Estimating hepatitis C prevalence in England and Wales by synthesizing evidence from multiple data sources. Assessing data conflict and model fit

Multiparameter evidence synthesis is becoming widely used as a way of combining evidence from multiple and often disparate sources of information concerning a number of parameters. Synthesizing data in one encompassing model allows propagation of evidence and learning. We demonstrate the use of such an approach in estimating the number of people infected with the hepatitis C virus (HCV) in England and Wales. Data are obtained from seroprevalence studies conducted in different subpopulations. Each subpopulation is modeled as a composition of 3 main HCV risk groups (current injecting drug users (IDUs), ex-IDUs, and non-IDUs). Further, data obtained on the prevalence (size) of each risk group provide an estimate of the prevalence of HCV in the whole population. We simultaneously estimate all model parameters through the use of Bayesian Markov chain Monte Carlo techniques. The main emphasis of this paper is the assessment of evidence consistency and investigation of the main drivers for model inferences. We consider a cross-validation technique to reveal data conflict and leverage when each data source is in turn removed from the model.