DOES THE COVARIANCE STRUCTURE MATTER IN LONGITUDINAL MODELING FOR THEPREDICTION OF FUTURE CD4 COUNTS

We investigate the importance of the assumed covariance structure for longitudinal modelling of CD4 counts. We examine how individual predic tions of future CD4 counts are affected by the covariance structure. W e consider four covariance structures: one based on an integrated Orns tein-Uhlenbeck stochastic process; one based on Brownian motion, and t wo derived from standard linear and quadratic random-effects models. U sing data from the Multicenter AIDS Cohort Study and from a simulation study, we show that there is a noticeable deterioration in the covera ge rate of confidence intervals if we assume the wrong covariance. The re is also a loss in efficiency. The quadratic random-effects model is found to be the best in terms of correctly calibrated prediction inte rvals, but is substantially less efficient than the others. Incorrectl y specifying the covariance structure as linear random effects gives t oo narrow prediction intervals with poor coverage rates. Fitting using the model based on the integrated Ornstein-Uhlenbeck stochastic proce ss is the preferred one of the four considered because of its efficien cy and robustness properties. We also use the difference between the f uture predicted and observed CD4 counts to assess an appropriate trans formation of CD4 counts; a fourth root, cube root and square root all appear reasonable choices. (C) 1998 John Wiley & Sons, Ltd.