Interval-censored Cox proportional hazards models

[gvelasq]

Thank you for this package, it is exciting to see its rapid development.

Stata 17 added support for interval-censored semiparametric Cox proportional hazards models in the stintcox function. The closest I've been able to find in R is the ic_sp() function in the icenReg package, though I believe the details of the implementations differ. I have filed an issue in pistacliffcho/icenReg_devel#12 to see if the Stata implementation can be reimplemented in R, based on:

• Zeng, D., L. Mao, and D. Y. Lin. 2016. Maximum likelihood estimation for semiparametric transformation models
  with interval-censored data. Biometrika 103: 253–271. https://doi.org/10.1093/biomet/asw013
• Zeng, D., F. Gao, and D. Y. Lin. 2017. Maximum likelihood estimation for semiparametric regression models with
  multivariate interval-censored data. Biometrika 104: 505–525. https://doi.org/10.1093/biomet/asx029

Would you consider adding support for interval-censored Cox proportional hazards models?

[hfrick]

Thanks for the suggestion! Let's see what the maintainer of icenReg says. Are you interested only in this particular implementation or interval-censored Cox models more general? It is not one of our top priorities right now but I'll leave the issue open to track interest.

[gvelasq]

Thank you for considering. I'd be interested in interval-censored Cox models in general.

[hfrick]

The CRAN Task View for survival also lists the ICsurv package for interval-censored Cox models but this hasn't been updated in a long while. Are you currently fitting those models in R? Or stata?

[gvelasq]

I'm fitting these models in Stata 17.

On a quick Google Scholar search, I found 18 hits for 'icenReg ic_sp' (there are 175 hits for 'icenReg' alone, but it's more difficult to disentangle how many use ic_sp() specifically), 49 hits for 'ICsurv', and 1 for 'stintcox'. That said, stintcox was released with Stata 17 in April 2021, so given their promotional materials for the new feature I suspect its adoption will rise in the coming years. I am less familiar with how to search for whether this is implemented in SAS. On a first look it does not appear to be implemented in Python's scikit-survival. So in my opinion this is a great growth opportunity for tidymodels and the censoring package! The downside is that it would require a brand new implementation, as opposed to bindings to an existing R package.