Linear discriminant models for unbalanced longitudinal data

This paper discusses statistical methods for the classification of observat ions into one of two or more groups based on longitudinal observations. Mea surements on subjects in longitudinal medical studies are often collected a t different times and on a different number of occasions. Classical multiva riate methods for linear discriminant analysis are difficult to apply to re peated measurements due to the highly unbalanced structure observed in thes e data. Linear models for the analysis of longitudinal data proposed by Lai rd and Ware and non-linear models proposed by Lindstrom and Bates can be us ed to estimate population parameters for a discriminant model that classifi es individuals into distinct predefined groups or populations. An example i s presented using data from a study in 150 pregnant women in Santiago,Chile , in order to predict normal versus abnormal pregnancy outcomes. Copyright (C) 2000 John Wiley & Sons, Ltd.