I recently got interested in evidence sinthesis and, while studying, a little doubt came up to my mind:

Is it possible to do Network Meta-analysis using poisson regressions?

Also, is it possible to calculate NNTs from the Incidence Rate Ratios of the overall pairwise comparison? If not, is there a way to calculate NNT of the pairwise comparisons?

Thanks for your suggestions and comments


1 Answer 1


Yes, a network meta-analysis is quite general and can be applied to any probability model including Poisson. The underlying assumption is that multiple interventions have been applied to different samples from the same patient population and we are interested in making indirect comparisons between the interventions (comparing effects across samples) on a population parameter such as the mean.

It might be challenging to work out the number needed to treat from only a rate ratio. This ratio tells you the disparity between two means, but there are any number of means that would result in that ratio. You may need to also know at least one of the incident rates to work out number needed to treat.

  • $\begingroup$ Thanks for your kind answer! Thus, if I have the incidence rates of the single arms of the studies, could I be able to use them to calculate NNTs? $\endgroup$ Aug 11, 2021 at 17:55
  • 1
    $\begingroup$ I believe so. You may want to consider working out the NNT using a lower confidence limit for the unknown fixed true incidence rate, a point estimate for the incidence rate, as well as an upper confidence limit. To be conservative you could use the largest NNT of these three. $\endgroup$ Aug 12, 2021 at 13:33
  • $\begingroup$ Thanks again for your answer. I think it would be easier to go for the classic Risk Difference way than using IRR (even if I lose the follow-up standardization) $\endgroup$ Aug 12, 2021 at 18:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.