Background: Advancements in mobile health technologies and machine learning approaches have expanded the framework of behavioral phenotypes in obesity treatment to explore the dynamics of temporal changes.
Objective: This study aimed to investigate the dynamics of behavioral changes during obesity intervention and identify behavioral phenotypes associated with weight change using a hybrid machine learning approach.
Methods: In total, 88 children and adolescents (ages 8-16 years; 62/88, 71% male) with age- and sex-specific BMI ≥85th percentile participated in the study. Behavioral phenotypes were identified using a hybrid 2-stage procedure based on the temporal dynamics of adherence to the 5 behavioral goals during the intervention. Functional principal component analysis was used to determine behavioral phenotypes by extracting principal component factors from the functional data of each participant. Elastic net regression was used to investigate the association between behavioral phenotypes and weight change.
Results: Functional principal component analysis identified 2 distinctive behavioral phenotypes, which were named the high or low adherence level and late or early behavior change. The first phenotype explained 47% to 69% of each factor, whereas the second phenotype explained 11% to 17% of the total behavioral dynamics. High or low adherence level was associated with weight change for adherence to screen time (β=-.0766, 95% CI -.1245 to -.0312), fruit and vegetable intake (β=.1770, 95% CI .0642-.2561), exercise (β=-.0711, 95% CI -.0892 to -.0363), drinking water (β=-.0203, 95% CI -.0218 to -.0123), and sleep duration. Late or early behavioral changes were significantly associated with weight loss for changes in screen time (β=.0440, 95% CI .0186-.0550), fruit and vegetable intake (β=-.1177, 95% CI -.1441 to -.0680), and sleep duration (β=-.0991, 95% CI -.1254 to -.0597).
Conclusions: Overall level of adherence, or the high or low adherence level, and a gradual improvement or deterioration in health-related behaviors, or the late or early behavior change, were differently associated with weight loss for distinctive obesity-related lifestyle behaviors. A large proportion of health-related behaviors remained stable throughout the intervention, which indicates that health care professionals should closely monitor changes made during the early stages of the intervention.
Trial registration: Clinical Research Information Science KCT0004137; https://tinyurl.com/ytxr83ay.
Keywords: FDA; behavioral dynamics; behavioral phenotype; functional data analysis; mHealth; machine learning analysis; mobile health; mobile phone; obesity intervention; pediatric obesity.
©Sarah Woo, Sunho Jung, Hyunjung Lim, YoonMyung Kim, Kyung Hee Park. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 17.08.2023.