Can we report NNT from a post-hoc publication?

kshulist

Hi there,
I've read Question 508 from the PAAB Q&A (https://www.paab.ca/ask/question-508/) on NNT and I have a couple follow-up questions:

Question 1: Let's say the pivotal trial publication did not calculate ARR, but does provide the response rates for treatment vs. control (with confidence intervals or p-values). A second publication then calculates NNT from the ARR using the statistically significant response rates from the pivotal trial. Are we allowed to present the NNT from this second publication? I understand that the second publication is technically "post-hoc" but I'm unsure if the same concern about post-hoc bias applies here since it seems like it's just math done on previously existing data.

Question 2: Similar situation to Question 1, however, in this case, the statistics done on the treatment vs. control response rates in the pivotal trial were post-hoc. My guess is since the stats on the original pivotal trial data were post-hoc, we cannot present any calculated ARR or NNT. Is that correct?

Thanks in advance for your help!!

Jennifer Carroll

Hi @kshulist

It would be difficult to determine if the second publication could be used solely based on this information. We would need to look at the entirety of the publication to ensure the promotion of the publication does not contravene the code. Fortunately, per the prior question, you can likely calculate the NNT from the original trial.
Correct

kshulist

@jennifer-carroll Thank you!

Just to clarify I'm understanding you're last sentence of (1) correctly: (let's forget about the second publication) in general, if just the response rates (with statistical significance) are presented in the PM (or a PAAB-acceptable publication of the pivotal trial), then anyone can (likely) use those response rates to calculate ARR / NNT?

Thank you and have a great weekend!

Jennifer Carroll

Hey @kshulist

As per the principles noted in Q&A 508, if the published paper conveys a statistically significant improvement in the absolute risk reduction for the response rates (with multiplicity adjustments performed when required to manage risk of type I error), the NNT may be calculated.