Non-inferiority margins: how they are chosen and how they are abused — a second dataset — setting out what I have, and where I think it stops being reliable.
Posting a small dataset on Non-inferiority margins. It is mine, it is uncontrolled, and the method is stated so it can be discounted appropriately.
What I would like is not agreement but a second dataset collected by someone with no stake in mine. If one exists I would rather read it than argue for this one.
Non-inferiority margins is one of those subjects where the general answer and the answer for a specific case diverge, and the thread will go in circles until someone says which one is being asked for.
Risk of bias: structured appraisal of internal validity. Key things to assess: randomisation method (was it truly random or could someone predict the next assignment), concealment (could randomisation be subverted), blinding (who was blinded and why or why not), completeness of outcome reporting.
I would rather say I do not know than round it up to an answer.
Before the thread moves on from Non-inferiority margins — what is the sample size behind the claim? I am not being difficult; I have seen the same figure quoted from an n of four and from an n of four hundred.
Everything in post #7 holds. The case it does not cover is the one I have.
Absolute and relative effects answer different questions. Write down the event rate in each arm and the difference between them; everything quotable is derived from those two numbers.
Marking that as an opinion rather than a finding.
Narrowing post #7, because the general version has more than one answer.
What would change my mind on Non-inferiority margins is a second dataset collected by someone with no stake in the first. Until then I hold it loosely and I would rather say so than pretend to more.
Picking up post #8: that is the part I would want checked first.
Non-inferiority margins are chosen, and the choice is an argument rather than a fact. A wide margin can make a worse treatment look acceptable.
It is worth stating the boring hypothesis before the interesting one.
Appreciated. The plain phrasing does more work here than a longer post would.
Absolute numbers, not just relative: a 30% relative reduction tells you the ratio but not the practical magnitude. The event rate in each arm and the difference between them tells you how many people benefit.
I have kept the units in throughout, for the obvious reason.
Risk of bias: structured appraisal of internal validity. Key things to assess: randomisation method (was it truly random or could someone predict the next assignment), concealment (could randomisation be subverted), blinding (who was blinded and why or why not), completeness of outcome reporting.
That is a description of practice, not a recommendation of it.
The reason Non-inferiority margins keeps being re-asked is that the answer is conditional and people quote it without the condition. It is not that the answer is unknown.
I read post #14 twice before replying, because I had assumed the opposite.
Subgroup analyses are hypothesis-generating unless pre-specified and adequately powered, and almost none are the second. The interaction test matters more than the subgroup point estimate.
Post #19 describes the usual case. This is about the unusual one.
Two claims get bundled together under Non-inferiority margins and they need separating. The descriptive one — this is what was observed — is usually well supported. The causal one — this is why — usually is not.
Almost every disagreement in threads like this one dissolves once you say which of the two you are making.
On Non-inferiority margins, the part that usually goes wrong is that the question is asked as though it has one answer. It has a range, and the width of the range is the interesting bit.
If you can post the two or three numbers you are working from, several people here will check the arithmetic rather than argue about the conclusion.
A treatment-policy estimand asks what happens to people assigned to a strategy, including those who abandon it. A hypothetical estimand asks what would have happened had everyone continued. Both are legitimate and they give different numbers.
I have no interest in any supplier named above.
Confirming post #23 from a second method, which matters more than confirming it from a second person.
Pre-specification is the property that makes a primary endpoint trustworthy. An endpoint chosen after seeing the data can be the best endpoint in the world and it no longer carries the same guarantee.
Absolute and relative effects answer different questions. Write down the event rate in each arm and the difference between them; everything quotable is derived from those two numbers.
The short version is the first sentence; the rest is why.
Coming back to post #23, because the follow-up matters more than the original answer.
The useful distinction on Non-inferiority margins is between what was measured and what was inferred from it. Both end up in the same sentence and only one of them has error bars.
Adding a data point of agreement rather than a data point.
Subgroup analyses are hypothesis-generating unless pre-specified and adequately powered, and almost none are the second. The interaction test matters more than the subgroup point estimate.
I would want to see it done twice before believing it once.
Narrowing post #27, because the general version has more than one answer.
Non-inferiority margins would be much easier to settle if anyone reported the denominator. Almost nobody reports the denominator.