DeaR users. <framework> These days i'm working on fitting an extended Cox model with time-dependent covariables and possibly time-varying effects. My data are in counting process format as described in Therneau&Grambsh's `Modeling Survival Data', page 68. I'm trying to follow Harrell's `Regression Modeling Strategies' advices for the choice of my model. This study aims to the development of a prognostic model, so it'is primary predictive.
I have to do stepwise model selection and provide a measure of predictive accuracy. I'm using rms's cph and validate function with bw=TRUE option. </framwork> <questions> 1. Is validate good at resampling from a counting process format database? Or should i use a somewhat modified version? 2. Why fastbw(fit,"aic") and step(fit) don't select the same model? step() appears to stop first. I can't manage to get the stopping rule in the help files. 3. cph seems to be a bit less "permissive" than coxph in parsing the model formula. Particularly i have some difficulty in modeling interactions between covariables and time. Am I totally misguided? Is there any reference on this topic? Now a theoretical one: 4. Is it somewhat sensible to use cox.zph() and schoenfeld residuals to investigate which time dependent variables could need a time interaction parameter for estimating a time-varying effect? </questions> Thanks in advance for any advice. ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
