Distributional modeling, also known as normative modeling, allows for the creation of centile curves to visualize the variation of a developmental phenotype as a function of age. Previous studies have utilized a variety of modeling approaches, including Bayesian regression, Gaussian processes, and Generalized Additive Models for Location, Scale, and Shape (GAMLSS) to model trajectories of cortical thickness and cortical surface area over the course of the human lifespan. Such studies typically incorporate a cross-sectional design, generally discarding non-baseline data from longitudinal data sources and failing to account for within-person changes across time. Additionally, while many studies plot centile curves separately for each sex and utilize random effects to account for site- and/or study-related variability, other demographic factors, such as race, are not considered in the distributional models.
This project addresses these limitations through the creation of longitudinal distributional models for cortical thickness and surface area of adolescents ages 9 to 15, stratified by race and sex, for 70 different brain regions. Spatial and longitudinal results are visualized in this interactive dashboard – explore the dashboard to learn more!
This project was completed as part of the 2024 Equitable Data Science in Adolescent Development REU through the University of Minnesota School of Public Health Division of Biostatistics and Health Data Science. Special thanks to our advisor, Mark Fiecas.