"Noninferior Means It's No Better": Appraising a Noninferiority Trial for a 2-Monthly Antipsychotic

Yes - noninferiority was met, and that is not the same as "no better."

• The noninferiority margin is a pre-specified threshold representing the largest loss of effect that would be clinically acceptable relative to the comparator (here, HR 1.30).
• The upper bound of the 95% CI, 1.28, is below 1.30. Therefore the data are statistically compatible with X retaining an acceptable share of the comparator's effect - noninferiority is demonstrated.
• The point estimate (0.89) is numerically favourable to X, but the CI crosses 1.0 (0.62–1.28), so superiority is not established. Do not present this as "X reduced relapse."

What it does not establish:

• No claim of greater efficacy.
• No information on populations excluded from the trial.
• Noninferiority trials are sensitive to assay sensitivity - if the comparator was underdosed or adherence was poor, "noninferior" can mean "equally mediocre." Check the comparator dose and level/adherence data.
• Analysis set matters: for noninferiority, the intention-to-treat analysis is the conservative one (dropouts dilute differences), and per-protocol results must also be shown and consistent.
• With 18% lost to follow-up, missing-data handling (how were dropouts counted for relapse?) must be scrutinized; TI or worst-case imputation can flip the conclusion.

Answer to the colleague: noninferiority is exactly the result a convenience product is designed to show. The decision is about the dosing interval, not about comparative efficacy.

Absolute numbers:

• Risk with comparator: 22.3%; risk with X: 20.1%.
• Absolute risk reduction = 22.3% − 20.1% = 2.2 percentage points.
• NNT = 1 / 0.022 ≈ 45 - about 45 patients treated with X for 12 months to prevent one additional relapse.
• A 95% CI around that difference spans from roughly −1.4 to +5.8 percentage points (i.e., it includes zero), consistent with the hazard ratio CI crossing 1.

What this adds:

1. A statistically noninferior result can still hide a clinically meaningful absolute difference in either direction - and here the CI's lower bound suggests X could plausibly be slightly worse.
2. The NNT of ~45 is modest; the real value proposition of X is six injections per year instead of twelve, fewer clinic visits, and potentially better adherence - not fewer relapses.
3. Contextualize the baseline risk: a 22% annual relapse rate implies a relatively stable, adherent, mostly outpatient population. If your patient's untreated relapse risk is much higher, the same relative effect maps to a larger absolute benefit - or to a larger absolute loss.
4. Also weigh harms: 31% vs 29% discontinuation means roughly a third of patients stopped either drug within a year - a reminder that tolerability, not efficacy, often drives outcomes.

Bottom line to communicate: "The drugs performed similarly on relapse; X offers a longer interval. That is the argument - and it is a legitimate one if the interval solves a real adherence problem."

Who in your practice actually resembles this trial?

• Eligible: adults with schizophrenia, stable enough to consent and complete a 12-month trial, without active substance use disorder, and able to attend visits for injection.
• Your frequently relapsing patient with daily cannabis use and a 90-minute commute would have been excluded - the trial says nothing directly about him, and that is precisely the patient for whom the interval argument is most attractive.

How to reason about the extrapolation honestly:

1. There is no biological reason the drug would work differently in a cannabis user; the uncertainty is about adherence and follow-up, not mechanism.
2. The trial's own advantage (fewer visits) is largest exactly where trial conditions (frequent visits, research nurse contact) are absent - so the real-world benefit may be larger than the trial shows, but also less well measured.
3. Practical unknowns: cost/coverage, cold-chain handling, clinic capacity to give a 2-monthly depot, and whether the manufacturer's patient-support program creates a hidden switching incentive.

Communicating with the colleague:

• Correct the misreading: noninferiority is a finding of acceptable efficacy, not proof of equivalence or inferiority.
• Reframe the question: "Given similar relapse rates, which product is this patient more likely to keep taking for a year?" For a patient who has already relapsed after stopping oral therapy and misses clinic visits, halving the visit burden is a plausible adherence intervention.
• Acknowledge the counter-arguments: manufacturer funding, open-label elements in scheduling, 18% attrition, and a wide CI that includes harm.
• Propose a decision rule: if the patient's barrier is visit frequency or transport, the 2-monthly product is reasonable; if the barrier is cost, coverage, or tolerability, it is not. That is a patient-level judgment the trial cannot make for you.
• Suggest auditing your own clinic's on-time injection rate after switching - the policy question is testable locally.