Random Forest Machine Learning Matches Human Expert Accuracy in Trauma Severity Scoring.

Area of Science:

• Trauma Care and Research
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Accurate Abbreviated Injury Scale (AIS) and Injury Severity Score (ISS) are crucial for trauma care and research.
• Manual scoring in trauma registries often results in incomplete data due to omissions.
• The hybrid electronic medical registry (HEMR) is utilized by Level 1 trauma services for recording AIS and ISS.

Purpose of the Study:

• To evaluate the performance of machine learning algorithms in predicting missing AIS and ISS scores.
• To assess the impact of ML-driven data imputation on trauma registry completeness.
• To determine if ML prediction maintains clinical accuracy comparable to human scoring.

Main Methods:

• Analysis of 21,704 trauma patient records from the HEMR.
• Application of four machine learning (ML) algorithms to predict missing AIS scores per body region.
• Mathematical derivation of ISS from predicted AIS scores and performance evaluation using R², RMSE, MAE, sensitivity, specificity, and Cohen's kappa.

Main Results:

• Random forest models demonstrated high accuracy with R²=0.847, MAE=1.87, and Cohen's kappa=0.893.
• Sensitivity for high-severity cases was 87.1%, and specificity for low-severity cases was 100.0%.
• Trauma registry data completeness significantly improved from 75.3% to 88.3%, recovering 2815 missing scores.

Conclusions:

• Random forest ML algorithms reliably predict missing AIS and ISS scores.
• The use of ML significantly enhances trauma registry data completeness.
• ML prediction achieves clinical accuracy equivalent to human expert scoring, vital for trauma research and care.