R help - May 2011 - Cumulative dose in survival models

Dear List,
?
I want to fit a CPH survival model with a time-dependent cumulative exposure
variable. More specifically, I have exposure measurements for all cohort members
at discrete times (yearly measurements) during follow-up.
?
I suppose that when estimating the model parameters the most correct would be to
have the cumulative exposure for everybody at risk at each death time, as I
cannot assume the cumulative exposure to be constant between the yearly
measurements . Thus, I need to calculate (interpolate) cumulative exposure for
everybody at risk at each death time and pass this to the estimation routine.
?
Does anyone know how to do this in R? Any help would be much appreciated.
?
Klaus Andersen

On May 10, 2011, at 10:52 , klaus at kaae-consult.dk wrote:
> 
> Dear List,
>  
> I want to fit a CPH survival model with a time-dependent cumulative
exposure variable. More specifically, I have exposure measurements for all
cohort members at discrete times (yearly measurements) during follow-up.
>  
> I suppose that when estimating the model parameters the most correct would
be to have the cumulative exposure for everybody at risk at each death time, as
I cannot assume the cumulative exposure to be constant between the yearly
measurements . Thus, I need to calculate (interpolate) cumulative exposure for
everybody at risk at each death time and pass this to the estimation routine.
>  
> Does anyone know how to do this in R? Any help would be much appreciated.
You didn't say what your data layout is. One record per yearly measurement?

Generic time splitting is discussed in the 2nd ed. of ISwR (or peek into the
scripts in the ISwR package). That is for a simplified situation (5 year age
groups) but should generalize to arbitrary cutpoints with some elaboration. Then
you "just" need to figure out the exposure. If you can things narrowed
down to

- cumulative exposure at last measurement
- current exposure
- time since last measurement at end of observation interval

then it would be obvious how to proceed. One worry would be whether you have the
current exposure for those who die. I.e., you do not want to use interpolations
that depend on _next_ year's measurements.


-- 
Peter Dalgaard
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com