eventTime: Add an observed event time outcome to a latent variable model.

Description

For example, if the model 'm' includes latent event time variables are called 'T1' and 'T2' and 'C' is the end of follow-up (right censored), then one can specify

Usage

1
eventTime(object, formula, eventName, ...)

Arguments

object

Model object

formula

Formula (see details)

eventName

Event names

...

Additional arguments to lower levels functions

Details

eventTime(object=m,formula=ObsTime~min(T1=a,T2=b,C=0,"ObsEvent"))

when data are simulated from the model one gets 2 new columns:

- "ObsTime": the smallest of T1, T2 and C - "ObsEvent": 'a' if T1 is smallest, 'b' if T2 is smallest and '0' if C is smallest

Note that "ObsEvent" and "ObsTime" are names specified by the user.

Author(s)

Thomas A. Gerds

Examples

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
# Right censored survival data without covariates
m0 <- lvm()
distribution(m0,"eventtime") <- coxWeibull.lvm(scale=1/100,shape=2)
distribution(m0,"censtime") <- coxExponential.lvm(rate=10)
m0 <- eventTime(m0,time~min(eventtime=1,censtime=0),"status")
sim(m0,10)

# Alternative specification of the right censored survival outcome
## eventTime(m,"Status") <- ~min(eventtime=1,censtime=0)

# Cox regression:
# lava implements two different parametrizations of the same
# Weibull regression model. The first specifies
# the effects of covariates as proportional hazard ratios
# and works as follows:
m <- lvm()
distribution(m,"eventtime") <- coxWeibull.lvm(scale=1/100,shape=2)
distribution(m,"censtime") <- coxWeibull.lvm(scale=1/100,shape=2)
m <- eventTime(m,time~min(eventtime=1,censtime=0),"status")
distribution(m,"sex") <- binomial.lvm(p=0.4)
distribution(m,"sbp") <- normal.lvm(mean=120,sd=20)
regression(m,from="sex",to="eventtime") <- 0.4
regression(m,from="sbp",to="eventtime") <- -0.01
sim(m,6)
# The parameters can be recovered using a Cox regression
# routine or a Weibull regression model. E.g.,
## Not run: 
    set.seed(18)
    d <- sim(m,1000)
    library(survival)
    coxph(Surv(time,status)~sex+sbp,data=d)

    sr <- survreg(Surv(time,status)~sex+sbp,data=d)
    library(SurvRegCensCov)
    ConvertWeibull(sr)


## End(Not run)

# The second parametrization is an accelerated failure time
# regression model and uses the function weibull.lvm instead
# of coxWeibull.lvm to specify the event time distributions.
# Here is an example:

ma <- lvm()
distribution(ma,"eventtime") <- weibull.lvm(scale=3,shape=0.7)
distribution(ma,"censtime") <- weibull.lvm(scale=2,shape=0.7)
ma <- eventTime(ma,time~min(eventtime=1,censtime=0),"status")
distribution(ma,"sex") <- binomial.lvm(p=0.4)
distribution(ma,"sbp") <- normal.lvm(mean=120,sd=20)
regression(ma,from="sex",to="eventtime") <- 0.7
regression(ma,from="sbp",to="eventtime") <- -0.008
set.seed(17)
sim(ma,6)
# The regression coefficients of the AFT model
# can be tranformed into log(hazard ratios):
#  coef.coxWeibull = - coef.weibull / shape.weibull
## Not run: 
    set.seed(17)
    da <- sim(ma,1000)
    library(survival)
    fa <- coxph(Surv(time,status)~sex+sbp,data=da)
    coef(fa)
    c(0.7,-0.008)/0.7

## End(Not run)


# The Weibull parameters are related as follows:
# shape.coxWeibull = 1/shape.weibull
# scale.coxWeibull = exp(-scale.weibull/shape.weibull)
# scale.AFT = log(scale.coxWeibull) / shape.coxWeibull
# Thus, the following are equivalent parametrizations
# which produce exactly the same random numbers:

model.aft <- lvm()
distribution(model.aft,"eventtime") <- weibull.lvm(scale=-log(1/100)/2,shape=0.5)
distribution(model.aft,"censtime") <- weibull.lvm(scale=-log(1/100)/2,shape=0.5)
set.seed(17)
sim(model.aft,6)

model.cox <- lvm()
distribution(model.cox,"eventtime") <- coxWeibull.lvm(scale=1/100,shape=2)
distribution(model.cox,"censtime") <- coxWeibull.lvm(scale=1/100,shape=2)
set.seed(17)
sim(model.cox,6)

# The minimum of multiple latent times one of them still
# being a censoring time, yield
# right censored competing risks data

mc <- lvm()
distribution(mc,~X2) <- binomial.lvm()
regression(mc) <- T1~f(X1,-.5)+f(X2,0.3)
regression(mc) <- T2~f(X2,0.6)
distribution(mc,~T1) <- coxWeibull.lvm(scale=1/100)
distribution(mc,~T2) <- coxWeibull.lvm(scale=1/100)
distribution(mc,~C) <- coxWeibull.lvm(scale=1/100)
mc <- eventTime(mc,time~min(T1=1,T2=2,C=0),"event")
sim(mc,6)

Questions? Problems? Suggestions? or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.