When validating cox model: NaN when using rms::validate, Divergence or singularity in [all] samples

Question

I would like to fit a Cox model for prediction of 1-year survival using a specific set of predictors. The data contains 78 events in 620 subjects.

I don't think my problem is related to my data, so here is an example using R data in which I am also unable to use rms::validate().

library(survival)
library(rms)

data(cancer, package="survival")
dd <- datadist(cancer) 
options(datadist= 'dd')

# fit
s1yr <- Surv(cancer$time, cancer$status)
cph.euro.log <- cph(s1yr ~ log(age), data = cancer, method = 'exact',
                   x = TRUE, y = TRUE, surv = TRUE)

# validate
validate(cph.euro.log, B=20)

Divergence or singularity in 20 samples

	index.orig	training	test	optimism	index.corrected	n
Dxy	0.1005	NaN	NaN	NaN	NaN	0
R2	0.0179	NaN	NaN	NaN	NaN	0
Slope	1.0000	NaN	NaN	NaN	NaN	0

The output from validate above is similar to what I get for my real data. The function gives point estimates for the predictive performance measures but nothing more (just NaN). What could be the issue?

Answer 1

Straight from the author's mouth: The function is not yet available for survival models fit using the 'exact' method. If one drops 'exact' and uses, for example, the default method 'efron', it works.