[R] Trend test for survival data
Terry Therneau
therneau at mayo.edu
Tue Apr 22 14:52:16 CEST 2008
> Hello,
> is there a R package that provides a log rank trend test
> for survival data in >=3 treatment groups?
> Or are there any comparable trend tests for survival data in R?
The log-rank test is equivalent to a Cox model with a factor variable as the
predictor. To do a trend test, simply fit the Cox model with the variable
scored as 1,2,3,... Or fit it as a factor and do a post-hoc trend test:
> fit1 <- coxph(Surv(time, status) ~ ph.ecog, data=lung)
> fit1
Call:
coxph(formula = Surv(time, status) ~ ph.ecog, data = lung)
coef exp(coef) se(coef) z p
ph.ecog 0.476 1.61 0.113 4.2 2.7e-05
Likelihood ratio test=17.6 on 1 df, p=2.77e-05 n=227 (1 observation deleted
due to missingness)
> fit2 <- coxph(Surv(time, status) ~ factor(ph.ecog), data=lung)
> fit2
coef exp(coef) se(coef) z p
factor(ph.ecog)1 0.369 1.45 0.199 1.86 6.3e-02
factor(ph.ecog)2 0.916 2.50 0.225 4.08 4.5e-05
factor(ph.ecog)3 2.208 9.10 1.026 2.15 3.1e-02
Likelihood ratio test=18.4 on 3 df, p=0.000356 n=227 (1 observation deleted
due to missingness)
> zz <- c(1,2,3)
> test.num <- zz %*% coef(fit2)
> test.var <- zz %*% fit2$var %*% zz
> test.num/sqrt(test.var)
[,1]
[1,] 2.74323
Note that ecog performace score=0 is implicitly part of the contrast. The
full coefficient vector is (0, .369, .916, 2.208) and my linear contrast zz is
0,1,2,3.
Terry Therneau
More information about the R-help
mailing list