Important Predictors of In-Hospital Mortality in Patients With Acute Subdural Hematomas Using Machine Learning

Abstract

Abstract INTRODUCTION Severe traumatic brain injury (TBI) associated with acute subdural hematomas (aSDH) is common and represents around 10% to 20% of all TBI. Predictive models have been used in an attempt to modulate the morbidity and mortality of patient outcomes. We used machine learning (ML) to identify risk factors predictive of in-hospital mortality in the severe TBI patient population with aSDH. METHODS We included 74 patients with severe TBI and aSDH in the analysis. Random forest, ML architecture, was used to create a predictive model of in-hospital mortality with a pre-set precision of 97.4% (RStudio-3.5). A total of 133 input variables including demographics, in-hospital laboratory values, and outcome measures were included and mean accuracy ranks were assessed RESULTS The highest scoring input variables were length of stay, last sodium value collected, last platelet value collected, and Glasgow Coma Scale (GCS) motor exam score on day two. Mean length of stay was significantly shorter in the group of patients that died (4.114 ± 4.241 d vs 22.72 ± 11.72; P < .0001) The mean sodium that was last collected was significantly more elevated in the group of patients who died (139.9 ± 3.299 vs 148.9 ± 8.825 mEq/L; P < .0001). The mean platelet values last collected during the hospitalization were significantly lower in the group of patients who died (440.8 ± 240.4 vs 165.6 ± 113.7 × 109/L; P < .0001). GCS motor exam score at day 2 following the injury was also significantly greater in the survival group (4.872 ± 1.005 vs 2.143 ± 1.574; P < .0001). CONCLUSION ML is an efficient tool that can provide a reasonable level of accuracy in predicting mortality in our patient population. Adequate monitoring of sodium and platelet levels, as well as the GCS motor examination, can promote goal-directed therapy. Integration of ML into the severe TBI algorithm may help early identification of high-risk patients.