Survival Regression using Stratification vs Partial Effects in Lifelines

Question

I have been using the Lifelines survival estimation package: https://lifelines.readthedocs.io/en/latest/

I am interested in the effect that a specific variable (number of pain sites) has on the survival model (using the CoxPH model). I tried introducing this model as a covariate using plot_partial_effects_on_outcome(): 

and as a stratification variable: 

I am wondering if why these two methods present inverse survival ratings for the values of the variables? For example, 4 pain sites seems to decrease survival in the covariate model but increases survival in the stratification model, and the opposite for 1 pain site.

It is my understanding that the stratification method includes the variable in the model without estimating its effect. So would this be the same as plotting 4 different survival functions with different variable values? Any help is greatly appreciated! Thank you!

Answer 1

It is my understanding that the stratification method includes the variable in the model without estimating its effect. So would this be the same as plotting 4 different survival functions with different variable values?

That's pretty close to what is going on, with one important note: there are other variables in the regression being accounted for. This may or may not have a large effect on the inference. It's up to you to decide if they should be included or not.