Hi Jan
Wednesday, May 16, 2007, 5:29:20 PM, You wrote:
kj> I have a cox model and I want to extrapolate survival beyond
kj> the available time window to predict survival times.
kj> How I can do it with spss
Ouch!
First of all, I suppose you know the risks of extrapolation in any
regression model. General advice is usually "don't do it". Assuming
you must do that, there's a problem inherent to Cox regression that
might make the task quite difficult: the model is SEMIparametric. The
covariates effect are modelized (the parametric half), but the
baseline hazard function is empiric, it is not a mathematical
function, but it is obtained directly from the data instead (the non
parametric half) like in Kaplan-Meier:
Lambda(t)=Lambda(0)*EXP(b1X1+b2X2+...+bkXk)
This means that, although you can extrapolate for a given covariates
pattern, and get estimated HR, and survivals, you can't get survivals
for extrapolated survival times, since you lack infomation on the non
parametric part of the model.
A possible way (sorry I don't have much time now to give details and
look for references in books) to overcome the problem would be:
- Ask SPSS to save the baseline hazard function.
- Use NLR to try to estimate a model (Weibull, negative exponential...)
for the survival function.
- If the fit is really good, then use the model to extrapolate for the
survival time you want.
- Use the Cox model to estimate the survival probability at that
extrapolated time with a given covariates pattern.
--
Regards,
Dr. Marta García-Granero,PhD mailto:
[hidden email]
Statistician
---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".
(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)
Locked
