBT remain poorly characterized. We sought to develop predictive models for PHBT among motor vehicle crash (MVC) patients using National Emergency Medical Services Information System (NEMSIS) data and to estimate the national distribution of patients with a high predicted probability of receiving PHBT.

We conducted a retrospective cohort study using the 2024 and 2025 NEMSIS data. Eligible encounters included 9-1-1 EMS responses for MVCs among patients aged 8-100 years. A derivation cohort was constructed from EMS agencies with active PHBT programs and at least 10 documented transfusions. Candidate predictors included demographic, physiologic, clinical, and operational variables. Random forest and ridge-penalized logistic regression models were developed to identify covariates associated with PHBT. Variables selected through machine learning approaches were subsequently evaluated in multivariable logistic regression models. Model performance was assessed using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and calibration metrics. Final models were applied to the remaining national cohort to estimate patients with a high predicted probability of PHBT.

Among 2,078,452 eligible EMS activations, 303,383 comprised the derivation cohort. The random forest-selected logistic regression model demonstrated strong discrimination with an AUC of 0.901, sensitivity of 75.3%, and specificity of 96.5%. The ridge-selected logistic regression model achieved an AUC of 0.942, sensitivity of 63.9%, and specificity of 97.7%. Both models identified composite physiologic measures, critical hemorrhagic shock designation, and unspecified traumatic shock as significant predictors of PHBT. Application of the final models to the national cohort identified 73,754 and 70,599 predicted PHBT recipients, representing 3.6% and 3.4% of the evaluated population, respectively.

Predictive modeling applied to NEMSIS data accurately identified patients with a high probability of receiving PHBT following MVC trauma. These findings provide a framework for estimating geographic demand for prehospital blood products and may inform data-driven expansion of PHBT programs.