The Trauma and Injury Severity Score (TRISS) revised

Area of Science:

• Trauma care and injury outcome prediction.
• Biostatistics and predictive modeling.

Background:

• The Trauma and Injury Severity Score (TRISS) is the standard for trauma outcome benchmarking.
• Previous research suggests re-classifying TRISS variables can enhance predictive power.

Purpose of the Study:

• To develop and validate revised TRISS models using re-classified variables.
• To compare the predictive performance of revised TRISS models against existing ones.
• To identify an optimal TRISS model for trauma benchmarking.

Main Methods:

• Developed revised TRISS models using 5-category variable classifications and weighted logistic regression on the National Trauma Data Bank National Sample Project (NSP) data.
• Assessed predictive performance against existing TRISS models using Area Under the ROC Curve (AUC) and Bayesian Information Criterion (BIC) on NSP, NTDB, and New Zealand Database (NZDB) samples.

Main Results:

• Revised TRISS models with main-effects and two-factor interactions showed superior AUC and BIC statistics compared to existing models across all tested datasets.
• Model performance remained robust even with missing predictor variables, generally outperforming existing TRISS models.

Conclusions:

• Revised TRISS models offer significantly improved predictive capacity for trauma outcomes.
• These models are computationally simple, use existing data variables, and handle missing data effectively.
• A revised TRISS model incorporating main-effects, two-factor interactions, and allowing for missing values is recommended for replacing current systems.