The Impact of a Bayesian Network for Pre-Hospital Decision-Support after Trauma

In patients with major traumatic injuries, early intervention can be lifesaving. However, identifying high-risk patients can be difficult, and judgement errors may compromise optimal care. Prediction models can be used to augment clinical judgment. The aim of the thesis was to assess whether Bayesian Networks (BN), which are causal probabilistic models able to fuse knowledge and data, can augment clinical judgment in pre-hospital trauma care. The thesis focuses on decisions relating to Trauma Induced Coagulopathy (TIC), a difficult to diagnose condition that is central to resuscitation decisions, and haemorrhage. To assess decision making, thematic analysis of interviews with expert clinicians was performed. The interviews revealed that situational recognition is key to effective decision making, and uncertain information results in difficult decisions. An example of uncertainty was evident in clinician’s decisions to activate the major haemorrhage protocol. In such decisions, the predictive performance of an existing BN model (“ED TIC BN”) was not significantly different to native clinical judgement (at equal specificity, sensitivity = clinicians: 86% vs BN: 96%, p = 0.13). The ED TIC BN risk prediction model was updated for pre-hospital use to enable earlier decision support, and the PH TIC BN performed better than any other TIC model. The impact of the pre-hospital model on decision-making was analysed in a prospective, multicentre study. The discriminative performance of the BN was not significantly different from clinical assessment. The performance of clinical assessment did not improve after decision support information. However, there was evidence of the positive influence of the PH TIC BN on clinical assessment and increased ease of decision making. This thesis advances the understanding of impact analysis in pre-hospital decision support after injury. The knowledge generated will be used to design subsequent trials to assess the impact of decision support on patient and health system outcomes