What procedure should I use for essentially a longitudinal/repeated measures logistic regression?

[sas_user124]

Hello! I am using Health and Retirement Study data for a project.

Essentially, I am measuring the impact of antidepressant use on cognition over time. My main analysis is a linear mixed model coded as:

proc mixed data=longssri method=reml;
class SSRI(ref='No SSRI');
model cog=SSRI Time SSRI*Time/s chisq;
random intercept / solution type=un subject=ID;
run;

where i am determining the magnitude/significance of the change in cognition score (cog) over time (6 waves of followup)

I can divide cognition into impaired/not impaired and would like to perform effectively the same test, but with the binary outcome instead of continuous.

Can you help me code that correctly? My advisor doesn't typically work in SAS, so wasn't quite sure the exact coding. Would I use proc genmod? or something else? I have very limited experience within longitudinal data, so I would greatly appreciate advice. Thanks!

[StatDave]

If you want to make inferences at the population level, you could use PROC GEE to fit the population-averaged, Generalized Estimating Equations model. (You can also use GENMOD for GEE models, but PROC GEE is the newer procedure and is recommended for such models). See the binary response example in the Getting Started section of the PROC GEE documentation for syntax. If you want your inferences at the individual level, then you can stay with a mixed model approach with random effects by using PROC GLIMMIX. See the logistic regression example in the GLIMMIX documentation. 

[SteveDenham]

One thing to consider - dichotomizing your data that is continuous (or nearly so) will certainly reduce the power of any tests you apply, and widen confidence intervals. It may be common to do so, especially with clinical data, but once upon a time it was common to have a few drinks before driving...

SteveDenham