ABSTRACT
Obesity affects 2 out of 5 American adults and leads to high individual and societal costs. Behavioral
interventions focusing on diet and exercise can yield clinically significant weight loss of at least 5%. However,
the proportion of people who achieve this benchmark in behavioral weight loss trials is limited due to variable
and waning intervention adherence. One strategy to improve intervention adherence and weight loss is to
provide small monetary incentives for behaviors such as calorie logging or outcomes such as interim weight
loss. Current intervention structures are uniform, providing the same incentive timing and amount to all people.
Thus, some receive incentives even though they do not need them, while others do not receive enough
incentives to change their behavior. This uniform structure taxes limited budgets available for incentives. There
is an urgent need for a precision medicine approach that distributes incentives to people who respond to them.
To address this need, we apply reinforcement learning, a machine learning method, to create a precision
medicine intervention whereby each participant receives an individualized sequence of incentives to increase
the probability they achieve clinically significant weight loss. This novel approach takes data from cellular
scales and a dietary logging application to inform the prediction of participant behavior in response to incentive
amount. Using data from a previous incentives trial, we developed an algorithm with high predictive accuracy. We
plan to conduct a future randomized trial comparing the efficacy of this precision approach to a uniform approach on
weight loss. To prepare for the future trial, we will conduct a planning study with three aims: 1) Develop a digital
health platform to incorporate an existing commercial mathematical optimization modeling software for
prescriptive analytics to implement our algorithm. The platform will provide real-time analytics and deliver a
weekly incentive based on past behavior. We will develop a manual of procedures for detecting and resolving
errors for real-time data capture and processing.
2) Evaluate the feasibility of applying the algorithm
prospectively in a clinical trial. We will enroll two successive cohorts of adults with obesity in a single-arm
feasibility study and implement the personalized incentives intervention over six months. We will establish
logistical feasibility of executing the protocol and provide estimates of screening-to-enrollment, retention, and
incentives intervention response rates.
3) Characterize participant intervention acceptability as indicated by
intervention adherence rates, safety criteria, and feedback from qualitative interviews. Our findings will be used
to support a future, adequately powered randomized trial. If efficacious, our personalized intervention could
maximize return on investment for healthcare payers by optimally distributing a preset budget across a
population to maximize long-term, population-level weight loss in real-world settings.