Objective: To develop a statistically rigorous trauma mortality prediction model based on empiric estimates of severity for each injury in the abbreviated injury scale (AIS) and compare the performance of this new model with the injury severity score (ISS).
Summary Background Data: Mortality rates at trauma centers should only be compared after adjusting for differences in injury severity, but no reliable measure of injury severity currently exists. The ISS has served as the standard measure of anatomic injury for 30 years. However, it relies on the individual injury severities assigned by experts in the AIS, is nonmonotonic with respect to mortality, and fails to perform even as well as a far simpler model based on the single worst injury a patient has sustained.
Methods: This study is based on data from 702,229 injured patients in the National Trauma Data Bank (NTDB 6.1) hospitalized between 2001 and 2005. Sixty percent of the data was used to derive an empiric measure of severity of each of the 1322 injuries in the AIS lexicon by taking the weighted average of coefficients estimated using 2 separate regression models. The remaining 40% of the data was use to create 3 exploratory mortality prediction models and compare their performance with the ISS using measures of discrimination (C statistic), calibration (Hosmer Lemeshow statistic and calibration curves), and the Akaike information criterion.
Results: Three new models based on empiric AIS injury severities were developed. All of these new models discriminated survivors from nonsurvivors better than the ISS, but one, the trauma mortality prediction model (TMPM), had both better discrimination [ROCTMPM = 0.901 (0.898–0.905), ROCISS = 0.871 (0.866–0.877)] and better calibration [HLTMPM = 58 (35–91), HLISS = 296 (228–357)] than the ISS. The addition of age, gender, and mechanism of injury improved all models, but the augmented TMPM dominated ISS by every measure [ROCTMPM = 0.925(0.921–0.928), ROCISS = 0.904(0.901–0.909), HLTMPM = 18 (12–31), HLISS = 54 (30–64)].
Conclusions: Trauma mortality models based on empirical estimates of individual injury severity better discriminate between survivors and nonsurvivors than does the current standard, ISS. One such model, the TMPM, has both superior discrimination and calibration when compared with the ISS. The TMPM should replace the ISS as the standard measure of overall injury severity.
The injury severity score (ISS) has served as the standard summary measure for trauma for 30 years but it has serious shortcomings. We propose a trauma mortality prediction model that is based on empirical estimates of individual injury severities and provides substantially better mortality predictions than the ISS.
From the *Department of Surgery, University of Vermont, Burlington, VT; †Department of Anesthesia, University of Rochester, Rochester, NY; ‡Department of Mathematics and Statistics, University of Vermont, Burlington, VT; §School of Medicine, University of California, Irvine, CA; and ¶RAND-University of Pittsburg Health Institute, Pittsburgh, PA.
Supported by a grant from the Agency for Healthcare Research and Quality (RO1 HS 016737).
The National Trauma Data Bank data set (NTDB Version 6.1, 2007) was made available by the American College of Surgeons Committee on Trauma, Chicago, IL.
The views presented in this manuscript are those of the authors and may not reflect those of Agency for Healthcare Research and Quality. The content reproduced from the NTDB data remains the full and exclusive copyrighted property of the American College of Surgeons. The American College of Surgeons is not responsible for any claims arising from works based on the original data.
The 1322 MARC values calculated for this manuscript are available on request from the authors. Instructions for calculating the TMPM from these MARC values are given in Appendix C. An Excel spread sheet to automate the calculation of the TMPM from any data set is available as shareware from the authors and will be available at http://www.facs.org/trauma/ntdb.html.
Reprints: Turner Osler, MD, MSc, Department of Surgery, University of Vermont, 789 Orchard Shore Road, Colchester, VT 05446. E-mail: firstname.lastname@example.org.