[R-sig-Epi] Competing risks with time-varying covariates

[Erik Lampa]

Dear list,

This is a somewhat off-topic question as I don't have a specific issue
with an R-function, but rather seek advice on how to proceed with my
analyses.

I have survival data on approximately 2000 subjects followed for about
35 years. The subjects had their first medical examination at the age
of 50 and were examined almost every 10th year after that, so the data
is in long form with multiple records per subject.

The event of interest now is cardiovascular mortality with mortality
from other causes being a competing event. I consider myself a novice
when it comes to analysis of these kinds of data, so as far as I
understand estimating the survival function is not of much use as I
have time-varying covariates. I therefore turned my attention to the
Poisson representation of the Cox model. I have created a dataset with
yearly increments of time (with age as the time-scale) using the Lexis
and splitLexis functions in the Epi package v. 1.1.17. Additionaly the
dataset has been duplicated and the status variable changed according
to type of event. The main issue for me now is the interpretation of
the effects. My model looks like this:

mymodel <- glm(status ~ type*( bs(age, 3)*( ... covariates
...))+offset(log(duration)), family=poisson(), data = mydata)

where status is a dummy for event / no event and type =  the type of
event. My intention is to model the effect of the covariates and age
on the type of event (the type * rest interaction) and allow for
possible time-dependent effects of the (time-varying) covariates (the
age * covariates interaction).

My question is whether one can estimate age-specific incidence rates
for some values of the covariates this way? (for example. the
incidence rates for cardiovascular mortality among males aged 50 are
x% y% and z% and among males aged 60 a% b% and c% for bmi values of
25, 30 and 35 ...)

Thank you for your time. /Erik