Non-Inferiority & Equivalence Trials

A non-inferiority trial asks whether a new treatment retains enough of the standard’s benefit to be acceptable, allowing a small pre-specified loss; equivalence asks whether the two are close in BOTH directions. The verdict hinges on a margin (Δ), chosen in advance, and on where the confidence interval falls relative to it. Used when the new option is cheaper, safer, or easier to give.

Three new-versus-standard estimates against two reference lines: “no difference” and the pre-specified margin Δ. The position of the whole confidence interval — not the dot alone — decides the verdict.

If the entire CI lies on the acceptable side of Δ, the new treatment is non-inferior. If it also clears the “no difference” line in the new treatment’s favour, it is superior. If the CI crosses Δ, non-inferiority is simply not demonstrated — the result is inconclusive, not a proof of harm.

In a non-inferiority trial, intention-to-treat is anti-conservative: by blurring the groups (non-adherence, crossover) it pushes results toward “no difference” — which here means toward declaring non-inferiority. So per-protocol analysis is reported alongside ITT, and the margin must be clinically defensible, or “non-inferior” is meaningless.

• Non-inferior = whole CI on the acceptable side of Δ
• Equivalence = CI within ±Δ (both directions)
• Show ITT and per-protocol; ITT is anti-conservative here

EINSTEIN-PE (NEJM 2012) — 4,832 patients with acute symptomatic pulmonary embolism, oral rivaroxaban vs enoxaparin-plus-vitamin-K-antagonist. Recurrent VTE 2.1% vs 1.8% (HR 1.12, 95% CI 0.75–1.68). The pre-specified non-inferiority margin was an upper CI limit of 2.0; 1.68 lies below it, so rivaroxaban was non-inferior — with less major bleeding (1.1% vs 2.2%) and a single-drug oral regimen. “Non-inferior” did NOT mean “better at preventing clots” — the appeal was an equally effective but simpler, safer regimen. The whole CI had to clear the margin, not just the point estimate.

• Forgetting ITT is anti-conservative in NI trials — it biases toward declaring non-inferiority, so per-protocol matters.
• Accepting an unjustified or over-wide margin Δ without asking how it was chosen.
• Reading “no significant difference” as equivalence — absence of evidence is not evidence of sameness.