The standard approach for determining the contribution of risk variables to CHD incidence is the efficient estimation of the multivariate logistic function. This strategy suffers from the dficulty that the multiple logistic function views the disease mechanism as a "shock" process and, as a consequence, confounds the dynamic development or change of the risk variable with the risk or hazard of the disease. To alleviate this problem a continuous time-continuous state model of chronic disease was proposed by the authors and estimated as a discrete time process for longitudinal data on CHD risk in the Framingham study. In the present study the methodology is extended through the use of multivariate strategies to investigate the basic components of risk variable dynamics and of the association of risk variables with CHD incidence. These components, which do not necessarily constitute isomorphic sets for dynamics and risk prediction, are derived from the covariance structure of the risk variables and their temporal behavior. A multivariate analysis of both the dynamics and risk of CHD over 8 years of the Framingham study is presented.
I. INTRODUCTION
Analyses of coronary heart disease (CHD) in longitudinal studies frequently involve modeling of the accumulated incidence of CHD over the entire study as a 109