A Few Pearls for Noninferiority Trials

March 22-23, 2016

Objectives
Describe the null hypothesis and alternative hypothesis specific to noninferiority trials.
Define the purpose of noninferiority trials.
List reasons for conducting noninferiority trials.
Define and recognize noninferiority margins.
Interpret the meaning of confidence intervals.
Explain the implications of using an ITT or PP analysis population in an RCT compared to in a
noninferiority trial.

Hypotheses
Null hypothesis difference; not noninferior
Alternative hypothesis no difference; noninferior (i.e., no worse)
NOTE: not noninferior does not mean inferior

Test Types of Interventional Studies

Superiority Is one therapy better than another? Most common.
Equivalency Are these therapies similar with little difference (no better and no worse)? Look for
an upper and lower threshold of equivalence (range of little clinically meaningful difference)
Noninferiority Is one therapy no worse than another? Look only for a lower threshold of
acceptance (noninferiority margin). Check to see if you agree with margin.

Reasons for Noninferiority Trials

Placebo control is unethical.
Potential risk of decreased efficacy is worth it (e.g., more convenient in route, frequency, and
monitoring).

Interpreting Confidence Intervals

Line of no difference
Set at 0 for continuous data endpoints (e.g., change in weight or in blood pressure)
Set at 1 for ratio endpoints (e.g., RR, OR, HR)
If the 95% CI of the point estimate crosses the line of no difference, then the result is not statistically
significantwhich is acceptable for the purposes of noninferiority trials.
If the 95% CI of the point estimate is above or below the line of no difference, then the results shows a
statistically significant difference (i.e., superior or inferior depending on which side favors
intervention, but this is not the main purpose of a noninferiority trial.)

Noninferiority margin lowest acceptable treatment effect from the investigational drug
If the 95% CI of the point estimate is within the noninferiority margin, then the results show
noninferiority.
If the 95% CI of the point estimate crosses the noninferiority margin, then the results did not show
noninferiority (i.e., results are not noninferior).
Potential Issues
Noninferiority trials do not assess drug effectiveness just its similarity to control. But, the
control may be less effective than it used to be, which may make it easier to prove
noninferiority. Therefore, not only consider the noninferiority margin, but also consider the
comparator.
A larger N may be required for noninferiority studies since the difference between the 2
treatment groups may be smaller than the treatment difference from a superiority study.
ITT can produce false noninferiority results. A well-conducted noninferiority trial will show
that both ITT and PP analyses demonstrated noninferiority.