Choosing superiority, noninferiority or equivalence
Choose whether you need evidence of benefit, acceptable loss, or similarity within agreed bounds.
Fit
When it matters, and when it does not
It matters when
- You must decide what claim a treatment comparison needs to support.
- You can justify how much loss or difference would matter to patients.
It matters less when
- For a simple estimate without a decision margin, report the effect and its uncertainty. Read: Confidence intervals and p-values
Before you start
What to check in your study
- Direction of benefitState whether larger or smaller outcomes are better.
- Clinical marginChoose the largest acceptable loss, or both similarity bounds, before looking at results.
- Treatment questionSpecify the outcome, follow-up and handling of treatment changes.
- The margin has a clinical and historical justification.How to check: Check that the active control would still work in the current setting.
- Similarity reflects treatments rather than poor trial conduct.How to check: Examine adherence, missing outcomes and both assigned-group and per-protocol analyses.
Plain language
How it works
Superiority asks whether a treatment is better than its comparator. Noninferiority asks whether any loss is smaller than a justified margin. Equivalence asks whether the whole plausible difference lies within both bounds. None means the treatments are identical.
For a benefit difference, superiority needs the chosen confidence interval above zero. Noninferiority needs its lower bound above the negative margin. Equivalence needs both bounds inside the equivalence limits. Reverse the direction for harmful outcomes. Fix the test and error rate first. Schuirmann describes two one-sided tests; Piaggio and colleagues distinguish the trial claims.
Worked example
A worked example
- Setting
- A team compares cure rates for a simpler treatment.
- Question
- Is a small loss in cure acceptable?
- Approach
- The benefit scale is new minus standard. The margin is 5 percentage points. Assume a normally distributed estimate with standard error 1.5 points.
- Result
- The reference difference is 0 points; the estimate is -1 point, a change of -1. Its 95% interval is -3.94 to 1.94 points.
- What it means
- The interval clears -5, supporting noninferiority under this prespecified rule. It crosses zero, so superiority is not shown. Equivalence requires a separately prespecified two-sided margin and test.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A nonsignificant superiority test proves neither noninferiority nor equivalence. Set clinically justified margins before seeing results; poor adherence can make treatments look misleadingly similar.
- Choosing a margin after seeing the interval.Instead: Justify the margin in the protocol.
- Reading no significant difference as equivalence.Instead: Test the planned bounds with the chosen error rate.
- Trusting assigned-group similarity alone.Instead: Examine adherence and sensitivity analyses for the target question.
Inside Mallard
How Mallard uses this
Mallard includes this comparison in Learn search and the Atlas. Plans do not link to this comparison automatically.
concept.choosing_trial_hypotheses, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- A comparison of the two one-sided tests procedure and the power approach for assessing the equivalence of average bioavailability (opens in a new tab) · J Pharmacokinet Biopharm · doi 10.1007/BF01068419
- 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