Investigation on factors related to poor CPAP adherence using machine learning: a pilot study

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Respiratory Medicine

Background:

• Continuous positive airway pressure (CPAP) therapy is crucial for managing obstructive sleep apnea (OSA).
• Patient adherence to CPAP therapy remains a significant challenge, impacting treatment efficacy.
• Identifying predictors of poor adherence is essential for improving patient outcomes.

Purpose of the Study:

• To determine if machine learning can identify factors associated with poor CPAP adherence.
• To develop and evaluate a predictive model for CPAP adherence.
• To investigate the top factors correlating with suboptimal CPAP usage.

Main Methods:

• Utilized logistic regression and learn-to-rank machine learning with a pairwise approach.
• Analyzed CPAP usage logs from 219 OSA patients treated at Kyoto University Hospital.
• Assessed adherence prediction performance over a 12-week period.

Main Results:

• Achieved a peak adherence prediction accuracy with an F1 score of 0.864.
• Identified top ten factors correlating with poor CPAP adherence.
• Four of the top ten factors aligned with established clinical knowledge regarding CPAP use.

Conclusions:

• Machine learning models are effective for investigating factors related to poor CPAP adherence.
• Key factors for better adherence include minimizing air leakage, maintaining stable mask pressure, and ensuring consistent, longer CPAP usage duration.
• Findings can inform interventions to enhance patient compliance with CPAP therapy.