View PDF Version - RePub - Erasmus Universiteit Rotterdam
View PDF Version - RePub - Erasmus Universiteit Rotterdam
View PDF Version - RePub - Erasmus Universiteit Rotterdam
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Chapter 3.2<br />
144<br />
multiple markers was considered by Morrell et al. 7 and Marshall et al. 8 who based<br />
their discrimination rule on a multivariate (non)linear mixed model (M(N)LMM).<br />
Finally, Fieuws et al. 9 describe a discrimination procedure based on a multivariate<br />
generalized linear mixed model (MGLMM) which allowed them to include discrete<br />
markers as well.<br />
All of the above approaches assume a normal distribution for the random effects in<br />
the underlying mixed model. Nevertheless, it is known that it is difficult to check this<br />
assumption which cannot be evaluated using commonly used empirical Bayes estimates<br />
of individual random effects due to their shrinkage (Verbeke and Lesaffre10 ).<br />
Consequently, Verbeke and Molenberghs11 , Chapter 7 conclude that non-normality<br />
of the random effects can only be detected by comparing the results obtained under<br />
the normality assumption with results obtained from fitting a mixed model with<br />
relaxed distributional assumptions for the random effects. Moreover, according to<br />
Komárek et al. 12 , the most promising approach for discrimination based on longitudinal<br />
profiles is based on predictors of individual random effects and the distribution<br />
of these random effects. It is therefore natural that the correct specification of the<br />
random effects distribution plays an important role.<br />
For these reasons, we are targeting a method in which the normality assumption<br />
of random effects is relaxed. A suitable semi-parametric model for an unknown<br />
distribution is a normal mixture. Verbeke and Lesaffre10 used the homoscedastic<br />
version (variances of the mixture components are equal) as a model for random<br />
effects in LMM for a single marker. This approach relaxes the strong parametric<br />
assumption of the normal random effects distribution and also allows to cluster the<br />
longitudinal profiles in the absence of a training data set. The first objective of<br />
this paper is to generalize the model of Verbeke and Lesaffre10 to (a) allow for<br />
multiple longitudinal markers in a computationally tractable manner; (b) consider<br />
more general heteroscedastic normal mixtures (variances of the mixture components<br />
are unequal) in the random effect distribution. The second objective of this paper<br />
is to apply the developed model to the training (Dutch PBC) data set and to<br />
discriminate future patients using their multivariate longitudinal profiles.<br />
The paper proceeds as follows. The first Section describes the multivariate linear<br />
mixed model with a normal mixture in the random effects distribution. This approach<br />
will be used to model in each prognostic group the longitudinal evolution of the<br />
markers and their dependence on possible covariates. The estimation procedure for<br />
the proposed model is based on the Markov chain Monte Carlo methodology and<br />
is given in Section ’Estimation’. Section ’Discrimination procedure’ explains how<br />
the fitted mixed models can be used to discriminate future patients into prognostic