Homecare interventions as a Service model for Obstructive sleep Apnea: Delivering personalised phone call using patient profiling and adherence predictions

Abstract

Background and ObjectiveObstructive Sleep Apnea (OSA) is a sleep disorder that leads to different pathologies like depression and cardiovascular problems. The first-line medical treatment for OSA is Continuous Positive Airway Pressure (CPAP) therapy. However, this therapy has the lowest adherence level when compared to other homecare therapies. Consequently, the main objective of this paper is to increase this adherence level with methods that can be replicated in a large number of patients. MethodsThe Homecare Intervention as a Service model can build, verify, and deliver per-sonalised home care interventions. With the Homecare Intervention as a Service model, we build and provide on-demand personalised interventions according to the patient’s needs. The 2 core components of this model are patient clustering and CPAP adherence predictions. To define the patient profiles and predict the adherence level, we apply the K-means and the Logistic Regression algorithm respectively. To support these algorithms, we use the CPAP monitoring data and qualitative data on the patients. ResultsWe demonstrate that there are 3 patient profiles (non-adherent, attempter, and adherent). We draw a comparison with multiple machine learning algorithms to predict CPAP adherence at 30, 60 and 90 days. In this case, the Logistic Regression gives the best results with a f1-score of 0.84 for30 days, 0.79 for 60 days and 0.76 for 90 days. These newly build profiles were to be used to deliver personalised phone call interventions. The phone call intervention shows an increase in adherence by 1.02 h/night for non-adherent patients and 0.69 h/night for attempter patients. ConclusionsThis is the first study in CPAP therapy that formalises the process of transforming raw data into effective home care interventions that can be delivered directly to the patients. In fact,it is the first time that both patient characterisation and predictions based on data are used to provide personalised patient management for CPAP therapy. Our model is flexible to be extended to new types of interventions and other homecare therapies.