lcmm (version 1.5.7)

epoce: Computation of estimators of the Expected Prognostic Observed Cross-Entropy for evaluating predictive accuracy of joint latent class models estimated using Jointlcmm

Description

This functions computes estimators of the Expected Prognostic Observed Cross-Entropy for evaluating the predictive accuracy of joint latent class models estimated using Jointlcmm. On the same data as used for estimation of the Jointlcmm object, this function computes both the Mean Prognostic Observed Log-Likelihood (MPOL) and the Cross-Validated Observed Log-Likelihood (CVPOL). The latter corrects for over-optimism of the MPOL by approximated cross-validation. On external data, this function only computes the Mean Prognostic Observed Log-Likelihood (MPOL).

Usage

epoce(model, pred.times, var.time, newdata = NULL, subset=NULL, na.action=1)

Arguments

model

an object inheriting from class Jointlcmm

pred.times

Vector of times of prediction, from which predictive accuracy is evaluated (only subjects still at risk at the time of prediction are included in the computation, and only information before the time of prediction is considered

var.time

Name of the variable indicating time in the dataset

newdata

optional. When missing, the data used for estimating the Jointlcmm object are used, and CVPOL and MPOL are computed (internal validation). When newdata is specified, only MPOL is computed on this newdataset (external validation).

subset

a specification of the rows to be used: defaults to all rows. This can be any valid indexing vector for the rows of data or if that is not supplied, a data frame made up of the variable used in formula.

na.action

Integer indicating how NAs are managed. The default is 1 for 'na.omit'. The alternative is 2 for 'na.fail'. Other options such as 'na.pass' or 'na.exclude' are not implemented in the current version.

Value

• call.Jointlcmm
  the Jointlcmm call
• call.epoce
  the matched call
• EPOCE
  Dataframe containing, for each prediction time s, the number of subjects still at risk at s (and with at least one measure before s), the number of events after s, the MPOL and the CVPOL when computation is done on the dataset used for Jointlcmm estimation
• IndivContrib
  Individual contributions to the prognostic observed log-likelihood at each time of prediction. Used for computing tracking intervals of EPOCE differences between models.
• new.data
  a boolean for internal use only, which is FALSE if computation is done on the same data as for code{Jointlcmm} estimation, and TRUE otherwise.

Details

MPOL computed at time s equals minus the mean individual contribution to the conditional log-likelihood of the time to event given the longitudinal data up to time of prediction s and that the subject is still at risk of event in s.

CVPOL computed in time s equals MPOL in s plus a penalty term that corrects for over-optimism when computing predictive accuracy measures on the same dataset as used for estimation. This penalty term is computed from the inverse of the Hessian of the joint log-likelihood and the product of the gradients of the contributions to respectively the joint log-likelihood and the conditional log-likelihood.

The theory of EPOCE and its estimators MPOL and CVPOL is given in Commenges et al. (2012), and further detailed and illustrated for joint models in Proust-Lima et al. (2012).

References

Commenges, Liquet and Proust-Lima (2012). Choice of prognostic estimators in joint models by estimating differences of expected conditional {K}ullback-{L}eibler risks. Biometrics - in press

Proust-Lima, Sene, Taylor and Jacqmin-Gadda (2012). Joint latent class models of longitudinal and time-to-event data: a review. Statistical Methods in Medical Research - in press

See Also

Jointlcmm,print.epoce,summary.epoce,plot.epoce

Examples

#### estimation of a joint latent class model with 2 latent classes (ng=2)
# (see the example section of Jointlcmm for details about the model specification

data(data_Jointlcmm)
m <- Jointlcmm(fixed= Ydep1~Time*X1,random=~Time,mixture=~Time,subject='ID'
,survival = Surv(Tevent,Event)~ X1+X2 ,hazard="Weibull"
,hazardtype="PH",ng=2,data=data_Jointlcmm,
B=c(0.7608, -9.4974 , 1.0242,  1.4331 , 0.1063 , 0.6714, 10.4679, 11.3178,
 -2.5671, -0.5386,  1.4616, -0.0605,  0.9489,  0.1020 , 0.2079,  1.5045))
summary(m)

## Computation of the EPOCE on the same dataset as used for estimation of m
# with times at predictions from 1 to 15 
VecTime <- c(1,3,5,7,9,11,13,15)
cvpl <- epoce(m,var.time="Time",pred.times=VecTime)
summary(cvpl)
plot(cvpl,bty="l",ylim=c(0,2))