2016-12-02

Doug Lorenz, University of Louisville

"A marginal log rank test for clustered data under informative within-cluster group size"

Clustered data - data organized into independent units called clusters within which dependencies may exist - are common in biomedical applications.  Well-known methods for the marginal analysis of clustered data, such as generalized estimating equations, are adept at estimating marginal parameters while dealing with dependencies within cluster.  Additionally problematic characteristics of clustered data include the informativeness of cluster size and within-cluster covariate distributions, wherein the number of observations per cluster and the distribution of covariates within each cluster are correlated with an outcome variable to be analyzed.  We review methods for the marginal analysis of clustered data that account for these potentially biasing characteristics, and use them to develop a marginal log rank test for clustered data resistant to these effects.  We illustrate the new test's properties with a simulation study and demonstrate its use on a spinal cord injury data set.

Stay connected TwitterFacebookLinkedInYouTubeInstagram