목적: Although interest in the early diagnosis of Obstructive Sleep Apnea (OSA) using wearable devices is increasing, there is a lack of independent validation comparing the estimated Apnea-Hypopnea Index (AHI) of smart rings with Polysomnography (PSG) results in real world patients. This study aimed to prospectively evaluate the accuracy and real-world validity of the estimated AHI by a smart ring against the AHI from PSG. 방법:Patients with suspected OSA were prospectively recruited to undergo Type I PSG while simultaneously wearing a smart ring. The PSG recordings were manually scored by a physician to determine the AHI (pAHI), which served as the reference standard for analyzing the estimated AHI by smart ring (eAHI). The association and agreement between the two values were assessed using Pearson correlation analysis and the Intraclass Correlation Coefficient (ICC). Factors influencing the error between eAHI and pAHI were analyzed using simple linear regression, and predictive performance for moderate-to-severe (pAHI ≥15) and severe (pAHI ≥30) OSA was evaluated using Receiver Operating Characteristic (ROC) analysis. 결과:Data from a total of 85 participants (53 males, 62.4%; mean age 45.8 ± 11.4 years; mean BMI 25.4 ± 4.3 kg/m²) were analyzed. The mean pAHI was 22.9 ± 19.2, while the mean eAHI was 12.1 ± 16.2. The eAHI showed a general tendency to underestimate compared to the PSG AHI (mean error -10.8 ± 9.4). Pearson correlation analysis indicated a strong positive correlation between eAHI and pAHI (rho=0.874, p<0.001). The ICC was 0.727, indicating moderate-to-good agreement between the eAHI and pAHI. Simple linear regression identified BMI (p=0.003), neck circumference (B=-0.790, p=0.002), abdominal circumference (B=-0.219, p=0.004), and sex (B=4.647, p=0.026) as variables significantly affecting the error between eAHI and pAHI. In the ROC analysis, the area under the curve (AUC) for predicting moderate-to-severe OSA (pAHI ≥15) was 0.923 (p<0.001). At the optimal eAHI threshold of 5.0, sensitivity was 83.3%, specificity was 100.0%, and accuracy was 90.6%. For predicting severe OSA (pAHI ≥30), the AUC was 0.967 (p<0.001). At the optimal threshold of 14.7, sensitivity was 88.5%, specificity was 96.6%, and accuracy was 94.1%. 결론:The eAHI estimated by the smart ring demonstrated a strong correlation (rho=0.874) and satisfactory agreement (ICC = 0.727) with the physician-scored PSG AHI. However, the eAHI tended to underestimate the pAHI, resulting in optimal thresholds being set lower than OSA severity criteria. ROC analysis showed excellent diagnostic performance with an AUC of 0.923 for moderate-to-severe OSA and 0.967 for severe OSA. This study confirms the clinical potential and real- world validity of the smart ring-based OSA detection algorithm.