Blood Biomarker-Based Machine Learning Model for Predicting Cognitive Impairment in Stroke Patients

BACKGROUND: Cognitive impairment (CI) is common in stroke patients and is associated with a poor prognosis. Early and accurate identification of high-risk patients is crucial. This study aims to develop a predictive model for CI risk in stroke patients using machine learning (ML). METHODS: Data were sourced from the China Health and Retirement Longitudinal Study for stroke patients between 2011 and 2018. Lasso regression and the Boruta algorithm were used to select key feature variables. Ten ML algorithms were developed: traditional logistic regression, Extreme Gradient Boosting, Support Vector Machine, k-Nearest Neighbors, Gradient Boosting Machine, Adaptive Boosting, Neural Networks, Light Gradient Boosting Machine, Random Forest, and CatBoost. Model performance was evaluated using the area under the curve and decision curve analysis. The SHapley Additive exPlanations method was employed for model interpretation. RESULTS: In total, 2505 stroke patients were included, of whom 779 had CI. Ten key feature variables were selected. Among the prediction models, logistic regression model demonstrated the best performance, with an area under the curve of 0.824 [95% confidence interval: 0.794-0.854)]. Decision curve analysis showed that this model provided the highest net benefit. SHapley Additive exPlanations analysis identified education, age, pain, depression, hemoglobin A1c, and blood urea nitrogen as important predictive factors. CONCLUSIONS: ML-based prediction models demonstrate high accuracy in assessing the risk of CI in stroke patients, enabling early intervention to improve outcomes.