The alarming rise in childhood obesity and its health consequences are serious public health challenges. Our ability to address these challenges depends on understanding the multi-factorial etiology of obesity which involves factors which can be intrinsic (e.g., metabolic, cellular, genetic) or extrinsic (e.g. environmental, demographic). Data are available at several levels (e.g., temporal, individual, school, neighborhood, and community). Existing methods may not suffice for characterizing complex inter-level relationships of these factors with obesity during the rapid growth period of puberty. To address this gap, we propose new methods that incorporate both age and sex related changes in the dynamic relationship between weight, height and obesity as well as the complex multi-level relationships of determinants of obesity. The research is motivated by needs identified in the USC based Children's Health Study and the Smart Growth Study.
In Aim 1, we develop new flexible multi-level quantile regression models based on our work on functional-based multi-level growth curves. Bayesian inference is conducted for concurrent assessment of effects of various factors, across several levels, on important features of the BMI trajectories of each child, relative to an appropriate reference. This model is further extended via hierarchical modeling of regression coefficients to assess patterns of effect estimates across reference quantiles and to integrate prior characteristics among risk factors. Advances in genomics present opportunities for assessing joint effects of genetics and the environment to the development of obesity, but new methods are needed to fully exploit family based and genome wide data.
In Aim 2, we develop methods for analysis of genes and gene-environment (GxE) interactions, using parent-offspring data in the context of growth curves (including quantile regression) with focus on longitudinal versions of the quantitative transmission disequilibrium test (L-QTDT). We also develop a method to find obesity-related genes involved in a GxE interaction (e.g. genes that modify the effect of dietary intake) in the context of a candidate gene or genome wide association study. For intervention purposes, one needs to understand the role of mediating factors on development of obesity in the context of causal pathways.
In Aim 3, we develop structural equation models and latent growth curve models to incorporate such mediational effects via a new multi-level quantile regression approach. We finally develop a latent variable approach for integrating information across all environmental, genetic and biomarker factors. Theoretical work on estimation and inferential procedures will be followed by extensive simulation studies to evaluate their performance and to investigate statistical properties of model parameters. We anticipate that the new methods will play a significant role in our capacity to understand the impact of risk factors on childhood obesity, and the use of information gained from analysis of risk factors to inform future obesity prevention efforts.
The development of the proposed novel statistical modeling techniques will play a significant role in enhancing our ability to understand the multifactorial etiology of childhood obesity, potentially leading to development of effective management and preventive measures against the rising tide of childhood obesity and related health consequences.
