Improving Diabetes and Depression Self-Management Via Adaptive Mobile Messaging

Project Final Report (PDF, 544.36 KB) Disclaimer

The use of reinforcement learning algorithms for personalizing text messaging interventions can increase physical activity in a diverse sample of people with diabetes and depression.

Project Details -
Completed

Diabetes and depression are more prevalent and have poorer outcomes in certain populations, such as those with low incomes, limited health literacy, and ethnic minority status. Research shows that mobile health applications (mHealth apps) can assist patients in engaging in healthy activities. Because most lower-income Americans use mobile phones, mHealth apps have great potential to reach individuals who have limited access to healthcare. Thus, the use of effective mHealth apps might reduce current health inequities. However, most mHealth apps that target behavioral changes are not personalized, which limits their effectiveness. Personalization of mHealth app interventions may be done by computer tailoring, which involves adapting treatments to the participant's observed behavior and attributes. One promising strategy is implementing adaptive learning, which enables the prediction of which information may be useful for users.

The Diabetes and Mental Health Adaptive Notification and Tracking Evaluation (DIAMANTE) study tested whether a digital physical activity intervention using personalized text messaging via reinforcement learning algorithms could increase step counts in a diverse, multilingual sample of people with diabetes and depression symptoms.

The specific aims were as follows:

  • Conduct user testing of the DIAMANTE platform in English and Spanish.
  • Evaluate the effectiveness of adaptive versus static messaging interventions.
  • Evaluate the effectiveness of phone outreach for nonresponsive participants.

This research used machine learning algorithms to customize and distribute health messages to people with low income through text messaging. As part of the DIAMENTE three-armed randomized control trial, researchers recruited low-income minority patients with depression and diabetes within the San Francisco Health Network. Participants were randomized to one of three arms: 1) a control group receiving a weekly mood monitoring message; 2) a random messaging group receiving randomly selected feedback and motivational text messages daily; and 3) an adaptive messaging group receiving text messages selected by a reinforcement learning algorithm daily.

The evaluation found that participants in the adaptive messaging arm showed a significant step count increase of 19 percent in contrast to 1.6 percent and 3.9 percent increases in the random and control arms, respectively. This supports the use of reinforcement learning algorithms for personalizing text messaging interventions to increase physical activity in a diverse sample of people with diabetes and depression. These findings are significant because most digital health studies in vulnerable populations to date have been pilot studies, and the use of machine learning methods for personalization is rarely applied to low-income and Spanish-speaking populations.