Data-Driven Analysis to Inform National Strategies on Improving Post-Crash Care

Abstract

Reducing prehospital emergency medical service (EMS) time is essential for decreasing the severity of crash outcomes. Recently, South Korea has faced prolonged prehospital EMS time, prompting the Korea National Fire Agency (KFA) to announce strategies to reduce prehospital EMS time. This study aimed to provide a methodological process and quantitative basis for implementing the KFA’s strategies. To achieve this, the study employed an eXtreme Gradient Boosting (XGBoost) approach with SHapley Additive exPlanations (SHAP) value analysis to identify key factors influencing crash severity and to examine their global and local impacts. The resultant key factors included spatiotemporal attributes, including on-scene time, freeway mainline, and nighttime, which were used to quantitatively validate the KFA’s strategies. This research is the first attempt to apply an interpretable machine learning algorithm to improve post-crash care, using crash data and EMS infrastructure information from an entire country. The methodology employed can be applied to quantitatively support decision-making on similar issues in various regions and countries.