Comprehensive evaluation of odds ratio product and spectral slope as continuous sleep depth measures: advancing sleep staging and clinical applications.

STUDY OBJECTIVES: This study aims to validate two innovative continuous sleep depth measures, odds ratio product (ORP) and spectral slope (SS), using large-scale datasets, including MESA, SHHS, CHAT, and WSC, as alternatives to the traditional discrete gold-standard hypnogram.

METHODS: The relationship between ORP, SS, and the gold-standard hypnogram was evaluated through Pearson correlation analysis. Feature distributions across sleep stages and transitions, the Arousability Index, and sleep cycles were systematically analyzed. To evaluate their potential for automated applications, a pilot bi-LSTM framework was employed, with min-max scaling applied to improve neural network training. The ability of ORP and SS to characterize sleep architecture was also explored through their capacity to predict the Apnea-Hypopnea Index (AHI), a key indicator of sleep quality.

RESULTS: Both ORP and SS showed strong correlations with the gold-standard hypnogram. ORP excelled in binary Awake/Asleep classification, whereas SS demonstrated superior resolution of transitional stages and alignment with traditional sleep cycles. SS also outperformed ORP in automated sleep staging for unseen datasets, exhibiting higher generalizability and robustness. In addition, SS proved more effective than ORP in predicting AHI and showed significant differences across obstructive sleep apnea severity levels.

CONCLUSIONS: ORP and SS offer complementary strengths for continuous sleep depth assessment. ORP provides precise binary classification suitable for clinical diagnostics, while SS's robust generalizability and alignment with sleep dynamics present transformative potential for automated applications.