Competing risk for AFT models in R

I’m an amateur when it comes to stats but recently been having to do more of it for my masters
I’m looking at factors (Age, frailty, Social factors etc…) and how it effects outcomes such as death and entry into aged residential care (ARC)
I’m looking at AFT models for some of my data - For both death and entry into aged residential care.
For death I initially used a cox model but didn’t meet proportional hazard(pH) assumption but had better luck with an AFT(weibull model) which seems to fit well.
Similar goes for the other endpoint (Entry into aged residential care (ARC))

The issue is that Death is a competing risk when it comes to entry into ARC.
A few questions on this issue
a. Can I use the Fine and Gray model to model Competing risk is given the PH assumptions weren’t met for Cox (my understanding is that the F&G model is a semi-parametric model and an extension of Cox)
b. Is there a way of running a competing risk model for weibull distributions?



The Fine & Gray model has a proportional hazards assumption that Terry Therneau has shown to be very unrealistic under even ideal conditions. Take a look at this.


see section 3.1 Cause-specific AFT model. It seems straightforward for a cause-specific AFT model, ie: “this model can be estimated by treating all observations with failure due to all other causes except for k as censored”. That could done easily in any standard stat software, although they say they used R package “aftsrr function in a contributed R package aftgee” [warning; I’m not very familiar with the model or this R package]. Let us know if you go this route and confront any problems …

1 Like

That sounds good for parameter estimation, maybe not for estimating cumulative incidence.

1 Like

Thank you for all your replies

I also came across this

Not sure if anyone has any experience using this package