A1946
Title: Precision physical activity prescription via reinforcement learning for functional actions
Authors: Gefei Lin - George Washington University (United States)
Rui Miao - University of Texas at Dallas (United States)
Jennifer Sacheck - Brown University (United States)
Xiaoke Zhang - George Washington University (United States) [presenting]
Abstract: Physical activity (PA) plays an important role in maintaining and improving health. Daily steps have been a key measure for PA which are easily accessible with common wearable devices. However, methods are lacking that can recommend a personalized optimal distribution of daily steps for an individual to follow over a period of time for the best outcomes of certain health biomarkers. A new offline reinforcement learning (RL) algorithm is developed to learn personalized and optimal PA distributions associated with cardiometabolic risk using data from the All of Us Research Program, which includes months of step counts as well as repeated measurements of key health biomarkers. In this approach, the action is a function representing the daily step distribution over a period of time. Simulation studies demonstrate the advantage of the proposed approach over existing continuous-action RL methods. The learned optimal policies from the All of Us Research Program data generally suggest people follow more consistent patterns of PA over time while offering tailored recommendations for subgroups stratified by blood glucose level, body mass index, blood pressure, age, and sex.
