Developing and validating an interpretable machine learning model for frailty risk prediction in patients with...

Chronic diseases and frailty are both prevalent among middle-aged and older adults. Individuals with chronic disorders are at a higher risk of developing frailty. This study aims to develop and validate a machine learning model for predicting frailty risk among middle-aged and older adults with chronic diseases, and to assess their generalizability across Chinese and American populations.

Data from the China Health and Retirement Longitudinal Study (CHARLS) were utilized and split into a training set and an internal validation set at a ratio of 7:3. A subset from the National Health and Nutrition Examination Survey (NHANES) served as the external validation set. Seven machine learning algorithms were employed to predict frailty risk, with feature selection performed using LASSO regression. The SHapley Additive explanation (SHAP) analysis was applied to enhance model interpretability.

Among the 8,535 CHARLS participants, frailty was identified in 23.7% of patients. The Gradient Boosting Decision Tree (GBDT) model demonstrated the best decision-making performance compared to other models, with an AUC of 0.815 (95% CI: 0.797, 0.833) in the internal validation set, accuracy of 0.735 (95% CI: 0.681, 0.752), sensitivity of 0.751 (95% CI: 0.722, 0.844), and specificity of 0.730 (95% CI: 0.633, 0.6752). In the external validation set, the GBDT model achieved an AUC of 0.748 (95% CI: 0.721, 0.775). SHAP revealed the five most important predictors influencing frailty risk: sleep duration, education level, cognitive impairment, age, and visual impairment.

The GBDT model developed effectively predicts the risk of frailty in middle-aged and older adult patients and may serve as a practical tool for the early identification of high-risk populations. In the future, the generalizability of this model requires validation across more diverse cultural contexts to enhance its broader applicability.