Recombining survival curves when stratifying a time varying categorical variable what is the best practice

Hi

I wanted the experts to weigh in on the best method reporting out final survival curve after time splitting to address a time dependence issue with a cox-hazard model.

When you do the split you have 3 curves, which is confusing when you want to communicate to a broader audience the results of the model. I made a stackoverflow post on this issue on how to tackle this but from a statistical standpoint I would like some eyes on the answer.

see post below
post link

Time splitting is good for understanding time-dependent covariate effects but allows all predictors to operate in non-proportional hazards so may be a bit noisy. It may be better to have a smooth time-dependent covariate effect (say in \log(t)) for a single covariate in question.

1 Like