> m1 <- coxph(Surv(time, y) ~  x1+x2, data=a)
> m2 <- coxph(Surv(time, y) ~  x1+x2+x3, data=a)
> anova(m1, m2)
Analysis of Deviance Table
 Cox model: response is  Surv(time, y)
 Model 1: ~ x1 + x2
 Model 2: ~ x1 + x2 + x3
   loglik  Chisq Df P(>|Chi|)
1 -319.85
2 -317.17 5.3526  1   0.02069 *