Interpretable machine learning model for classification of mild cognitive impairment in patients with coronary artery disease and hypertension: development and external validation using UK Biobank and Chinese hospital data

Mild cognitive impairment (MCI) commonly coexists with coronary artery disease (CAD) and hypertension (HBP), reflecting vascular-cognitive interactions. Early identification of MCI is crucial for targeted prevention. This study aimed to develop and externally validate interpretable machine learning (ML) models for identifying prevalent MCI in CAD and HBP patients using UK Biobank and hospital data. In this cross-sectional study, 1,204 patients with CAD and HBP from the UK Biobank were used for model development and internal validation. MCI was identified using ICD-10-based health record codes in the UK Biobank and guideline-based clinical assessment in the external validation cohort. External validation involved 384 patients from the Third Hospital of Mianyang (Sichuan Mental Health Center) ( N = 252) and Yilong County People’s Hospital ( N = 132). Eight ML algorithms, including random forest (RF), logistic regression, and support vector machine, were compared. Model discrimination, calibration, and interpretability were assessed using the area under the receiver operating characteristic curve (AUC), brier score, and SHapley Additive exPlanations (SHAP). The RF model showed the best performance (AUC = 0.8694, in internal validation; AUC = 0.8130, in external validation). Calibration indicated good agreement (Brier score: 0.036 internal validation, 0.029 external validation). SHAP identified key contributors: increased values of LDL (around 4.007 mmol/L), systolic blood pressure (around 150 mmHg), diastolic blood pressure (around 84 mmHg), total cholesterol (around 6.06 mmol/L), age (around 62 years), and decreased value of HDL (around 1.046 mmol/L). Furthermore, an online calculator was developed (https://predictingmildcognitiveimpairment.shinyapps.io/predictingMCI/). We developed and externally validated an interpretable RF-based ML model to identify prevalent MCI in patients with CAD and HBP. The tool provides transparent, data-driven insights into modifiable risk factors and facilitates early risk stratification and decision support in cardiovascular–cognitive management.