If these are 1:1 matches, the traditional approach is to do a paired analysis for univariate (either a paired t-test, McNemar's test, or one of the non-parametric tests that PROC UNIVARIATE produces). If you are doing modelling, you could use the difference score as the outcome.
If these are many:1 matches (as you might see with propensity score matching), then people often break the match for analysis. You lose some statistical power in doing that, but you have a much simpler job in the analysis phase. You could even use the propensity score as a covariate (there is a large literature on the do's and don't's here).
Doc Muhlbaier
Duke