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: -TVC- option in the -stcox- command


From   Joerg Luedicke <joerg.luedicke@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: -TVC- option in the -stcox- command
Date   Fri, 15 Apr 2011 18:20:48 -0400

On Fri, Apr 15, 2011 at 4:34 PM, Pete de Blank <pdeblank@gmail.com> wrote:

>
> I include all drug classes that have a p value >0.2 on log rank test,
> according to the command
>     Sts test drugclass1, strata(hospitalsite)

I guess you mean p-value<0.2 (or "excluded" instead of "included")?
Anyway, this is a questionable approach as you are capitalizing on
chance which makes any further consideration of p-values a rather
nonsensical endeavor. Some other issues can be found here:

http://www.stata.com/support/faqs/stat/stepwise.html

(Though this is for "stepwise" regression, but still.)

>
> And my initial model is:
>     Stcox drugclass1 drugclass2 drugclass3 drugclass4 drugclass5,
> tvc(drugclass1 drugclass2 drugclass3 drugclass4 drugclass5)
> strata(hospitalsite)
>
> I have three questions:
> (1)  How do you choose the best multiplier for the time-varying
> covariates (with -texp(exp)-)?  How do you know if ln(t) is better
> than t?

Given that you did the log-rank test for "drugclass" I assume that
those drugclass variables are binary, not continuous? The  -texp(exp)-
option is for continuous time varying covariates. In case of
categorical variables, I would do spell splitting with -stsplit-.
>
> (2)  Under what criteria can non-significant covariates be eliminated
> from the model?  If the non-time-varying p-value is >0.05?  Only if
> both the time-varying and non-time-varying covariates have
> p-values>0.05?  If the time-varying portion (but not the
> non-time-varying portion) of a covariate has a p-value > 0.05, can I
> eliminate just that covariate in the tvc option?

Basically, there are no such criteria. See my comments above. You
should have some theory about why you think certain types of drugs are
causing an infection and then you include all of those drugs and do
not exclude any of them.

>
> (3)  Finally, how do you conceptualize the interpretation of the
> results?  Do I think of this as an interaction with time?
>

In the Stata 11 manual for stcox, page 129ff. are examples given that
discuss those things.


J.

*
*   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