Re: st: Sample size for equivalence trials in Stata

[Philip Ryan <philip.ryan@adelaide.edu.au>]

In a recent contribution to this thread Joseph Coveney wrote:

<snip>

> You'd need to check with Philip Ryan to be sure, but -equivsize- looks like
> it's based upon Blackwelder's* formula.  It's not necessarily intended to be
> an endorsement of the practice, but a Web search turns up examples of sample
> size estimation based upon the Blackwelder formula with proportions smaller
> than 0.2 (larger than 0.8).
>
> Joseph Coveney
>
> *W. C. Blackwelder, Proving the null hypothesis in clinical trials.
> _Controlled Clinical Trials_ 3:345-53, 1982.
<snip>

I based -equivsize- on the formulae in Machin & Campbells' Statistical 
tables for the design of clinical trials; Blackwell 1987.  In turn these 
authors quote Makuch & Simon's Sample size requirements for evaluating a 
conservative therapy in Cancer Treatment Reports Vol 62 pp1037-40.  The 
formula given in Machin & Campbell is also identical to that in Table 1 of 
Blackwelder, as Joseph surmised.

As I recall, equivsize- was written as a hurried reply to another Lister's 
request for help. I didn't even put it on SSC as it has no proper help 
file, nor was it exhaustively debugged or tidied. (Nonetheless it appears 
to work. As nQuery Advisor also quotes Machin & Campbell, and -equivsize- 
yields results identical to nQA, I seem to have coded things OK, 
algebraically if not aesthetically).

 I appended a caution regarding the normal approximation since Blackwelder 
states:

"Assuming sufficiently large samples to justify the normal approximation
to the binomial, the statistic z' indicated in Table 1 can be used to test the
null hypothesis H0'; the same statistic, without delta subtracted in the 
numerator,
can be used to test H0. (Note that the denominator of z' is slightly different
from what might be used in testing the usual hypothesis H0, but for com-
parison the same estimate of variance is used for both cases. Also, a 
correction
for continuity can easily be made in either case.) Dunnett and Gent [11]
compared several asymptotically appropriate test statistics for H0', including
z' corrected for continuity, and suggested use of a chi-square statistic. How-
ever, the statistic z' is especially useful since it is easily calculated 
and leads
to explicit sample size and confidence interval expressions. It will be appro-
priate in most practical situations, since the specified difference delta 
will usually
be relatively small and a sample large enough to justify the normal approx-
imation will therefore be required in order to provide acceptable statistical
power. If, however, the study is too small for use of the normal approximation,
the problem of an appropriate test statistic is more difficult. One approach,
discussed by Dunnett and Gent [11] and useful particularly when the prob-
ability _pi_ of success with standard therapy is known fairly accurately, 
involves
expressing the hypothesis in terms of an odds ratio and constructing confi-
dence intervals."

Having read that, I (probably over-conservatively) threw in the comment on 
proportions less than 0.2, but I freely confess being uncertain as to the 
performance of the asymptotic  normal-based tests in this circumstance. 
Wellek, in his book: Testing statistical hypotheses of equivalence,Chapman 
& Hall 2003, provides a detailed discussion of exact tests in his chapter 
2, and also provides a Fortran program to calculate sample sizes for  exact 
Fisher type tests of non-inferiority. see 
http://zima04.zi-mannheim.de/wktsheq/  and look for the bi2ste2 
program.  This is source code, so you will need a compiler.  I am not into 
Fortran so have not used it.

Phil



Philip Ryan
Associate Professor,
Department of Public Health
Associate Dean (Information Technology)
Head, Data Management & Analysis Centre
Faculty of Health Sciences
University of Adelaide 5005
South Australia
tel 61 8 8303 3570
fax 61 8 8223 4075
http://www.public-health.adelaide.edu.au/
CRICOS Provider Number 00123M
-----------------------------------------------------------
This email message is intended only for the addressee(s)
and contains information that may be confidential and/or
copyright. If you are not the intended recipient please
notify the sender by reply email and immediately delete
this email. Use, disclosure or reproduction of this email
by anyone other than the intended recipient(s) is strictly
prohibited. No representation is made that this email or
any attachments are free of viruses. Virus scanning is
recommended and is the responsibility of the recipient.

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/