Dynamic Association Based on Time-Dependent Risk Factors in Longitudinal Studies

• Activity Number: 608 - Novel Methods for Longitudinal Analysis in Large Cohort Studies
• Type: Topic Contributed
• Date/Time: Thursday, August 3, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Lihui Zhao* and Kiang Liu and Colin O. Wu and Donald Lloyd-Jones and Norrina Allen and Sejong Bae and Lu Tian
• Companies: Northwestern University and Northwestern University and Office of Biostatistics Research, National Heart, Lung and Blood Institute, NIH and Northwestern University and Northwestern University and University of Alabama, Birmingham and Stanford University
• Keywords: dynamic prediction ; longitudinal data ; random effects model ; time-dependent covariates

Abstract

In multiple large cohort studies, it has been observed that the risk factors levels at remote past could be more predictive to the current cardiovascular risk than the more recent risk factor levels. In general, the predictiveness of historical risk factor levels varies with their measurement time. It has been hypothesized that the early cumulative exposure to suboptimal risk factor levels such as higher blood pressure may cause irreversible organ damage and permanently elevates the long term cardiovascular risk regardless of the subsequent change in the levels of the same risk factors due to various intervention at later stage of the life. Therefore it is important to understand the prospective association between the underlying trajectory of the past risk factor levels and a response variable reflecting one's recent cardiovascular health status. In this talk, we present a set of statistical models allowing estimating patterns of the trajectory of the risk factor levels for predicting the outcome of interest. We apply the new proposal to analyze data recently collected in the coronary artery risk development in young adult (CARDIA) study.