Abstract
Objective: To examine the performance of the Corticosteroid Randomization After Significant Head injury (CRASH) trial prognostic model in older patients with traumatic brain injury.
Setting: The National Study on Costs and Outcomes of Trauma cohort, established at 69 hospitals in the United States in 2001 and 2002.
Participants: Adults with traumatic brain injury and an initial Glasgow Coma Scale score of 14 or less.
Design: The CRASH-CT model predicting death within 14 days was deployed in all patients. Model performance in older patients (aged 65-84 years) was compared with that in younger patients (aged 18-64 years).
Main Measures: Model discrimination (as defined by the c-statistic) and calibration (as defined by the Hosmer-Lemeshow P value).
Results: CRASH-CT model discrimination was not significantly different between the older (n = 356; weighted n = 524) and younger patients (n = 981; weighted n = 2602) and was generally adequate (c-statistic 0.83 vs 0.87, respectively; P = .11). CRASH-CT model calibration was adequate for the older patients and inadequate for younger patients (Hosmer-Lemeshow P values .12 and .001, respectively), possibly reflecting differences in sample size. Calibration-in-the-large showed no systematic under- or overprediction in either stratum.
Conclusion: The CRASH-CT model may be valid for use in a geriatric population.