Application of the SUDAAN Software Package to Clustered Data Problems: Pharmaceutical Research
Gayle S. Bieler
and Rick L. Williams, Research Triangle Institute
Presented to the
US Food and Drug Administration, February and June,1996,
at the 1996
Joint Statistical Meetings, and to the New Jersey Chapter of the ASA.
Abstract
In the
pharmaceutical sciences, researchers often encounter data which are
observed in clusters. Individual responses may represent multiple
outcomes from the same patient (such as sets of teeth, pairs of eyes,
or longitudinal outcomes on the same individual) or from multiple
patients within a larger cluster, such as a physician clinic or an
animal litter. Intracluster correlation, or the potential for
clustermates to respond similarly, poses special problems for
statistical analysis. This occurs because experimental units from the
same cluster are not statistically independent. Failure to account
for the cluster effect in the statistical analysis can result in
underestimated standard errors and false-positive test results. In
addition, cross-over clinical trials will not yield the associated
increase in statistical power if the design is ignored in the
analysis. This workshop will cover the statistical theory used in
SUDAAN to fit marginal or population-averaged models using
generalized estimating equations (GEE) with robust variance estimates
which fully account for intracluster correlation. Attendees should be
familiar with pharmaceutical research as well as fitting regression models.