Imagine a Cox proportional hazards regression computed on a binary covariate
Z. As a result one obtains in standard fashion estimates of beta, the log
hazard ratio (Z=1 vs. Z=0), and its standard error.
My question: a good approximation to the standard error appears to be
sqrt(1/e0+1/e1), where e0 and e1 are the number of events in the two
subgroups Z=0 and Z=1. Can anyone provide me with a literature reference to
this approximation? Are there other flavours of it?
I would be most grateful for help.
Thanks
Patrick.
Prof Patrick Royston
MRC Clinical Trials Unit
London NW1 2DA.
[log in to unmask]
|