CONSTRUCTION OF MULTIVARIATE GENTILE CHARTS FOR LONGITUDINAL MEASUREMENTS

Reference centiles are commonly used as a means of identifying whether some measure of interest on an individual lies within a 'normal' rang e. Individuals outside the range may be at risk for some problem outco me. The information needed to screen for problem outcomes is often mul tivariate in nature. We develop a multivariate approach to centile est imation which allows for updating of centiles based on the prior path of the individual as well as adjustments to the centiles according to individual covariate values. It is also possible to consider several v ariables jointly as a screen for problem outcomes.