Updates on ordinal longitudinal modeling: <hbiostat.org/R/rms/blrm.html>
Hi Dr. Harrell,
I’m wondering if you can clarify why and how blrm is not designed to handle absorbing states (death or any other form of patient censoring). It’s not clear to me how “breaking” these states are to the model from looking at http://hbiostat.org/R/rms/blrm.html – would you recommend I use this model at all with my covid19 dataset (where many patients die or are discharged)?
Thanks,
Eliza
Data Scientist, Mount Sinai NYC
Our best current thinking, which you can see from a link above is this:
- patients who die have their last data record on the day of death
- get the odds ratio from this “truncated at death” dataset
- use de-conditioning to get things like cumulative incidence out to the end of planned follow-up. Jonathan Schildcrout in my department is working on the math for that. This is a post-estimation process once you’ve developed the overall model. This formally handles absorbing states when getting derived parameters of interest.
2 Likes
Dear prof,
you are a rich resource for this topics really,
I would like to ask if you can do a paid consultation by email on the Bayesian phase I/II dose-finding trial?