Is artificial intelligence superior to traditional regression methods in predicting prognosis of adult traumatic brain injury?

Abstract

Traumatic brain injury (TBI) is a significant global health issue with high morbidity and mortality rates. Recent studies have shown that machine learning algorithms outperform traditional logistic regression models in predicting functional outcomes for TBI patients. This research aimed to compare the accuracy of the binomial logistic regression model with the Extreme Gradient Boosting (XGBoost) machine learning model. The study included 5056 adult TBI patients evaluated using the Glasgow Outcome Scale (GOS). The XGBoost model was trained on 80% of the sample and tested on the remaining 20%. The logistic regression model accurately predicted 59.7% of unfavorable outcomes, with a significant impact of variables like age and Glasgow Coma Scale (GCS). The ROC curve analysis showed an Area Under the Curve (AUC) of 0.942, indicating the model's predictive ability. The XGBoost algorithm achieved an accuracy of 0.89, AUC of 0.83. The most critical variables in the XGBoost model were days of hospitalization, age, systolic blood pressure, ICU length of stay, GCS and respiratory rate. The XGBoost algorithm performed better in accuracy for predicting unfavorable outcomes, while logistic regression was superior in terms of the ROC curve. Further studies are needed to fine-tune the algorithm's hyperparameters and develop models applicable in clinical settings. Clinical trial number Not applicable.