Predicting prolonged length of stay in patients with traumatic brain injury: A machine learning approach

Abstract

BackgroundProviding timely care while keeping an efficient bed turnover rate is a challenge hospitals usually face. Early prediction of prolonged hospital length of stay may help devise personalized plans of care that facilitate early discharge and ensure the timely availability of a bed for the next patient. ObjectivesThis study aims to design an Artificial Neural Network (ANN) machine learning model to help early predict patients at risk for prolonged hospital length of stay (PLOS) following Traumatic Brain Injury (TBI). MethodsPLOS was defined as the 75th percentile of the in-hospital length of stay of the entire patient cohort (PLOS ≥23 days). The study targeted adult patients with TBI who were admitted to the trauma surgery between January 2014 and February 2019 with head abbreviated injury score (HAIS) ≥ 3.1417 eligible patients were included (PLOS = 350 and non-PLOS = 1067). ResultsANN achieved good performance with accuracy of 84.3%, area under receiver operating characteristic curve (AUROC) 91.5%, precision 69%, negative predictive value 89%, sensitivity 69%, specificity 89% and F-score 69%. ConclusionThe study discusses the health economic aspects of PLOS and the potential benefits of utilizing machine learning models in enhancing hospital bed utilization.