0464 Deep Learning to Predict PAP Adherence in Obstructive Sleep Apnea

Abstract Introduction Machine Learning (ML) algorithms to predict Positive Airway Pressure (PAP) adherence may support personalized clinical management. Models were developed at various time-points after PAP initiation and in moving time windows. Methods Deep neural network (DNN) models trained utilizing daily data (Kaiser Permanente, Southern California). The DNN was evaluated with 10-fold cross-validation on N=21,397 patients. Algorithms included (a) 1 2 which utilized early usage 90-days 1-year respectively, (b) Model 3 14 30-day windows subsequent usage. Regression analyses compared ML Naïve (i.e., future use equals previous use) predictions versus Actual adherence. Results predicted “% days without usage” for first based 7, 14, 21, 30-days of input (90-day window) 30, 60, 90, 180-days input. superior predicting [R Actuals different days— 0.495-vs-0.193; 0.660-vs-0.465; 0.748-vs-0.607; 0.828-vs-0.735 0.362-vs-0.104; 0.463-vs- 0.247; 0.513-vs-0.339; 0.680-vs-0.547 1-year; all p< 0.05]. “hours/night” use—ML did not outperform the prediction similar R ; however, when < hours/night, nearly patients had “no significant (comparatively, naïve model no differentiating threshold this outcome.) predictive accuracy using or vs. used ≥4 hours” 0.687, 0.701, 0.699 14- 0.582, 0.702, 0.77 input; 0.05.] Conclusion can adherence, potentially supporting treatment decisions pre-emptive interventions upcoming non-adherence is predicted. Support (if any) AASM Foundation SRA205-SR-19