0114 Dynamic Instability of Breathing During Sleep Predicts Cognition, Disease, and Mortality

Abstract Introduction Breathing during sleep fluctuates across time and sleep stage, yet clinical assessment relies almost entirely on event counts and desaturation metrics. We developed an interpretable measure of nocturnal respiratory dynamics—the Breathing Stability Index (BSI)—and evaluated whether it predicts cognition, clinical disease, and mortality independent of traditional sleep-disordered breathing indices. Methods We analyzed 15,000 overnight polysomnograms from a clinical cohort at Massachusetts General Hospital and a community-based cohort from the Osteoporotic Fractures in Men study. From thoraco-abdominal effort signals we computed BSI, a continuous dimensionless time series quantifying how rapidly and strongly breathing amplitude deviates from its local baseline, along with nightly summary features. We examined relationships between BSI, apnea–hypopnea index (AHI), hypoxic burden, and sleep stage. Predictive models for cognition and 12 EHR-derived diseases were trained using cross-validated splits with strict subject separation. All-cause mortality risk was estimated using a cross-validated Cox proportional hazards model. Results BSI increased stepwise across AHI severity strata but showed only moderate correlations with AHI and hypoxic burden, indicating complementary physiological information. Within individuals, instability declined progressively from N1 to N2 to N3 and increased again in REM sleep. Using BSI features alone, out-of-fold correlations reached r=0.24 for fluid cognition and r=0.20 for total cognition, with minimal improvement when adding AHI or hypoxic burden. For disease prediction, BSI improved discrimination over demographics and over models using AHI or hypoxic burden alone (ROC-AUC 0.56–0.65 across conditions). Each 1-SD increase in instability predicted higher mortality in MGH (HR 1.67, 1.49–1.88) and MrOS (HR 1.11, 1.06–1.16). Age adjustment reduced but did not remove the effect in MGH (HR 1.25, 1.12–1.40) and eliminated it in MrOS (HR 1.02, 0.97–1.07). Conclusion Nocturnal breathing stability reflects a dynamic, clinically relevant dimension of sleep-disordered breathing. It complements event-based and desaturation metrics and provides independent information related to cognition, multimorbidity, and survival. Incorporating breathing-stability measures may improve individualized risk assessment and inform interventional studies targeting respiratory-control dynamics. Support (if any) NIH grants R01AG073410 and R01HL161253.