Predicting adherence to fully-automated, chatbot-delivered digital cognitive behavioral therapy for insomnia (dCBT-I)...

Area of Science:

• Digital health
• Machine learning applications
• Sleep medicine

Background:

• Digital cognitive behavioral therapy for insomnia (dCBT-I) shows effectiveness but faces adherence challenges in real-world settings.
• Machine learning (ML) presents a promising approach for predicting healthcare utilization and patient engagement.
• Understanding adherence predictors is crucial for optimizing dCBT-I interventions.

Purpose of the Study:

• To apply ML techniques to predict adherence to dCBT-I using baseline participant characteristics.
• To identify key factors influencing adherence to a chatbot-delivered dCBT-I program.
• To explore the potential of ML in personalizing dCBT-I for improved treatment outcomes.

Main Methods:

• A pilot real-world study involving 75 individuals with insomnia symptoms using a 28-day chatbot-delivered dCBT-I program.
• Application of ML models (logistic regression, SVM, random forest, gradient boosting) to predict adherence based on baseline data.
• Utilized grid search for model tuning, cross-validation for evaluation, and synthetic minority over-sampling technique for data imbalance.

Main Results:

• Baseline depressive symptoms were the strongest predictor of non-adherence and shorter usage duration.
• Previous night's sleep onset latency and wake time after sleep onset predicted increased next-day usage.
• ML models demonstrated significant predictive power for overall adherence (AUCs 0.65-0.91) and next-day adherence (AUCs 0.56-0.74).

Conclusions:

• ML models can effectively predict user adherence to dCBT-I in real-world applications.
• Depressive symptoms and specific sleep parameters are critical factors influencing dCBT-I engagement.
• These findings support the development of personalized dCBT-I interventions to enhance treatment effectiveness.