Non-inferiority trials test whether a new treatment is not meaningfully worse than an established active control, rather than whether it is better than placebo. The null hypothesis is that the new treatment is inferior by more than a pre-specified non-inferiority margin (delta); rejecting this null provides evidence that the new treatment preserves at least some of the active control's benefit. The non-inferiority margin represents the largest clinically acceptable loss of efficacy — often a fraction (e.g., 50%) of the active control's established effect over placebo. These trials are appropriate when placebo control is unethical (an effective treatment exists), and the new treatment offers advantages in safety, cost, convenience, or adherence. The per-protocol analysis is typically primary (unlike superiority trials where intention-to-treat is primary) because ITT analysis in non-inferiority trials is anti-conservative — non-compliance and treatment crossover bias toward finding non-inferiority.
Most clinical trials ask: "Is the new treatment better than placebo (or the current standard)?" But sometimes the relevant question is different: "Is the new treatment at least as good as what we already have?" If a new antibiotic has fewer side effects, costs less, or can be taken orally instead of intravenously, it would be valuable even if its cure rate were slightly lower — provided the difference is not clinically meaningful. Non-inferiority trials address this by testing whether the new treatment's efficacy falls within an acceptable margin of the active control.
The statistical setup inverts the usual null and alternative hypotheses. The null hypothesis is that the new treatment is inferior to the standard by more than the non-inferiority margin delta. The alternative is that the difference is within the margin. Non-inferiority is demonstrated when the confidence interval for the treatment difference excludes the margin — specifically, when the lower bound of the confidence interval (for a beneficial direction) is above -delta. This is a one-sided test: you are testing only for inferiority, not for superiority.
The non-inferiority margin is the single most important design parameter. It must satisfy two constraints. First, it should be clinically meaningful — a margin of 30 percentage points for a life-saving treatment is absurdly permissive, while 0.1 percentage points is impractically strict. Second, it should be small enough that a drug passing the non-inferiority test is still demonstrably better than placebo. This requires knowledge of the active control's effect over placebo from prior trials. If the active control reduces mortality by 10% versus placebo, a non-inferiority margin of 10% would allow the new drug to be equivalent to placebo — clearly unacceptable. Regulatory guidelines typically require the margin to preserve at least 50% of the active control's historical benefit.
A subtle but important feature of non-inferiority trials is the reversal of the ITT principle. In superiority trials, ITT analysis is conservative because treatment contamination and non-compliance dilute the true difference, making it harder to reject the null. In non-inferiority trials, dilution works the opposite way: it makes the treatments appear more similar, biasing toward a finding of non-inferiority. A trial where many patients in the new drug group switched to the standard treatment would appear non-inferior simply because the groups received similar treatments. For this reason, the per-protocol analysis (restricted to patients who complied with the protocol) is the primary analysis, and non-inferiority should ideally be demonstrated in both ITT and per-protocol populations.
No topics depend on this one yet.