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 />
164<br />
effects correct the mean structure to get the patient specific profile. Nevertheless,<br />
a further improvement of the mean structure may lead to a further improvement of<br />
the discriminant procedure. Further, we have also fitted models with K = 3 mixture<br />
components and compared the resulting discriminant rule to these described above.<br />
However, the solution with K = 3 performed worse than the above procedures based<br />
on K =1orK = 2 implying that the structure of the models with K = 3 is already<br />
overparametrized. Note that with K = 3, the dimension of the parameter space<br />
increases with 1 + 6 + 21 = 28 in each prognostic group.<br />
Discussion<br />
In this paper, we have generalized the discriminant analysis of multivariate longitudinal<br />
profiles by assuming a normal mixture in the random effects distribution in<br />
the mixed model. The application of our approach to the PBC Dutch Study data<br />
showed some improvements compared to the methodology based on mixed models<br />
with normal random effects. Due to the fact that the normal mixture serves as<br />
a semi-parametric model for the unknown random effects distribution, the first obvious<br />
question is how to choose K, the number of mixture components. In general,<br />
models with a different number of components can be fitted and then compared<br />
by means of a suitable measure of model complexity and fit like the deviance information<br />
criterion (DIC, Spiegelhalter et al. 25 ), or the penalized expected deviance<br />
(PED, Plummer27 ). Alternatively, posterior distributions of deviances under different<br />
models can be compared (Aitkin, Liu and Chadwick28 ). Nevertheless, when<br />
discrimination is of primary interest and a training data set is available then it is<br />
preferable to choose the optimal model by evaluating the resulting discrimination<br />
rule by, e.g. means of cross-validation, as was done in Section ’Application to PBC<br />
data’.<br />
In our specification of the MLMM, we assumed that the errors εi,r,j (i =1,...,N, r =<br />
1,...,R, j =1,...,ni,r) are independent and hence the markers Yi,r,j are conditionally<br />
independent given the random effects. Hence, one can further generalize<br />
the proposed model by using a more general covariance structure for the vectors<br />
of errors εi,r (i =1,...,N, r =1,...,R as was done, e.g., by Shah, Laird and<br />
Schoenfeld 29 or Morrell et al. 7 With such generalization, the results of Section ’Ap-<br />
plication to PBC data’ can even improve. Further, it is certainly possible to relax<br />
the normality assumption on random effects in several other directions than was<br />
done in this paper. For example, a multivariate t-distribution (see, e.g., Pinheiro,<br />
Liu and Wu 30 ) or a mixture of multivariate t-distributions (Lin, Lee and Ni 31 )for