Refinement and Adaption of Reinforcement Learning to Personalize Behavioral Messaging for Healthy Habits (REINFORCE2)

Refinement and Adaption of Reinforcement Learning to Personalize Behavioral Messaging for Healthy Habits

Refinement and Adaption of Reinforcement Learning to Personalize Behavioral Messaging for Healthy Habits

Reinforcement learning is an advanced analytic method that discovers each individual's pattern of responsiveness by observing their actions and then implements a personalized strategy to optimize individuals' behaviors using trial and error. The goal of the proposed research is to refine, adapt and perform efficacy testing of a novel reinforcement learning-based text messaging intervention to support medication adherence for patients with type 2 diabetes within a community health center setting. This study will be a parallel randomized pragmatic trial comparing medication adherence and clinical outcomes for adults in a community setting aged 18-84 with type 2 diabetes who are prescribed 1-3 daily oral medications for this disease. Participants will be randomized to one of two arms for the duration of the study period: (1) a reinforcement learning intervention arm with up to daily, tailored text messages based on time-varying treatment-response patterns; or (2) a control arm with up to daily, un-tailored text messages. Outcomes of interest will be medication adherence, as measured by electronic pill bottles, and HbA1c levels over 6 months.

The goal of the proposed research is to refine, adapt and perform efficacy testing of a novel reinforcement learning-based text messaging intervention to support medication adherence for patients with type 2 diabetes within a community setting. Type 2 diabetes is an optimal condition in which to refine this program, as it is one of the most prevalent chronic conditions in the US adult population and requires most patients to be on daily or twice daily doses of medications. This study will be a parallel randomized pragmatic trial comparing medication adherence and clinical outcomes for adults in a community setting aged 18-84 with type 2 diabetes who are prescribed 1-3 daily oral medications for this disease. Participants will be randomized to one of two arms for the duration of the study period: (1) a reinforcement learning intervention arm with up to daily, tailored text messages based on time-varying treatment-response patterns; or (2) a control arm with up to daily, un-tailored text messages. Outcomes of interest will be medication adherence, as measured by electronic pill bottles, and HbA1c levels over 6 months.

Participants in the intervention arm will receive up to daily, tailored text messages based on their electronic pill bottle-measured adherence. Given the participants' baseline characteristics and time-varying responses to the messages, a reinforcement learning algorithm will deliver different text messages and adapt over time to determine which type of messaging works best for each individual participant.

Proportion of correct doses recorded by electronic pill bottles in the 6-month follow-up period, averaged across study medications

Change between baseline HbA1c used for identification and the 6-month intervention period, using laboratory values in the EHR

Inclusion Criteria: Diagnosis of Type 2 Diabetes Mellitus (T2DM) Prescribed between 1-3 daily oral medications for diabetes Most recent HbA1c level of 7.5% or greater Suboptimal adherence, defined by proportion of days covered (PDC) < 0.80 based on chart review Must have a smartphone for which they are the sole user Must have a basic working knowledge of English or Spanish Exclusion Criteria: Currently using a pillbox and/or not willing to use electronic pill bottles for 6 months Receive help at home on a daily basis with taking medications