Paper 275-25 I Introduction Ii Dunnett's Test in Proc Glm Statistics and Data Analysis Multiple Comparisons with a Control in Gee Models Using the Sas System

In toxicology, it is often of interest to perform a Dunnett’s test to determine whether a pesticide or toxic substance is different at various levels from a control. The problem is complicated by the fact that often a traditional analysis of variance (ANOVA) model cannot be used because of a complex variance structure. For example, it might be of interest to determine whether rats receiving different doses of a toxic substance are statistically different with respect to certain outcome measures from a control. Because of concerns that animals from the same litter might be correlated, generalized estimating equations (GEE) methods are often used in modeling these types of data. In SAS, PROC GENMOD allows GEE fitting but does not provide p-values for multiple comparisons with a control. In this paper, we discuss how the algorithm proposed by Hsu (1992) for performing the Dunnett’s test and implemented in PROC GLM in the LSMEANS statement can be extended to GEE models. A SAS macro is provided for computing p-values and Confidence Intervals (CI) similar to those obtained in PROC GLM for the two-sided Hsu-Dunnett test.