View PDF Version - RePub - Erasmus Universiteit Rotterdam
View PDF Version - RePub - Erasmus Universiteit Rotterdam
View PDF Version - RePub - Erasmus Universiteit Rotterdam
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Introduction<br />
Discriminant analysis using a MLMM with a normal mixture 143<br />
The Dutch Multicenter Primary Biliary Cirrhosis (PBC) Study is a prospective cohort<br />
study of patients with PBC with participation of 7 university hospitals and 39 general<br />
hospitals. Recruitment of patients started in January 1990 and follow-up data until<br />
April 2007 were available for analysis. Follow-up data were collected at approximately<br />
3-monthly intervals in the first year and yearly intervals thereafter. In total, 375<br />
patients were recruited with a median follow-up of 9.7 years. See Kuiper et al. 1 for<br />
details of the study. It is of clinical interest to predict the future patients’ status.<br />
In this paper, we are interested in predicting whether the patient will suffer from<br />
a serious disease progression (liver related death or liver transplantation) in the first<br />
T = 10 years after the start of treatment by ursodeoxycholic acid (UDCA, 13–15<br />
mg/day/kg of weight). Many factors are known to be related to progression of PBC<br />
and hence could be used to establish the prognosis. Here we consider the following<br />
three markers whose longitudinal profiles are available: bilirubin, albumin, alkaline<br />
phosphatase (AP). A prognostic group (status at T = 10 years) is also known for<br />
many patients. Our aim is to use these patients as a training data set, model the<br />
dependence of these markers on time and possibly other covariates (e.g., age) and<br />
subsequently use the developed models to predict longitudinally the prognostic group<br />
of future patients.<br />
We now introduce the following notation. Let Y i,r =(Yi,r,1,...,Yi,r,ni,r )′ be a random<br />
vector of the r-th marker (r =1,...,R) observed for the i-th patient (i =1,...,N)<br />
at time points 0 ≤ ti,r,1 < ···