mrmr.cindex: Function to compute the concordance index for survival or...
In bhklab/survcomp: Performance Assessment and Comparison for Survival Analysis

Description Usage Arguments Value Note Author(s) References See Also Examples

Function to compute the minimum redundancy - maximum relevance (mRMR) ranking for a risk prediction or a binary classification task based on the concordance index. The mRMR feature selection has been adapted to use the concordance index to estimate the correlation between a variable and the output (binary or survival) data.

1
2
3

mrmr.cindex(x, surv.time, surv.event, cl, weights, comppairs=10, strat,
alpha = 0.05, outx = TRUE, method = c("conservative", "noether", "nam"),
alternative = c("two.sided", "less", "greater"), na.rm = FALSE)

`x`	a vector of risk predictions.
`surv.time`	a vector of event times.
`surv.event`	a vector of event occurence indicators.
`cl`	a vector of binary class indicators.
`weights`	weight of each sample.
`comppairs`	threshold for compairable patients.
`strat`	stratification indicator.
`alpha`	apha level to compute confidence interval.
`outx`	set to `TRUE` to not count pairs of observations tied on `x` as a relevant pair. This results in a Goodman-Kruskal gamma type rank correlation.
`method`	can take the value `conservative`, `noether` or `name` (see paper Pencina et al. for details).
`alternative`	a character string specifying the alternative hypothesis, must be one of `"two.sided"` (default), `"greater"` (concordance index is greater than 0.5) or `"less"` (concordance index is less than 0.5). You can specify just the initial letter.
`na.rm`	`TRUE` if missing values should be removed.

A mRMR ranking

The "direction" of the concordance index (< 0.5 or > 0.5) is the opposite than the rcorr.cens function in the Hmisc package. So you can easily get the same results than rcorr.cens by changing the sign of x.

Benjamin Haibe-Kains, Markus Schroeder

Harrel Jr, F. E. and Lee, K. L. and Mark, D. B. (1996) "Tutorial in biostatistics: multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing error", Statistics in Medicine, 15, pages 361–387.

Pencina, M. J. and D'Agostino, R. B. (2004) "Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation", Statistics in Medicine, 23, pages 2109–2123, 2004.

rcorr.cens, coxphCPE

set.seed(12345)
age <- rnorm(100, 50, 10)
sex <- sample(0:1, 100, replace=TRUE)
stime <- rexp(100)
cens   <- runif(100,.5,2)
sevent  <- as.numeric(stime <= cens)
stime <- pmin(stime, cens)
strat <- sample(1:3, 100, replace=TRUE)
weight <- runif(100, min=0, max=1)
comppairs <- 10
xx <- data.frame("age"=age, "sex"=sex)
cat("survival prediction:\n")
mrmr.cindex(x=xx, surv.time=stime, surv.event=sevent, strat=strat, weights=weight,
method="noether", comppairs=comppairs)