Noninferiority and equivalence
What question does this help answer?
Whether a treatment that is cheaper, safer, simpler or more tolerable can replace the standard, accepting that it may be slightly less effective.
The margin is the study
The margin is the largest loss of efficacy that would still be acceptable given the new treatment's other advantages. It is set from clinical judgement and from the historical evidence that the active comparator beats no treatment at all, and it must preserve a reasonable fraction of that historical benefit. A margin chosen because it makes the trial feasible is not a margin, and a wide one can license a treatment that is genuinely worse.
How the conclusion is read
Compute the confidence interval for the difference. Noninferiority is concluded when the whole interval lies on the acceptable side of the margin. Equivalence requires the interval to lie inside the margin in BOTH directions, which is a stronger claim and needs a larger study. A conventional p-value against no difference plays no part in the conclusion.
Intervals read against the margin
Requirements that superiority trials do not have
- Assay sensitivity: the trial must be capable of detecting a difference if one exists, or everything is noninferior to everything.
- The active comparator must be given at its proper dose and schedule, since a weakened comparator manufactures noninferiority.
- Both intention-to-treat and per-protocol analyses are reported, and agreement between them is part of the evidence.
- Poor adherence, measurement error and loss to follow-up all push toward the noninferiority conclusion, so trial conduct matters more here, not less.
- A switch to a superiority claim after the fact is permitted; a switch to noninferiority after a failed superiority trial, with a margin chosen afterwards, is not.
How it is sized
Sizing is driven by the margin, not by an expected difference, and a small margin can demand a very large trial. There is no closed form in Mallard's catalog that stands in for this: the honest route is to state the margin-based method and its inputs rather than borrow a two-group superiority formula, which answers a different question.
What to watch out for
A non-significant superiority test does not show equivalence: absence of evidence of a difference is not evidence of no difference, and reading a null trial that way is the single most common error in this area. The margin is the whole study, and it is a clinical judgement that must be justified and fixed in the protocol, not chosen to be achievable. Note also that intention-to-treat is CONSERVATIVE for superiority and ANTI-conservative here, because anything that blurs the groups pushes toward showing noninferiority.
Clinical example
Question: Is a once-daily oral anticoagulant noninferior to a twice-daily one? Design: Fix the margin from the historical benefit of the comparator over placebo and the clinical value of once-daily dosing, then conclude noninferiority only if the entire confidence interval for the difference sits on the acceptable side of it.
- Reporting of noninferiority and equivalence randomized trials: extension of the CONSORT 2010 statement (opens in a new tab) · JAMA · doi 10.1001/jama.2012.87802
- Per-Protocol Analyses of Pragmatic Trials (opens in a new tab) · N Engl J Med · doi 10.1056/NEJMsm1605385