Real-world generalizability of sleep apnea–related polysomnographic metrics and their associations with clinical comorbidities

Abstract Rationale Advanced polysomnographic (PSG) metrics reflecting the physiological causes and consequences of sleep apnea may enable precision medicine in research settings, but their feasibility in routine clinical practice has yet to be demonstrated. Objective Assess (1) the generalizability of PSG metrics from research to clinical cohort and (2) their associations with a broad range of comorbid diseases, many of which have not been previously examined. Methods Polysomnographic metrics, including endotypes (eg, loop gain) and physiological burdens (eg, hypoxic burden), were estimated from diagnostic polysomnographs of 6427 participants at Mass General Brigham (MGB). Comorbid conditions analyzed from MGB’s medical record system included 9 representative cardiometabolic and respiratory diseases, as well as 408 prevalent diseases. Associations were assessed using modified Poisson and least absolute shrinkage and selection operator (LASSO) regression. Results The sample included 62% females, aged 52.9 ± 16.8 years, and an apnea-hypopnea index (AHI) of 21.4 ± 15.9 events/h. Associations between endotypes and demographics/obesity-related factors were consistent with prior observational studies (median difference in β = 0.03 SD). After adjusting for AHI, older age was associated with lower heart rate (−0.40 SD) and arousal burdens (−0.23 SD), while higher body mass index was associated with increased hypoxic burden (0.25 SD). Having demonstrated reasonable concordance with published data, our subsequent analysis identified distinct and clinically meaningful associations between advanced PSG metrics and comorbid conditions. Specifically, elevated loop gain, ventilatory burden, and hypoxic burden were associated with hypertension, diabetes, and renal failure; increased ventilatory instability was associated with cardiovascular disease; and reduced collapsibility and ventilatory instability were associated with chronic airway obstruction. Even after LASSO-based selection, no single PSG metric consistently predicted risk across all comorbidities; ventilatory instability showed the most associations among endotypic traits and heart rate burden among physiological burdens, underscoring the heterogeneity of obstructive sleep apnea pathophysiology. Conclusions Phenome-wide analyses of a large clinical cohort demonstrate the real-world feasibility and clinical relevance of extracting advanced PSG metrics, supporting their potential to identify personalized, mechanism-specific intervention targets for sleep apnea.