Adherence Forecasting for Guided Internet-Delivered Cognitive Behavioral Therapy: A Minimally Data-Sensitive Approach

Area of Science:

• Digital mental health
• Machine learning in healthcare
• Behavioral science

Background:

• Internet-delivered psychological treatments (IDPT) enhance mental healthcare accessibility but face adherence challenges due to limited patient-provider interaction.
• Data privacy regulations like GDPR necessitate data minimization for real-world IDPT implementation.
• Predicting patient adherence is crucial for optimizing treatment outcomes and resource allocation in digital mental health.

Purpose of the Study:

• To develop and evaluate a deep learning model for automatic adherence forecasting in Guided Internet-delivered Cognitive Behavioral Therapy (G-ICBT).
• To assess the model's performance using only minimally sensitive login/logout timestamp data.
• To demonstrate the feasibility of adherence prediction within privacy-preserving digital mental health frameworks.

Main Methods:

• A Self-Attention-based deep learning model was employed for adherence forecasting.
• The model utilized minimally sensitive login/logout timestamp data from 342 G-ICBT patients.
• Adherence was defined as at least eight platform connections exceeding one minute over 56 days; 101 patients (approx. 30%) were non-adherent.

Main Results:

• The proposed model achieved over 70% average balanced accuracy in predicting adherence.
• Accurate adherence forecasting was accomplished using approximately one-third of the treatment duration data (20 out of 56 days).
• The findings indicate that adherence prediction is feasible even with highly minimized, non-sensitive user data.

Conclusions:

• Automatic adherence forecasting for G-ICBT is achievable using only minimally sensitive login/logout data.
• This approach supports the development of privacy-preserving tools for real-world IDPT platforms.
• The study highlights the potential of AI to improve adherence monitoring and patient outcomes in digital mental healthcare.