Abstract Background Many large data sets are limited in their usefulness for trauma epidemiology by a paucity of clinical data reflecting injury severity. Software packages have been developed for extrapolation of diagnostic codes to Injury Severity Scores commonly used in trauma research. The aim of this project was to evaluate one recently upgraded package, ICDPICR2. Methods Data were taken from the Trauma Quality Improvement Program (TQIP) for 2022. The inclusion criterion was an Abbreviated Injury Scale (AIS) head score of 1 or greater. The ICDPICR2 package in R was used to calculate AIS body region scores and Injury Severity Scores (ISS) from the diagnostic codes in TQIP. Performance of actual and mapped scores in common epidemiological applications were compared. Results There were 337,148 files with valid AIS head scores and known hospital discharge status. Cohen’s kappa statistics for AIS head scores and binned ISS were only fair at 0.310 and 0.280, respectively. Beyond an ISS of 15, mapped ISS underestimated actual ISS. Mapped AIS head scores and ISS had low sensitivity in identification of severe injuries, 54% and 23%, respectively. Discrimination of withdrawal of care, hospital morality and home discharge was poorer for mapped scores than for actual, and models of length of stay based on mapped scores had worse fit. Mapped scores yielded very different parameter estimates from actual scores in illustrative regression models. Conclusions Limitations of the reliability of ICDPICR2 in administrative data sets must be anticipated. Parameter estimates from regression models constructed from mapped scores are not a reliable basis for hypothesis testing.
Joseph H. Piatt (2026) studied this question.