Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: Fw: subgroups in cox regression

 From catherine duggan To statalist@hsphsun2.harvard.edu Subject Re: st: Fw: subgroups in cox regression Date Sun, 1 May 2011 18:36:57 +0100 (BST)

```Many thanks!

----- Original Message ----
From: Steven Samuels <sjsamuels@gmail.com>
To: statalist@hsphsun2.harvard.edu
Sent: Sun, 1 May, 2011 7:32:59
Subject: Re: st: Fw: subgroups in cox regression

Sorry- ignore the snipped 'There is no "lik" at the end.

Steve

On May 1, 2011, at 10:19 AM, Steven Samuels wrote:

Catherine:

Effect modification means that the coefficient the exposure  might differ
between stages. To test it, you add an interaction or test for an interaction.
One consequence of  using the -if- statement is that it doesn't isolate a
differential effect of your exposure variable, because all components of the
model differ: 1) the baseline hazard; 2) the coefficient for race; and 3) the
coefficient for your exposure.  This is certainly a possible model and it is a
good idea to check for proportionality of the baseline hazard by stage and race
by using the strata() option in Stata. I'm also wondering whether you have
checked that a linear term in your exposure variable is adequate; you can use
the -fracpoly- command to find out.  There is no "lik

******************************************
// If you have Stata Version < 11 (I presume so, since you use -xi-

xi: stcox  i.race i.stage*igf

// Otherwise with Stata 11
stcox i.race c.igf##i.stage   // add ", robust" to get robust standard error

//The interaction term estimates the difference in igf slopes between the two
stages.

// You can get the slopes directly by:

tab stage, gen(stage)  // create indicators stage1 stage2
gen igf1= igf*stage1
gen igf2= igf*stage2
xi: stcox i.race  stage1 igf1 stage2 igf2, nocons
lincom igf1 - igf2  // estimate and test differences with Wald test
****************************

Steve
sjsamuels@gmail.com

On Apr 30, 2011, at 10:40 PM, catherine duggan wrote:

Hello
I have what is probably a basic question, which is currently defeating me.

I have run Cox regression with death as my outcome: my main exposure of interest

is a a continuous variable (IGF)

I have other covariates in the model - I will just refer to two of them, race (5

level) and tumor stage (2 level)

Xi: stcox igf i.race i.stage

I did the main analysis, and then did some exploratory subgroup  analysis, with
the usual caveats of smaller sample sizes and using a  more conservative P value
Eg

Xi: stcox igf i.race if stage==1

However a reviewer gave the comment

' what you might want to do is to construct a model for effect  modification -
that is compare the model with no subgroup to one with  trends fit for each
subgroup - a 1 dgree of freedom likelihood ratio  test - do see if the trends
differ statistically - a test for  homogeneity of trends.  This would be more
informative'

Unfortunately I have no idea how to accomplish this - I wasn't sure whether this

was a test for interaction or not.

I'd appreciate any guidance

Many thanks
Catherine

*
*   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/

*
*   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/
```