Predicting clinical outcomes in patients with traumatic bleeding: A secondary analysis of the CRASH-2 dataset

Severe bleeding is one of the main causes of death in hospitals for patients with trauma. Early treatment using tranexamic acid, timely transfer to the intensive care unit and prompt surgical interventions are key factors determining short-term survival and clinical outcomes. The aim of this research is to apply machine learning methods to predict clinical outcomes for patients with severe bleeding from trauma, in order to inform clinical decision making in the hospital setting. The presented study consists in a secondary analysis of the CRASH-2 (Clinical Randomisation of an Antifibrinolytic in Significant Haemorrhage) study data. This dataset contains 20,207 patient entry and outcome data. Machine learning methods have been used to create prognostic models for the prediction of outcomes such as death, significant head injury, need for a surgical operation and admission into the ICU. Results show that patients admitted in the ICU have a higher mortality rate as compared to other patients, highlighting the need for a more detailed analysis of the causes of death in the ICU. Another meaningful result is that a significant head injury can be predicted from a patient's hospital entry data, which may have a significant impact on early treatment decisions and, eventually, improve outcomes.