1262 Personalizing PAP Comfort Through AI-enabled Modeling: Retrospective Insights and Future Directions

Abstract Introduction Personalizing positive airway pressure (PAP) comfort through machine-learning (ML)–derived insights may help improve early therapy comfort and long-term adherence. Personalized Therapy Comfort Settings (PTCS), FDA-cleared and marketed as Smart Comfort, is an AI-enabled medical device that provides personalized PAP comfort setting recommendations to support patients when using their compatible therapy device. We conducted retrospective analyses to evaluate whether model-guided comfort setting recommendations, developed using data from 1.4 million patients, were associated with improved PAP device usage and patient-reported early therapy experience. Methods Retrospective analyses using de-identified Resmed PAP therapy datasets were conducted to evaluate whether comfort settings aligned with model-recommended configurations (PTCS-Matched) were associated with improved outcomes when compared with Default or Other (non-PTCS/non-Default) settings. Propensity score matching (SMD < 0.10) controlled for imbalances between comparison groups. Outcomes included nightly usage, days used, residual AHI (rAHI), mask leak, and long-term discontinuation at 90 days and 1 year. Day-3 patient-reported outcome measures (PROMs) assessing early therapy experience were evaluated in patients who completed them. Results Datasets of over 900,000 AirSense11 device users between October 2022 – May 2025 were analyzed. Across retrospective comparisons, patients whose settings aligned with model recommendations (PTCS-Matched) showed improved PAP device usage when compared with Default configurations. For AS11 devices, PTCS-matched users demonstrated +20 minutes/night at 90 days, +15 minutes/night at 1 year, and +21 usage days at 1 year (all p< 0.001). Residual AHI remained well within clinical limits (< 5), mask leak differences were minimal, and AS11 discontinuation was slightly lower (–2.8%, p< 0.001). Findings were directionally consistent when compared against Other settings. Day-3 PROMs showed PTCS-Matched users were less likely to find therapy challenging than those on Default comfort settings (19.8% vs 26.5%; –5.9 pp; p< 0.001) and more likely to rate therapy as great (+2.7 pp; p=0.003). Combined pressure/breath-in discomfort was modestly lower (–2.2 pp; p=0.012). Conclusion These large-scale real-world analyses show that individualizing comfort settings from the start of PAP therapy, aligning with ML-derived recommendations, is associated with higher PAP usage and better early therapy experience without compromising safety, supporting personalized PAP comfort settings at scale. Support (if any) Funded by Resmed.