Practical considerations when designing an online learning algorithm for an app-based mHealth intervention

Area of Science:

• Digital health interventions
• Clinical trial methodology
• Behavioral science

Background:

• Mobile health (mHealth) technology is increasingly integrated into clinical trials.
• Reinforcement learning (RL) offers potential for dynamic, individualized treatment policies.
• The LOWSALT4LIFE 2 (LS4L2) trial investigated app-based sodium reduction for hypertension.

Purpose of the Study:

• To implement and evaluate a reinforcement learning algorithm within an mHealth clinical trial.
• To optimize the timing of app notifications for improved participant engagement and behavior change.
• To present solutions for common challenges encountered when deploying RL in mHealth trials.

Main Methods:

• A reinforcement learning algorithm was developed to personalize reminder notifications.
• The algorithm predicted optimal notification times based on participant engagement likelihood.
• Key implementation challenges were identified and addressed using data from the LS4L2 trial.

Main Results:

• The RL algorithm aimed to enhance app engagement by delivering timely, effective notifications.
• Challenges included defining rewards, timescales, statistical models, computational balance, and missing data.
• Template solutions were developed based on the LS4L2 trial experience.

Conclusions:

• Reinforcement learning can enhance mHealth interventions by optimizing engagement and reducing participant burden.
• Addressing implementation challenges is crucial for successful RL deployment in future clinical trials.
• This work provides a framework for integrating RL into app-based mHealth research.