Predicting specific injury patterns from crash data: a GIDAS-based and manufacturer-independent logistic regression approach for next generation eCall

Abstract

Background

While prehospital traffic fatalities remain a critical challenge, current eCall systems transmit limited injury-relevant data, and existing prediction models often fail to support targeted prehospital triage. This study developed and validated a manufacturer-independent, data-driven framework to predict tactically relevant, region-specific injury patterns to enhance early rescue chain activation.

Methods

Logistic regression models were developed using GIDAS data for front-row occupants, stratified by impact direction. Clinically relevant injury categories were defined based on prehospital tactical criteria. Predictors included crash severity (Energy Equivalent Speed, EES), restraint use, and vehicle characteristics. Performance was assessed via cross-validation and external validation on a recent GIDAS subset, utilizing metrics robust to class imbalance.

Results

Across all injury groups, the models demonstrated high discrimination (AROC > 0.85 for most groups), robust calibration, and biomechanically plausible associations. Energy-equivalent speed (EES) consistently emerged as the strongest and most reliable predictor of injury severity. Optimized classification thresholds supported the prediction of both global and region-specific trauma. External validation confirmed reproducibility for thorax, head, and overall severe injuries, maintaining high sensitivity and specificity under optimized thresholds.

Conclusion

Ultimately, this study provides a scientifically robust framework for predicting region-specific injury patterns, with results that can be translated into next-generation eCall systems to inform prehospital triage and decision-support. Clinically, this predictive framework translates crash data into triage intelligence, enabling emergency dispatchers to allocate specialized trauma resources more precisely. While the models demonstrate strong predictive performance on the GIDAS dataset, prospective multicenter validation is ongoing to assess their real-world operational utility.