Introduction:
Anatomic derangement is a critical step in the pathway between mechanism and outcomes of injury. Injury severity benchmarking allows for principled comparison of severity distributions, patterns of injury, and surveillance. The Trauma Mortality Prediction Model (TMPM) is superior to previous injury severity metrics, though it was derived for the Abbreviated Injury Scale (AIS) and the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) which is only used in the United States. Objectives: We seek to derive the TMPM for use with the ICD-10 lexicon from the World Health Organization (ICD-10-WHO) for global applications.

Methods:
ICD-10-CM injury codes (S- and T- codes, N=8,538 codes) were mapped to the four digit ICD-10-WHO (N=513) using data from the US National Trauma Data Bank (NTDB) for years 2016 - 2019 for the derivation and validation datasets. The ICD-10-WHO injury codes were parameterized as dichotomous independent variables and as body region-severity dyads in two separate probit regression models. Patient-level probabilities of death (pDeath) were computed from the Model Averaged Regression Coefficients (MARC). The area under the receiver operating characteristic curve (AUROC), the Akaike Information Criterion (AIC), and calibration curves characterized model performance. Further validation was tested using the Patient Episode Database for Wales (PEDW). Of note, the ICD-10-CM lacks codes in the T00-T14 range which are present in the ICD-10-WHO and common in trauma registries outside the US, including the PEDW.

Results:
The dataset included 2,343,954 patients, including 52,327 (2.23%) deaths. TMPM demonstrated very good discrimination in the derivation and validation datasets mapped from ICD-10-CM, with AUROC values of 0.854 and 0.862, respectively. However, the TMPM revealed diminished performance in the PEDW, AUROC 0.772. Interestingly, the AIC for the PEDW data was lower and thus preferable compared to the ICD-10-CM mapped codes (PEDW 85269 vs. 390460 and 275455 for derivation and validation, respectively). Calibration curves indicated the performance of the TMPM in data mapped from ICD-10-CM and the consequence of the missing MARC values for T00-T14 codes in the PEDW data. Conclusion: The TMPM demonstrates very good discrimination and calibration in data mapped from ICD-10-CM, however the lack of MARC values for T-codes T00-T14 diminish its performance in data which include such codes. Further work is needed to derive MARC values for the T00-T14 codes.

Keywords: Injury severity, Benchmarking, ICD-10, TMPM

Other info: University of Texas at Tyler - Tyler - Texas - United StatesAngharad Walters, MS, Jens Lauritsen, PhD, MD, Ronan Lyons, MD