Hi,
I have several question related to survival analysis and Cox
models, albeit not all strictly related to S. I use S+ 3.3/W95.
1) survfit cannot handle interval-censored data (as it itself says).
Is that a mere limitation of this function or is there a theoretical
reason for that?
2) I am not sure of how to interpret Cox model with a time-varying
covariate.
I am studying the effect of (some kind of) vaccination on mortality.
People are born unvaccinated, then may be vaccinated then die
(or die unvaccinated). Very young people (under 3 months, say)
are never vaccinated, so what can mean the proportionnality
hypothesis in that age interval?
3) in my data, vaccinal status is known by interview, that is,
mothers are interviewed every three months and are asked
whether and when their children have been vaccinated. One
problem is that, when a child dies, that question is not asked
in subsequent interviews, so that if this child has been
vaccinated after the previous interview, it is nevertheless
classified as not vaccinated. Thus, if care is not taken, there
will be a bias towards a protective effect of vaccination.
I plan to fit a Cox model, with vaccination status as a
time-varying covariate.
Someone suggested me to set the vaccination date to the
date of the first subsequent interview, arguing that in the time
interval between vaccination and interview, the child was not
at risk of dying.
This sounds sensible, but is that standard practice? Could
someone point me to some references on the topic?
4) I would like to plot the hazard function, instead of the survival
curve. I am able to devise a naive estimate, which consists in
numerically derivating a smoothed version of the Kaplan-Meier
curve, but is there a better estimate?
thanks,
Eric Elguero
Epidémiologie et Prévention
I.R.D.
Montpellier - France
|