OBJECTIVE: To examine the performance of the International Mission for Prognosis and Clinical Trial Design in Traumatic Brain Injury (IMPACT) prognostic models in older patients.
METHODS: Using data from the National Study on Costs and Outcomes of Trauma (NSCOT), this study identified adult patients presenting to US hospitals in 2001 and 2002 with non-penetrating moderate or severe traumatic brain injury (GCS ≤ 12). IMPACT model calibration and discrimination in the older stratum (65-84 years) was compared to that in the younger stratum (18-64 years).
RESULTS: IMPACT model discrimination did not differ significantly between the older (n = 202; weighted n = 268) and younger strata (n = 613; weighted n = 1632) and was generally adequate (c-statistic for the core-death model = 0.81 [0.77-0.84] vs 0.75 [0.66-0.84], respectively; p = 0.26). IMPACT model calibration was poor for both older and younger strata (Hosmer-Lemeshow p-value for the core-death model = 0.01 vs < 0.0001, respectively). Pre-specified qualitative graphical evaluation suggested substantial under-prediction of mortality in the oldest decades of life, but not among younger patients.
CONCLUSIONS: The examined IMPACT prognostic models demonstrated adequate discrimination and poor calibration in both older and younger patients, yet particular caution may be required when applying these models to the elderly.