Predicting mortality in non-traumatic intracerebral hemorrhage with glucose and lipid data

This study integrated dynamic glucose variation indicators and lipid profiles to develop and validate a machine learning-based predictive model for in-hospital mortality in patients with non-traumatic intracerebral hemorrhage (NTICH).

Data of this study were derived from the Medical Information Mart for Intensive Care-IV (MIMIC -IV) database (2008-2019, which was split into training and internal validation sets at a 7:3 ratio) and from NTICH cases from the Second Affiliated Hospital of Harbin Medical University (for external validation). The Boruta algorithm was used to evaluate feature importance. Nine machine learning algorithms were used to develop predictive models for in-hospital mortality in NTICH patients. Model discrimination was assessed using the area under the curve (AUC) and clinical utility was evaluated with decision curve analysis. The SHapley Additive exPlanations (SHAP) method was used to rank feature importance.

A total of 2,323 patients were included from the MIMIC-IV database, with an in-hospital mortality rate of 14.03%. The Boruta algorithm identified 20 relevant features. The K-Nearest Neighbors model achieved the highest AUC in the training set (AUC = 0.980), the Light Gradient Boosting Machine (LightGBM) model performed best in the internal validation set (AUC = 0.851), and the eXtreme Gradient Boosting (XGBoost) model yielded the highest performance in the external validation set (AUC = 0.814). SHAP analysis identified that the Sequential Organ Failure Assessment score, Glasgow Coma Scale score, age, invasive mechanical ventilation, and mean glucose were the most important predictors.

The XGBoost and LightGBM models demonstrated excellent performance for predicting in-hospital mortality in patients with NTICH. This study highlights the critical value of integrating dynamic glucose variation indices and comprehensive lipid profiles in improving the prognostic prediction of patients with NTICH. The identified key predictive factors provide actionable targets for early risk stratification and individualized intervention strategies, such as precise glucose regulation, thereby facilitating the optimization of resource allocation in neurocritical care and improving clinical outcomes.