# Development and validation of a predictive model for cognitive impairment after first-episode acute ischemic stroke without reperfusion therapy

## TL;DR

This study developed a machine learning model to predict cognitive impairment after first-time stroke, using factors like education and brain infarction location.

## Contribution

A novel predictive model for cognitive impairment after stroke was developed and validated using machine learning and clinical factors.

## Key findings

- The model identified number of infarcts, C-reactive protein levels, and frontal lobe infarction as key risk factors for cognitive impairment.
- Logistic regression performed best with AUCs of 0.925 (internal) and 0.897 (external validation).
- Higher education level was found to be a protective factor against cognitive impairment.

## Abstract

To construct and validate a predictive model of cognitive impairment after first-episode acute ischemic stroke (AIS) based on multiple factors.

A total of 627 patients with first-episode AIS admitted to the Affiliated Huai’an Hospital of Xuzhou Medical University between January 2023 and June 2024 were enrolled in this study. Patients were followed up 6 months after discharge. Cognitive function was assessed using the Mini-Mental State Examination (MMSE). Among them, 188 patients with cognitive impairment were assigned to the cognitively impaired group, and 439 cognitively normal patients were assigned to the non-impaired group. These 627 patients were randomly split 8:2 into a training cohort (n = 502) and an internal validation cohort (n = 125). Additionally, 184 patients from the Huai’an First People’s Hospital served as the external validation cohort. The least absolute shrinkage and selection operator (LASSO) regression was used to screen for variables significantly associated with cognitive impairment, and multivariate logistic regression was used to further verify their independence. Python and Dataiku DSS platforms were used to build seven machine learning models, including gradient boosted trees, random forest, support vector machine and so on, and the prediction performance of the model was evaluated by AUC value of ROC.

LASSO regression finally identified 10 predictors, such as education level, C-reactive protein (CRP), frontal lobe infarction, number of infarcts. Subsequent multivariable logistic regression analysis revealed that the number of infarcts (OR = 2.17, 95% CI: 1.65–2.86, p < 0.001), C-reactive protein level (OR = 1.02, 95% CI: 1.01–1.03, p = 0.002), and left-sided frontal lobe infarction (OR = 4.56, 95% CI: 1.97–10.55, p < 0.001), among others, were independent risk factors for cognitive impairment after acute ischemic stroke (AIS). In contrast, a higher educational level was identified as a protective factor. Among the machine learning models evaluated, the logistic regression model demonstrated the best performance. It achieved an area under the curve (AUC) of 0.925 (accuracy: 0.871) in the internal validation and an AUC of 0.897 (accuracy: 0.817) in the external validation. The model achieved high predictive accuracy and demonstrated excellent discrimination.

This study successfully constructed a predictive model of cognitive impairment after AIS based on multi-factor analysis and machine learning. By visualizing the model in the form of the nomogram, clinicians can quickly assess the risk based on the individual patient data, thus enabling early identification and individualized intervention.