Improving sleep monitoring in athletes: a comparative evaluation of four automated sleep scoring algorithms on 204...

Objective.Automated sleep scoring offers the potential for fast, precise, and cost-effective analysis. This study evaluated the accuracy, robustness, and clinical agreement of four algorithms against manual polysomnography (PSG) in athletes, and assessed how signal degradation and individual physiological traits affect automated scoring in ambulatory settings.Approach.Data from 75 healthy athletes (91.1% male, mean age 26.3 ± 5.4 years, range [18-41], 204 PSG recordings) were retrospectively collected from studies conducted in our laboratory. Manual scoring was compared to four automated sleep staging algorithms: Luna, U-Sleep, Yet Another Spindle Algorithm (YASA), and Greifswald Sleep Stage Classifier (GSSC). Performance was evaluated on accuracy (per stage, macro and weightedF1), sensitivity (confusion matrices), robustness (signal quality:R²; individual differences: beta coefficients) and clinical relevance (prediction of total sleep time (TST), sleep onset latency (SOL), wake after sleep onset (WASO) and sleep efficiency (SE)).Main results.All algorithms demonstrated strong agreement with manual scoring, with macroF1 scores of 0.76 (GSSC), 0.71 (U-Sleep), 0.70 (YASA), and 0.69 (Luna). GSSC was the most accurate and robust across sport, physiological, behavioral, and anthropometric variables. It showed minimal bias and narrow limits of agreement for TST, WASO and SE. GSSC was also the most resilient to signal degradation (marginalR²= 0.05), compared to YASA and Luna (0.06) and U-Sleep (0.24).Significance.Automated sleep staging algorithms demonstrated accuracy comparable to typical inter-rater agreement, supporting broader use in athletes. These algorithms could make PSG more accessible for both monitoring and research, providing deeper insights into sleep and its influence on athletic performance.