Future diagnostics: Lessons from Sleep Revolution - Timo Leppänen

Background

Current diagnosis of obstructive sleep apnea (OSA) relies heavily on the apnea–hypopnea index (AHI), an outdated metric that fails to adequately reflect patient-reported outcome measures or OSA-related comorbidities. Polysomnography (PSG) captures multiple physiological signals, which are traditionally reviewed manually by trained sleep technologists or specialists. This manual approach is costly, time-consuming, and overlooks the rich information contained in the recordings. This hinders the estimation of the true severity of OSA, limiting both diagnostic accuracy and treatment optimization.

Objective

One central aim of the Sleep Revolution project was to modernize diagnostic strategies for OSA. The project sought to create automated analysis tools for automatic PSG scoring and detect the most informative signals to be measured to simplify the measurement setup.

Methods

Our efforts centered on developing algorithms for PSG data that can automatically identify sleep stages, microstructural sleep features, respiratory disturbances, oxygen desaturation events, and arousals, while still allowing expert oversight when necessary.

Results

Different physiological signals—and signal combinations—provide different types of information. Therefore, the intended purpose of the assessment (e.g., screening vs. full diagnostic evaluation) should guide the selection of signals to record and analyze, while clinical symptoms and individual patient characteristics must also be considered. Wearable devices may be adequate for screening in many cases, whereas more comprehensive recordings, including electroencephalography and electrooculography, offer deeper insight into sleep microstructure and disease manifestation.

Conclusion

Human-in-the-loop approaches, where sleep experts oversee and take responsibility for the analysis, remain essential, and measurement setups should be tailored for the specific purpose of the assessment. After thorough clinical validation, medical-grade oximetry-based devices and wearables may be suitable for OSA screening and monitoring treatment adherence. However, when a detailed evaluation of sleep architecture is required, at least one electroencephalography channel should be included. Finally, similarly to the medical devices, automated algorithms must undergo rigorous validation in the appropriate populations and for their intended clinical use.