Development and validation of a prediction model for long-term cognitive frailty risk in stroke patients based on CHARLS data.

The prevalence of cognitive frailty among stroke survivors in the study sample was 29.59%.

• A total of 2,325 stroke patients were included in the study.
• 688 of the 2,325 stroke patients (29.59%) exhibited symptoms of cognitive frailty.
• Data were drawn from the China Health and Retirement Longitudinal Study (CHARLS), conducted between 2018 and 2020.
• The study population consisted of community-dwelling elderly adults with stroke.

XGBoost achieved the highest predictive performance among eight machine learning models evaluated for predicting cognitive frailty in stroke survivors.

• XGBoost achieved an AUC of 0.810.
• Random Forest was the second-best performing model with an AUC of 0.795.
• The eight models evaluated were: Logistic Regression, Decision Tree, XGBoost, Support Vector Machine, k-Nearest Neighbors, Naïve Bayes, Random Forest, and LightGBM.
• LASSO regression was employed to identify predictive factors prior to model development.

Age and education level were identified as the most significant predictors of cognitive frailty risk in stroke survivors based on SHAP values.

• Age had a SHAP value of 0.32, the highest among all predictors.
• Education had a SHAP value of 0.28.
• IADL (Instrumental Activities of Daily Living) had a SHAP value of 0.21.
• Nutritional status had a SHAP value of 0.18, and physical exercise had a SHAP value of 0.16.
• SHapley Additive exPlanations (SHAP) values were applied to interpret the contributions of the variables.

Five key predictors of cognitive frailty in stroke survivors were identified through LASSO regression and SHAP analysis.

• The key predictors were education, nutritional status, physical exercise, Instrumental Activities of Daily Living (IADL), and age.
• 22 behavioral variables were initially examined, encompassing indicators from the sociodemographic, physical, psychological, cognitive, and social domains.
• LASSO regression was used to reduce the variable set to the most predictive factors.
• These predictors were described as 'readily available clinical and demographic indicators.'

The optimized XGBoost model was identified as a practical tool for early screening of cognitive frailty risk in primary care settings.

• The model was described as suitable for use 'particularly within primary care settings.'
• The model leverages 'readily available clinical and demographic indicators.'
• The authors noted the model 'can aid clinicians in devising targeted intervention strategies to mitigate disease progression.'
• Further external validation was identified as necessary to confirm generalizability across various clinical contexts.