A stochastic model for estimating the ratio between a fecal indicator and a pathogen based on leftcensored data, which includes a substantially high number of non-detects, was constructed. River water samples were taken for 16 months at six points in a river watershed, and conventional fecal indicators (total coliforms and general Escherichia coli), genetic markers (Bacteroides spp.), and virulence genes (eaeA of enteropathogenic E. coli and ciaB of Campylobacter jejuni) were quantified. The quantification of general E. coli failed to predict the presence of the virulence gene from enteropathogenic E. coli, different from what happened with genetic markers (Total Bac and Human Bac). A Bayesian model that was adapted to left-censored data with a varying analytical quantification limit was applied to the quantitative data, and the posterior predictive distributions of the concentration ratio were predicted. When the sample size was 144, simulations conducted in this study suggested that 39 detects were enough to accurately estimate the distribution of the concentration ratio, when combined with a dataset with a positive rate higher than 99%. To evaluate the level of accuracy in the estimation, it is desirable to perform a simulation using an artificially generated left-censored dataset that has the identical number of non-detects as the actual data.
ASJC Scopus subject areas