Solved – How to get the p-value for the full model from R’s coxph

cox-modelr

In a coxph model with 2 variables (let's say age and sex), how can I get the p-values for LR, Wald and score tests which I see in the summary(coxphobject) as below?

Concordance= 0.653  (se = 0.058 )
Rsquare= 0.085   (max possible= 0.924 )
Likelihood ratio test= 30.09  on 12 df,   p=0.002708
Wald test            = 34.73  on 12 df,   p=0.0005169
Score (logrank) test = 34  on 12 df,   p=0.0006736,   Robust = 17.61  p=0.1279

I think those values represented in the summary as p=... for each model are the p-values for the full model, but I cannot figure out how to retrieve them. If there is only one variate, let's say age, then I can retrieve the p-value using
summary(coxphobject)$coefficients[5], which is the same as the one I see for the Wald test in the summary(coxphobject). But when there are two variables (age and sex), summary(coxphobject)$coefficients[1,5] and summary(coxphobject)$coefficients[2,5] give the p-values for age and sex separately, but they both are different from the p-values I see for the full model in the summary(coxphobject) for LR, Wald and score tests.

I appreciate any help. Thanks.

Best Answer

1) Put summary(coxphobject) into a variable

summcph <- summary(coxphobject)

2) examine it with str()

str(summcph)

Values! Values everywhere!

so we find, (proceeding line by line in your above output):

a) the Concordance values

summcph$concordance

b) the Rsquare values

summcph$rsq

c) The Likelihood ratio test values

summcph$logtest

d) The Wald test values

summcph$waldtest

e) The score test values

summcph$sctest

f) The robust values

summcph$robscore

It really helps if you post a reproducible example, rather than make us go find your data set in order to check we're doing all the options the same. For example, you didn't mention you had a cluster term. (It would take an extra few moments for you, and would have saved me ten minutes while I tried to figure out why I couldn't get the last couple of values. At the least you could have mentioned which example you ran in the help!)

Related Question