Development and Validation of a Korean Trauma and Injury Severity Score (K-TRISS) Model for Predicting Trauma Outcomes.

Abstract

Background: Since the Trauma and Injury Severity Score (TRISS) was first developed in 1987, it has been widely used to predict trauma outcomes. However, numerous attempts have been made to adjust coefficients or develop new prediction models, as TRISS may not align with the local conditions. This study aimed to develop a Korean-TRISS (K-TRISS) model suitable for the Korean population.

Methods: We analyzed data from adult patients with blunt trauma in the Korea Trauma Data Bank (KTDB) from January 2017 to December 2021. A new set of TRISS coefficients (K-TRISS-1) was derived from the study data using a logistic regression method. We compared the predictive ability of the K-TRISS-1 with the previous Major Trauma Outcome Study (MTOS) in 1987 and the National Trauma Data Bank (NTDB) model in 2009. The predictive power of the models was evaluated with the area under receiver operating characteristic (AUROC) curves, and the validity was evaluated with the C-statistic and bootstrap verification methods. Additionally, we enhanced the model construction (K-TRISS-2) by converting the age variable from a categorical format to a continuous one.

Results: Overall, 128,534 cases were included in the statistical analysis. The comparison of AUROC values indicated that K-TRISS-1 (0.9232) outperformed MTOS (0.9210) and NTDB (0.9190), with their 95% confidence intervals showing some tendency (0.9196-0.9267, 0.9200-0.9230, and 0.9180-0.9210, respectively). However, K-TRISS-2 (0.9336, 0.9305-0.9367) had a significantly higher AUROC value compared to those of the others and showed excellent predictive power in the C-statistic and bootstrap tests.

Conclusion: This study proposes the K-TRISS model, derived from the KTDB, which does not significantly differ from the previous models in terms of predictive power. Furthermore, we were able to construct a model demonstrating improved predictive power when converting age to a continuous variable.