Physiology Guided Machine Learning Approach to Predict Short and Long Term Outcomes of Obstructive Sleep Apnea

Obstructive Sleep Apnea(OSA) is a chronic condition that affects 1 billion people worldwide. Apnea Hypopnea Index(AHI) is the clinical gold standard to measure the severity of OSA. This study highlights limitations in the apnea-hypopnea index as a predictor for obstructive sleep apnea (OSA) outcomes. Instead, a physiology-guided machine learning (ML) approach was developed using features from ventilatory, hypoxic, and arousal domains, based on polysomnography data from the Sleep Heart Health Study (SHHS). The ML model demonstrated superior predictive performance for all-cause mortality (AUROC-0.93) and daytime sleepiness (AUROC-0.81) compared to AHI. Explainable AI techniques, such as SHAP analysis, provided insights into feature importance, offering a clinically interpretable and scalable tool for OSA outcome prediction.