Novel Statistical Modeling Advances for Obesity Research

• Activity Number: 62
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Edward H. Ip*+ and Kiros Berhane
• Companies: Wake Forest University School of Medicine and University of Southern California
• Keywords: hidden Markov model ; Louisiana Health Study ; quantile regression ; BMI trajectory ; Bayesian analysis ; eco-social model

Abstract

In the first paper, we present two examples of how statistical modeling can be applied to complex, multi-level data in the study of childhood obesity. The first example uses a multiple-chain hidden Markov model framework that aims to operationalize an eco-social conceptual model proposed by Thomas Glass and Matthew McAtee. A subset of data from the Louisiana Health study (N=2,201)is used to illustrate the model. Neighborhood information collected from Google Map were used to augment the data set. In the second example, we present results from methodologic advances that extend quantile regression to allow for flexible multi-level growth models and also for assessing quantile specific mediational effects. Spline based approaches, with rich random effects structure to allow for subject-specific heterogeneity, will be used to characterize BMI trajectories in the quantile setting, along with Bayesian techniques for the multi-level and mediation modeling extensions. We use data from the Southern California Children's Health Study (N=11,700) and the Health Places Study (N=376) to illustrate insights gained in assessing impacts of environmental and genetic factors on childhood obesity.