Objective:

To develop and assess a deep learning model for predicting coronary heart disease (CHD) risk in hypertensive patients using 24-hour ambulatory blood pressure monitoring (ABPM) data.

Approach:

• Study Design: A single-center retrospective cohort study involving 1,026 patients, with 718 for model development and 308 for internal validation.
• Model Development: A deep neural network with three hidden layers was created and compared to eight conventional machine learning algorithms.
• Feature Selection: Thirty-two clinical variables were evaluated, with a two-step feature selection process using the Boruta algorithm and LASSO regression.
• Performance Evaluation: Model performance was assessed based on discrimination (AUC), calibration (Brier score), and clinical utility (decision curve analysis).
• Interpretability: SHAP values were used to evaluate model interpretability.

Key Findings:

• The deep neural network model achieved an AUC of 0.822 in the training cohort and 0.796 in the validation cohort.
• The model had the lowest Brier score of 0.172, indicating superior calibration.
• Nine predictors were identified as significant: diabetes mellitus, mean systolic blood pressure, time in target range of systolic blood pressure, left atrial diameter, left ventricular end-systolic diameter, left ventricular ejection fraction, use of antihypertensive medications, calcium channel blockers, and β-blockers.

Interpretation:

The deep neural network model enables identification of high-risk CHD patients with hypertension using routinely available clinical variables.

Limitations:

• The study is limited to a single center, which may affect generalizability.
• The model requires prospective multicenter external validation.

Conclusion:

The developed model requires further validation to assess its effectiveness in risk stratification for hypertensive patients at risk for CHD.