Machine Learning Approaches to Stroke Risk Prediction in Atrial Fibrillation patients with Obstructive Sleep Apnea

Obstructive sleep apnea (OSA) is believed to increase the risk of stroke in patients with atrial fibrillation (AF), though evidence remains inconsistent. This retrospective study evaluates OSA in AF and non-AF patients to assess stroke risk using large-scale data analytics and machine learning. Two million unique patients’ health records were extracted from the Cerner Corporation’s HealthFacts® data warehouse using the International Classification of Diseases. The CHA2DS2-VASc scoring tool was used to assess stroke risk in all patients. Three-factor ANOVA assessed statistical significance and effect size, while logistic regression, decision trees, and random forest models evaluated contributing risk factors. AF patients with OSA have a statistically significant (p < 0.0001) and uniformly higher stroke risk score. However, OSA was not an independent risk factor and primarily amplified existing risks, particularly hypertension and age. These findings demonstrate the value of EHR-driven analytics in efficiently generating clinically relevant evidence.