Development and validation of a risk identification model for frailty in stroke survivors: new evidence from CHARLS

Abstract

Background: Stroke survivors with frailty exhibit elevated rates of complications, mortality, disability, and hospital readmission. As frailty represents an early, reversible, and preventable stage of disability, developing a reliable risk identification model is essential. This study aimed to develop and validate a risk model for frailty among stroke survivors using data from the China Health and Retirement Longitudinal Study (CHARLS).

Methods: Data were extracted from the CHARLS database. Stroke survivors were identified and assessed across 30 indicators, including socio-demographic, physical, psychological, cognitive, and social variables. The data were divided by year, with 2013 and 2015 as the development set and 2018 and 2020 as the validation set. Least Absolute Shrinkage and Selection Operator (LASSO) regression was employed for variable selection. Logistic regression models were then developed based on univariate and LASSO-selected predictors. A nomogram was constructed to facilitate risk visualization. Calibration curves and decision curve analysis were used to evaluate model calibration and clinical utility.

Findings: A total of 2,188 stroke survivors from the 2013, 2015, 2018, and 2020 follow-ups were included. Approximately 68% exhibited symptoms of frailty. Significant group differences were found by age, marital status, living alone, hypertension, and self-reported health status (all p < 0.05). Age, poor sleep quality, impaired balance, nervousness/anxiety, and living alone emerged as independent risk factors for frailty. The area under the receiver operating characteristic (ROC) curve for the development and validation sets was 0.833 and 0.838, respectively.

Interpretation: The model derived from CHARLS data identified 5 readily assessable predictors (age, sleep quality, balance, anxiety, and living alone), allowing for early screening of frailty without specialized instruments. It demonstrated superior discriminatory performance compared to models from smaller-sample studies, supporting targeted interventions and providing valuable insights for identifying high-risk stroke survivors.

Interpretation: The model derived from CHARLS data identified 5 readily assessable predictors (age, sleep quality, balance, anxiety, and living alone), allowing for early screening of frailty without specialized instruments. It demonstrated superior discriminatory performance compared to models from smaller-sample studies, supporting targeted interventions and providing valuable insights for identifying high-risk stroke survivors.

Keywords: CHARLS; Frailty; Logistic regression; Risk identification model; Stroke.

Fig. 1

Flowchart of this study

Fig. 2

Subject selection flow chart of this study

Fig. 3

Correlation Heatmap

Fig. 4

Results of the LASSO regression model for the selection of demographic and clinical characteristics (a) Coefficient curve based on the logarithmic (lambda) series: this curve shows the variation in coefficients corresponding to different lambda values. Non-zero coefficients are generated from the best lambda, indicating characteristics important for predicting the risk of frailty. b Best parameter (lambda) in LASSO: this plot shows the variation in log(lambda) relative to the partial likelihood deviation (binomial deviation) curve. The best lambda value within a range of one standard error (1-SE criterion) was found by tenfold cross-validation and minimum criterion selection. The position corresponding to the best lambda is indicated by a virtual vertical line

Fig. 5

Bubble Chart

Fig. 6

Nomogram

Fig. 7

Nomogram ROC curve: A Generated from the development dataset. B Validation dataset Generated from the validation dataset

Fig. 8

Calibration plot: A Development dataset. B Validation dataset

Fig. 9

DCA curves: A Development dataset. B Validation dataset