Big Data/Machine Learning/AI
Methods for Deployment of a Dynamic Ecological Momentary Digital Intervention (EMI) to Improve Uptake of a 3-arm Behavioral Intervention Christian Okitondo* Christian Okitondo Angela Fertig Junia N de Brito Amanda Trofholz Jerica Berge Allan Tate
Introduction: Previous childhood obesity interventions have had limited effectiveness, partially due to inattention to competing real-time factors that influence health behaviors. Ecological momentary intervention (EMI) is a scalable, cost-effective approach to adapt to the experiences of users through real-time feedback. The ongoing Family Matters clinical trial deploys a dynamic EMI to interrupt stress processes influencing behavioral patterns shaping the family home food environment.
Objective: This study will describe an adaptive algorithm design feature of an EMI implementation, including content challenges, adaptive adjustments based on stress type and message library size, and responsiveness to user feedback, ensuring a comprehensive understanding for effective deployment.
Methodology: The Family Matters intervention study used machine learning to develop a dynamic algorithm for choosing personalized EMI messages throughout the day, aimed at enhancing family meals and reducing stress based on the stress source. The development of the algorithm was guided by two types of analyses:
1. Time-Series Analysis: Machine learning identifies and analyzes parental stress patterns influencing family meal dynamics.
2. Geospatial Analysis: Stress location (at-home vs. out-of-home) and its effect on EMI effectiveness are examined using machine learning.
Using the algorithm, morning stress report and location are then used to inform the EMI message delivered late in the day. The algorithm adapts to stress type, message library size, and user feedback. Data collected demonstrates the algorithm’s adaptation, focusing on periods with consistent stress patterns.
Significance and Implications: Machine learning in EMIs facilitates adaptive interventions. Tailoring messages based on real-time stress and location sets the stage for scalable solutions enhancing home food environments and child health outcomes.