Project V - Clinical prediction modelling for trauma care

One of the most central parts of trauma care is triage; defined as the prioritization of the severely injured according to their needs, chances of survival, and available resources. The process of triage requires a highly developed clinical ability to adequately judge patients' chances of survival. To aid trauma care providers, extensive research has focused on the development of accurate mortality prediction models. In the trauma care context, a prediction model can also be referred as a prognostic model, risk score, and triage score, and aims to provide trauma care providers with an estimate of a patient's probability of survival, to understand prognosis, guide treatment and improve resource utilization. This estimate is often based on several different parameters, such as physiological, anatomical injury, and mechanism of injury.

Most established prediction models have now been criticized for methodological limitations. The literature of prediction modelling is growing, and this development has driven demands for increased methodological rigor. Key issues in trauma prediction modelling include methods to derive and validate models, including how to deal with continuous varibales such as vital signs and missing data. Another current hot topic is the transferability of a model developed in one setting to use in another setting without adjustment.

In this project, we aim to generate knowledge on methodological issues in trauma prediction modelling, such as transferability, validation, updating and adjustment for regional versus national use, and handling of continuous vital sign data, as well as missing data. This year, we will build on our previous results by using more data from the US National Trauma Data Bank, start working with the Swedish trauma registry, and expand our our Indian databank

Objectives 2016

  1. To assess if the performance of prediction models for trauma care is affected by transfer within Sweden, i.e. if such models and transferred between high- and low-volume centers, or rural and urban centers
  2. To assess if international models may be adjusted to fit the Swedish trauma context using outcome prevalence data
  3. To compare the performance of simple vital sign based models for trauma care with more complex triage algorithms in naive contexts