Non-inferiority trials
From TrialTree Wiki
Non-Inferiority Trials
A non-inferiority trial is designed to determine whether a new intervention is not unacceptably worse than an existing standard treatment, based on a predefined threshold known as the non-inferiority margin. These trials are especially important when the new treatment may be slightly less effective but offers other clinical or logistical advantages—such as fewer side effects, lower cost, or easier administration.
Defining the Non-Inferiority Margin (Δ)
The non-inferiority margin (Δ) is the largest difference from the standard treatment that would still be considered clinically acceptable. This margin must be carefully chosen based on prior research and expert judgment. It should be smaller than the established effect size of the standard treatment to ensure the new treatment retains a meaningful portion of the benefit.
Example: If the standard treatment reduces mortality by 10%, a Δ of 3% means the new treatment would be considered non-inferior if it reduces mortality by at least 7%.
Choice of Control Group
A non-inferiority trial must use an active comparator with proven efficacy—placebos are generally not appropriate. The comparator should be a treatment with established benefits from prior randomized controlled trials, ideally reflecting the current standard of care.
Study Design Considerations
As with other trial types, randomization is used to minimize bias and confounding. Blinding is strongly preferred to prevent performance or measurement bias, though it may not always be feasible in all clinical settings.
Parallel-group designs are most common in non-inferiority trials, particularly for pharmaceutical studies. Cross-over designs may be used in cases where treatment effects are short-term and reversible, with no expected carryover between periods.
Sample Size Calculation
Non-inferiority trials generally require larger sample sizes than superiority trials. This is because they must be powered to detect small differences within a narrow margin. Typical power levels range from 80% to 90%, and statistical significance is usually assessed using a one-sided alpha of 2.5%, as opposed to the traditional two-sided 5% used in superiority designs.
Statistical Analysis
The primary analysis involves estimating the confidence interval (CI) for the difference between treatments. Non-inferiority is concluded if the entire CI falls within the predefined margin (Δ).
Example:
- Difference in efficacy = -2%
- 95% CI = (-4%, 0%)
- Δ = 5%
Since the upper bound (0%) is within Δ, the new treatment is considered non-inferior.
Both per-protocol (PP) and intention-to-treat (ITT) analyses should be reported. The PP analysis is often emphasized because it better reflects the true efficacy among adherent participants. However, ITT analysis is more conservative and should not contradict the findings from PP analysis.
Handling Missing Data
Missing data can bias results and lead to erroneous conclusions of non-inferiority. Appropriate methods such as multiple imputation or worst-case scenario analyses should be used to minimize this risk. Sensitivity analyses are also recommended to assess the impact of missing outcomes.
Ethical Considerations
Ethical justification is essential when choosing a non-inferiority design. Investigators must explain why demonstrating “not worse than” is clinically meaningful, and ensure that the active control is the best available standard. Continuous monitoring for adverse effects is especially important when the goal is to replace a well-established treatment.
Interpretation of Results
Once non-inferiority is established, additional benefits of the new treatment—such as safety, tolerability, or cost-effectiveness—should be evaluated. If the CI exceeds the non-inferiority margin, the treatment cannot be declared non-inferior. In some cases, if non-inferiority is demonstrated, a follow-up superiority analysis may be conducted to test whether the new treatment is actually better.
Reporting
Results from non-inferiority trials should be reported in accordance with the CONSORT extension for non-inferiority and equivalence trials. The predefined non-inferiority margin (Δ), analysis populations, confidence intervals, and all relevant results should be clearly described.
Example conclusion statement: "The new drug was non-inferior to the standard treatment, with a difference of -2% (95% CI: -4%, 0%), within the predefined non-inferiority margin of 5%. Given its improved safety profile, it may be considered a viable alternative."
Conclusion
Non-inferiority trials are a valuable tool for evaluating new interventions that may offer advantages beyond efficacy. However, they require rigorous design, thoughtful interpretation, and robust analysis. The non-inferiority margin should be predefined based on clinical judgment, and both ITT and PP analyses should be reported to support the findings. Proper handling of missing data and clear communication of the results are essential to avoid false claims of equivalence.
See also: