Modelling outcomes after paediatric brain injury with admission laboratory values: a machine-learning approach

BACKGROUND: Severe traumatic brain injury (TBI) is a leading cause of mortality in children, but the accurate prediction of outcomes at the point of admission remains very challenging. Admission laboratory results are a promising potential source of prognostic data, but have not been widely explored in paediatric cohorts. Herein, we use machine-learning methods to analyse 14 different serum parameters together and develop a prognostic model to predict 6-month outcomes in children with severe TBI. METHODS: A retrospective review of patients admitted to Cambridge University Hospital's Paediatric Intensive Care Unit between 2009 and 2013 with a TBI. The data for 14 admission serum parameters were recorded. Logistic regression and a support vector machine (SVM) were trained with these data against dichotimised outcomes from the recorded 6-month Glasgow Outcome Scale. RESULTS: Ninety-four patients were identified. Admission levels of lactate, H+, and glucose were identified as being the most informative of 6-month outcomes. Four different models were produced. The SVM using just the three most informative parameters was the best able to predict favourable outcomes at 6 months (sensitivity = 80%, specificity = 99%). CONCLUSIONS: Our results demonstrate the potential for highly accurate outcome prediction after severe paediatric TBI using admission laboratory data.