timereg (version 2.0.1)

rchaz: Simulation of Piecewise constant hazard model (Cox).

Description

Simulates data from piecwise constant baseline hazard that can also be of Cox type. Censor data at highest value of the break points.

Usage

rchaz(
  cumhazard,
  rr,
  n = NULL,
  entry = NULL,
  cum.hazard = TRUE,
  cause = 1,
  extend = FALSE
)

Arguments

cumhazard

cumulative hazard, or piece-constant rates for periods defined by first column of input.

rr

relative risk for simulations, alternatively when rr=1 specify n

n

number of simulation if rr not given

entry

delayed entry time for simuations.

cum.hazard

specifies wheter input is cumulative hazard or rates.

cause

name of cause

extend

to extend piecewise constant with constant rate. Default is average rate over time from cumulative (when TRUE), if numeric then uses given rate.

Details

For a piecewise linear cumulative hazard the inverse is easy to compute with and delayed entry x we compute $$\Lambda^{-1}(\Lambda(x) + E/RR)$$, where RR are the relative risks and E is exponential with mean 1. This quantity has survival function $$P(T > t | T>x) = exp(-RR (\Lambda(t) - \Lambda(x)))$$.

Examples

# NOT RUN {
rates <-  c(0,0.01,0.052,0.01,0.04)
breaks <- c(0,10,   20,  30,   40)
haz <- cbind(breaks,rates)
n <- 1000
X <- rbinom(n,1,0.5)
beta <- 0.2
rrcox <- exp(X * beta)
cumhaz <- cumsum(c(0,diff(breaks)*rates[-1]))
cumhaz <- cbind(breaks,cumhaz)

pctime <- rchaz(haz,n=1000,cum.hazard=FALSE)

par(mfrow=c(1,2))
ss <- aalen(Surv(time,status)~+1,data=pctime,robust=0)
plot(ss)
lines(cumhaz,col=2,lwd=2)

pctimecox <- rchaz(cumhaz,rrcox)
pctime <- cbind(pctime,X)
ssx <- cox.aalen(Surv(time,status)~+prop(X),data=pctimecox,robust=0)
plot(ssx)
lines(cumhaz,col=2,lwd=2)

### simulating data with hazard as real data 
data(TRACE)

par(mfrow=c(1,2))
ss <- cox.aalen(Surv(time,status==9)~+prop(vf),data=TRACE,robust=0)
par(mfrow=c(1,2))
plot(ss)
###
pctime <- rchaz(ss$cum,n=1000)
###
sss <- aalen(Surv(time,status)~+1,data=pctime,robust=0)
lines(sss$cum,col=2,lwd=2)

pctime <- rchaz(ss$cum,rrcox)
pctime <- cbind(pctime,X)
###
sss <- cox.aalen(Surv(time,status)~+prop(X),data=pctime,robust=0)
summary(sss)
plot(ss)
lines(sss$cum,col=3,lwd=3)

# }