A note of caution about NNT's.   Like predictive values in diagnostic testing, NNT's are sensitive to baseline event rates (prevalence for tests).
consider the following data for 10,000 patients randomized to heparin or placebo
 
 

We determine the following values:

CER= 50/5000 = .02
EER= 100/5000= .01
AAR= .01
RR= .01/.02 = 0.5
RRR= 1- 0.5 = 0.5
NNT= 1/.01= 100

In this case a reduced risk of 50% isn't as great as it seems.  You need to treat 100 patients, instead of 10 to avoid an adverse event.  Again, note the difference in baseline adverse event rates between the first example (20%) and this example (10%).   This highlights why NNT's can be more informative than RR's and RRR's

However, you must be careful when applying NNT's from other studies to your patient population.  If the baseline event rate (CER) is different, the NNT will not be the same, even assuming all other factors are equal.

This is where the second NNT formula above comes in handy.  RRR's tend to be more similar, for a given treatment, between studies.   Thus it is often safer to assume that a reported RRR is applicable to your population than the reported NNT.   You can re-calculate the NNT based on a known, or estimated, CER for your population.   Try doing this calculation over a range of plausible CER's for your setting.

Confidence Intervals on NNT

Like RR's, the NNT as a measure of effect size is a point estimate and confidence intervals should be considered.   If the CI for the NNT includes 0, than the treatment may have no benefit.