Estimates of Age-Specific Reductions in HIV Prevalence in Uganda: Bayesian Melding Estimation and Probabilistic Population Forecast with an HIV-enabled Cohort Component Projection Model

We estimate age-specific HIV incidence and prevalence in Tanzania and Uganda in the late 1990s and forecast forward assuming no change in incidence. Comparisons between our forecasts of HIV prevalence and direct measures from the HIV/AIDS Indicator and Demographic and Health Surveys in the mid-2000s provide an age-specific measure of changes in HIV prevalence. In Tanzania our forecast accurately predicts age-specific HIV prevalence, suggesting little change in HIV incidence in Tanzania over the intervening decade. In Uganda our forecasts significantly overstate HIV prevalence. The age pattern of our forecast errors reflects the age-specific reductions in HIV prevalence and incidence in Uganda. Our estimates and forecasts are produced using an HIV-enabled cohort component model of population projection first proposed by \citet{heuveline-dem-2003}. We refine that model (\citeauthor{thomas-clark-dr-2011} in press) and implement the -Bayesian melding with IMIS estimation method \citep{raftery-bao-bio-2010}. This method allows us to estimate the parameters of the Heuveline model with robust measures of uncertainty and to quantify uncertainty in the model outputs, e.g. forecasts. We validate the model and estimation procedure by comparing maximum likelihood estimates with our Bayesian estimates and by confirming the calibration of the model outputs.