Re: st: equivalence trial power calculation

[Constantine Daskalakis <c_daskalakis@entwhistle.jci.tju.edu>]

At 04:38 PM 10/7/2003, you wrote:
> Dear list
>
> Using Stata 8.1
>
> I'm unsure on how to calculate the required sample for the following trial,
> been searching by findit for equivalence ado's without finding my answer:
>
> We are planning to do a 2 year equivalence trial.The current standard of
> treatment has a 2 year failure rate of 7%: In our trial of a new drug, the
> treatments will be considered equivalent if the 95% confidence interval are
> within the bound of -12% to 12%, which we believe is the  the largest
> difference that would be considered clinically acceptable.
First off, it would seem that what is appropriate here is a non-inferiority 
setup (1-sided) rather than (bio)equivalence (2-sided). If the new drug is 
much much better, that's good and acceptable, no? Why would you want to 
bound the efficacy difference from both sides? It's only when the new drug 
is worse that you might need to impose bounds. It comes down to what 
exactly you want to prove:

(i) equivalence -- TO PROVE that drug and standard treatment have "similar" 
failure probabilities; but
(ii) non-inferiority -- TO PROVE that drug is "at least as good as" ("no 
worse than") standard treatment.

If the difference in true failure probabilities is  p = p{drug} - 
p{standard},  then

(i)
H0:  |p|  >=  0.12
H1:  |p|  <  0.12

(ii)
H0:  p  >=  0.12
H1:  p  <  0.12

In both cases, the "delta" (0.12 here) determines what is "similar" or "as 
good as".

You need equivalence power calculations for (i) but the usual 1-sided 
superiority/inferiority power calculations for (ii).

In (i), you give up the opportunity to infer anything about differences 
(ie, if one treatment is much different than the other, you'll still simply 
infer that they are not "similar" enough -- end of story).

In (ii), you give up the opportunity to infer that the drug is "better" 
than standard treatment if that turns out to be the case (ie, you'll simply 
have to say that the drug is not worse than treatment). If this "better" 
kind of statement is of interest, then the usual 2-sided 
superiority/inferiority setup should be used instead.

That said, I don't even see how the 12% bound for the difference is 
reasonable for the one side (drug better). Bounds refer to the TRUE 
parameter values. If the hypothesized failure probability for standard 
treatment is 7%, the biggest difference in one direction can only be 7% 
(ie, 0% failure for drug, 7% failure for standard treatment, p = -7%). On 
the other hand, the 12% bound for the other side seems huge (ie, 19% 
failure for the drug vs. 7% failure for standard tretament, an almost 
3-fold increase!). I'll take your word that this would be an "acceptable" 
drug... I suppose if the disease is not serious but the standard treatment 
has serious side effects (but the new drug doesn't), it could be ok...


> any suggestions?
>
> Jannik Helweg-Larsen


The documents accompanying this transmission may contain confidential 
health or business information. This information is intended for the use of 
the individual or entity named above. If you have received this information 
in error, please notify the sender immediately and arrange for the return 
or destruction of these documents.
________________________________________________________________

Constantine Daskalakis, ScD
Assistant Professor,
Biostatistics Section, Thomas Jefferson University,
211 S. 9th St. #602, Philadelphia, PA 19107
   Tel: 215-955-5695
   Fax: 215-503-3804
   Email: c_daskalakis@mail.jci.tju.edu
   Webpage: http://www.kcc.tju.edu/Science/SharedFacilities/Biostatistics  
*
*   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/