Beyond the apnea-hypopnea index: clinical phenotypic heterogeneity of obstructive sleep apnea in an Indian tertiary referral cohort using level III polysomnography

Obstructive sleep apnea (OSA) is usually classified by a single severity index, the Apnea-Hypopnea Index (AHI), which does not capture the clinical heterogeneity described in other populations. To characterize clinically distinct OSA phenotypes by body habitus, symptom–severity relationship, and by respiratory event type in an Indian tertiary referral cohort, using level III polysomnography. Retrospective analysis of 309 patients with OSA (AHI≥5) from consecutive level III polysomnography studies at a tertiary Indian center. Phenotypes were defined as follows: 1) lean [body mass index (BMI)<25 kg/m²] versus non-lean OSA; 2) symptomatic vs. asymptomatic disease, using an Epworth Sleepiness Scale (ESS) score of 10 as a threshold, crossed with AHI severity; 3) apnea-predominant vs. hypopnea-predominant disease, from the respective event indices. An exploratory unsupervised cluster analysis on physiological variables was performed as a comparator. A lean phenotype was present in 12.3% of patients (95% confidence interval 9.1-16.5) and had lower AHI, less hypoxemia, and fewer daytime symptoms than non-lean OSA, with less moderate-to-severe disease (40.5% vs. 68.4%, p=0.002). Symptom burden was poorly matched to severity: 52.7% of severe OSA patients were not sleepy (ESS<10), while 21.5% of mild OSA patients were. Sleepy patients had a greater hypoxemic burden but similar body habitus. Apnea-predominant disease (54.3% of patients with usable indices) occurred in younger, leaner patients but with a higher AHI than hypopnea-predominant disease. Both the lean-phenotype association with less severe disease and the apnea-predominant AHI difference persisted after adjustment for age, sex, and BMI. STOP-Bang score did not reliably discriminate moderate-to-severe disease in this already high-risk referral population (AUC 0.56), and hypertension and diabetes prevalence did not differ between lean and non-lean phenotypes. ODI graded on AHI's own thresholds agreed with the AHI category in 67.4% of patients (quadratic-weighted κ 0.82) and identified moderate-to-severe disease with 89% sensitivity and 92% specificity. Unsupervised clustering on AHI-correlated physiological variables did not identify phenotypes independent of severity. OSA in this referral cohort separates into body-habitus, symptom-discordance and event-type phenotypes that are not interchangeable with AHI severity alone. Recognizing these patterns may help identify lean patients, underrecognized asymptomatic severe disease, and mechanistically distinct event patterns.