A Bayesian analysis of mouse infectivity data to evaluate the effectiveness of using ultraviolet light as a drinking water disinfectant

Abstract

Modelling disinfectant performance using Bayesian hierarchical methods can overcome problems with traditional methods and lead to improved estimates. Animal and cell-culture assays are used to estimate the degree of inactivation of a microorganism produced by a given disinfectant dose. Assay data traditionally are analyzed with logistic model or most probable number (MPN) method. These methods are limited particularly when assays show all (or no) animals or cells to be infected—estimates are reported as greater than (or less than) a measurement limit (i.e., censored data). The proposed Bayesian approach (1) properly models the propagation of uncertainty through the data analysis/modelling process, resulting in reduced model uncertainty, and (2) uses appropriate probability distribution models for the response variables, avoiding the censored data problem and more accurately describing statistical error when estimating dose–response behavior. This paper applies the Bayesian hierarchical models to logistic and MPN data from published papers for the ultraviolet (UV) inactivation of Cryptosporidium. Results are compared to those from three alternative models. The Bayesian model estimates a significantly lower UV dose for a given level of Cryptosporidium inactivation than the alternative models, due mainly to the reduced model uncertainty.