Application of interpretable machine learning to predict activities of daily living disability in sarcopenia: insights from the CHARLS dataset.

PURPOSE: The decline in activities of daily living (ADL) among older persons is a significant public health concern. Sarcopenia is a major risk factor for ADL disability. This study aimed to develop and validate an interpretable machine learning (IML)-based model to predict ADL disability in sarcopenia patients using the China Health and Retirement Longitudinal Study (CHARLS) dataset.

METHODS

Participants diagnosed with sarcopenia between 2015 and 2018 were included and divided into training and test cohorts (8:2 ratio). Data were pre-processed using the edited nearest neighbor method (ENN). The Boruta algorithm, XGBoost model, and SHAP method were employed for feature selection, prediction, and interpretation. Model performance was assessed using AUC, accuracy, sensitivity, specificity, and F1 score.

RESULTS

ADL disability occurred in 258 (22.5%) of 1146 sarcopenia patients. Key predictors identified by SHAP analysis included depression severity, walking speed, grip strength, and age. The XGBoost model achieved an AUC of 0.803 and 0.738 in the training and test sets, respectively, with good interpretability.

DISCUSSION: The model demonstrated strong predictive performance and identified critical factors influencing ADL disability, underscoring the potential of IML in geriatric risk stratification.

CONCLUSION

The IML-based model provides a scientific basis for developing health management strategies aimed at reducing ADL disability in older adults with sarcopenia, thereby improving quality of life and easing societal and familial care burdens.