Posting: # 1245
What is the importance of Null ratio (=test/reference) in sample size calculations?
It should be 1 (test/reference=1) or 0.95 or 1.05 (5% variability in test/reference).
I am using paried equivalence test in SAS for sample size calculation where it asked for null ratio. Here I am confused whether to use 1 or 0.95 or to be consider actual variability (geometric mean ratio of test/ref untransformed data) received from the reference/pilot data.
I gone through many guidelines but none has mentioned clearly about null ratio. ANVISA says take difference 0% or 5%. USFDA suggested to consider 5% variability. pls guide what is appropriate?
Is it appropriate to consider actual variability as a null ratio
Posting: # 1266
» What is the importance of Null ratio (=test/reference) in sample size calculations?
Since power curves become very steep if moving the test/reference ratio away from 1 (for examples see this post) it's quite important to use reasonable estimates (of both the expected deviation of test from reference and CVintra).
» I gone through many guidelines but none has mentioned clearly about null ratio. ANVISA says take difference 0% or 5%. USFDA suggested to consider 5% variability. pls guide what is appropriate?
» Is it appropriate to consider actual variability as a null ratio
The magical number ±5% has its origin in the following reasoning:
Specifications for batch release commonly are set to ±5% of the declared content. If both products show identical BA, you may expect in your study a deviation of ±5% (let’s say, test’s content -2.5% and reference’s content +2.5% of nominal).
Nwakama P, Haidar S, Yang Y, Davit B, Conner D, Yu L. Generic Drug Products Demonstrate Small Differences in Bioavailability Relative to the Brand Name Counterparts: A Review of ANDAs Approved 1996 – 2004. AAPS J. 2005;7/S2:Abstract M1262. online abstract
... report from 1411 studies submitted to the FDA deviations of 3.13% (±2.71) for AUCt, 3.05% (±2.62) for AUCinf, and 4.52% (±3.57) for Cmax.
A more recent metaanalysis of 1636 BE studies submitted to the FDA within 1996-2005 showed deviations of 3.19% (±2.72) for AUCt, 3.12% (±2.66) for AUCinf, and 4.50% (±3.57) for Cmax.
If you have no clue, expect the true ratio to be 1 and want to be very conservative you may even go with an expected deviation of ±10%.
Power curves are not symmetrical; if you want to go with a 5% deviation of test from reference and have no clue about which direction the deviation will point to, it's wise to use a ratio of 0.95 instead of 1.05. Power is lower for ratios <1 than for ratios >1; therefore you should apply the most conservative approach.
At least some guidelines are specific:
Canada calls for a sample size estimation based on the expected deviation, but wants to see two evaluations - one with original data and one with data corrected for actual content of study batches.
The European guideline calls for a sample size estimation based on the expected deviation; a potency correction is currently under discussion for the next revision.
The WHO’s guideline also suggests the expected deviation; a dose correction (which was suggested in earlier versions) is currently not recommended.
As guidelines are ambiguous in their recommendations about potency correction, some people opt for this procedure:
The quality of responses received is directly proportional to the quality of the question asked. ☼