Transferability of Clinical Prediction Models for Early Trauma Care in a Swedish Setting

Trauma is one of the top threats to population health globally. A substantial body of research has been devoted to the development of clinical prediction models to aid early decision making in trauma care. Often these models are applied outside the context in which they were originally developed. In contrast, very little research has focused on the effects on model performance of such transfers despite the fact that a potential loss in performance could have devastating consequences. Data from the Swedish trauma registry SweTrau will be used to study the effects on model performance of transfers between different contexts within a single conceptually homogenous setting. Using logistic regression models for predicting all-cause mortality within 30 days of trauma in adult patients will be developed, focusing on systolic blood pressure, respiratory rate, and Glasgow coma scale as predictors. Four different sets of data sampled from SweTrau will be used to simulate transfer of models between high and low volume centres, metropolitan and non-metropolitan centres, multicentre and single centre data, and finally between individual centres. Measures of overall performance, clinical usefulness, discrimination, calibration, and recalibration will be used to quantify the effects on performance of model transfers. The study will provide evidence to help clinicians and policy makers in deciding on whether it is appropriate to apply models developed in other contexts. For example, the results of this study may inform decisions on the development and implementation of models intended to be applied on a national or regional level, and ultimately help designing better trauma care and improve the outcomes of trauma patients.