Displaying 1 result from an estimated 1 matches for "creat6mo".
Did you mean:
creat64
2010 Dec 07
0
coxph failure
...7627778
Unfortunately exp(900) = Inf, and that pretty much does in my C routine
right there; within another iteration there are NA values littered about
and the return value for the coefficients is all NA.
If instead you fit a smoothing spline to the data
coxph(Surv(TTFAIL, FFAIL) ~ pspline(CREAT6MO, df=4), data=anomaly1)
the fitted response is essentially linear from 0 to 5 before turning
upward, but it never gets outrageous due to the shrinkage towards
linearity.
Catching this type of issue with an extremely skewed variable
(creatinine^4) would be very hard to do reliably.
The primary mes...