Markov models for treatment adherence in obstructive sleep apnea

Nonadherence to continuous positive airway pressure therapy (CPAP) in patients with obstructive sleep apnea (OSA) hinders effective treatment of this type of sleep disorder. Markov models can be a useful tool to analyze such behavior by revealing the underlying dynamics. In this paper, we build Markov models based on data obtained from CPAP devices and examine how such dynamics of adherence change over time. Specifically, we define CPAP usage level as a state, and estimate transition probabilities among the states from the CPAP usage data to build a Markov chain. For comparison purposes, we also classify subjects into subgroups in terms of descriptive information or clustering results, and build a Markov chain for each group. By revealing such dynamics, we hope to obtain insights for developing strategies for improving CPAP adherence.