Article Text
Abstract
Background Comparison of trauma centres requires accurate injury severity metrics. The Trauma Audit and Research Network (TARN) in the UK, and the Trauma Injury Severity Score (TRISS) provide such risk adjustment. These models incorporate the Injury Severity Score (ISS) as the measure of anatomic injury severity but categorise it differently. The Trauma Mortality Prediction Model (TMPM) better predicts mortality than the ISS. We compared the anatomic injury components of TARN, and TRISS models to TMPM.
Methods Data from the National Trauma Data Bank for 2002–2012. Probability of death was estimated for TARN, and TRISS from ISS values according to each model’s treatment and compared these to TMPM using measures of discrimination (area under ROC curves), proximity to the true model (Akaike Information Criterion (AIC)), and calibration (Hosmer-Lemeshow statistic (HL)). Confidence intervals (CI) were estimated by bootstrapping. ISS and TMPM values were estimated for the Abbreviated Injury Scale (AIS) and the International Classification of Diseases, 9th Revision (ICD-9) lexicons. Data are shown as medians and 95% CI.
Results N = 1,145,959
Conclusions TMPM better predicted mortality than did the anatomic injury component of TRISS, or TARN and this was true whether individual injuries were coded in the AIS lexicon or in the ICD-9. As trauma care evolves, so must models of mortality prediction that best capture injury severity.
- Trauma
- Injury Severity
- Mortality Prediction
- Anatomic Injury