Analysis of longitudinally observed irregularly timed multivariate outcomes: regression with focus on cross-component correlation

Components of repeatedly observed multivariate outcomes (for example, the t wo components of blood pressure measures (SBPit, DBPit), obtained on subjec t i at arbitrarily spaced times t) are often analysed separately. We presen t a unified approach to regression analysis of such irregularly timed multi variate longitudinal data, with particular attention to assessment of the m agnitude and durability of cross-component correlation. Maximum likelihood estimates are presented for component-specific regression parameters and au tocorrelation and cross-correlation functions. The component-specific autoc orrelation function has the 'damped exponential' form. (corr(Y-it, Y-i,Y- t +s) = gamma (\s\0)), which generalizes the AR(1), MA(1) and random intercep t models for univariate longitudinal outcomes. The cross-component correlat ion function (CCCF) has an analogous form, allowing damped-exponential deca y of cross-component correlation as time between repeated measures elapses. Finite sample performance is assessed through simulation studies. The meth ods are illustrated through blood pressure modelling and construction of mu ltivariate prediction regions. Copyright (C) 2001 John Wiley & Sons, Ltd.