🚀 go-pugleaf

I am involved in a study of critically ill patients for whom we
wish to compare two groups with regard to the timing of
"Do Not Resussitate" (DNR) orders. Not all patients had DNRs
and many have died. The risk of DNR increases as the risk of
death increases, hence there is informative censoring.
In the spirit of the Mantel-Haenszel-logrank statistic (and
related to the work of Pepe in Stat. in Med. and JASA), I can
conceive of a cascade of 2x2 tables of the form:

	For the n patients still alive at time t,
	what is the fraction, in each treatment group,
	who had DNR before time t.

Vary t from 0 to the largest follow-up time.
I need assistance in combining these tables into a test
statistic, or a pointer to an existing statistic suitable
for comparing non-fatal endpoints in the presence of informative
censoring (a terminating event, death).

Frank Harrell
Division of Biometry
Duke U.
Durham NC USA
feh@biostat.mc.duke.edu (our mail header gives an erroneous address)


-- 
----------------------------------------------------------------------------
Frank E Harrell Jr			feh@biostat.mc.duke.edu
Associate Professor of Biostatistics
Division of Biometry                    Duke University Medical Center