Unadjusted p-values
-
Random question, but wondering if an unadjusted p-value can ever be used to support a claim in an APS directed at HCPs?
-
@COREkarentaylor, yes, there are circumstances where an unadjusted p-value can be used to support a claim in an APS directed at HCPs. In fact, even when adjustments would have been required to mitigate familywise type 1 error rate, a claim-neutral presentation of the data may be acceptable. Here are some of the key points you’d want to keep in mind.
When the study and secondary endpoint do not (in and of themselves) contravene the code, data may be presented in a factual unembellished manner. When there is no adjustment for multiplicity, the claim should appear neutral. Simply present data with corresponding stats (text or graphic form) without the addition of elements to the data presentation which suggest an improvement or difference (e.g. relative reduction callouts/arrows, NNT). Claim neutrality also applies to context (e.g. headlines & tabs). Introduce the data presentations with “topics” (e.g. “efficacy” as opposed to conclusions (e.g. demonstrated powerful efficacy). The presentation must state that the presented CI or p-value is unadjusted. For example, it suffices to insert “(unadjusted)” next to the CI or p-value. Note that p-values & CI cannot be relegated to weblink destinations. Essentially, provided statistical analysis is presented to satisfy s5.9, we’ll allow a data presentation. But we’ll reject overt and contextually implied claims of efficacy/safety/difference.
For subgroups, in addition to the above, one must present the subgroup of interest along with all of the complementary subgroups with equal emphasis (such that none stands out). E.g. Can’t just present male data, must present both male and female data3.
Note that this response assumes the primary endpoint was statistically significant. Secondary endpoints cannot be used to salvage a failed study.
-
Thanks @Jennifer-Carroll!
-
Thanks @Jennifer-Carroll!
Hi there,
I had a follow up question to the one above re: unadjusted p-values. Do the same rules apply to nominal p-values? I've seen that nominal p-values / nominally significant p-values are unadjusted p-values that are <0.05 - I'm not sure if my understanding is correct. But if it is, then I assume the rules for unadjusted p-values include nominal p-values as well?
Thanks so much in advance for your help!
-
Hi there,
I had a follow up question to the one above re: unadjusted p-values. Do the same rules apply to nominal p-values? I've seen that nominal p-values / nominally significant p-values are unadjusted p-values that are <0.05 - I'm not sure if my understanding is correct. But if it is, then I assume the rules for unadjusted p-values include nominal p-values as well?
Thanks so much in advance for your help!
Hello @kshulist
This answer is provided with the understanding that the nominal p-value is not in the product’s TMA. The application of the above would depend on why the p-value was nominal. It can be from different reasons and the nominal p-value may become a meaningless number if the assumptions of the statistical model used to compute it do not hold. Violating any of the varying prerequisites of a significance test might render the nominal p-value not acceptable. As such, the ability to present the data and the potential types of claims in an APS would require consideration of the statistical analysis. For the example in question, if a statistically significant p-value is nominal simply because it was unadjusted, a claim neutral presentation could be considered per the prior Q&A.
