Purpose <p>REM-related obstructive sleep apnea (OSA) presents a paradox where a low apnea-hypopnea index (AHI) may mask significant disease. We compared polysomnography, comorbidities, and oxygenation metrics across OSA phenotypes to determine if AHI-based severity classifications underestimate the nocturnal hypoxemia in REM-related OSA.</p> Methods <p>Retrospectively, 378 OSA patients were classified as REM-related (AHI_REM ≥2× AHI_NREM; AHI_NREM &lt;5·h⁻¹, <i>n</i> = 30), REM-aggravated (AHI_REM ≥2× AHI_NREM; AHI_NREM ≥5·h⁻¹, <i>n</i> = 63), or stage-independent (<i>n</i> = 285). We analyzed sleep architecture, AHI values, oxygen desaturation index (ODI; ≥3% dips·h⁻¹), and T90 (% sleep time with SpO₂&lt;90%). Multivariable linear regression was used to assess the independent contribution of phenotype to cumulative nocturnal hypoxemia after adjusting for key confounders.</p> Results <p>The REM-related group had the lowest AHI_Total (6.3 ± 1.6) but a comorbidity profile indistinguishable from the stage-independent group (AHI_Total=32.3 ± 25.4), where 43.5% had severe OSA. Mean nocturnal SpO₂ was comparable across phenotypes (<i>p</i> = 0.15). After adjustment, the phenotype independently predicted T90. Compared with the REM-related group, T90 was significantly lower in both the REM-aggravated (β=–15.6; <i>p</i> = 0.010) and stage-independent (β=–13.1; <i>p</i> = 0.016) groups. In contrast, phenotype was not associated with the ODI; this metric was influenced primarily by body weight (<i>p</i> = 0.004) and AHI_Total (<i>p</i> &lt; 0.001).</p> Conclusion <p>REM-related OSA is a distinct phenotype where a low AHI_Total underestimates risk, obscuring significant comorbidities and substantial cumulative hypoxemia. Time-based metrics such as T90, rather than event-based indices such as ODI, are essential for capturing physiological strain in this phenotype. These findings support moving beyond AHI-centric assessments toward a phenotype-specific diagnostic framework to improve risk stratification in OSA.</p>