Integrating machine learning and SHAP analysis to identify the risk of cognitive frailty in older adults with chronic heart failure.

BACKGROUND: Accurately identifying older adults with chronic heart failure (CHF) at high risk of cognitive frailty is crucial for timely preventive interventions, yet current approaches to detect this risk are under-researched. This study aimed to construct and test machine learning models to identify cognitive frailty risk in older CHF patients. METHODS: This cross-sectional study continuously enrolled 807 older CHF patients (aged ≥60 years) in a tertiary hospital from April 2023 to May 2024. The data were randomly split into training and testing sets (8:2 ratio). Feature selection was performed using least absolute shrinkage and selection operator regression. Six machine learning algorithms were used to construct the models. Model performance was evaluated based on discrimination, calibration, and clinical application. The SHapley Additive exPlanations (SHAP) was employed to attain both global and individual-level interpretation for the output of an optimal model. RESULTS: The training and testing sets comprised 645 and 162 patients respectively. Among these models, categorical boosting (CatBoost) achieved best comprehensive performance, with AUROC of 0.9157 (95% CI, 0.8775-0.9522) on the testing set. The SHAP demonstrated that depression, age, malnutrition, hospitalization for CHF, and education level were among the top five important features for cognitive frailty. At the patient level, the waterfall plot exhibited a clinically meaningful explanation of the CatBoost algorithm. CONCLUSIONS: We constructed a ML model (CatBoost) to identify the risk of cognitive frailty in older adults with CHF. Moreover, SHAP enhances model interpretability, offering insights into contributors to cognitive frailty.