Comparing COXPH models, one with age as a continuous variable

The "anova" process is only statistically valid for nested models, i.e., where
one includes the other as a proper subset.  A model with continuous age and one
with discrete age are not nested; your p-value will be meaningless.

 Terry Therneau