Noninferiority margins and intention-to-treat: reading the CI against the margin

Study cases written by our editors, not BPS items. The real BCPS exam is computer-based multiple choice, answered in order with no going back.

1. Case 01

   A noninferiority trial compares an oral step-down antibiotic with continued IV therapy for complicated urinary tract infection. The prespecified noninferiority margin is −10%. Clinical cure was 84.0% with oral therapy and 86.5% with IV therapy, a difference of −2.5% (95% CI −12.0% to 7.0%). Which conclusion is correct?

   1. AOption A: Noninferior, since the point estimate lies inside the margin
   2. BOption B: Equivalent, since the confidence interval includes zero
   3. COption C: Inferior, since the lower bound extends past the margin
   4. DOption D: Noninferiority is not shown, so the result is inconclusive

   Pick an option: the verdict and a note on every option open here.

   Best answer: D.

   In a noninferiority trial, the new treatment is noninferior only if the entire confidence interval of the difference (new minus standard) lies above the prespecified margin. Here the lower bound, −12%, is below −10%, so noninferiority is not shown. The interval also includes zero and positive values, so inferiority is not proven either: the trial is inconclusive. A point estimate inside the margin is not enough. A CI that includes zero is not equivalence, which needs the whole CI inside a two-sided margin. Sources: CONSORT noninferiority extension (2012) and the FDA Non-Inferiority Clinical Trials guidance (2016).

   A

   Noninferiority is judged on the whole confidence interval, not the point estimate. Its lower bound (−12%) crosses −10%.

   B

   A CI that includes zero does not show equivalence. Equivalence needs the whole CI inside both margins.

   C

   Inferiority is not shown either: the CI still includes zero and values above it.

   D

   Correct. The lower bound (−12%) falls below the −10% margin, so noninferiority is not demonstrated, and inferiority is not proven.

2. Case 02

   A noninferiority trial compares a new antibiotic with vancomycin for acute bacterial skin and skin-structure infection, using a prespecified margin of −10%. The difference in clinical response (new minus vancomycin) is −3.5% (95% CI −8.2% to 1.2%). Which conclusion is most appropriate?

   1. AOption A: Inconclusive, because the confidence interval crosses zero
   2. BOption B: Inferior, because the point estimate favors vancomycin
   3. COption C: Superior, because the lower bound lies above −10%
   4. DOption D: Noninferior to vancomycin, but not shown to be superior

   Pick an option: the verdict and a note on every option open here.

   Best answer: D.

   Noninferiority is shown when the lower bound of the CI (−8.2%) lies above the prespecified margin (−10%), and superiority would require the whole CI to lie above 0, which it does not. A point estimate slightly favoring the comparator does not make the new drug inferior. A CI crossing zero matters for a superiority test, not for the noninferiority conclusion.

   A

   Crossing zero matters for a superiority test; noninferiority is judged against the −10% margin.

   B

   A point estimate slightly favoring vancomycin does not make the new drug inferior when the whole CI stays above the margin.

   C

   Superiority would need the entire CI above 0, and it reaches down to −8.2%.

   D

   Correct. The lower bound of −8.2% lies above the −10% margin, so noninferiority holds, but the CI includes 0, so superiority is not shown.

3. Case 03

   A sponsor plans a noninferiority trial of a new anticoagulant against warfarin for stroke prevention in atrial fibrillation. Historical placebo-controlled trials show that warfarin substantially reduces stroke. A conservative estimate of that effect (from the confidence bound closest to no effect) is taken as M1, the whole effect warfarin is assumed to have in the new trial. How should the noninferiority margin be chosen?

   1. AOption A: Prespecify a fraction of M1, so part of warfarin’s effect is kept
   2. BOption B: Set it equal to M1, since beating placebo is all that is required
   3. COption C: Choose it after unblinding, based on the observed event rates
   4. DOption D: Make it as wide as needed to keep the sample size feasible

   Pick an option: the verdict and a note on every option open here.

   Best answer: A.

   The FDA Non-Inferiority Clinical Trials guidance (2016) defines M1 as the entire effect of the active control assumed to be present in the new trial, estimated conservatively from historical placebo-controlled data. Ruling out M1 only shows that the new drug has some effect over placebo. The clinical margin, M2, is the largest loss of effect that is clinically acceptable. It is usually a fraction of M1, so part of the control’s benefit is preserved; this matters for a serious outcome like stroke. Both margins must be prespecified, and the CONSORT noninferiority extension (2012) asks reports to state the margin and its rationale. Margins chosen after unblinding or to keep the sample size small are invalid.

   A

   Correct. The margin (M2) is fixed in advance and is smaller than M1, so a clinically important share of warfarin’s effect is preserved.

   B

   Ruling out M1 only shows the drug beats placebo. For stroke, losing all of warfarin’s benefit is not clinically acceptable.

   C

   A margin picked after seeing the data invites bias and cannot support a noninferiority claim; it must be prespecified.

   D

   Widening the margin for feasibility lets a clearly worse drug pass as noninferior. The margin rests on clinical and historical evidence.

4. Case 04

   A noninferiority trial compares a new drug with a standard drug for treatment success, with a prespecified margin of −8%. By 6 months, 18% of patients in each arm had stopped their assigned drug or crossed over to the other one. The intention-to-treat (ITT) difference is −3.0% (95% CI −7.5% to 1.5%). The per-protocol difference is −4.0% (95% CI −10.2% to 2.2%). Which conclusion is best supported?

   1. AOption A: Noninferior, because ITT is the conservative primary analysis
   2. BOption B: Noninferior, because per-protocol results are biased and set aside
   3. COption C: Not established, because ITT can bias toward finding no difference
   4. DOption D: Inferior, because the per-protocol lower bound crosses the margin

   Pick an option: the verdict and a note on every option open here.

   Best answer: C.

   In a superiority trial, ITT analysis is conservative: crossovers and stopped drugs dilute the difference between arms and make it harder to show superiority. In a noninferiority trial the same dilution works the other way. It pulls the arms together and makes a false noninferiority claim easier. That is why the CONSORT noninferiority extension (2012) and the FDA Non-Inferiority guidance (2016) call for both ITT and per-protocol analyses, and noninferiority is persuasive only if both agree. Here the ITT CI stays above −8%, but the per-protocol lower bound (−10.2%) does not, so noninferiority is not established. Inferiority is not shown either, because both CIs include zero.

   A

   ITT is conservative for superiority, not for noninferiority. With many crossovers it pulls the arms together and favors noninferiority.

   B

   Per-protocol analysis has its own biases, but in a noninferiority trial it is needed as a check, not discarded.

   C

   Correct. The per-protocol lower bound (−10.2%) crosses −8%, and the two analyses disagree, so noninferiority is not established.

   D

   Crossing the margin means noninferiority is not shown. It does not prove inferiority, since the CI still includes zero.

5. Case 05

   A trial compares a new antifungal with a standard azole for invasive candidiasis. The protocol prespecifies a noninferiority margin of −10% for global response and a fixed testing order: noninferiority first, then superiority. In the ITT population, the difference in response (new minus standard) is +6.0% (95% CI +1.4% to +10.6%). Which conclusion is correct?

   1. AOption A: Noninferior only; a noninferiority trial cannot claim superiority
   2. BOption B: Noninferior, and the prespecified superiority test is also met
   3. COption C: Equivalent, since the difference lies within ±10% of zero
   4. DOption D: Superiority is suggested but needs a new trial to be claimed

   Pick an option: the verdict and a note on every option open here.

   Best answer: B.

   The whole CI (+1.4% to +10.6%) lies above the −10% margin, so noninferiority is shown. It also lies entirely above zero. The CONSORT noninferiority extension (2012) states that, once noninferiority is shown, it is acceptable to test for superiority if that test is prespecified and done in the ITT population. A fixed order (noninferiority first, then superiority) needs no extra alpha adjustment. So both claims hold. Superiority could not be claimed if the CI had crossed zero, for example −2% to +5%. Noninferiority is also not equivalence: equivalence needs a two-sided margin with the whole CI inside it.

   A

   Superiority can be claimed after noninferiority is shown, if the test was prespecified and done in the ITT population.

   B

   Correct. The lower bound (+1.4%) is above −10% (noninferior) and above 0 (superior), and the stepwise test was planned in advance.

   C

   Not inferior is not the same as equivalent, and the upper bound (+10.6%) lies outside a ±10% range anyway.

   D

   No new trial is needed: a prespecified, hierarchical superiority test in the same trial is valid once noninferiority is shown.

6. Case 06

   A noninferiority trial compares a new oral anticoagulant with a standard drug for preventing venous thromboembolism after knee replacement. The −10% margin was based on placebo-controlled trials of the standard drug. In the new trial, however, the standard drug was given at half the dose used in those placebo-controlled trials. Protection from VTE (new minus standard) differed by −1.5% (95% CI −6.0% to 3.0%). Which conclusion is best supported?

   1. AOption A: Not reliable, since the underdosed control may lack its usual benefit
   2. BOption B: Noninferior, since the lower bound of −6.0% lies above the −10% margin
   3. COption C: Superior to the standard drug, since the interval reaches above zero
   4. DOption D: Inconclusive, but only because the confidence interval includes zero

   Pick an option: the verdict and a note on every option open here.

   Best answer: A.

   A noninferiority margin is built from the effect the control drug showed against placebo in earlier trials. The FDA guidance on noninferiority trials (2016) calls the belief that this effect still holds in the new trial the constancy assumption. It lists the dose of the control drug among the design features that must match the earlier trials. Here the control was given at half its studied dose, so its true effect may be smaller. Then the margin is too wide, and a trial that seems to show noninferiority could be wrong: the new drug might work no better than an underdosed control. The interval does sit above the margin, but the conclusion is not reliable.

   A

   Correct. The margin assumes the control works as well as it did in the trials behind it. At half that dose its effect may be smaller, so the result may mislead.

   B

   The numbers do pass the margin. But the margin only means something if the control drug keeps its historical effect, and the lower dose breaks that.

   C

   Superiority needs the whole interval above zero. This interval runs from −6.0% to 3.0%, so superiority is not shown.

   D

   Crossing zero is expected in a noninferiority trial and does not make it inconclusive; the interval is judged against the margin, not zero.