A predictive model for mortality in massively transfused trauma patients

Abstract

Background: Improvements in trauma systems and resuscitation have increased survival in severely injured patients. Massive transfusion has been increasingly used in the civilian setting. Objective predictors of mortality have not been well described. This study examined data available in the early postinjury period to identify variables that are predictive of 24-hour- and 30-day mortality in massively transfused trauma patients.

Methods: Massively transfused trauma patients from 23 Level I centers were studied. Variables available on patient arrival that were predictive of mortality at 24 hours were entered into a logistic regression model. A second model was created adding data available 6 hours after injury. A third model evaluated mortality at 30 days. Receiver operating characteristic curves and the Hosmer-Lemeshow test were used to assess model quality.

Results: Seven hundred four massively transfused patients were analyzed. The model best able to predict 24-hour mortality included pH, Glasgow Coma Scale score, and heart rate, with an area under the receiver operating characteristic curve (AUROC) of 0.747. Addition of the 6-hour red blood cell requirement increased the AUROC to 0.769. The model best able to predict 30-day mortality included the above variables plus age and Injury Severity Score with an AUROC of 0.828.

Conclusion: Glasgow Coma Scale score, pH, heart rate, age, Injury Severity Score, and 6-hour red blood cell transfusion requirement independently predict mortality in massively transfused trauma patients. Models incorporating these data have only a modest ability to predict mortality and should not be used to justify withholding massive transfusion in individual cases.