Reinforcement Learning on Dyads to Enhance Medication Adherence

Area of Science:

• Adolescent and Young Adult (AYA) Health
• Digital Health Interventions
• Computational Health

Background:

• Medication adherence is crucial for hematopoietic cell transplantation (HCT) recovery in AYAs.
• AYAs face individual and interpersonal barriers to adherence post-discharge.
• Existing interventions may not fully address dyadic and relational challenges.

Purpose of the Study:

• To optimize a digital intervention for AYA medication adherence post-HCT.
• To personalize intervention delivery using a novel Multi-Agent Reinforcement Learning (MARL) approach.
• To target both AYAs and their care partners within a dyadic framework.

Main Methods:

• Proposed a Multi-Agent Reinforcement Learning (MARL) framework for personalized intervention delivery.
• Each agent within the MARL framework delivered a specific intervention component.
• Evaluated the MARL approach using a dyadic simulator based on real clinical data.

Main Results:

• The MARL approach demonstrated faster learning compared to a flattened agent model.
• Simulations showed a significant improvement in medication adherence (approximately 3%) versus random delivery.
• The personalized approach showed potential for enhancing adherence in a complex patient population.

Conclusions:

• A MARL framework can effectively personalize digital interventions for AYA medication adherence.
• This approach addresses dyadic and relational factors impacting adherence post-HCT.
• Further evaluation in a clinical trial is warranted to confirm effectiveness.