A machine learning-based risk prediction model for cognitive frailty in middle-aged and elderly participants with circadian syndrome in China: a longitudinal study

Middle-aged and elderly individuals with circadian syndrome (CircS) are prone to cognitive frailty (CF). Early identification of high-risk individuals is critical for optimizing clinical decision-making and timely intervention. Machine learning (ML) has been frequently applied in the field of risk prediction research. However, the interpretation and translation of predictive models into clinical applications are equally crucial. This study aims to develop an interpretable machine learning-based risk prediction model for CF in middle-aged and elderly participants with CircS. Feature selection was conducted using five algorithms: LASSO, Boruta, Random Forest importance ranking, univariate logistic regression, and recursive feature elimination (RFE). Six ML models, including K-nearest neighbors (KNN), decision tree (DT), support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and light gradient boosting machine (LightGBM) were developed to predict CF. Model performance was assessed using area under the receiver operating characteristic (AUROC), accuracy, sensitivity, specificity, F1 score, positive predictive value, and negative predictive value. Calibration curve and Brier score were used to evaluate the degree of calibration of the model, and decision curve analysis was generated to assess the net clinical benefit. Additionally, the Shapley additive explanation (SHAP) method was employed to evaluate the risk of CF among middle-aged and elderly participants with CircS. A total of 2,894 middle-aged and elderly participants with CircS were included in this study, among whom 761 (26.3%) were identified as having CF. Nine predictors were identified and ranked by SHAP analysis from highest to lowest importance: Education, age, BMI, activities of daily living score (BADL/IADL), residence, gender, self-report health, grip strength, and marry. Among the six models, the SVM model demonstrated the most robust predictive performance, achieving AUROC values of 0.858 and 0.849 in the training and testing dataset, respectively. The calibration curve of SVM model also revealed a high consistency between observations and predictions. Decision curve analysis showed that the SVM model had the greatest net benefit rate when the prediction probability threshold was 0.10–0.80. In addition, we developed a web-based risk calculator for middle-aged and elderly participants with CircS. The SVM model developed reliably predicts CF in middle-aged and elderly participants with CircS, offering a valuable tool for early intervention and enhanced patient management.