A Deterministic PPG-Based Algorithm for Obstructive Sleep Apnea Detection

Obstructive sleep apnea (OSA) remains substantially underdiagnosed, largely due to the logistical constraints of polysomnography (PSG), the current clinical reference standard. Although consumer smartwatches provide a scalable alternative for home-based monitoring, many existing approaches rely on deep learning models that are difficult to interpret and may be challenging to deploy in resource-constrained settings. In this study, we propose a deterministic white-box pipeline for OSA screening based on heart rate (HR) and peripheral oxygen saturation $({S p O}_{{2}})$ signals. The method is explicitly designed as a high-sensitivity screening tool and is grounded in clinically relevant physiological patterns, including cyclic variation of heart rate (CVHR) and dynamic ${S p O}_{2}$ desaturation behavior, together with an autonomic thresholding mechanism. Evaluation on 29 subjects showed a binary classification accuracy of 89.7%, with 100.0% sensitivity and 78.6% specificity, compared to medical doctor PSG analysis. All OSA cases were correctly identified in this cohort, supporting the use of the proposed method as a conservative first-line screening tool. Agreement with the reference apnea-hypopnea index (AHI) classification was strong $({I C C}={0. 9 6})$, and severity misclassifications were limited to adjacent categories: mild overestimations in healthy subjects and minor underestimations in lower-moderate cases. These results show that a physiologically grounded, computationally efficient approach can reliably identify at-risk individuals and prioritize them for clinical PSG evaluation.