Joel Dubin, University of California, Davis

Dynamical Correlation for Multivariate Longitudinal Data

Longitudinal data with a single response are commonplace in biostatistical studies. Some examples include monitoring blood glucose levels over time for diabetes patients, evaluating tumor size growth for cancer patients, checking CD4+ cell counts over time for HIV positive patients, etc. A more difficult situation arises when the responses over time are considered multivariate, where data on several time courses are recorded per subject. We will discuss one approach to handling this type of data from a study of hemodialysis patients in which several acute phase blood proteins per patient were measured longitudinally. In particular, we will present a non-parametric method for describing the dynamical correlation between multivariate longitudinal response curves, including consideration of derivatives and time-shifted versions of curves.