# No clear difference is not the same as demonstrated equivalence

Draft • Research checked 4 October 2026 • AI-assisted DomDNA editorial content. No independent clinical review or publication is claimed. General education, not individual medical advice.

Two interventions produce similar-looking results, and a summary calls them equally effective. That conclusion may go beyond what the study tested.

A conventional superiority analysis can fail to detect a difference for several reasons, including imprecision. Non-inferiority and equivalence designs ask different, deliberately specified questions. Their conclusions depend on margins and uncertainty, not merely on the absence of a significant result.

## The question changes the test

A superiority trial asks whether one option performs better on a specified outcome. A non-inferiority trial asks whether an option is not worse than its comparator by more than a chosen margin. An equivalence trial asks whether differences fall within specified bounds in both directions.

None of these means that the two options are identical in every respect. They concern a defined outcome, population and analysis.

[FDA's final non-inferiority guidance](https://www.fda.gov/regulatory-information/search-fda-guidance-documents/non-inferiority-clinical-trials) addresses the design and interpretation of these comparisons. A margin has to be justified; it cannot be selected after seeing the result merely to make the preferred conclusion fit.

## An invented scheduling comparison

Imagine a hypothetical study compares a simpler appointment system with an established system. The average attendance difference is close to zero, but the uncertainty is wide enough to include a meaningful loss.

A superiority analysis that does not find a clear difference cannot establish that the simpler system preserves attendance closely enough. The study may simply be unable to distinguish several practically different possibilities.

Now imagine a prospectively designed non-inferiority study states how much loss would be acceptable for its particular service question and explains why. Its analysis examines the result and uncertainty against that margin.

The example deliberately supplies no numerical acceptable loss. These are fictional service comparisons, not clinical thresholds or a recommendation for how much reduced effectiveness someone should accept.

## Read the margin in units

When a report claims non-inferiority, locate the margin and the outcome to which it applies. A percentage-point margin is not the same as a relative percentage, and a margin for one outcome does not establish equivalence for every benefit and harm.

Also check which direction indicates worse performance. For some measures higher is better; for others lower is better. An unexplained “within the margin” statement can be confusing without that orientation.

The [Piaggio and colleagues reporting extension](https://pubmed.ncbi.nlm.nih.gov/23268518/) addresses non-inferiority and equivalence reporting. Published in 2012, it is a design-specific extension that predates the 2025 base CONSORT update. Its presence in a citation list does not prove that the chosen margin was clinically defensible.

## Regulatory documents have versions too

The [EMA page on these comparisons](https://www.ema.europa.eu/en/scientific-guidelines/non-inferiority-equivalence-comparisons-clinical-trials-0), checked for this article, labels its newer document as draft and shows a closed consultation. That status matters. A closed consultation is not the same as an adopted final guideline.

This article uses the page as evidence of the current document status, not as binding policy or an authority for a personal acceptable-loss rule. The distinction is useful whenever an article cites “new guidance” without naming whether it is draft, final or superseded.

Official guidance also addresses particular research settings. It does not decide which trade-offs a person should make between interventions.

## Separate preserved benefit from other advantages

An option may offer convenience, lower cost or a different burden. Those possible advantages do not erase uncertainty about the outcome being compared. They should be described separately and supported by their own measurements.

Likewise, demonstrating non-inferiority on the main outcome does not prove that adverse effects, implementation effort or long-term outcomes are equivalent. A broad “just as good” headline often removes those boundaries.

## One practical next step

When saving a comparison, write the design question explicitly: superiority, non-inferiority or equivalence. Then add the stated margin and the uncertainty interval, if the report provides them.

If the study only reports no statistically significant difference, keep the conclusion narrow. “No clear difference detected in this analysis” is more faithful than “the options are equal”.

This distinction prevents a quiet inversion of evidence. A study that was too imprecise to show a difference has not thereby shown that important differences are absent. The design has to ask the intended question, and the results have to support its answer.

For an educational starting point, the [DomDNA lifestyle quiz](https://domdna.com/quiz) records lifestyle answers, not DNA analysis or a clinical diagnosis.