0602 Personalized CPAP Adherence with SleepWell24: Early Outcomes of an Adaptive Smartphone Application Using Control Systems Engineering Approaches

Abstract Introduction Adherence to continuous positive airway pressure (CPAP) therapy is difficult for many patients with obstructive sleep apnea. Clinical outcomes are improved with CPAP adherence ≥6 hr/night, yet scalable approaches to personalized support are limited. We developed a theory-based, clinic experience–driven conceptual dynamical model for optimizing CPAP use and applied it to a ‘just-in-time’ adaptive version of SleepWell24, a smartphone application integrating near–real-time wearable data, CPAP use, and patient-reported symptoms and CPAP-related problems with evidence-based behavior change interventions. In this feasibility optimization trial, we examined patient acceptability and initial outcomes of the adaptive SleepWell24 platform, and guided by the conceptual model, evaluated a control systems engineering approach to demonstrate, in simulation, improved CPAP adherence. Methods Aim 1: Naïve CPAP users (N=10; M age=61.7, 60% female) participated in this 60-day ongoing trial. Patients were encouraged to interact daily with SleepWell24 to report symptoms, receive personalized troubleshooting recommendations, set goals, and receive feedback on CPAP use (via Wi-Fi-enabled smart plug assessing CPAP ‘on/off state’) and sleep/activity (Fitbit). We assessed SleepWell24 use, post-intervention acceptability, and consistent CPAP use defined as ≥6hr, for ≥6nights/week for ≥2 consecutive weeks. Aim 2: The conceptual model of CPAP adherence produced a dynamical system, informed by Model Predictive Control (MPC) as a decision-making algorithm to optimize CPAP use to ≥6hr/night by tailoring actionable behavioral recommendations using patient data integrated in real-time. Results Aim 1: Across the 60-day trial, average daily SleepWell24 engagement was 72.7%. Median CPAP adherence of ≥6hr/night was 65.8%. All participants (100%) rated SleepWell24 as moderately-to-totally acceptable and would recommend it; perceived helpfulness averaged 7.0 (of 10). Eighty percent achieved consistent CPAP use. Aim 2: In simulation, the MPC framework managed CPAP-associated symptoms (e.g., nasal, anxiety, leak) and adjusted CPAP use goals, improving predicted CPAP adherence and sleep duration and quality. Conclusion SleepWell24 was engaging, acceptable, and helpful, with most users reaching optimal CPAP use. We will next estimate dynamical models using data from trial patients, a necessary step towards full automation of SleepWell24. Support (if any) AASMF #285-SR-22