A reinforcement learning based algorithm for personalization of digital, just-in-time, adaptive interventions.

Area of Science:

• Digital health
• Artificial Intelligence
• Behavioral science

Background:

• Chronic diseases and unhealthy behaviors cause most global deaths.
• Personalized patient support improves health outcomes.
• Digital, just-in-time, and adaptive interventions (JITAI) offer mobile-based support.

Purpose of the Study:

• To develop a reinforcement learning (RL) mechanism for personalizing JITAI.
• To optimize intervention timing, frequency, and type based on user context.
• To enhance RL performance using accelerator techniques.

Main Methods:

• Employed two RL models: intervention-selection and opportune-moment-identification.
• Intervention-selection adapts delivery based on type and frequency.
• Opportune-moment-identification finds optimal delivery times.
• Utilized customized eligibility traces and transfer learning for acceleration.

Main Results:

• The proposed RL approach demonstrated superior performance compared to standard RL algorithms in simulations.
• The system effectively captured variations in user behavior and preferences across simulated personas.
• Personalized intervention strategies led to improved outcomes in simulated scenarios.

Conclusions:

• RL-based personalization of JITAI is effective for improving health behaviors.
• Optimizing intervention delivery through adaptive algorithms enhances user engagement and outcomes.
• The proposed methods offer a promising approach for scalable digital health interventions.