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 at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: stcrreg postestimation


From   rgutierrez@stata.com (Roberto G. Gutierrez, StataCorp)
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: stcrreg postestimation
Date   Mon, 26 Jul 2010 17:03:40 -0500

Caroline Savage <savagec@mskcc.org> asks:

> I'm using STATA's new competing risk regression command, stcrreg.  I'm
> trying to use the postestimation commands, but am running into problems. 
> My aim is to calculate the 10-year predicted cumulative incidence of
> cause-specific death (non-cause spec ific death is the competing risk) by
> age and adjusted to the mean of another covariate in the model.

> To accomplish this, I ran the competing risks regression model:
> . stset ttdeath, f(outcome==1)
> . stcrreg age covariate1, compete(outcome==2)

> Then I set covariate to the mean and asked STATA for the linear predictor
> (predict xb, xb). So far so good.

> The problem came when I tried to calculate the baseline cumulative incidence
> or cumulative subhazard.  The issues I noticed were:

> - Baseline cumulative incidence and subhazard predictions changed depending
> on whether I predicted them before or after setting the covariates to the
> mean.  My understanding of these functions was that they are calculated for
> subjects who have zero-valued covariates - so why would manually changing
> the covariate have any impact?  Moreover, if I manually set the covariate
> to zero, I actually get the same predictions as when I had set it to the
> mean (which was different than when I did nothing to the covariate). 

>[...]

Accurately predicting the baseline cumulative incidence function (CIF) with
-predict, basecif- after -stcrreg- requires that the estimation data (time
variables, censoring/competing variables, and covariates) remain intact.  This
is necessary because -predict- needs to reconstruct all the risk sets from the
original estimation before any prediction can be made.

Once you change the estimation data, baseline function estimates as obtained
from -predict- are no longer what you think they will be.  As such, it would
probably be better if -predict- generated an error in this case, and we'll
look into making that change.

If you need to obtain predicted CIFs for specified sets of covariate values
then, keeping the estimation data intact, type

   . predict cif0, basecif

and use the following relationship to transform this baseline CIF to one for a
specified covariate pattern.

   . generate cif1 = 1 - (1 - cif0)^exp(xbeta)

where -xbeta- is the linear predictor for the new covariate pattern.  

Of course, Caroline should feel free to email me privately if she has any
questions.

--Bobby
rgutierrez@stata.com
*
*   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