Machine learning model for unfavorable outcome prediction in neurosurgical patients: the potential role of liver function markers.

Introduction: Early prediction of prognosis for neurosurgical diseases remains challenging. This study aimed to develop a machine learning-based model to predict unfavorable outcomes in neurosurgical patients. Methods: We conducted a retrospective cohort study of patients with traumatic brain injury, intracerebral hemorrhage, or aneurysmal subarachnoid hemorrhage between 2018 and 2020. The primary outcome was functional status at discharge, assessed via the modified Rankin Scale. Feature selection used LASSO regression and the Boruta algorithm, with overlapping selected features retained for model development. Six machine learning algorithms were trained with 5-fold cross-validation for hyperparameter optimization via Optuna. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Shapley additive explanations were used for interpretability. Results: The CatBoost model performed best (AUC = 0.932, accuracy = 0.879, precision = 0.872, recall = 0.810, F1 score = 0.840, Brier score = 0.116), balancing discriminative power and clinical relevance. Key predictive features included Glasgow Coma Scale (GCS) score at admission, age, and liver function markers including aspartate transaminase (AST) Discussion: The CatBoost model showed excellent performance in predicting the prognosis of neurosurgical patients by integrating neurological and liver function markers. Future studies are needed for external validation through multicenter investigations, and explore mechanistic associations between liver dysfunction and neurological deterioration.