Open CamDavidsonPilon opened 6 years ago
Related: https://www.ncbi.nlm.nih.gov/pubmed/15158046 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4983650/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4310506/
Also related to reference 5 in your linked article, are there any plans to implement accelerated failure time models in lifelines?
Also related to reference 5 in your linked article, are there any plans to implement accelerated failure time models in lifelines?
yup, it's a priority now.
Here's a quick roadmap:
Robust errors for cox models
Does the robust errors for cox models include adding weights to the time varying cox model? Ultimately, the marginal structural models I am interested in estimate risk difference/ratio, however clinicians do like hazard ratios
AFTmodels
Based on my Biostatistics class, AFT shouldn't be terrible to implement. Collett's "Modeling Survival Data in Medical Research" gives a pretty good description
survival curves adjusted by baseline covariates
What's the strategy for adjusting survival curves by baseline covariates? Inverse probability weights?
For the weights in cox model, the robust errors are a prerequisite, so that needs to get done first. (Adding weights will be in that addition too)
For the survival curves, hernan in the paper I linked has a “recipe” I want to try.
Wanted to follow-up with this since I recently reread the "Hazards of Hazard Ratios". The 'Hernan' approach is just a disguished version of the g-formula / standardization. I am happy to chat further about it sometime. Basically, it is a special case of the MonteCarloGFormula
I have written
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3653612/#R5