A non-inferiority study is a controlled clinical trial intended to show that a new treatment is not worse than an active comparator treatment by more than a pre-specified clinically relevant margin. This boundary is called the non-inferiority margin. The design is only meaningful if the comparator treatment has a proven effect that can be expected under study conditions.
Objective and prerequisites
A non-inferiority study can be appropriate if a new treatment does not claim a higher efficacy effect, but suggests other relevant advantages, such as better tolerability, easier application or less treatment burden. It does not replace the question of whether there is a benefit compared to no treatment. Rather, it supports an indirect conclusion by comparing the effect against an active reference within an acceptable boundary.
The EMA guideline on the choice of the non-inferiority margin requires a well-founded choice of comparator treatment and margin. Historical data must convincingly prove the effect of the active control compared to placebo or no treatment. At the same time, study population, endpoints, dose, concomitant therapies and conduct must be so similar that this effect is plausibly maintained in the new study. This constancy assumption and assay sensitivity are crucial for an interpretable study.
Non-inferiority margin and analysis
The non-inferiority margin defines the maximum acceptable loss of effect. It is not a threshold retroactively adjusted to the data. Its justification combines the historical efficacy of the active control with clinical judgment about what proportion of this effect must be maintained. Excessively generous margins can lead to a less effective treatment being falsely accepted; repeated comparisons of such treatments can gradually dilute the effect.
The analysis uses an effect measure and confidence interval appropriate for the estimand. Depending on the specified direction, non-inferiority is demonstrated if the unfavorable boundary of the interval does not exceed the pre-defined margin. ICH E9(R1) requires the clinical question, intercurrent events and sensitivity analyses to be planned coherently. In non-inferiority studies, various relevant analysis populations and assumptions regarding protocol adherence or missing data should be tested, because biases towards a lack of difference can facilitate the proof.
The presentation of the results should therefore immediately reveal the effect measure, direction of the margin, confidence interval, analysis population and assumptions. A p-value without these details does not sufficiently answer the clinical and methodological question. Deviating results between primary and sensitivity analyses require a cautious interpretation.
Differentiation from superiority and equivalence study
A superiority study tests whether a treatment is superior to the comparator treatment in the pre-defined direction. It thus answers a different question than a non-inferiority study. A statistically non-significant difference in a superiority study does not prove non-inferiority, because margin, design, control effect and analysis must be specifically planned for this.
An equivalence study requires that the difference in both directions lies within a pre-specified interval. It thus tests both against a clinically relevant inferiority and against a clinically relevant superiority. The non-inferiority study is one-sided: It only excludes an excessively large disadvantage of the new treatment. Therefore, non-inferiority is not synonymous with equality or equivalence. Superiority study and equivalence study remain separate terms.
Relevance for clinical trials
In study planning, clinical objective, active comparator treatment, endpoint, margin, sample size and analysis method must be brought together early on. The protocol and statistical analysis plan should comprehensibly record the historical evidence, the assumptions regarding the efficacy of the control, the analysis populations and sensitivity analyses. Missing data, deviations from the treatment regimen and change of therapy can particularly influence the interpretation and must be planned for the chosen estimand.
Full-service CROs such as Mediconomics support the methodological justification of comparator treatment and non-inferiority margin, the preparation of protocol and statistical analysis plan as well as the sample size planning and programming of the confidence interval analyses. Biostatistics, data management, monitoring and medical writing work together to consistently document assumptions, data quality, sensitivity analyses and result presentation.
Frequently Asked Questions (FAQ)
Why does a non-inferiority study need an active control?
The conclusion refers to the difference to the established comparator treatment. Its effect compared to placebo or no treatment must be known from suitable historical evidence so that the maintained effect can be contextualized.
Does non-inferiority mean that the treatments are equally effective?
No. It only means that a disadvantage beyond the defined margin has been excluded. An equivalence test makes a different, two-sided demand.
Can superiority be tested subsequently?
This can be possible if a corresponding testing strategy with control of the type I error was defined in advance in the protocol and analysis plan. A retroactive reinterpretation of the objective is not sufficient.
Regulatory references
- EMA/CPMP/EWP/2158/99 “Guideline on the Choice of the Non-Inferiority Margin” – describes the justification of margin and comparator treatment.
- ICH E10 “Choice of Control Group and Related Issues in Clinical Trials” – deals with the informative value of active control groups and assay sensitivity.
- ICH E9 “Statistical Principles for Clinical Trials” – contains statistical principles for controlled clinical trials.
- ICH E9(R1) “Addendum on Estimands and Sensitivity Analysis” – connects estimand, intercurrent events and sensitivity analyses.
- EMA/CPMP/EWP/1776/99 Rev. 1 “Guideline on Missing Data in Confirmatory Clinical Trials” – deals with missing data in confirmatory studies.