Aims, Objectives and Background Trauma audit has used lived/died as an outcome for 30 years, but Patient Reported Outcome Measures (PROMS) have also been collected by the Trauma Audit and Research Network (TARN) for the past 5 years across major trauma centres. These are measured at 6 months after injury and include two measures of health-related quality of life, EQ5D-5 and GOSE, employment status and three patient experience questions. It is not known if 6-month PROMS can be predicted after major trauma.

Method and Design The TARN PROMS data was extracted and randomly divided into a model development (training) and a model testing (test) dataset. There is no standard way of using this type of complex data, so three different modelling approaches were used: (1) conventional logistic regression, (2) artificial intelligence (AI) selection of ‘nearest neighbours’, and (3) AI decision trees. The performance of each model was evaluated using the test dataset.

Results and Conclusion There were 5791 patients in the training set and 1447 patients in the test set. All three of the methods achieved an ROC AUC between 0.69 and 0.77 – implying that this might be the limit of prediction based on this type of data. When tested against the binary Glasgow Outcome Score the results are shown in the table 1. The AI method of ‘k Nearest Neighbours’ achieved a balance between sensitivity (72%) and specificity (71%).

Conclusion patient reported outcomes at 6 months after injury can be predicted from in-hospital data. This potentially gives a new method for clinical audit and comparison of trauma outcomes using a measure that is more relevant to survivors of major trauma than the current ‘lived/died’.