BookmarkSubscribeRSS Feed
DGrint
Obsidian | Level 7

Hello,

 

I have data in a series of stratified 2x2 tables, I want to calculate a p-value for a non-inferiority hypothesis test based on the common risk difference. There appears to be a way to do this using the riskdiff option, but not using the commonriskdiff option as far as I can see.

 

My proc freq tables statement looks like this:

tables site*trt*outcome / commonriskdiff(column=1 cl=(mh) printwts=(mh)  test=(mh)) alpha=0.1;

 

The test=(mh) option is testing whether the common risk difference is different from 0, but in a non-inferioity setting we want to know if it is below a margin of non-inferiority, not 0.

 

If using the riskdiff option there are additional options (noninf margin=0.1 method=wald), which are used to define the margin of non-inferiority. But those options apply to the risk difference from each of the 2x2 tables, not the common risk difference.

 

Does anyone know who to test a non-inferiority hypothesis on a common risk difference?

 

Thanks,

D

 

4 REPLIES 4
Ksharp
Super User

That is totally two different method.
"commonriskdiff" is like CMH test for nx2x2 table(combine n riskdiff into one riskdiff).
But for Noninferiority test,you cann't do that ,Noninferiority is ONLY for 2x2 table.
Maybe @StatDave  and @Rick_SAS knew the solution ?

DGrint
Obsidian | Level 7

Exactly, I want a non-inferiority hypothesis p-value from a common risk difference (CMH). It appears there isn't a way to do this.

 

I've found a workaround, but it would be good if there were a simple option to calculate the non-inferiority test p-value.

 

Workaround: alter the level of significance (alpha) for the common risk difference until the limit of the CI matches the non-inferiorty margin. The p-value can then be calculated based on the alpha value.

Ksharp
Super User
Clever idea .
But that is still risk difference test , NOT non-inferiority test.
DGrint
Obsidian | Level 7

I disagree.

 

Non-inferiority in this case is determined at the upper bound of the 95% CI for the common risk difference. If the CI for the risk difference is below a certain threshold (non-inferiority margin) then the experimental treatment is considtered non-inferior.

 

Risk difference and non-inferiority are not mutually exclusive. (Common) risk difference is the outcome, non-inferiority is hypothesis, rather than the standard scenario where this would be superiority (a comparison against the null of no difference).

SAS Innovate 2025: Save the Date

 SAS Innovate 2025 is scheduled for May 6-9 in Orlando, FL. Sign up to be first to learn about the agenda and registration!

Save the date!

What is ANOVA?

ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 4 replies
  • 1310 views
  • 1 like
  • 2 in conversation