Personalized Digital Care Program Allocation for Older Adults: Reinforcement Learning-Based Simulation Study.

Area of Science:

• Gerontology
• Artificial Intelligence
• Digital Health

Background:

• Growing demand for older adult care and workforce shortages necessitate innovative, personalized solutions.
• Personalization is crucial for optimizing services and ensuring the long-term sustainability of care.
• This study proposes an adaptive reinforcement learning (RL)-based framework for precision digital care.

Purpose of the Study:

• To develop and evaluate an RL-based model for personalizing digital care program allocation.
• To optimize care engagement and health outcomes for low-income older adults living alone.
• To address the need for dynamic and individualized care strategies.

Main Methods:

• Developed a framework using contextual bandits, specifically Thompson Sampling, to maximize user outcomes.
• Tested four program allocation strategies: systematic, population average, idealized personalized, and precision digital care (Thompson Sampling).
• Conducted simulation-based experiments using synthetic datasets and multiple iterations to assess adaptive learning and optimization.

Main Results:

• Precision digital care using Thompson Sampling significantly outperformed baseline and population-average methods.
• Achieved comparable outcomes to the theoretical upper bound, with substantial improvements in call success rates, depression scores, and self-reported health.
• Demonstrated improved learning efficiency, dynamically refining assignments based on user responses.

Conclusions:

• Personalization is vital for effective digital care delivery.
• The RL-based framework offers a scalable and effective approach to advance precision digital care.
• Future validation through a publicly funded AI care program for community-dwelling older adults is planned.