Random Forest Prognostication of Survival and 6-Month Outcome in Pediatric Patients Following Decompressive Craniectomy for Traumatic Brain Injury.

Abstract

Background: There is a dearth of literature regarding prognostic and predictive factors for outcome following pediatric decompressive craniectomy (DC) performed after traumatic brain injury (TBI). The aim of this study was to develop a random forest machine learning algorithm to predict outcomes following DC in pediatric patients.

Methods: This multi-institutional retrospective study assessed the 6-month postoperative outcome in pediatric patients who underwent DC. We developed a machine learning model using classification random forest (CRF) and survival random forest (SRF) algorithms for prediction of outcomes. Data on clinical signs, radiographic studies, and laboratory studies were collected. Outcome measures for the CRF model were mortality and good or bad outcome based on Glasgow Outcome Scale at 6 months. A Glasgow Outcome Scale score of ≥4 indicated a good outcome. Outcome for the SRF model was mortality during the follow-up period.

Results: The study included 40 pediatric patients. Hospital mortality rate was 27.5%, and 75.8% of survivors had a good outcome at 6-month follow up. The CRF model for 6-month mortality had a receiver operating characteristic area under the curve of 0.984, whereas, 6-month good and bad outcomes had a receiver operating characteristic area under the curve of 0.873. The SRF model was trained at the 6-month time point with a receiver operating characteristic area under the curve of 0.921.

Conclusions: CRF and SRF models successfully predicted 6-month outcomes and mortality following DC in pediatric patients with TBI. These results suggest that random forest models may be efficacious for predicting outcome in this patient population.