Invasive Species Compendium

Detailed coverage of invasive species threatening livelihoods and the environment worldwide

Abstract

A statistical modelling approach for source attribution meta-analysis of sporadic infection with foodborne pathogens.

Abstract

Numerous source attribution studies for foodborne pathogens based on epidemiological and microbiological methods are available. These studies provide empirical data for modelling frameworks that synthetize the quantitative evidence at our disposal and reduce reliance on expert elicitations. Here, we develop a statistical model within a Bayesian estimation framework to integrate attribution estimates from expert elicitations with estimates from microbial subtyping and case-control studies for sporadic infections with four major bacterial zoonotic pathogens in the Netherlands (Campylobacter, Salmonella, Shiga toxin-producing E. coli [STEC] O157 and Listeria). For each pathogen, we pooled the published fractions of human cases attributable to each animal reservoir from the microbial subtyping studies, accounting for the uncertainty arising from the different typing methods, attribution models, and year(s) of data collection. We then combined the population attributable fractions (PAFs) from the case-control studies according to five transmission pathways (domestic food, environment, direct animal contact, human-human transmission and travel) and 11 groups within the foodborne pathway (beef/lamb, pork, poultry meat, eggs, dairy, fish/shellfish, fruit/vegetables, beverages, grains, composite foods and food handlers/vermin). The attribution estimates were biologically plausible, allowing the human cases to be attributed in several ways according to reservoirs, transmission pathways and food groups. All pathogens were predominantly foodborne, with Campylobacter being mostly attributable to the chicken reservoir, Salmonella to pigs (albeit closely followed by layers), and Listeria and STEC O157 to cattle. Food-wise, the attributions reflected those at the reservoir level in terms of ranking. We provided a modelling solution to reach consensus attribution estimates reflecting the empirical evidence in the literature that is particularly useful for policy-making and is extensible to other pathogens and domains.