Assumptions and pitfalls in competing risks model

Question

Question: What are the major pitfalls and assumptions for competing risk analysis?

Background

I'm trying to learn competing risks and I use the cmprsk package in R based upon Gray's algorithm. I've read Melania Pintilie's book, Competing Risks: A practical perspective, but although it's a very nice introduction to the field I lack a list of the common pitfalls. My mentor has warned me about the competing risk analysis still being a developing field and I would therefore really appreciate if I could get your opinions on this.

In the answer to this question it's mentioned:

just make sure that you understand what assumptions you are making

As I understand it the major assumption is the proportional hazards, similar to the Cox model, but other than that assumption I'm not sure that I know what other assumptions and pitfalls I'm to look out for.

Answer 1

Pintile's book is an excellent book for understanding competing risks but if you want to study theoretical side of the competing risks, take a look at Martin J. Crowder's book, Classical Competing Risks.

I wrote my master's thesis on competing risks and from what I remember, they are some drawbacks/disadvantages when competing risks analysis is conducted. The most well-known problem is an issue of identifiability. There were many papers and journals regarding this issue when I was preparing my thesis and I believe many scholars are still writing papers about this.

The issue of identifiability arises due to the nature of competing risks modelling -- when death or failure has observed from an identifiable cause-of-failure it cannot occur again later from another case.

Another issue arises when the cause of failure for a subject or unit is NOT precisely identified, it can be narrowed down to a subset of the potential risks. This phenomenon is called masking.