topic Re: main interest variable violated Cox proportional-hazards assumption in Statistical Procedures

main interest variable violated Cox proportional-hazards assumption

qiali — Mon, 09 Jul 2018 23:23:30 GMT

Hi Everyone,

Among certain cancer patients, I am running a cox model to examine Effect of Treatment Era (year of cancer diagnosis) on Hazard of developing cardiovascular disease (CVD) accounting for Competing Risk of Death. Actually, it is the Fine-Gray model.

The problem is that Treatment Era (year of cancer diagnosis) violated the PH assumption. Since it is our main interest variable, we can not stratify by it. I have checked the interaction between treatment Era (year of cancer diagnosis) and chemotherapy/ radiation, none of them is significant. The cumulative incidence of CVD between 1990-2007 is higher than 2008-2014 for sure.

So my question is for the Fine-Gray model, does PH assumption matter? Can I ignore it? If we can not ignore, anything else can I do?

I feel the year of cancer diagnosis is kind of time variable...

Any suggestion will be appreciated! Thanks a lot!

Re: main interest variable violated Cox proportional-hazards assumption

Reeza — Tue, 10 Jul 2018 01:18:07 GMT

Yeah, it’s time dependent variable, it would likely have dependencies. I bet the age distribution in your groups are not equivalent.

Re: main interest variable violated Cox proportional-hazards assumption

pau13rown — Tue, 10 Jul 2018 03:17:43 GMT

a more powerful analysis could be a parametric, joint frailty model. Rogers and Pocock compare alternatives and declare the joint frailty to be a superior option, although they consider the case of recurrent events (which you don't have): https://www.ncbi.nlm.nih.gov/pubmed/24453096