Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: RE: Power calculation for competing risks regression


From   Steven Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: Power calculation for competing risks regression
Date   Mon, 22 Aug 2011 18:20:54 -0400

Actually, the code can be shortened.

Steve

*************CODE BEGINS*************
webuse hypoxia, clear
stset dftime, failure(failtype==1)
stcrreg ifp tumsize, compete(failtype==2)
stcrreg, noshr
capture net install st0074.pkg, replace
powercal pow1, nunit(e(N_fail))  ///
    delta(log(1.06)) sdinf(sqrt(e(N_fail))*_se[ifp]) alpha(.05)
list pow1 in 1
**************CODE ENDS**************



I agree with Rich.  Post-Hoc power calculations are too late: if your study had found a p-value of 0.05 exactly, for example, the post-hoc power to detect the observed effect would have been 50%. But arguing with reviewers who don't know better is fruitless. The following is approximate. Simulation would be better, but for your purpose, I don't think it matters.

Steve

*************CODE BEGINS*************
webuse hypoxia, clear
stset dftime, failure(failtype==1)
stcrreg ifp tumsize, compete(failtype==2)
stcrreg, noshr
di _se[ifp]
scalar nfail = e(N_fail)
scalar sd = sqrt(e(N_fail))*_se[ifp]
scalar list sd
scalar  log_eff1 = log(1.06) // HR = 1.06
capture net install st0074.pkg, replace
// Roger Newson's -powercal- package
powercal pow1, nunit(nfail)   ///
 delta(log_eff1) sdinf(sd) alpha(.05)
list pow1 in 1
**************CODE ENDS**************




On Aug 22, 2011, at 11:36 AM, Richard Goldstein wrote:

I disagree that a post-hoc power analysis is ever a good idea (see
http://www.stat.uiowa.edu/~rlenth/Power/ for more on this)

instead I recommend looking at the CI's and seeing whether anything of
substantive interest is included in the CI; if yes, you have
insufficient power; if no, I see no problem

Rich

On 8/22/11 11:29 AM, Hackney, David wrote:
> Hello all,
> 
> Does anyone have advice on power (or sample size) calculations for competing risks (stcrreg) in Stata? We did a competing risks regression and found that the HR for the variable of interest was non-significant and the reviewers (reasonably) asked for a post-hoc power analysis to provide support for our conclusion. There is no competing risk model specific power calculation which appears to be available, though I didn't know if one of the other power commands could be modified to accommodate competing events, or if a user-written program existed (I couldn't find one with the search function). Thanks.
> 
> David Hackney
> 
> 
> 
> PS.  My apologies if I have sent this multiple times, I'm just not sure if I have done it correctly.  Thanks.
> 
> 
> David N Hackney MD, MS
> Assistant Professor
> Division of Maternal-Fetal Medicine
> Department of Obstetrics and Gynecology
> University of Rochester Medical Center
> Phone (admin assistant): 585-275-7480 
> Fax: 585-256-1416
> david_hackney@URMC.rochester.edu
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index