I am analysing data from a cluster RCT looking at the effect of an intervention (int) in health centres (centre) to reduce inappropriate antibiotic prescribing (abu). We initially planned to present results as ORs using mixed models with random effects for health centre. This is the code:

Code:
meglm abu int || centre: , family(binomial) link(logit) eform
However, antibiotic prescribing is over 90% and the OR showed a very strong effect (OR=0.25) while the absolute difference was small (5%). This seems misleading, and we thought maybe this was because the outcome is not rare, so we decided to present the results as RRs rather than ORs using this code:

Code:
meglm abu int || centre: , family(poisson) link(log) eform
This gives an RR of 0.93, which seems more reasonable.

We also wanted to present Risk Difference (RD) and have tried a few different ways to get this. Binomial with identity or log link as suggested in this post
HTML Code:
https://www.statalist.org/forums/forum/general-stata-discussion/general/1693010-adjusted-risk-ratios-and-risk-differences-after-logistic-regression
gives error messages saying that identity and log links are not allowed with binomial distribution. We also tried:

Code:
meglm abu int || centre: , family(gaussian) link(identity)
But this gives an error saying that initial values are not feasible.

Can anyone suggest another way of doing this? Or is the best way to use margins?