Subject: | |
From: | |
Reply To: | |
Date: | Tue, 7 May 2013 11:49:09 +0900 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Hi Kiyoung
There are basically two common methods of dealing with competing risks in semi-parametric proportional hazards models. One is to run separate Cox PH models for each event of interest and censor units that experience competing risks. Basically you are saying you don't know why they left the risk set. The other approach (e.g. Fine and Gray) essentially counts the units in the risk set after they've failed so as to keep information on the units that experienced another event and cannot experience the event of interest. However, they are not "at risk" in the sense that they could in the future experience the event of interest. A basic Fine and Gray model is not a repeated event survival model, so it should be fine for what you are proposing. For more info see the Stata Survival Analysis Manual discussion of Fine and Gray.
"think of [the Fine and Gray hazard] as that which generates failure events of interest while keeping subjects who experience competing events “at risk” so that they can be adequately counted as not having any chance of failing. (2009, 200)"
I think the key issue for you is whether a Fine and Gray set up is more useful than running separate Cox models for each event type. Bakoyannis and Touloumi (2012)(http://smm.sagepub.com/content/21/3/257.short) show that if covariates only affect one event and have no effect on the others, then the two models will give equivalent results. However, if covariates have different affects on different events, then separate Cox models will produce biased estimates and Fine and Gray will be better. I confirm this in a political science paper I have forthcoming in IPSR. (The preprint version is available at:http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2155986)
Hope that helps.
Best
Christopher
--------------------------------------------------
Christopher Gandrud, PhD
Lecturer
International Relations
Yonsei University
208 Jeongui Hall, 1 Yonseidae-gil
Wonju, Gangwon-do, 220-710
Republic of Korea
[log in to unmask]
http://christophergandrud.blogspot.com/
On 7 May 2013, at 03:40, Kiyoung Chang <[log in to unmask]> wrote:
> Kiyoung
**********************************************************
Political Methodology E-Mail List
Editors: Ethan Porter <[log in to unmask]>
Gregory Whitfield <[log in to unmask]>
**********************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
http://polmeth.wustl.edu/polmeth.php
**********************************************************
|
|
|