Date of Completion

2022

Document Type

Honors College Thesis

Department

Mathematics

Thesis Type

Honors College

First Advisor

Nicholas Allgaier

Second Advisor

Matthew Price

Keywords

Machine learning, health behaviors, dietary patterns, adolescent cognition, obesity, physical activity

Abstract

Adolescence is a critical period during which the brain undergoes dramatic changes that ultimately influence life-long behavior and health. Concerningly, this period of maturation is increasingly accompanied by declines in physical activity, poor diets, and the development of obesity, type II diabetes, and cardiovascular disease (Andrade et al., 2016; Dumith et al., 2011; Goran et al., 2003). Investigations into the causes and consequences of (and the brain’s role in) such trends is essential to overcoming the escalating public health crises. Central to these problems is an imbalance in caloric intake and expenditure, or a disruption in energy homeostasis (World Health Organization [WHO], 2021). We therefore set out to explore the association between health behaviors related to energy homeostasis and cognition in adolescents. Our study comprised 3,143 children between the ages of 10-13 years old from the large cohort enrolled in the Adolescent Brain Cognitive Development® (ABCD) study; the analysis employed nested, 5-fold, cross-validated elastic-net regression models to test the hypothesis that two behaviors underlying energy homeostasis—diet and physical activity—are associated with crystallized and fluid cognitive outcomes in adolescents and brain activity in a number of regions of interest (ROIs) during the n-back working memory task. Further, we predicted that these findings would be generalizable to a held-out (“lockbox”) validation set containing 20% the original sample. After controlling for potential covariates, these predictive models suggest fluid cognition may be positively associated with physical activity and unrelated to diet (R2train = 0.048; R2validate =0.025), while crystallized cognition may be unassociated with physical activity, negatively associated with Western-style diets, and positively associated with healthy diets (R2 train = 0.059; R2 validate =0.056). However, models did not reliably predict brain activity during the n-back during cross-validation. The approach yielded generalizable non-imaging cognitive correlates of dietary patterns and physical activity levels, ultimately emphasizing the significance of these adaptable health behaviors to cognitive outcomes. Such research helps identify multimodal targets for public health policies required to combat these trends and minimize health inequity.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Share

COinS