Skip to main content
ARS Home » Plains Area » Houston, Texas » Children's Nutrition Research Center » Research » Publications at this Location » Publication #420687
Research Project: Preventing the Development of Childhood Obesity

Location: Children's Nutrition Research Center

Title: Machine learning strategies for assessing intervention groups to predict obesity in an online obesity prevention program

Author
item MUSAAD, SALMA - Children'S Nutrition Research Center (CNRC)
item HAZAN, HANANEL - Tufts University
item BARANOWSKI, TOM - Children'S Nutrition Research Center (CNRC)
item CALLENDER, CHISHINGA - Children'S Nutrition Research Center (CNRC)
item Thompson, Deborah - Debbe

Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 10/25/2024
Publication Date: 11/9/2024
Citation: Musaad, S.M., Hazan, H., Baranowski, T., Callender, C., Thompson, D.J. 2024. Machine learning strategies for assessing intervention groups to predict obesity in an online obesity prevention program [abstract]. 2nd Annual Houston Nutrition and Obesity Research Symposium. November 9, 2024; Houston, TX. Oral and Poster Presentation.

Interpretive Summary:

Technical Abstract: Butterfly Girls was a randomized controlled trial of an 8-episode online program promoting healthy diet and physical activity of 342 8–10-year-old African American girls. The girls were randomized to one of three groups (treatment (story and goal setting), comparison (story only), wait list control). Body mass index (BMI), dietary recall, activity behaviors, and food environment measures were collected at three time points (baseline, 3 months, 6 months). We aimed to understand the impact of using the intervention exposure variables (randomized group, number of episodes watched by the girls) in different ways. Decision trees were used to examine the importance of sociodemographic, number of steps (as a measure of physical activity), and environmental features in predicting change in BMI under three intervention-testing scenarios: a) use randomized group and number of episodes, b) use of number of episodes without randomized groups, c) stratifying by randomized group. Selected features varied based on the method of using the intervention variables in the decision trees. We are exploring the reason for the differences in selection of some features, including differences in randomized group characteristics, decision tree identification issue, or true differences in intervention effect. Given how online health-promoting programs have become common, researchers need to be aware of the different ways that the intervention exposure can be examined during analysis. Knowing that will inform study design and accurate treatment effect estimation.