Projecting the future size and health status of the U.S. elderly population

Federal demographic projections are often based on discrete state, discrete time models of population change. Uncertainty is assessed by systematically varying parameters of the projection/state transition matrix and determining the effect on outcomes. The number of covariates that can be represented in such projections is limited by the number of discrete state parameters, produced by stratifying on covariates, that can be reliably estimated. A projection model based on a multivariate continuous state, stochastic process is presented. The model allows multiple time-varying covariates to be used so parameters can be estimated from time series information on health changes and mortality, and their interaction. Health changes are simulated by altering parameters controlling the age trajectory and diffusion of risk factor means, variances, and covariances. This is important in assessing the effects of health interventions and stochasticity on projections. By increasing the information used in projections it may be possible to better (a) anticipate the state of health at extreme ages, (b) forecast changes in health at specific ages over time, (c) stimulate the effects of specific interventions, and (d) determine the sensitivity of outcomes to a range of interventions.