s-news
[Top] [All Lists]

Model comparison metric for multiple event survival models

To: <s-news@lists.biostat.wustl.edu>
Subject: Model comparison metric for multiple event survival models
From: "Zainab Jamal" <zjamal@anderson.ucla.edu>
Date: Thu, 25 Dec 2003 13:35:24 -0800
Organization: UCLA
Reply-to: "Zainab Jamal" <zjamal@anderson.ucla.edu>
Hi
I am working on splus 2000 windows version. I have a panel data set with each user experiencing multiple events.
I am using a conditional cox proportional hazard model with stratification based on number of recurring events.
The sample of users with large number of events,say 10 or 12 is small. So I am truncating the data at a max of 6 events for each user as some of my covariates are significant till this level.
I have three qeries:
1) If I need to decide at what level to truncate, is there a metric like BIC to decide the best fitting model? Currently my sample size varies if I truncate at different number of visits and I don't think I can use BIC.
 
2)Is there anyway I can use all the information I have about each user and still limit the actual estimation of parameters and survival curves to say 6 events.
 
3)If I want to incorporate a different frailty distribution (parametric is fine) for individuals in each strata how do I do this?
Really appreciate all ur help as I am new to both splus and survival models.
Regards
Zainab Jamal
<Prev in Thread] Current Thread [Next in Thread>
  • Model comparison metric for multiple event survival models, Zainab Jamal <=