You could be reading the full-text of this article now if you...

If you have access to this article through your institution,
you can view this article in

Improving the TRISS Methodology by Restructuring Age Categories and Adding Comorbidities

Bergeron, Eric MD; Rossignol, Michel MD; Osler, Turner MD; Clas, David MD; PhD, Andre Lavoie

Journal of Trauma-Injury Infection & Critical Care:
Original Articles
Abstract

Background: The Trauma and Injury Severity Score (TRISS) methodology was developed to predict the probability of survival after trauma. Despite many criticisms, this methodology remains in common use. The purpose of this study was to show that improving the stratification for age and adding an adjustment for comorbidity significantly increases the predictive accuracy of the TRISS model.

Methods: The trauma registry and the hospital administrative database of a regional trauma center were used to identify all blunt trauma patients older than 14 years of age admitted with International Classification of Diseases, Ninth Revision codes 800 to 959 from April 1993 to March 2001. Each individual medical record was then reviewed to ascertain the Revised Trauma Score, the Injury Severity Score, the age of the patients, and the presence of eight comorbidities. The outcome variable was the status at discharge: alive or dead. The study population was divided into two subsamples of equal size using a random sampling method. Logistic regression was used to develop models on the first subsample; a second sub-sample was used for cross-validation of the models. The original TRISS and three TRISS-derived models were created using different categorizations of Revised Trauma Score, Injury Severity Score, and age. A new model labeled TRISSCOM was created that included an additional term for the presence of comorbidity.

Results: There were 5,672 blunt trauma patients, 2,836 in each group. For original TRISS, the Hosmer-Lemeshow statistic (HL) was 179.1 and the area under the receiver operating characteristic (AUROC) curve was 0.873. Sensitivity and specificity were 99.0% and 27.8%, respectively. For the best modified TRISS model, the HL statistic was 20.35, the AUROC curve was 0.902, the sensitivity was 99.0%, and the specificity was 27.8%. For TRISSCOM, the HL statistic was 14.95 and the AUROC curve was 0.918. Sensitivity and specificity were 99.0% and 29.7%, respectively. The difference between the two models almost reached statistical significance (p = 0.086). When TRISSCOM was applied to the cross-validation group, the HL statistic was 10.48 and the AUROC curve was 0.914. The sensitivity was 98.6% and the specificity was 34.9%.

Conclusion: TRISSCOM can predict survival more accurately than models that do not include comorbidity. A better categorization of age and the inclusion of co-morbid conditions in the logistic model significantly improves the predictive performance of TRISS.

Author Information

From the Department of Social and Preventive Medicine (E.B., M.R.), University of Montreal, Montreal, Quebec, Canada; the Department of Trau-matology (E.B., A.L., D.C.), Charles-LeMoyne Hospital, University of Sher-brooke, Greenfield Park, Quebec, Canada; the Traumatology Program (A.L.), Laval University of Vermont, Burlington, Vermont.

Submitted for publication September 23, 2003.

Accepted for publication December 11, 2003.

Poster presentation at the 62nd Annual Meeting of the American Association for the Surgery of Trauma, September 11–13, 2003, Minneapolis, Minnesota.

Address for reprints: Eric Bergeron, MD, MSc, Hôpital Charles-LeMoyne, 3120 Blvd. Taschereau, Greenfield Park, Quebec J4V 2H1, Canada; email: eric.bergeron@traumaquebec.org.

© 2004 Lippincott Williams & Wilkins, Inc.