| Alexis Wardell, Allison Deal |
Research studies often involve continuous outcomes, for example survey instrument scores measured repeatedly over
time. We see this in our cancer research when patients complete surveys before, during, and after an intervention. Our
motivating example evaluates an intervention vs. usual care, within a randomized trial, and compares scores between
groups at 3 timepoints. PROC GENMOD is used to fit marginal linear models with generalized estimating equation
methods to account for potential clustering within patients and allow for missingness. The longitudinal model includes
time, group (intervention vs. control), and the interaction between time and group. There are many interesting
comparisons that can be made from this type of study, including 1) how the outcome scores change over time, 2) do the
outcome scores change differently over time between groups, and 3) are scores different between groups at different
timepoints. Additionally, consideration must be made in treating time as continuous or categorical. PROC GENMOD is a
useful tool to make these comparisons, but special considerations must be made in order to write all the necessary
ESTIMATE statements. While tools like LSMEANS are useful, understanding how to write ESTIMATE statements for
comparisons is helpful for cleaning output and making comparisons that are unable to be made via LSMEANS. We lay
out the underlying model structure and equations to clearly illustrate how to write these ESTIMATE statements. |