A Novel Big Data Analysis of Longitudinal CPAP Compliance Patterns

Introduction: We have recently reported on big data analysis of early (90 day) and mid-term (one year) compliance and efficacy data from connected CPAP devices. Longitudinal patterns of compliance may provide insights into predicting long-term compliance and into identifying actionable situations for reprogramming adherence trajectories by early interventions. We sought to characterize real world temporal patterns of compliance using novel approaches. Methods: A US based telemonitoring database (AirView, ResMed) was queried for patients starting CPAP/APAP (AirSense 10 device; ResMed) between Aug 2014 and Dec 2016. Longitudinal patterns of compliance in the first 90 days were examined using run-length analysis and frequency analysis. Summary statistics are reported as median and interquartile range (except for age). The study was exempt from IRB oversight. Results: Approximately 1.7 million patients were included in the current analysis. Patient age averaged 55.7 ± 13.5 years. Average daily usage in the first 90 days was 5.6 hours (3.7, 7.0) while usage per session was 6.2 hours (4.8, 7.3). Maximum consecutive days with usage ≥ 4 hours was 20.0 (7.0, 49.0) while the maximum consecutive days with usage > 0 hours was 43.0 (17.0, 84.0). The maximum consecutive days with usage equal to zero was 2.0 (1.0, 5.0). Frequency analysis allowed the identification of specific phenotypes such as on-off behaviour and good compliance with high or low variability (figure 1). Conclusions: Temporal patterns of compliance in large data sets require novel techniques to identify different patient phenotypes. Future research will focus on clustering patterns and linking them to clinical phenotypes as well as predicting long-term adherence and outcomes.