SURVDIFF()

How to note (in code) a few (!) adjusting covariates for cox
regression. I had an example for one covariate, and tried (according
to my own understanding) two variantes of code  (pls, see below), and
got ... a different p-value. What is the right code? Many thanks!!!

(1)
survdiff (Surv(survt,status)~clinic+strata(prison, dose, gender),data=addicts)
N Observed Expected (O-E)^2/E (O-E)^2/V
clinic=1 163      122    106.2      2.35        14
clinic=2  75       28     43.8      5.70        14
Chisq= 14  on 1 degrees of freedom, p= 2e-04

(2)
survdiff (Surv(survt,status)~clinic+strata(prison+dose+gender),data=addicts)
N Observed Expected (O-E)^2/E (O-E)^2/V
clinic=1 163      122    106.2      2.35      12.1
clinic=2  75       28     43.8      5.69      12.1
Chisq= 12.1  on 1 degrees of freedom, p= 5e-04

Read the help page:

?strata


Usage

strata(..., na.group=FALSE, shortlabel, sep=', ')
Arguments

...
any number of variables. All must be the same length.

So the function is documented to accept a list of variable names, but 
_not_ as accepting a formula. So I read the help page as endorsing your 
first option. Since you didn't include a suitable dataset for testing I 
borrowed the ovarian dataframe and checked to see whether the sum of 
variables submitted with a "+" sign was being calculated and can confirm 
that it is:


 >? coxph(Surv(futime, fustat) ~ age + strata(I(rx+ecog.ps)), data=ovarian)
Call:
coxph(formula = Surv(futime, fustat) ~ age + strata(I(rx + ecog.ps)),
 ??? data = ovarian)

 ?????? coef exp(coef) se(coef)???? z?????? p
age 0.11942?? 1.12684? 0.04528 2.637 0.00836

Likelihood ratio test=9.48? on 1 df, p=0.002073
n= 26, number of events= 12
 >? coxph(Surv(futime, fustat) ~ age + strata( rx+ecog.ps) , data=ovarian)
Call:
coxph(formula = Surv(futime, fustat) ~ age + strata(rx + ecog.ps),
 ??? data = ovarian)

 ?????? coef exp(coef) se(coef)???? z?????? p
age 0.11942?? 1.12684? 0.04528 2.637 0.00836

David.