R/rsq.R

Defines functions rsq.survfit rsq.coxph rsq

Documented in rsq rsq.coxph rsq.survfit

#' @name rsq
#' @title r^2 measures for a a \code{coxph} or \code{survfit} model
#'
#' @param x A \code{survfit} or \code{coxph} object
#' @param sigD \bold{sig}nificant \bold{d}igits (for ease of display).
#' If \code{sigD=NULL}, will return the original numbers.
#' @inheritParams sf.ten
#' 
#' @return A \code{list} with the following elements:
#'  \item{cod}{The \bold{c}oefficient \bold{o}f \bold{d}etermination, which is
#'             \deqn{R^2=1-\exp(\frac{2}{n}L_0-L_1)}{
#'                   R^2 = 1-exp((2/n).(L[0]-L[1]))}
#'            where \eqn{L_0}{L[0]} and \eqn{L_1}{L[1]} are the log partial
#'            likelihoods for the \emph{null} and \emph{full} models respectively
#'            and \eqn{n}
#'            is the number of observations in the data set.}
#' \item{mer}{The \bold{m}easure of \bold{e}xplained \bold{r}andomness, which is:
#'            \deqn{R^2_{mer}=1-\exp(\frac{2}{m}L_0-L_1)}{
#'                  R^2[mer] = 1-exp((2/m).(L[0]-L[1]))}
#'            where \eqn{m} is the number of observed \emph{events}.}
#' \item{mev}{The \bold{m}easure of \bold{e}xplained \bold{v}ariation (similar to
#'            that for linear regression), which is:
#'           \deqn{R^2=\frac{R^2_{mer}}{R^2_{mer} + \frac{\pi}{6}(1-R^2_{mer})}}{
#'                 R^2 = R^2[mer] / ( R^2[mer] + pi/6(1-R^2[mer]) )}
#' }
#'
#' @references Nagelkerke NJD, 1991.
#' A Note on a General Definition of the Coefficient of Determination.
#' \emph{Biometrika} \bold{78}(3):691--92.
#' \href{http://www.jstor.org/stable/2337038}{JSTOR}
#' @references
#' O'Quigley J, Xu R, Stare J, 2005.
#' Explained randomness in proportional hazards models.
#' \emph{Stat Med} \bold{24}(3):479--89.
#' \href{http://dx.doi.org/10.1002/sim.1946}{Wiley (paywall)}
#' \href{http://www.math.ucsd.edu/~rxu/igain2.pdf}{Available at UCSD}
#' @references
#' Royston P, 2006.
#' Explained variation for survival models.
#' \emph{The Stata Journal} \bold{6}(1):83--96.
#' \href{http://www.stata-journal.com/sjpdf.html?articlenum=st0098}{The Stata Journal}
#'
#' @rdname rsq
#' @export 
#'
rsq <- function(x, ...) UseMethod("rsq")
#'
#' @rdname rsq
#' @export
#' @method rsq coxph
#' @aliases rsq.coxph
#' @examples
#' data("kidney", package="KMsurv")
#' c1 <- coxph(Surv(time=time, event=delta) ~ type, data=kidney)
#' cbind(rsq(c1), rsq(c1, sigD=NULL))
#'
rsq.coxph <- function(x, ..., sigD=2){
  stopifnot(inherits(x, "coxph"))
  l0 <- x$loglik[1]
  l1 <- x$loglik[2]
  n1 <- x$n
  ne1 <- x$nevent
  res1 <- vector(mode="list", length = 3L)
  names(res1) <- c("cod", "mer", "mev")
  res1$cod <- 1 - exp((2 / n1) * (l0 - l1))
  res1$mer <- 1 - exp((2 / ne1) * (l0 - l1))
  res1$mev <- res1$mer / (res1$mer + pi^2 / 6 * (1 - res1$mer))
  if (is.null(sigD)) return(res1)
  res1 <- lapply(res1, function(N)
                 as.numeric(formatC(signif(N, digits = sigD),
                                    digits=sigD,
                                    format="fg", flag="#")))
  return(res1)
}
#'
#' @rdname rsq
#' @export
#' @method rsq survfit
#' @aliases rsq.survfit
#' 
rsq.survfit <- function(x, ..., sigD=2){
    c1 <- deparse(x$call)
    c1 <- sub("survfit", "coxph", c1)
    c1 <- eval(parse(text=c1))
    rsq(c1, sigD=sigD)
}

Try the survMisc package in your browser

Any scripts or data that you put into this service are public.

survMisc documentation built on May 2, 2019, 5:14 p.m.