Prof. Jaroslaw Harezlak
Dept. of Biostatistics, Fairbanks School of Public Health, Indiana University
Title :
Regression trees for longitudinal data (LongCART)
and their application in the biomarker neuroimaging study
Date and Time: April 17, 2014 - 4:15pm to 5:15pm
Location: Cupples I, Room 199
Abstract:
Mixed model methodology (Laird and Ware, 1982) have proven to be extremely useful in the longitudinal data settings. Usually, we
assume that a mean longitudinal model conditional on baseline covariates is applicable to the entire population. However in a
heterogeneous population the variation in the longitudinal trajectories can be hard to explain. This is usually the case in
observational studies with many possible predictors. In such cases, a group-averaged trajectory could mask important subgroup
differences. Our aim is to identify and characterize subgroups based on the combination of baseline covariates with differential
longitudinal behavior. We extend the CART methodology (Breiman et al., 1984) to identify the such subgroups via evaluation of a
goodness of fit criterion at all possible splits of partitioning variables. In such an approach, we encounter the problem of
multiple testing. We ameliorate this problem by performing a single test identifying the parameter stability of a longitudinal
model at all values of a partitioning variable. We propose a tree construction algorithm and obtain asymptotic results for the
instability tests. Simulation study is used to evaluate the numerical properties of our LongCART method. Finally, we apply the
LongCART to study the changes in the brain metabolite levels of chronically infected HIV patients.