Calculate nonparametric cumulative incidence functions and associated standard errors
This function computes nonparametric cumulative incidence functions and associated standard errors for each value of a group variable.
Cuminc(time, status, data, group, na.status = c("remove", "extra"), ...)
time: Either 1) a numeric vector containing the failure times or 2) a string containing the column name indicating these failure times
status: Either 1) a numeric, factor or character vector containing the failure codes or 2) a string containing the column name indicating these failure codes
data: When appropriate, a data frame containing time, status and/or group variables
group: Optionally, name of column in data indicating a grouping variable; cumulative incidence functions are calculated for each value or level of group. If missing no groups are considered
na.status: One of "remove" (default) or "extra", indicating whether subjects with missing cause of failure should be removed or whether missing cause of failure should be treated as a separate cause of failure
...: Allows extra arguments for future extensions, but for now just used for backwards compatibility (e.g. allowing use of defunct failcodes
argument in reverse dependencies).
An object of class "Cuminc", which is a data frame containing the estimated failure-free probabilities and cumulative incidences and their standard errors. The names of the dataframe are time, Surv, seSurv, and cuminc and secuminc followed by the values or levels of the failcodes. If group was specified, a group variable is included with the same name and values/levels as the original grouping variable, and with estimated cumulative incidences (SE) for each value/level of group.
Cuminc is now simply a wrapper around survfit of the survival package with type="mstate", only maintained for backward compatibility. The survfit object is kept as attribute (attr("survfit")), and the print, plot and summary functions are simply print, plot and summary applied to the survfit object. Subsetting the "Cuminc" object results in subsetting the data frame, not in subsetting the survfit object.
The estimated cumulative incidences are as described in Putter, Fiocco & Geskus (2007); the standard errors are the square roots of the Greenwood variance estimators, see eg. Andersen, Borgan, Gill & Keiding (1993), de Wreede, Fiocco & Putter (2009), and they correspond to the variances in eg. Marubini & Valsecchi (1995). In case of no censoring, the estimated cumulative incidences and variances reduce to simple binomial frequencies and their variances.
### These data were used in Putter, Fiocco & Geskus (2007) data(aidssi) ci <- Cuminc(time=aidssi$time, status=aidssi$status) head(ci); tail(ci) ci <- Cuminc(time="time", status="status", data=aidssi, group="ccr5") head(ci); tail(ci) ### Some fake data fake <- data.frame(surv=c(seq(2,10,by=2),seq(1,13,by=3),seq(1,9,by=2),seq(1,13,by=3)), stat=rep(0:3,5),Tstage=c(1:4,rep(1:4,rep(4,4)))) fake$stat[fake$stat==0 & fake$Tstage==2] <- 3 fake$stat[fake$stat==3 & fake$Tstage==1] <- 2 fake Cuminc(time="surv", status="stat", data=fake) # If we remove all entries with status=0, # we should get binomial sample probabilities and corresponding SEs fake0 <- fake[fake$stat!=0,] Cuminc(time="surv", status="stat", data=fake0)
Andersen PK, Borgan O, Gill RD, Keiding N (1993). Statistical Models Based on Counting Processes. Springer, New York.
Marubini E, Valsecchi MG (1995). Analysing Survival Data from Clinical Trials and Observational Studies. Wiley, New York.
Putter H, Fiocco M, Geskus RB (2007). Tutorial in biostatistics: Competing risks and multi-state models. Statistics in Medicine 26 , 2389--2430.
de Wreede L, Fiocco M, Putter H (2009). The mstate package for estimation and prediction in non- and semi-parametric multi-state models. Submitted.
Hein Putter H.Putter@lumc.nl