duplicate {xhaz}

duplicate function

Description

Duplicate data for survival analysis in the context of competing risks, where an individual can experience only one of alternative events, using the Lunn & McNeil (Biometrics, 1995) approaches. Duplication of data proceeds as follows: Suppose that we study J distinct types of events. Each observation concerning a given subject is duplicated J times, with one row for each type of event. In addition, (J-1) dummy variables are created, each indicating the type of event in relation with that observation (delta.j=1 if the event of type j is the observed one and 0 otherwise). Since, for a given subject, only the first occurring event is considered, the status indicator equals 1 for that event and 0 for all the others. In the case of a censored observation (dropout or administrative censoring), the same principle applies also: duplication of each subject's data is made J times with (J-1) dummy variables and a status indicator equal to 0 for all observations.

Usage

duplicate(status, event, data)

Arguments

status

the censoring status indicator (numeric vector), 0=alive, 1=dead.

event

the indicator of the event type (numeric vector). By default, the event==0 acts as the censoring indicator.

data

a data frame containing the data to duplicate.

Value

A data.frame containing the duplicated data with the new dummy variables, named delta.number_of_the_event, indicating the type of event.

Author(s)

Roch Giorgi

References

Lunn M and McNeil D. Applying Cox regression to competing risks. Biometrics 1995;51:524-532 (PubMed)

Examples

## Create the simplest test data set
data1 <- data.frame(futime     = c(1, 2, 5, 2, 1, 7, 3, 4, 8, 8),
                    fustat     = c(0, 1, 1, 1, 0, 0, 1, 0, 1, 1),
                    firstevent = c(0, 2, 1, 2, 0, 0, 1, 0, 2, 2),
                    sex        = c(1, 0, 0, 1, 0, 1, 1, 1, 0, 0))

## Duplicate data1 with firstevent == 0 as the censoring indicator.
library(xhaz)
dupli.data <- duplicate(status=fustat, event=firstevent, data=data1)


data2 <- data.frame(futime = c(10, 2, 7, 3, 4, 9, 13, 2, 5, 9),
                    fustat = c(0, 1, 1, 1, 0, 0, 1, 0, 1, 1),
                    firstevent = c(3, 2, 1, 2, 3, 3, 1, 3, 2, 2),
                    sex = c(1, 0, 0, 1, 0, 1, 1, 1, 0, 0))


## Duplicate data1 with firstevent == 3 as the censoring indicator.

dupli.data <- duplicate(status = fustat,
                        event = firstevent == 3,
                        data = data2)


# Joint modeling
coxph(Surv(futime, fustat) ~ delta.2 + sex + delta.2:(sex), data = dupli.data)

coxph(Surv(futime, fustat) ~ delta.1 + sex + delta.1:(sex), data = dupli.data)

# exemple using ccr.mevents data


ccr.mevents$loc.rec <- as.numeric(ccr.mevents$event == 1)
ccr.mevents$dist.rec <- as.numeric(ccr.mevents$event == 2)
ccr.mevents$death <- as.numeric(ccr.mevents$event == 3)
# Age centered to mean and scaled
ccr.mevents$agecr <-  scale(ccr.mevents$age, TRUE, TRUE)

## Duplication of the data with local recurrence as the reference
dupli.ccr.mevents <- duplicate(status = status,
                               event = event, data = ccr.mevents)
head(dupli.ccr.mevents)
# joint model including overall mortality modelling
fit <- coxph(Surv(time, status) ~ agecr + sexe + stage + delta.2 + delta.3,
             data = dupli.ccr.mevents)

fit

# add expected mortality from french life table to the data


library(survexp.fr)
fit.haz <- exphaz(formula = Surv(time, death) ~ 1,
                  data = dupli.ccr.mevents,
                  ratetable = survexp.fr, only_ehazard = TRUE,
                  rmap = list(age = 'age', sex = 'sexe', year = 'date_diag'))

 dupli.ccr.mevents$mua <- fit.haz$ehazard * dupli.ccr.mevents$delta.3

# joint model including excess hazard modelling
library(mexhaz)
fit.mort <- mexhaz(
    Surv(time, status) ~ delta.2 + delta.3,
    data = dupli.ccr.mevents, base = "exp.bs", degree = 3, knots = c(1),
    expected = "mua")

fit.mort