Quantitative Bayesian predictions of source water concentration for QMRA from presence/absence data for E. coli O157:H7

Abstract

A hierarchical Bayesian framework was applied for describing variability in pathogen concentration (with associated uncertainty) from presence/absence observations for E. coli O157:H7. Laboratory spiking experiments (method performance) and environmental sample assays were undertaken for a surface drinking water source in France. The concentration estimates were strongly dependent upon the assumed statistical model used (gamma, log-gamma or log-gamma constrained), highlighting the need for a solid theoretical basis for model choice. Bayesian methods facilitate the incorporation of additional data into the statistical analysis; this was illustrated using faecal indicator results of E. coli (Colilert) to reduce the posterior parameter uncertainty and improve model stability. While conceptually simple, application of these methods is still specialised, hence there is a need for the development of data analysis tools to make Bayesian simulation techniques more accessible for QMRA practitioners.