Skip to content
Prev 247963 / 398503 Next

Function comparable to cutpt.coxph from "Survival Analysis using S"

It is very uncommon for the assumptions underlying this method to be
satisfied.  These assumptions include (1) the relationship between X and log
relative hazard is discontinuous at X=c and only X=c; (2) c is correctly
found as the cutpoint; (3) X vs log hazard is flat to the left of c; (4) X
vs log hazard is flat to the right of c; (5) the 'optimal' cutpoint does not
depend on the values of other predictors.

These relationships rarely occur in nature unless X=time.  Failure to have
these assumptions satisfied will result in (1) great error in estimating c
(because c doesn't exist); (2) low predictive accuracy; (3) serious lack of
fit; (4) residual confounding; and (5) overestimation of effects of
remaining variables.

This non-existence of cutpoints is why in medical research no two
investigators seem to find the same cutpoint for the same predictor in
different datasets.

Frank


-----
Frank Harrell
Department of Biostatistics, Vanderbilt University