mstate (version 0.2.6)

crprep: Function to create weighted data set for competing risks analyses

Description

This function converts a dataset which is in short format (one subject per line, one column indicating type of end point at end of follow-up) into a weighted dataset in counting process style notation. With this data set, competing risks analyses can be performed.

Usage

crprep(Tstop, status, data, trans = 1, cens = 0, Tstart=0, id, keep, shorten = TRUE, rm.na = TRUE, origin = 0, prec.factor = 100)

Arguments

Tstop

Either 1) a vector containing the time at which the follow-up is ended, or 2) a character string indicating the column name that contains the end times. Missing values are allowed.

status

Either 1) a vector describing status at end of follow-up, or 2) a character string indicating the column name that contains this information. Missing values are allowed. See "Details".

data

Data frame in which to interpret "Tstart", "status", "trans", "cens", "Tstart", "id" and "keep", if appropriate.

trans

Values of "status" that are the event types of interest. Defaults to 1. See "Details".

cens

Value of "status" indicating censoring. Defaults to 0.

Tstart

Either 1) a vector containing the individual times at which the follow-up is started, or 2) a character string indicating the column name that contains the entry times, or 3) One numeric value in case it is the same for every subject. Missi

id

Either 1) a vector containing the subject identifiers, or 2) a character string indicating the column name containing these subject identifiers. If not provided, "id" will be assigned with values 1,...,n.

keep

Either 1) a data frame or matrix with n rows or a numeric or factor vector of length n containing covariate(s) that need to be retained in the output dataset, or 2) a character vector containing the column names of these covariates in data.

shorten

Logical. If true, number of rows in output is reduced by collapsing rows within a subject in which weights do not change.

rm.na

Logical. If true, rows for which any of "Tstart", "status" or "Tstart" is missing are deleted.

origin

Substract origin time units from all Tstop and Tstart times.

prec.factor

Factor by which to multiply the machine's precision. Censoring and truncation times are shifted by prec.factor*precision if event times and censoring/trunaction times are equal.

Value

• A data frame in long (counting process) format containing the covariates (replicated per subject) and the following columns
• patid
  subject identifier (1:n if argument "id" was missing)
• Tstart
  start dates of dataset (counting process notation)
• Tstop
  stop dates of dataset (counting process notation)
• status
  status of the subject at the end of his follow-up
• weight.cens
  weights due to censoring mechanism
• weight.trunc
  weights due to truncation mechanism (if present)
• count
  counter per subject and type of end point, 1 to number of rows per subject id and type of end point
• failcode
  type of end point, thus allowing to perform regression using the "long format" data set

Details

The function creates a data set that allows to perform analyses that are based on the subdistribution hazard. For each type of outcome as specified via "trans", individuals with a competing event remain in the risk set with a weight that depends on the censoring and truncation mechanisms in the data. Typically, their weights change over follow-up, and therefore such individuals are split into several rows. Several types of outcome can be specified at once, thus allowing for regression analyses using the "long format" data set (see Putter et al. 2007). A regression on the cause-specific hazard using the created data set can be performed by using "subset=count==0". If keep is a data.frame or a named matrix, the same names are used for the covariate columns in the output data set. If keep is a matrix without names, then the covariate columns are given the names "V1" until "Vk". If keep is a vector from a (sub)-list, e.g. obj$name2$name1, then the column name is based on the most inner part (i.e. "name1"). If keep is a vector of the form obj[,"name1"], then the column is named "name1". For all other specifications, the name is copied as is. The current function does not allow to create a weighted data set in which the censoring and /or truncation mechanisms depend on covariates. One option is to create a weighted data set for each level of a categorical covariate.

References

Geskus RB (2011). Cause-Specific Cumulative Incidence Estimation and the Fine and Gray Model Under Both Left Truncation and Right Censoring. Biometrics, to appear. Putter H, Fiocco M, Geskus RB (2007). Tutorial in biostatistics: Competing risks and multi-state models. Statistics in Medicine 26, 2389--2430.

Examples

# Create simulated data
set.seed(1234)
N <- 200
p <- 0.3
p.t <- 0.6
Z1 <- rnorm(N)
Z2 <- rnorm(N)
tmp <- runif(N)
Data <- data.frame(Tstart=runif(N,0,3)*rbinom(N,1,p.t),
          Tstop=ifelse( tmp>(1-p)^(exp(0.5*Z1+0.5*Z2)),
                  -log(1-(1-exp(log(tmp)/exp(0.5*Z1+0.5*Z2)))/p),
                  -log(tmp)/exp(-0.5*Z1+0.5*Z2)),
          stat=ifelse(tmp>(1-p)^(exp(0.5*Z1+0.5*Z2)),1,2),
          Z1=Z1, Z2=Z2, tstat=rep(0,N))
Data$type <- Data$stat
Data$tevent <- Data$Tstop
Data$cens <- runif(N,0.5,1)
Data[Data$cens<Data$Tstop,"stat"] <- 0
Data$Tstop <- pmin(Data$Tstop,Data$cens)
Data$tstat <- ifelse(Data$Tstart > Data$Tstop, 1, 0)
Data.weight <- crprep(Data$Tstop, Data$stat, trans=c(1,2))

# calculate cause-specific cumulative incidence, no truncation,
# compare with Cuminc (also from mstate)
ci <- Cuminc(Data$Tstop, Data$stat)
plot(ci$time,ci$CI.1,type="s",lwd=3,col="black",ylim=c(0,0.5),xlab="Time",ylab="Cumulative incidence")
lines(ci$time,ci$CI.2,type="s",lwd=3,col="red")
lines(ci$time,ci$CI.1-qnorm(0.975)*ci$seCI.1,type="s",lty=3)
lines(ci$time,ci$CI.1+qnorm(0.975)*ci$seCI.1,type="s",lty=3)
lines(survfit(Surv(Tstart,Tstop,status==1)~1,data=Data.weight,
   weight=weight.cens,subset=failcode==1),
   fun="event",col="lightblue",lwd=1,mark.time=FALSE)
lines(survfit(Surv(Tstart,Tstop,status==1)~1,data=Data.weight,
   weight=weight.cens,subset=failcode==1),
   fun="event",col="lightblue",mark.time=FALSE,conf.int="only",lty=2)


# Proportional hazards regression on subdistribution and cause-specific hazard,
# with truncation
Data.weight.trunc <- crprep("Tstop", "stat", Tstart="Tstart",
                       data=subset(Data,tstat==0), trans=1:2, keep=c("Z1","Z2"))
coxph(Surv(Tstart,Tstop,status==1)~Z1+Z2,data=Data.weight.trunc,
   weight=weight.cens*weight.trunc,subset=failcode==1) #cause 1
# Both end points, assume effect Z2 same for both
coxph(Surv(Tstart,Tstop,status==1)~strata(failcode)*Z1+Z2,
   data=Data.weight.trunc,weight=weight.cens*weight.trunc)
# Cause-specific hazard
coxph(Surv(Tstart,Tstop,stat==1)~Z1+Z2,data=subset(Data, tstat==0))
coxph(Surv(Tstart,Tstop,status==1)~Z1+Z2,data=Data.weight.trunc,
   subset=failcode==1&count==1)

## Not run: 
data(ebmt2)
ebmt2.long <- crprep("time", "status", data=ebmt2, trans=1:6,
                keep=c("dissub", "match", "tcd", "year", "age"))
# ebmt2.long <- with(ebmt2, crprep(time, cod, trans=levels(cod)[-1], cens="Alive",
#                  keep=c("dissub", "match", "tcd", "year", "age")))
plot(survfit(Surv(time, status>0)~1, data=ebmt2, etype=cod), mark.time=FALSE,
   col=2:7, fun="event",lwd=3)
lines(survfit(Surv(Tstart, Tstop, failcode==status) ~ failcode, data=ebmt2.long,
   weight=weight.cens), lwd=2, lty=2, fun="event", mark.time=FALSE)
## End(Not run)