Non inferiority study


A study that investigates whether one drug is not inferior to another (noninferiority study) can conclude that the drug is superior?

"Clinically significant cancer was detected in 95 men in the MRI-targeted biopsy group, as compared with 64 of 248 in the standard-biopsy group (adjusted difference, 12 percentage points; 95% confidence interval [CI], 4 to 20; P=0.005). It, was noninferior to standard biopsy, and the 95% confidence interval indicated the superiority of this strategy over standard biopsy"

Thank you!


Not a robit
I got a feeling the authors didn't perform a noninferiority test, just interpreted the 95% CIs that way. For a inferiority test you perform a one-sided test. Look at their methods section.
In material and methods exposes that if the lower boundary was greater than zero, superiority would be claimed. It´s correct?

"Using a noninferiority margin of 5 percentage
points that was agreed on at an expert consensus
group meeting10 and a one-sided alpha level
of 2.5%, we calculated that the randomization of
422 men would provide the trial with 90%
power to show the noninferiority of MRI, with
or without targeted biopsy, to standard biopsy,
assuming a detection rate of clinically significant
cancer of 40% in the group that underwent
MRI, with or without targeted biopsy, and 30%
in the standard-biopsy group.

The statistical analysis plan was prespecified
and approved by the data and safety monitoring
committee before the analysis of any data. For
the primary outcome, if the lower boundary of
the two-sided 95% confidence interval for the
difference in the rates of detection of clinically
significant cancer in the MRI-targeted biopsy
group relative to the standard-biopsy group was
greater than −5 percentage points, then MRI,
with or without targeted biopsy, would be
deemed to be noninferior. Furthermore, if the
lower boundary was greater than zero, superiority
would be claimed. The difference was estimated
with the use of a generalized linear mixed
model (with the use of an identity link function
with a binomial distribution) that included trial
center as a random effect"