Using Machine Learning to Predict PAP Adherence After Therapy Initiation
2019 Strategic Research Grant
Dennis Hwang, MD
Kaiser Foundation Research Institute
Key Project Outcomes
This grant facilitated the creation of an extensive multi-modal dataset from a large, integrated healthcare system, enhancing the development of machine learning models for predicting future PAP adherence. The dataset includes longitudinal PAP metrics, patient-reported questionnaires, diagnostic sleep study data, and electronic health records (EHR). We created models to forecast PAP adherence at various post-initiation intervals. Our findings indicate that past PAP usage is the most reliable predictor of future use, with other data types yielding modest results. Algorithms using just the initial 7 days of PAP data accurately predicted 3-month and 1-year adherence with 86% and 76% accuracy, respectively. Additionally, we developed models using 30-day rolling PAP metrics to anticipate adherence loss within the ensuing two weeks. These prediction models, designed for integration into clinical practice, can serve as decision support tools to optimize treatment decisions and enhance PAP monitoring by enabling proactive interventions for patients at risk of non-adherence. These advances represent potential progression towards enhancing personalized medicine. Future work will focus on clinical trials to assess the practical application of these models in clinical settings. Moreover, the data integration framework established in this study paves the way for further machine learning and epidemiological research.
Abstracts
American Thoracic Society International Conference
Impact of Obstructive Sleep Apnea and Positive Airway Pressure Therapy on COVID-19 Outcomes
SLEEP
Deep Learning Classification of Future PAP Adherence based on CMS and other Adherence Criteria
Deep Learning to Predict PAP Adherence in Obstructive Sleep Apnea