Methods for Studying Variability as a Predictor of Health Status

• Activity Number: 335
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Michael Elliott*+ and Bei Jiang and Naisyin Wang
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: Differential measurement error ; latent class ; dementia ; hot flash ; longitudinal data

Abstract

Means or other central tendency measures are by far the most common focus of statistical analyses. However, as Carroll (2003) noted, ``systematic dependence of variability on known factors'' may be ``fundamental to the proper solution of scientific problems'' in certain settings. We will discuss methods we have recently developed to assess the degree to which individual variability in a predictor variable predicts a health outcome of interest in a longitudinal setting. We focus on combining information from mean profiles and residual variance to predict categorical outcomes in a joint hierarchical modeling framework. We consider applications to predict dementia onset using word recall measures obtained over time from the Health and Retirement Survey (Elliott et. al. 2012), and hot flash severity in menopausal women from follicle stimulating hormone (FSH) measures during menopausal transition (Jiang et al. 2012).