This project examined the validity of the CRASH and IMPACT predictive models for traumatic brain injury (TBI) mortality.

Using data from the National Trauma Data Bank, we assessed the performance of both predictive models using a nationwide cohort of 26,228 (larger than both the original CRASH and IMPACT datasets combined).

Why do this study?

How do they work in the current clinical setting?

Both of these predictive scores were developed using patients from 15 to 35 years ago¹, and the CRASH score in particular was derived from a population that was mostly coming from low and middle-income countries. In the years following the development of these models, it’s likely that some factors have changed (innovations in TBI treatment, characteristics of injury mechanism), so it’s important to validate these models’ performance in a more recent cohort.

Interesting methodological paper

This was unlike most projects I’d worked on before, and was very much a learning experience (both in the analysis and in writing the paper). It highlighted the importance of case-mix and the sensitivity of predictive models to the underlying prevalence of disease. It was also a change of pace to write about something that didn’t have an “exposed/treated” group, and how to present these technical findings to a more clinically-oriented audience.