Development and validation of a prediction model for activities of daily living dysfunction among stroke survivors: insights from the CHARLS cohort.

Abstract

Background: Activities of daily living (ADL) dysfunction is prevalent in stroke survivors and places a significant burden on both patients and healthcare systems. Improved identification of individuals with ADL dysfunction may facilitate more targeted rehabilitation strategies.

Methods: The China Health and Retirement Longitudinal Study (CHARLS) provided the data. A training set (n = 906) and a validation set (n = 389) were randomly selected from a total of 1,295 stroke survivors. Least absolute shrinkage and selection operator (LASSO) regression and multivariable logistic regression were used to select predictors and develop a prediction model, which was visualized using a nomogram. SHapley Additive exPlanations (SHAP) were applied for model interpretation. The area under the receiver operating characteristic curve (AUC), calibration analysis, and decision curve analysis (DCA) were used to evaluate the model's performance.

Results: Ten predictors were identified, including CES-D scores, age, sleep time duration, drinking, lung disease, social contact, falls, hypertension, arthritis, and sex. SHAP analysis identified CES-D scores as the most influential predictors. The model demonstrated acceptable discriminative ability in both the training set (AUC: 0.76, 95% CI: 0.73-0.79) and validation set (AUC: 0.76, 95% CI: 0.72-0.81). Calibration was satisfactory in both the training and validation sets (Hosmer-Lemeshow test, P = 0.16 and P = 0.99, respectively). Positive clinical usefulness was suggested by DCA analysis.

Conclusions: The model demonstrated acceptable predictive performance and may assist in identifying individuals with prevalent ADL dysfunction. Further external validation is required before broader clinical application.