0

I am working on a revise and resubmit for a journal article. This is an observational, cross-sectional study. We compared endorsements of different behaviors across groups (e.g., 30.1% of Group 1 members endorsed this behavior, a significantly higher proportion the 15.1% of Group 2 members who endorsed this behavior).

It was requested that we supply NNT for these comparisons, which was described to us as "Number Needed to Take" outside the context of a treatment study. However, in researching online, I am struggling to find guidance about how to calculate this outside the clinical trial context. For instance, we don't have a control group. So, when comparing prevalences, do we just choose a reference group? Or is there another way to calculate this outside the clinical trial context? I looked at R packages, but just found one for meta-analyses and another that appears unrelated to comparing prevalence across groups. There are calculators online, but again, they require the specification of a control group. I tried using some of these calculators, but sometimes the NNT is really high, which makes me question its applicability to this use case.

Any help is appreciated!

Blair Burnette
  • 13
  • 2

1 Answers1

0

It's a bit of a trick question. The problem is not that the study is observational, it's that it's not longitudinal in design. There are quite a few websites and books about the relation between the NNT and the RR - the risk ratio - or consequently with the OR (odds ratio) - using the approximation of the OR to the RR.

However, with cross sectional studies, you don't actually estimate the risk because there's no follow-up, so you can estimate neither an RR or an OR. The correct association measure is described as a prevalence or proportion ratio (or difference depending on the analysis). So the NNT is not the correct association measure.

AdamO
  • 52,330
  • 5
  • 104
  • 209