Unsupervised Phenotyping to Stratify Surgical Outcomes in Obstructive Sleep Apnea.

Abstract

Surgical outcomes after upper-airway surgery for obstructive sleep apnea vary widely because of clinical heterogeneity. This study aimed to identify distinct obstructive sleep apnea phenotypes using unsupervised clustering and evaluate their association with surgical outcomes. Retrospective prognostic cohort study. Single-centre study. Two hundred fourteen adults who underwent upper-airway surgery for obstructive sleep apnea were analysed across 50 clinical and polysomnographic variables. The Leiden clustering algorithm was used to identify patient subgroups. Baseline characteristics and surgical outcomes, including changes in the apnea-hypopnea index and oxygen saturation, were compared across the identified clusters. The analysis revealed 3 distinct patient phenotypes. Cluster 0 (n = 91) comprised older patients with lower BMI and a high frequency of respiratory effort-related arousal. Cluster 1 (n = 73) was characterised by younger patients with higher BMI and severe apnea-dominant obstructive sleep apnea. Cluster 2 (n = 50) was defined by significant anatomical obstruction and hypopnea-dominant breathing patterns. The reductions in the apnea-hypopnea index and oxygen desaturation index were significantly less pronounced in Cluster 0 than in Clusters 1 and 2 (P < .01). Conversely, Cluster 1 showed the greatest improvement, with a significantly larger reduction in the apnea index and a greater increase in the lowest oxygen saturation compared to the other 2 clusters (P < .01). Unsupervised clustering of multidimensional data revealed obstructive sleep apnea phenotypes with distinct surgical prognoses. Integrating this approach into preoperative evaluations may help inform surgical counseling and shared decision-making.