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