Artificial intelligence for predicting adherence to CPAP therapy in obstructive sleep apnoea: A systematic review
Manuel Casal-Guisande, José-Benito Bouza-Rodríguez, Mar Mosteiro-Añón, Alberto Comesaña-Campos, Alberto Fernández-Villar, María Torres-DuránObjectives
Adherence to continuous positive airway pressure (CPAP) therapy in patients with obstructive sleep apnoea (OSA) remains a major clinical challenge that limits its effectiveness. This systematic review aimed to evaluate the application of artificial intelligence (AI) models for predicting adherence to CPAP therapy in adults with OSA.
Methods
A systematic review was conducted in accordance with PRISMA guidelines. A literature search was performed in PubMed (MEDLINE) and IEEE Xplore for studies published between January 2015 and March 2026. Original studies and conference proceedings using AI techniques to predict CPAP adherence in adults with OSA and reporting performance metrics were included. Data were extracted on study characteristics, predictive variables, models used, validation strategies, and outcomes. Risk of bias was assessed using PROBAST.
Results
A total of 104 unique records were identified, of which 5 studies were included. The studies showed heterogeneity in sample size (42–3,785 patients), data type, and methodological approaches. Models included logistic regression, support vector machine, random forest, neural networks, and learning-to-rank approaches. Two main strategies were identified: models based on baseline variables and models based on longitudinal CPAP device data. Predictive performance was moderate to good, with AUC values of up to 0.84 and F1-scores of up to 0.86. The PROBAST assessment showed that all included studies had relevant risk-of-bias concerns, mainly concentrated in the analysis domain. No study performed external validation.
Conclusions
AI shows potential for predicting CPAP adherence and supporting more personalised management. However, the available evidence is limited and methodologically heterogeneous. Further studies with external validation and evaluation in real-world clinical settings are required before implementation.
Protocol registration
PROSPERO CRD420261355690.