AUTHOR=May Anna M. , Dalton Jarrod E. 

TITLE=Comparison of machine learning approaches for positive airway pressure adherence prediction in a veteran cohort

JOURNAL=Frontiers in Sleep

VOLUME=Volume 3 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/sleep/articles/10.3389/frsle.2024.1278086

DOI=10.3389/frsle.2024.1278086

ISSN=2813-2890

ABSTRACT=Background: Adherence to positive airway pressure (PAP) therapy for sleep apnea is suboptimal, particularly in the Veteran population. Accurately identifying those best suited for other therapy or additional interventions may improve adherence. We evaluated various machine learning algorithms to predict 90-day adherence.The cohort of VA Northeast Ohio healthcare system patients who were issued a positive airway pressure machine (1/1/10-6/30/2015) had demographics, comorbidities, and medications at the time of polysomnography obtained from the electronic health record. Data was split 60:20:20 into training, calibration, and validation datasets with no use of validation data for model development. We constructed models for the first 90-day adherence period (% nights ≥4 hours use) using the following algorithms: linear regression, lasso, elastic net, ridge regression, gradient boosted machines, support vector machine regression, Bayes-based models, and neural nets. Prediction performance was evaluated in the validation dataset using root mean square error (RMSE).The 5047 participants were 38.3 ± 11.9 years old and 96.1% male with 36.8% having coronary artery disease and 52.6% with depression. Median adherence was 36.7% [IQR: 0%, 86.7%]. Gradient boosted machine was superior to other machine learning techniques (RMSE 37.2). However, performance was similar and not clinically useful for all models without 30-day data. 30day PAP data and using raw diagnoses and medications (vs. grouping by type) improved RMSE to 24.27.Comparing multiple prediction algorithms using electronic medical record information, we found that none have clinically meaningful performance. Better adherence predictive measures may offer opportunities for personalized tailoring of interventions.