Thread by @drjohnm

Need help from trial methods people. I ran across this amazing paper by @phlegmfighter et al looking at consequences of recommendations in the design and interpretation of Non-inferiority trials. pubmed.ncbi.nlm.nih.gov/28875400/
THEY SUGGEST SYSTEMIC BIASES.
Thread

If you put the new treatment on the left side of the interpretative diagram, there are four ways to make Non-inferiority. Scenarios 1-4. but....

They show that if the convention were to put the active control on the left side of the diagram, make the new treatment the control, you would bias strongly toward the control (thus making NI more difficult to reach). No change in data, just the mirror image here:

Now you could say the new treatment deserves that the left side of the diagram because it's 'new' but the authors give multiple examples of where this is not exactly true.

it seems to me that merely by convention, NI trials favor the new treatment. Indeed that is what their systematic review found: Nearly 80% of the 163 trials they looked at met NI. If you count 13% superior then only 7% miss the mark.
Are NI rigged to give positive results?

@VPrasadMDMPH has a great editorial pubmed.ncbi.nlm.nih.gov/28875400/
cc here @ADAlthousePhD @AndrewFoy82 @StatModeling @f2harrell @stephensenn @methodsmanmd @venkmurthy @AndrewFoy82 @rwyeh