A Multimodal Machine Learning Model for Framingham Risk Score–Based Cardiovascular Risk Stratification in Patients With Obstructive Sleep Apnea

ABSTRACT Machine learning provides a powerful tool to capture complex, nonlinear patterns in biomedical data and enhance precision risk stratification. In this study, we applied a multilayer perceptron (MLP) framework to classify Framingham Risk Score–based cardiovascular risk categories in patients with obstructive sleep apnea (OSA). By integrating polysomnographic, clinical, and haematological data, the model demonstrated robust discriminative performance (AUROC = 0.822). The findings suggest that OSA‐related phenotypes may provide additional information relevant to cardiovascular risk stratification beyond conventional risk factors, highlighting the added value of incorporating sleep‐specific markers into cardiovascular risk assessment. This work illustrates the potential of machine learning to deliver more comprehensive cardiovascular risk stratification in OSA populations. Future studies with larger, multi‐center cohorts, expanded biomarker panels, and explainable artificial intelligence approaches are needed to further refine predictive performance, improve interpretability, and support future validation and clinical application.