| 1 |
% Generated by roxygen2: do not edit by hand |
| 2 |
% Please edit documentation in R/crprep.R |
| 3 |
\name{crprep.default} |
| 4 |
\alias{crprep.default} |
| 5 |
\alias{crprep} |
| 6 |
\title{Function to create weighted data set for competing risks analyses} |
| 7 |
\usage{ |
| 8 |
\method{crprep}{default}( |
| 9 |
Tstop, |
| 10 |
status, |
| 11 |
data, |
| 12 |
trans = 1, |
| 13 |
cens = 0, |
| 14 |
Tstart = 0, |
| 15 |
id, |
| 16 |
strata, |
| 17 |
keep, |
| 18 |
shorten = TRUE, |
| 19 |
rm.na = TRUE, |
| 20 |
origin = 0, |
| 21 |
prec.factor = 1000, |
| 22 |
... |
| 23 |
) |
| 24 |
} |
| 25 |
\arguments{ |
| 26 |
\item{Tstop}{Either 1) a vector containing the time at which the follow-up |
| 27 |
is ended, or 2) a character string indicating the column name in \code{data} |
| 28 |
that contains the end times (see Details).} |
| 29 |
|
| 30 |
\item{status}{Either 1) a vector describing status at end of follow-up, |
| 31 |
having the same length as \code{Tstop}, or 2) a character string indicating |
| 32 |
the column name that contains this information.} |
| 33 |
|
| 34 |
\item{data}{Data frame in which to interpret \code{Tstart}, \code{status}, |
| 35 |
\code{Tstart}, \code{id}, \code{strata} and \code{keep}, if given as |
| 36 |
character value (specification 2, "by name").} |
| 37 |
|
| 38 |
\item{trans}{Values of \code{status} for which weights are to be calculated.} |
| 39 |
|
| 40 |
\item{cens}{Value that denotes censoring in \code{status} column.} |
| 41 |
|
| 42 |
\item{Tstart}{Either 1) a vector containing the time at which the follow-up |
| 43 |
is started, having the same length as \code{Tstop}, or 2) a character string |
| 44 |
indicating the column name that contains the entry times, or 3) one numeric |
| 45 |
value in case it is the same for every subject. Default is 0.} |
| 46 |
|
| 47 |
\item{id}{Either 1) a vector, having the same length as \code{Tstop}, |
| 48 |
containing the subject identifiers, or 2) a character string indicating the |
| 49 |
column name containing these subject identifiers. If not provided, a column |
| 50 |
\code{id} is created with subjects having values 1,...,n.} |
| 51 |
|
| 52 |
\item{strata}{Either 1) a vector of the same length as \code{Tstop}, or 2) a |
| 53 |
character string indicating the column name that contains this information. |
| 54 |
Weights are calculated for per value in this vector.} |
| 55 |
|
| 56 |
\item{keep}{Either 1) a data frame or matrix or a numeric or factor vector |
| 57 |
containing covariate(s) that need to be retained in the output dataset. |
| 58 |
Number of rows/length should correspond with \code{Tstop}, or 2) a character |
| 59 |
vector containing the column names of these covariates in \code{data}.} |
| 60 |
|
| 61 |
\item{shorten}{Logical. If true, number of rows in output is reduced by |
| 62 |
collapsing rows within a subject in which weights do not change.} |
| 63 |
|
| 64 |
\item{rm.na}{Logical. If true, rows for which \code{status} is missing are |
| 65 |
deleted.} |
| 66 |
|
| 67 |
\item{origin}{Substract origin time units from all Tstop and Tstart times.} |
| 68 |
|
| 69 |
\item{prec.factor}{Factor by which to multiply the machine's precision. |
| 70 |
Censoring and truncation times are shifted by prec.factor*precision if event |
| 71 |
times and censoring/truncation times are equal.} |
| 72 |
|
| 73 |
\item{\dots}{Further arguments to be passed to or from other methods. They |
| 74 |
are ignored in this function.} |
| 75 |
} |
| 76 |
\value{ |
| 77 |
A data frame in long (counting process) format containing the |
| 78 |
covariates (replicated per subject). The following column names are used: |
| 79 |
\item{Tstart}{start dates of dataset} \item{Tstop}{stop dates of dataset} |
| 80 |
\item{status}{status of the subject at the end of that row} |
| 81 |
\item{weight.cens}{weights due to censoring mechanism} |
| 82 |
\item{weight.trunc}{weights due to truncation mechanism (if present)} |
| 83 |
\item{count}{row number within subject and event type under consideration} |
| 84 |
\item{failcode}{event type under consideration} |
| 85 |
|
| 86 |
The first column is the subject identifier. If the argument "id" is missing, |
| 87 |
it has values 1:n and is named "id". Otherwise the information is taken from |
| 88 |
the \code{id} argument. |
| 89 |
|
| 90 |
Variables as specified in \code{strata} and/or \code{keep} are included as |
| 91 |
well (see Details). |
| 92 |
} |
| 93 |
\description{ |
| 94 |
This function converts a dataset that is in short format (one subject per |
| 95 |
line) into a counting process format with time-varying weights that correct |
| 96 |
for right censored and left truncated data. With this data set, analyses |
| 97 |
based on the subdistribution hazard can be performed. |
| 98 |
} |
| 99 |
\details{ |
| 100 |
For each event type as specified via \code{trans}, individuals with a |
| 101 |
competing event remain in the risk set with weights that are determined by |
| 102 |
the product-limit forms of the time-to-censoring and time-to-entry |
| 103 |
estimates. Typically, their weights change over follow-up, and therefore |
| 104 |
such individuals are split into several rows. Censoring weights are always |
| 105 |
computed. Truncation weights are computed only if \code{Tstart} is |
| 106 |
specified. |
| 107 |
|
| 108 |
If several event types are specified at once, regression analyses using the |
| 109 |
stacked format data set can be performed (see Putter et al. 2007 and Chapter |
| 110 |
4 in Geskus 2016). The data set can also be used for a regression on the |
| 111 |
cause-specific hazard by restricting to the subset \code{subset=count==0}. |
| 112 |
|
| 113 |
Missing values are allowed in \code{Tstop}, \code{status}, \code{Tstart}, |
| 114 |
\code{strata} and \code{keep}. Rows for which \code{Tstart} or \code{Tstart} |
| 115 |
is missing are deleted. |
| 116 |
|
| 117 |
There are two ways to supply the data. If given "by value" (option 1), the |
| 118 |
actual data vectors are used. If given "by name" (option 2), the column |
| 119 |
names are specified, which are read from the data set in \code{data}. In |
| 120 |
general, the second option is preferred. |
| 121 |
|
| 122 |
If data are given by value, the following holds for the naming of the |
| 123 |
columns in the output data set. If \code{keep}, \code{strata} or \code{id} |
| 124 |
is a vector from a (sub)-list, e.g. obj$name2$name1, then the column name is |
| 125 |
based on the most inner part (i.e.\ "name1"). If it is a vector of the form |
| 126 |
obj[,"name1"], then the column is named "name1". For all other vector |
| 127 |
specifications, the name is copied as is. If \code{keep} is a data.frame or |
| 128 |
a named matrix, the same names are used for the covariate columns in the |
| 129 |
output data set. If keep is a matrix without names, then the covariate |
| 130 |
columns are given the names "V1" until "Vk". |
| 131 |
|
| 132 |
The current function does not allow to create a weighted data set in which |
| 133 |
the censoring and/or truncation mechanisms depend on covariates via a |
| 134 |
regression model. |
| 135 |
} |
| 136 |
\examples{ |
| 137 |
|
| 138 |
data(aidssi) |
| 139 |
aidssi.w <- crprep("time", "cause", data=aidssi, trans=c("AIDS","SI"), |
| 140 |
cens="event-free", id="patnr", keep="ccr5") |
| 141 |
|
| 142 |
# calculate cause-specific cumulative incidence, no truncation, |
| 143 |
# compare with Cuminc (also from mstate) |
| 144 |
ci <- Cuminc(aidssi$time, aidssi$status) |
| 145 |
sf <- survfit(Surv(Tstart,Tstop,status=="AIDS")~1, data=aidssi.w, |
| 146 |
weight=weight.cens, subset=failcode=="AIDS") |
| 147 |
plot(sf, fun="event", mark.time=FALSE) |
| 148 |
lines(CI.1~time,data=ci,type="s",col="red") |
| 149 |
sf <- survfit(Surv(Tstart,Tstop,status=="SI")~1, data=aidssi.w, |
| 150 |
weight=weight.cens, subset=failcode=="SI") |
| 151 |
plot(sf, fun="event", mark.time=FALSE) |
| 152 |
lines(CI.2~time,data=ci,type="s",col="red") |
| 153 |
|
| 154 |
# Fine and Gray regression for cause 1 |
| 155 |
cw <- coxph(Surv(Tstart,Tstop,status=="AIDS")~ccr5, data=aidssi.w, |
| 156 |
weight=weight.cens, subset=failcode=="AIDS") |
| 157 |
cw |
| 158 |
# This can be checked with the results of crr (cmprsk) |
| 159 |
# crr(ftime=aidssi$time, fstatus=aidssi$status, cov1=as.numeric(aidssi$ccr5)) |
| 160 |
|
| 161 |
# Gray's log-rank test |
| 162 |
aidssi.wCCR <- crprep("time", "cause", data=aidssi, trans=c("AIDS","SI"), |
| 163 |
cens="event-free", id="patnr", strata="ccr5") |
| 164 |
test.AIDS <- coxph(Surv(Tstart,Tstop,status=="AIDS")~ccr5, data=aidssi.wCCR, |
| 165 |
weights=weight.cens, subset=failcode=="AIDS") |
| 166 |
test.SI <- coxph(Surv(Tstart,Tstop,status=="SI")~ccr5, data=aidssi.wCCR, |
| 167 |
weights=weight.cens, subset=failcode=="SI") |
| 168 |
## score test statistic and p-value |
| 169 |
c(test.AIDS$score, 1-pchisq(test.AIDS$score,1)) # AIDS |
| 170 |
c(test.SI$score, 1-pchisq(test.SI$score,1)) # SI |
| 171 |
# This can be compared with the results of cuminc (cmprsk) |
| 172 |
# with(aidssi, cuminc(time, status, group=ccr5)$Tests) |
| 173 |
# Note: results are not exactly the same |
| 174 |
|
| 175 |
} |
| 176 |
\references{ |
| 177 |
Geskus RB (2011). Cause-Specific Cumulative Incidence Estimation |
| 178 |
and the Fine and Gray Model Under Both Left Truncation and Right Censoring. |
| 179 |
\emph{Biometrics} \bold{67}, 39--49. |
| 180 |
|
| 181 |
Geskus, Ronald B. (2016). \emph{Data Analysis with Competing Risks and |
| 182 |
Intermediate States.} CRC Press, Boca Raton. |
| 183 |
|
| 184 |
Putter H, Fiocco M, Geskus RB (2007). Tutorial in biostatistics: Competing |
| 185 |
risks and multi-state models. \emph{Statistics in Medicine} \bold{26}, |
| 186 |
2389--2430. |
| 187 |
} |
| 188 |
\author{ |
| 189 |
Ronald Geskus |
| 190 |
} |
| 191 |
\keyword{datagen} |
| 192 |
\keyword{survival} |