목적: Obstructive sleep apnea (OSA) is a prevalent disorder marked by the collapse of the upper airway during sleep. In contrast to previous studies that relied on invasive electromyography (EMG) focusing on muscles such as the genioglossus—which limited sample sizes and precluded the detailed examination of gradual changes in OSA severity—this research employs non-invasive submental EMG to investigate compensatory upper airway muscle activity across varying OSA severities using a large-scale dataset. By analyzing natural, eupneic breathing conditions, the study aims to delineate how incremental increases in muscle activation correlate with the severity of OSA. Ultimately, it seeks to validate non-invasive EMG as a promising clinical tool and biomarker for assessing OSA severity. 방법:This study analyzed submental EMG data from 7,745 polysomnography (PSG) recordings in the Korean Image-Based Sleep Study (KISS) dataset, collected from four independent hospitals. Submental EMG signals were preprocessed using a bandpass filter (10–100 Hz) and a notch filter (50 Hz) to reduce noise and eliminate line interference. Eupneic breathing periods were extracted from supine and lateral sleep positions, excluding periods with arousal, apnea and hypopnea. EMG signals were log-transformed, and their mean absolute value (MAV) was calculated. Participants were categorized into five OSA severity groups based on their apnea-hypopnea index (AHI): normal (AHI < 5), mild (5 ≤ AHI < 15), moderate (15 ≤ AHI < 30), severe (30 ≤ AHI < 60), and very severe (AHI ≥ 60). Statistical comparisons were conducted using one-way ANOVA with Bonferroni-adjusted post-hoc tests. To assess the independent association between eupneic EMG amplitude and AHI, a Generalized Linear Model (GLM) was applied, adjusting for age, sex, and BMI. 결과:A total of 7,576 submental EMG recordings were analyzed after excluding individuals with missing AHI, age, sex, or BMI data. Overall, eupneic submental EMG amplitude (expressed in log(μV)) significantly increased with OSA severity, (normal: 0.81 ± 0.22, mild: 0.86 ± 0.22, moderate: 0.92 ± 0.23, severe: 1.01 ± 0.24, very severe: 1.27 ± 0.28) with all pairwise comparisons reaching statistical significance (p < 0.001) . A Generalized Linear Model further revealed a strong association between EMG amplitude and AHI (β = 42.05, p < 0.001), and also demonstrated that BMI (β = 2.16, p < 0.001), age (β = 0.28, p < 0.001), and male sex (β = 11.12, p < 0.001) were positively correlated with AHI. When stratified by sleep stage, the N1 stage showed an increase in eupneic submental EMG amplitude from 0.97 ± 0.25 in normal subjects to 1.28 ± 0.28 in the very severe group; here, the Normal–Mild comparison was non-significant (p = 1.00), the Mild–Moderate comparison was significant (p = 0.011), and all other pairwise comparisons were significant (p < 0.001). In the N2 stage, EMG amplitude increased progressively from 0.81 ± 0.21 in the normal group to 1.25 ± 0.35 in the very severe group, with all pairwise differences significant (p < 0.001). Similarly, during the N3 stage, values ranged from 0.76 ± 0.22 (normal) to 1.14 ± 0.40 (very severe), with the Normal–Mild comparison significant at p = 0.002 and all other comparisons significant at p < 0.001. Finally, in the REM stage, EMG amplitude increased from 0.61 ± 0.08 in the normal group to 0.91 ± 0.24 in the very severe group, with all pairwise comparisons achieving statistical significance (p < 0.001). 결론:Eupneic submental EMG activity emerges as a potential biomarker for OSA severity, independent of sleep stages and demographic factors. The robust association between eupneic submental EMG activity and AHI underscores the promise of non-invasive EMG as a clinical tool for the assessment of OSA.