The European Sleep Apnea Database was used to identify distinguishable obstructive sleep apnoea (OSA) phenotypes and to investigate the clinical outcome during positive airway pressure (PAP) treatment.

Prospective OSA patient data were recruited from 35 sleep clinics in 21 European countries. Unsupervised cluster analysis (anthropometrics, clinical variables) was performed in a random sample (n=5000). Subsequently, all patients were assigned to the clusters using a conditional inference tree classifier. Responses to PAP treatment change in apnoea severity and Epworth sleepiness scale (ESS) were assessed in relation to baseline patient clusters and at short- and long-term follow-up.

At baseline, 20?164 patients were assigned (mean age 54.1±12.2 years, 73% male, median apnoea–hypopnoea index (AHI) 27.3 (interquartile range (IQR) 14.1–49.3) events·h^-1, and ESS 9.8±5.3) to seven distinct clusters based on anthropometrics, comorbidities and symptoms. At PAP follow-up (median 210 [IQR 134–465] days), the observed AHI reduction (n=1075) was similar, whereas the ESS response (n=3938) varied: largest reduction in cluster 3 (young healthy symptomatic males) and 6 (symptomatic males with psychiatric disorders, -5.0 and -5.1 units, respectively (all p<0.01), limited reduction in clusters 2 (obese males with systemic hypertension) and 5 (elderly multimorbid obese males, -4.2 (p<0.05) and -3.7 (p<0.001), respectively). Residual sleepiness in cluster 5 was particularly evident at long-term follow-up (p<0.05).

OSA patients can be classified into clusters based on clinically identifiable features. Importantly, these clusters may be useful for prediction of both short- and long-term responses to PAP intervention.