Data mining for characterizing obstructive sleep apnea treatment adherence trends

Continuous positive airway pressure therapy (CPAP) is known to be one of the most effective treatments for obstructive sleep apnea (OSA). Non-adherence to CPAP, however, is often observed among CPAP-treated patients, and it contributes to significant health-related issues, leaving OSA untreated or sub-optimally treated. In this paper, we explore usage data obtained from CPAP devices to identify patterns in adherence trends. For this, we employ a variety of sequential data mining techniques, such as the sequence classification approach to identify any usage patterns from the data, and a sequence clustering approach to identify any sub-group trends. Furthermore, we build and suggest a classifier to predict a patient's adherence to CPAP during the pre-intervention stage. Through the analyses, we reveal that early intervention is crucial in order to prevent nonadherence to CPAP and effectively treat OSA. In particular, we observed that monitoring patients during their first week of treatment is sufficient to identify their CPAP usage patterns and to provide adjusted and tailored intervention according to the prediction. Characterizing treatment-adherence trend patterns will enable effective early preventive interventions to be developed in order to improve CPAP treatment adherence.