Characterizing High-risk Respiratory Events for Predicting Cardiovascular Disease and All-cause Mortality in Obstructive Sleep Apnea

Abstract Rationale: Obstructive sleep apnea (OSA) is associated with cardiovascular disease (CVD) and mortality. OSA severity is currently measured by summarizing all respiratory events and providing a summary metric per subject without considering the within-subject heterogeneity of events. This study proposes using an unsupervised clustering method to identify similar respiratory events based on their characteristics. We hypothesize that respiratory event subtypes vary in their associations with increased risk of CVD and all-cause mortality. Methods: Participants in the Sleep Heart Health Study who underwent baseline polysomnography (PSG)(N=5633) were included. K-means clustering was used to cluster respiratory events ( ≥30% reduction in airflow lasting ≥10 seconds regardless of desaturation/arousal) of all individuals (n= 1,015,521) based on severity features (i.e., event duration, desaturation depth, area under the desaturation curve)) and physiologic responses (heart rate response [difference between the minimum and maximum heart rate]). EEG features, including the ratio of EEG power (after/before event) in beta and delta bands, were tested in a secondary analysis. The elbow method was used to select the optimal number of clusters. Subsequently, the frequency of respiratory events (i.e. apnea-hypopnea index(AHI)) within each cluster was quantified, resulting in multiple AHIs per participant. Finally, we examined the associations of cluster-based AHIs with CVD and all-cause mortality, controlling for relevant confounders. Results: The sample had a median[IQR] age=63[55, 72]years, and mean±standard deviation(SD) of follow-up time=11±3.1years). The elbow method identified three clusters(NC1=624,324, NC2=143,183, NC3=248,014), resulting in three different AHIs per subject(mean(SD) AHIC1 =18.7(12.5), AHIC2=4.2 (5.0), AHIC3=7.5(10.6), events/hour) (preliminary results)). Cox regression models revealed a significant association between AHIC3 and all-cause mortality (HR=1.06[1.01,1.12]SD;p=0.024) and between AHIC1 and incident CVD (HR=1.08[1.01,1.15]SD;p=0.02)). The other AHIs were not associated with these outcomes. Respiratory events in cluster3(24% of events) were characterized by an average event duration of 24.7[18.6;33.0]seconds, desaturation depth of 6.0[4.0;9.0], desaturation area of 2.3[1.65; 3.3]%min, and heart rate response of 11.0[8.0; 16.0]beats/min. The cluster 1(14% of events) features were characterized by an average event duration= 16.20[13.30;21.10]seconds, desaturation area of 0.15[0.03;0.25]%min). Finally, adding EEG features did not change the number of clusters or cluster characteristics. Conclusion: In this study, we showed that specific clusters of respiratory events significantly predicted increased risk of CVD and mortality in a community-based study of middle-aged and older adults. Ongoing analyses are assessing generalizability in an independent dataset.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.