[Feature suggestion]: Feature selection (for dependent censoring)

[arturomoncadatorres]

One of the challenges in a CPH analysis is including the right features. The simplest approach consists in selecting a feature subset using univariate Cox regression. However, this approach relies on the assumption of independent censoring (survival time and censoring time need to be statistically independent at a given feature), which quite often is not true (e.g., young patients may be more likely to quite a treatment than older patients).

Recently, Emura and Chen proposed a copula-based method for feature selection in the case of dependent censoring. They showed that this method outperformed the classical univariate approach. Moreover, they have an implementation for R (repo here).

I think lifelines would benefit greatly from having such functionality. Would this be something you are interested in adding?

[CamDavidsonPilon]

Hi @arturomoncadatorres, looks very interesting. I'll give it a look later this week!