R survival analysis : surv_pvalue vs fit.coxph for log-rank-test pvalue

I have different value for test$pval and and other.pval that should return the same log-rank test p-value, no ?

surv_object <- Surv(time = as.numeric(final_group_annotated[,time]), event = final_group_annotated[,opt$endPoint] )it1 <- survfit(surv_object ~ group, data = final_group_annotated)

test=surv_pvalue(fit1, final_group_annotated)
print(test$pval)

0.3427874

fit.coxph <- coxph(surv_object ~ group, data = final_group_annotated)
print(summary(fit.coxph))

coxph(formula = surv_object ~ group, data = final_group_annotated)

  n= 188, number of events= 30

               coef exp(coef) se(coef)      z Pr(>|z|)
groupgroup2 -0.3482    0.7060   0.3689 -0.944    0.345

            exp(coef) exp(-coef) lower .95 upper .95
groupgroup2     0.706      1.417    0.3426     1.455

Concordance= 0.538  (se = 0.049 )
Likelihood ratio test= 0.87  on 1 df,   p=0.4
Wald test            = 0.89  on 1 df,   p=0.3
Score (logrank) test = 0.9  on 1 df,   p=0.3

other.pval <- coef(summary(fit.coxph))[,5]
print(other.pval)

0.3452066

survival analysis r

• 8,659 views • link

updated 6.7 years ago by Kevin Blighe

• written 6.7 years ago by ZheFrench

1 answer

The p-value returned by Surv() is the log-rank p-value from the score test; so, you should be comparing this to the log-rank p-value from the Cox PH model. I went over this recently in this thread: A: survfit(Surv()) P-value interpretation for 3 survival curves?

------------------------------------

The way that you are doing it, i.e.:

coef(summary(fit.coxph))[,5]

This will return multiple p-values when there are multiple strata, one p-value representing a single stratum / coefficient compared to the reference stratum.

To extract the log-rank p-value for the overall model, you need to use:

summary(coxfit)$sctest[3]

Kevin

• 0 views • link

written 6.7 years ago by Kevin Blighe