I am extrapolating out of sample. I don’t want the OS and RFS curves to cross. By definition they can’t. However, they can when you estimate and extrapolate them the SOC arm does. I set up the model so SOC has the same cure fraction fraction for RFS and OS. I added a d to make OS a little higher than RFS. I did the same for FLIN. So, I’m stuck with at least two likelihoods. My data is one record for each patient, and has an OS time, OS event, RFS time, RFS event, and there is a treatment variable that is either SOC or FLIN.
I want to get the advantage of having the OS and RFS correlated by the fact that i’m reading the data one row at a time, and a row has both the OS and RFS. I don’t want to model correlation or covariance, but I want to get some benefit from the hierarchal structure.
Given that I have four sets of likelihoods I suspect that I may have lost some benefit from doing OS and RFS in the same model.
I’m not 100% clear on if what I’m talking about is a real issue. I have some small issues with divergencies, but the diagnostic plots and traces look decent and when I set up the plots using the estimated parameters they seem to track the KM curves pretty well.
I would like to replicate what the author is doing in this paper:
The author is working in R and stan and made a website, but it not clear that the website is up-to-date and there is not an easy to follow example with data.