Related Experiment Video
Updated: Jun 23, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Machine learning-based prediction model for cognitive frailty in elderly patients with ischaemic stroke: a
Xuan Chen1, Linjie Zhou2, Ying Zhang3
1School of Nursing, Air Force Medical University, Xi'an, Shaanxi, China.
Background:
Cognitive frailty (CF), which is characterised by the coexistence of cognitive impairment and physical frailty, is common among older patients after ischaemic stroke (IS) and is associated with adverse functional outcomes. This study aimed to develop and internally validate a machine learning (ML)-based model for predicting 3-month CF risk in older patients with IS.
Methods:
In this prospective cohort study, 402 older patients with IS were enrolled. Baseline assessments included 26 candidate variables, such as demographic characteristics, stroke severity, nutritional status, psychosocial factors, and vascular markers. Feature selection was performed using least absolute shrinkage and selection operator regression within the training set. Ten supervised ML algorithms were evaluated, including random forest (RF), CatBoost, and XGBoost. Model interpretability was assessed using SHAP. Model performance was evaluated using the area under the receiver operating characteristic curve, accuracy, sensitivity, and decision curve analysis.
Results:
At the 3-month follow-up, 149 patients (37.1%) developed CF. Among the evaluated models, the RF model demonstrated the best overall performance on the held-out internal test set, with an AUC of 0.889, an accuracy of 0.798, and a sensitivity of 0.909. SHAP analysis revealed that the discharge National Institutes of Health Stroke Scale score, age, and white matter hyperintensity burden were major contributors to model prediction. Depression and social support also demonstrated notable interactive effects. The RF model demonstrated favourable calibration and net clinical benefit across a range of threshold probabilities.
Conclusion:
This study developed an interpretable ML-based model for estimating early CF risk in older patients after IS using routinely available clinical variables. These findings suggest that neurological, nutritional, and psychosocial factors may jointly contribute to poststroke CF risk. Although the model demonstrated promising performance in terms of internal validation, external validation is needed before clinical application.