XGBoost Machine Learning Algorism Performed Better Than Regression Models in Predicting Mortality of Moderate-to-Severe Traumatic Brain Injury

Abstract

Traumatic brain injury (TBI) brings severe mortality and morbidity risk to patients. Predicting the outcome of these patients is necessary for physicians to make suitable treatments to improve prognosis. The aim of this study is to develop a mortality prediction approach using XGBoost (extreme gradient boosting) in moderate-to-severe TBI. A total of 368 patients hospitalized in West China hospital for TBI with Glasgow Coma Scale (GCS) below 13 were identified. To construct the XGBoost prediction approach, patients were divided into training set and test set with a ratio of 7:3. A logistic regression prediction model was also constructed and compared with the XGBoost model. Area under the receiver operating characteristic curve, accuracy, sensitivity, and specificity were calculated to compare the prognostic value between XGBoost and logistic regression. A total of 205 patients suffered a poor outcome with a mortality of 55.7%. Nonsurvivors had a lower GCS (5 vs. 7, P < 0.001) and a higher Injury Severit Score (ISS) than survivors (25 vs. 16, P < 0.001). Platelet (P < 0.001), albumin (P < 0.001), and hemoglobin (P < 0.001) were significantly lower in nonsurvivors, whereas glucose (P < 0.001) and prothrombin time (P < 0.001) were significantly higher in nonsurvivors. Among the XGBoost approach, GCS, prothrombin time, and glucose had the most significant feature importance. The area under the receiver operating characteristic curve (0.955 vs. 0.805) and accuracy (0.955 vs. 0.70) of XGBoost were both higher than logistic regression. Predicting mortality of patients with moderate-to-severe TBI using the XGBoost algorism is more effective and precise than logistic regression. The XGBoost prediction approach is beneficial for physicians to evaluate patients with TBI at high risk of poor outcome.