STUDY OBJECTIVES: Obstructive sleep apnea is a prevalent condition effectively treated by continuous positive airway pressure (CPAP) therapy. CPAP adherence data, routinely gathered in clinical practice, include detailed information regarding both duration and timing of use. The purpose of the present study was to develop a systematic way to measure the diurnal pattern of CPAP adherence data and to see if distinct patterns exist in a clinical cohort. METHODS: Machine learning techniques were employed to analyze CPAP adherence data. A cohort of 200 unselected patients was assessed and a cluster analysis was subsequently performed. Application of this methodology to 17 patients with different visually noted patterns was carried out to further assess performance. RESULTS: Each 30-day period of CPAP use for each patient was characterized by 4 variables describing the time of day of initiation and discontinuation of CPAP use, as well as the consistency of use during those times. Further analysis identified 6 distinct clusters, reflecting different timing and adherence patterns. Specifically, clusters with relatively normal timing vs delayed timing were identified. Finally, application of this methodology showed generally good performance with limitations in the ability to characterize shift worker and non-24 rhythms. CONCLUSIONS: This study demonstrates a methodology for analysis of diurnal patterns from CPAP adherence data. Furthermore, distinct timing and adherence patterns are demonstrated. The potential impact of these patterns on the beneficial effects of CPAP requires elucidation. CITATION: Scharf MT, Androulakis IP. Novel assessment of CPAP adherence data reveals distinct diurnal patterns.