Background: Despite serious documented limitations, the Trauma Injury Severity Score (TRISS) is still used for risk adjustment in trauma system evaluation and clinical research. Several modifications have been proposed to address TRISS limitations. We aimed to assess the impact of proposed TRISS modifications on the accuracy of mortality prediction for blunt trauma.
Methods: The Quebec Trauma Registry (QTR), based on a mature, regionalized trauma system with mandatory participation of all trauma centers as well as standardized inclusion criteria and coding practices, was used to evaluate TRISS modifications. The National Trauma Data Bank was then used to validate our findings. Gains in predictive accuracy were evaluated in logistic regression models of hospital mortality with the area under the receiving operator curve and the Hosmer-Lemeshow statistic.
Results: When population-based weights, expanding age, modeling the Glasgow Coma Scale score as a quantitative variable, adding an indicator of comorbid status, and modeling quantitative variables with nonparametric functions to allow the expression of nonlinear relations to mortality were used, all were associated with a significant improvement in model discrimination.
Conclusions: Several modifications that have been proposed to address limitations of the TRISS lead to significant improvements in the accuracy of mortality prediction. This study provides valuable information in the quest to improve trauma mortality modeling.
From the Unité de traumatologie-urgence-soins intensifs (L.M., A.L., A.F.T., N.L.S., M.E., E.B.), Centre de recherche du CHA (Hôpital de l'Enfant-Jésus). Quebec City, Quebec, Canada; Département de médecine sociale et préventive (L.M., A.L., B.A.), Département d′Anethésiologie, Division de Soins Intensifs (A.F.T.), Département de médecine familiale (N.L.S., M.E.), Université Laval, Quebec City; Department of Surgery (M.L.), McGill University Health Center, McGill University, Montreal, Quebec, Canada; and Département de chirurgie (É.B.), hôpital Charles-Lemoyne, Greenfield Park, Université de Sherbrooke, Sherbrooke, Quebec, Canada.
Submitted for publication October 15, 2008.
Accepted for publication March 30, 2009.
Supported by Fonds de la recherche en Santé du Québec (grant number: 015102, A.F.T. is the recipient of a Clinical Research Career Award) and Canadian Institutes of Health Research (L.M. is the recipient of a Doctoral Research Award).
Address for reprints: Lynne Moore, PhD, Unité de traumatologie-médecine d'urgence-soins intensifs, Centre de recherche du CHA (Hôpital de l'Enfant-Jésus), 1401, 18eme rue, Quebec City, Quebec, Canada G1J 1Z4; email: firstname.lastname@example.org.