Online Program

Evaluating the Dynamic Association Between Bivariate Binary Variables Over Time (309917)

Keywords: Kernel density, Longitudinal data, Nonparametric estimation, Unstructured models, Time-varying odds ratio

In many medical studies, researchers have a particular interest in an association between two binary response variables. As longitudinal data where the responses are measured repeatedly over time arise frequently in the studies, it enables us to examine the association at a given time and explore changes in the association over time. However, assessing the association of the two variables at the given time can be challenging in longitudinal studies in which the two variables are collected at different time points. In our paper, we develop an unstructured nonparametric model that describes the dynamic association and proposes estimation procedures that examine the association from the longitudinal samples with two repeated outcomes measured at different times. In order to account for the nature of the longitudinal data, a kernel density function is used to yield a consistent estimate of the association at each time point. The performance of the proposed estimation procedure is studied in two scenarios where the two variables are measured concurrently and differently and confirm that the proposed estimation procedure performs similarly in both scenarios. We also illustrate the proposed approach by analyzing the Framingham Heart Study data.