Prediction of risk of death in severely injured patients: the revised injury severity classification score, version 3...

Trauma Registries with a focus on severely injured patients use survival as their primary outcome. In order to compare hospitals, interventions, and changes over time, a precise risk of death prediction is mandatory. The German TraumaRegister DGU^® uses the RISC II model (Revised Injury Severity Classification, version II) since 2013. The aging trauma population and an improved handling of missing data required the present revision.

A total of 53,738 seriously injured trauma patients documented in 2022-2023 served as basis for development and validation (3:1 ratio). Missing values should now be considered to be within normal physiological range except in patients with specific findings indicative for an altered physiology. These findings were suggested by clinical experts and validated by registry data. The increasing age of trauma patients was addressed by additional categories and higher point weights. A logistic regression analysis provided new point weights for all predictors. Precision (observed versus predicted mortality) and discrimination (area under the receiver operating characteristic curve, AUROC) were calculated in the development and validation dataset.

Patients in both datasets were well comparable, with a mean age of 55 years, 69% males, and an average Injury Severity Score (ISS) of 18 points. The rate of missing values ranged from 0% (compulsory data) to 17.5% (initial base excess). Missing pre-injury health status was imputed by age, missing pupil size, light reaction, and motor function were imputed by severity of head injury. In case of specific findings, blood pressure and initial laboratory values were imputed by injury severity (ISS), or blood transfusion, or catecholamines, or intake of anticoagulation drugs. The AUROC was 0.946 (95% confidence interval 0.944-0.949) for the new RISC III score which was confirmed in the validation data (0.949; CI 0.945-0.954). Observed and predicted mortality were 13.1% / 13.0% in the development dataset, and 13.2% / 13.0% in the validation dataset.

Risk of death estimates require repeated validations. The increasing number of elderly trauma patients, some of them with restrictions regarding the intensity of treatment, required this update of the RISC II model. New point weights for age were established now, especially for the elderly, in order to enhance the precision of prediction in this patient group. Patients with missing values showed on average a low injury severity, thus replacing a missing value with the normal category as a general rule (RISC III) seems to be superior than replacing it with an average value (RISC II). The new prediction model shows high discrimination and precision in both datasets, development and validation, and will replace the previous version in quality reports and scientific analyses.