Hello, all:
I am looking at the survival of patients following initiation of dialysis.
There is a high early mortality with some patient groups, but not with others,
and I am particularly interested in the factors that predict this early
mortality. I am examining this using coxph with a time dependent variable in
the model for "early," set to 1 during the first period of follow-up and 0
subsequently.
I would also like to look at whether a variable that clumps the patients is a
significant source of variability. There are too many clustering factors to
look at this by cluster, and I am not actually interested in how any specific
cluster does -- just whether cluster is a source of variation. A standard way
to do this would be to use a fraility() variable in the model. But in my model
this would look at whether the cluster factor is a source of variability in the
underlying, long term mortality. I am interested mostly in whether the
clustering factor is a source of variability in the rate of "early" mortality.
That is, I am really interested in an "interaction" of "early" with the
clustering factor.
Is there a way to enter an interaction of "early" with frailty(something) in
the coxph function? I have tried various ways to do this, but each seems to
bomb the function out with a comment that I have either vectors of unequal
length or a singular matrix.
Many thanks, in advance, to any of you that can help me with this problem. And
also generic thanks to Terry Therneau, who has provided us with such a flexible
coxph function.
Larry Hunsicker
|