Reward Design For An Online Reinforcement Learning Algorithm Supporting Oral Self-Care

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Behavioral Science

Background:

• Dental disease is preventable, yet patient adherence to oral hygiene advice is often poor.
• Personalized, timely interventions are needed to encourage consistent oral self-care behaviors.
• Mobile health (mHealth) platforms offer a scalable solution for delivering such interventions.

Purpose of the Study:

• To develop and evaluate an online reinforcement learning (RL) algorithm for optimizing mobile-based prompts to improve oral hygiene.
• To address challenges of delayed effects and noisy, sparse data in real-world mHealth settings.
• To design a reward function that balances desired health outcomes with user burden.

Main Methods:

• Development of an online reinforcement learning (RL) algorithm tailored for mHealth applications.
• Design of a quality reward function to maximize brushing quality and minimize user burden.
• Creation of a simulation environment test bed for optimizing RL hyperparameters.
• Deployment of the RL algorithm within the Oralytics mobile health study.

Main Results:

• The developed RL algorithm effectively optimizes the delivery of motivational messages for oral self-care.
• The quality reward function successfully balances health outcomes and user experience.
• The simulation test bed provided a robust method for hyperparameter tuning.
• Oralytics represents the first mHealth study using RL for dental disease prevention through optimized messaging.

Conclusions:

• Online reinforcement learning offers a promising approach for personalized behavior change interventions in mHealth.
• Optimized motivational messaging via RL can significantly improve adherence to oral hygiene practices.
• This work lays the foundation for future mHealth interventions leveraging AI for chronic disease prevention.