BAYESIAN-ANALYSIS OF 2-COMPONENT MIXTURE DISTRIBUTIONS APPLIED TO ESTIMATING MALARIA ATTRIBUTABLE FRACTIONS

Malaria illness can be diagnosed by the presence of fever and parasita emia. However, in highly endemic areas the diagnosis of clinical malar ia can be difficult since children may tolerate parasites without feve r and may have fever due to other causes. We propose a novel, simulati on based Bayesian approach for obtaining precise estimates of the prob abilities of children with different levels of parasitaemia having fev er due to malaria, by formulating the problem as a mixture of distribu tions. The methodology suggested is a general methodology for decompos ing any two-component mixture distribution nonparametrically, when an independent training sample is available from one of the components. i t is based on the assumption that one of the component distributions l ies on the left of the other but there is some overlap between the dis tributions.