Review and prediction of trauma mortality

Abstract

Quality management principles stipulate that outcome after injury is dependent upon patient factors, injury severity, structures and processes of care in a trauma system. Structures refers to the context in which care is delivered, including material resources, equipment and competence of involved personnel. Processes refers to what is literally done by the personnel involved in patient care. In this thesis, we examine the different aspects of this conceptual model with outcome as the main focus.

Historically, trauma mortality has been the standard quality outcome measure. However, nontrauma related deaths and patients that are dead on arrival (DOA) in registries, complicates the interpretation of trauma mortality statistics. In Paper I, we demonstrated by clinical review of all deaths during 2007-2011 in a Level I trauma centre (Karolinska University Hospital – Solna [KUH]), that 30-day trauma mortality included 10.5% of non-trauma related deaths and the exclusion of DOAs significantly reduced the mortality rate. We concluded that review of all trauma deaths was necessary to correctly interpret trauma mortality.

Analysis of preventable death (PD) is another quality outcome measure. The World Health Organization (WHO) has defined PD by the use of survival prediction models which calculates a probability of survival (Ps): non-PD with a Ps <25%, and potentially PD with a Ps >50%. In Paper II, we used a multidisciplinary peer review during 2012-2016, to identify the proportion of potentially PD and errors committed at KUH, and to evaluate the use of the WHO’s Ps cut-offs as a tool to identify the right patients to review, i.e., exclude non-PD from review or to focus review on potentially PD. We used the North American Trauma and Injury Severity Score (TRISS) and the Norwegian Survival Prediction Model in Trauma (NORMIT) to calculate the Ps. When applying the cut-off limits to the groups of non-PDs and potentially PDs for review, both models missed cases that otherwise needed to be reviewed. We concluded that peer review of all trauma deaths is essential in preventability analysis.

Survival prediction models, which adjust for case-mix, have been developed to allow comparisons of the quality of trauma care between centres and over time. In Paper III, we used TRISS based risk-adjusted survival to compare two Scandinavian Level I trauma centres (KUH and Oslo University Hospital – Ullevål) during 2009-2011 and concluded that the model had its shortcomings when applied in a Scandinavian setting. The model lacks adjustments for age as a continuous variable and does not include comorbidity which, if included, could improve survival prediction in Scandinavian trauma populations.

In Paper IV, we tested the accuracy of NORMIT and its later update (NORMIT 2), in regards to survival prediction, in two Swedish trauma populations; one national population including all hospitals admitting trauma patients in Sweden and one subpopulation of patients admitted to a single designated Level I trauma centre (KUH) during 2014-2016. We concluded that NORMIT 2 can be used to predict survival in a Swedish trauma centre population, but both NORMIT models performed poorly in a more heterogeneous national trauma population.