[R] Extrapolating Cox predicted risk

[Terry Therneau]

On 09/26/2012 05:00 AM, r-help-request at r-project.org wrote:
> I generated predicted risk of death for each subject in the study by
> means of Cox proportional hazards model at 8 year of follow-up, a time
> point at which follow-up was more than 90% complete. It is possible to
> extrapolate to 10-year the predicted risk of each subjet  by assuming
> an exponential distribution?
> Any help would be greatly appreciated.
> Thanks for your consideration.
>

The ability to extrapolate over long distances is exactly why industrial reliability work 
uses parametric models (weibull or log-normal, see survreg) instead of the Cox model.

If any of your data goes out to 10 years, then the predictions for coxph will go out that 
far, just like they would for a Kaplan-Meier.  But, just like the KM, if there are only a 
handful of people out that far the extrapolated curve will be very noisy.

Terry Therneau