IntroductionObstructive sleep apnea (OSA) is heterogeneous and complex, but its severity is still based on the apnea–hypoapnea index (AHI). The present study explores using cluster analysis (CA), the additional information provided from routine polysomnography (PSG) to optimize OSA categorization. MethodsCross-sectional study of OSA subjects diagnosed by PSG in a tertiary hospital sleep unit during 2016–2020. PSG, demographical, clinical variables, and comorbidities were recorded. Phenotypes were constructed from PSG variables using CA. Results are shown as median (interquartile range). Results981 subjects were studied: 41% females, age 56 years (45–66), overall AHI 23events/h (13–42) and body mass index (BMI) 30kg/m2 (27–34). Three PSG clusters were identified: Cluster 1: “Supine and obstructive apnea predominance” (433 patients, 44%). Cluster 2: “Central, REM and shorter-hypopnea predominance” (374 patients, 38%). Cluster 3: “Severe hypoxemic burden and higher wake after sleep onset” (174 patients, 18%). Based on classical OSA severity classification, subjects are distributed among the PSG clusters as severe OSA patients (AHI≥30events/h): 46% in cluster 1, 17% in cluster 2 and 36% in cluster 3; moderate OSA (15≤AHI<30events/h): 57% in cluster 1, 34% in cluster 2 and 9% in cluster 3; mild OSA (5≤AHI<15events/h): 28% in cluster 1, 68% in cluster 2 and 4% in cluster 3. ConclusionsThe CA identifies three specific PSG phenotypes that do not completely agree with classical OSA severity classification. This emphasized that using a simplistic AHI approach, the OSA severity is assessed by an incorrect or incomplete analysis of the heterogeneity of the disorder.