Data-driven phenotypic clustering of sleep-disordered breathing and long-term cardiovascular risk

Sleep-disordered breathing (SDB) has heterogeneous cardiovascular risks that are not fully captured by the apnea–hypopnea index (AHI). We aimed to determine whether data-driven sleep–cardiometabolic phenotypes identify long-term cardiovascular risk and support phenotype-based risk stratification. We analyzed 5682 adults from the Sleep Heart Health Study baseline cohort enrolled in 1995-1998 (mean [SD] age, 63.2 [11.2] years; 52.7% female). Fifteen prespecified variables were used for k-means clustering to define SDB phenotypes. Participants with adjudicated follow-up data (n = 4909) underwent time-to-event analyses with a primary composite endpoint of various cardiovascular events. Three phenotypes were identified: Cluster 1 (aging–hypertensive OSA), Cluster 2 (healthy sleep-CVD profile), and Cluster 3 (metabolic–hypoxic OSA). Clusters 1 and 3 showed higher risks of the composite endpoint than Cluster 2 (hazard ratios [HRs], 4.06 [95% confidence interval (CI), 3.50–4.71] and 3.62 [95% CI, 2.94–4.44], respectively) and consistently elevated cardiovascular death, myocardial infarction, and stroke risks. Heart failure risk was pronounced for Clusters 1 and 3 (HRs, 5.46 [95% CI, 4.41–6.76] and 5.55 [95% CI, 4.23–7.29], respectively) compared with Cluster 2. The multinomial classifier demonstrated high accuracy and discrimination in the internal test set. The phenotype-based risk score exhibited a clear stepwise gradient in predicted and observed cardiovascular risk and retained prognostic value in the external cohort, outperforming the AHI with acceptable calibration. Multidimensional phenotyping identifies SDB subgroups with divergent cardiovascular risk. A phenotype-based score using routine variables may improve risk stratification beyond event frequency alone. Sleep-disordered breathing (sleep apnea) is common and can increase the risk of heart disease and stroke. It is usually graded by the apnea–hypopnea index (AHI), but AHI does not fully capture risk. We analyzed 5682 adults in a cohort with sleep testing and long follow-up. A computer method grouped people into three profiles: older with high blood pressure, low-risk healthy profile, and obesity with low oxygen at night. Compared with the group at lower risk, the older and obesity profiles had about four times more cardiovascular events, especially heart failure. A score using routine measures predicted risk better than AHI and also worked in an external cohort. This approach may help identify people at higher-risk earlier and support more personalized prevention and treatment. Huang, Xiang et al. identify sleep-disordered breathing phenotypes using clinical and sleep data from a community cohort. The phenotypes show divergent cardiovascular risks, and a phenotype-based score improves risk stratification beyond the apnea–hypopnea index.