QMRA and decision-making: are we handling measurement errors associated with pathogen concentration data correctly?

Abstract

Knowledge of the variability in pathogen or indicator concentrations over time at a particular location (e.g. in drinking water sources) is essential in implementation of concentration-based regulations and in quantitative microbial risk assessment. Microbial enumeration methods, however, are known to yield highly variable counts (even among replicates) and some methods are prone to substantial losses (i.e. only a fraction of the target microorganisms in a sample are observed). Consequently, estimated microorganism concentrations may be biased and only a fraction of the variability that is observed in temporally distributed concentration estimates is due to variability in concentration itself. These issues have often been ignored in the past, and approaches to integrate knowledge about the measurement error associated with enumeration methods into decisions have not been standardized. Here, an existing model that describes variability in microorganism counts as a function of sample volume and the analytical recovery of the enumeration method is expanded to include temporal concentration variability and sample-specific recovery information. This model is used to demonstrate that microorganism counts and analytical recovery are not independent (as has often been assumed), even if the correlation is obscured by other sources of variability in the data. It is also used as an experimental design tool to evaluate strategies that may yield more accurate concentration estimates. Finally, the model is implemented in a Bayesian framework (with a Gibbs sampling algorithm) to quantify temporal concentration variability with appropriate consideration of measurement errors in the data and the analytical recovery of the enumeration method. We demonstrate by simulation that this statistical approach facilitates risk analyses that appropriately model variability in microorganism concentrations given the available data and that it enables decisions based on quantitative measures of uncertainty.