A Bayesian Method for Exposure Prevalence Comparison During Foodborne Disease Outbreak Investigations.

Abstract

CDC and health departments investigate foodborne disease outbreaks to identify a source. To generate and test hypotheses about vehicles, investigators typically compare exposure prevalence among case-patients with the general population using a one-sample binomial test. We propose a Bayesian alternative that also accounts for uncertainty in the estimate of exposure prevalence in the reference population. We compared exposure prevalence in a 2020 outbreak of Escherichia coli O157:H7 illnesses linked to leafy greens with 2018-2019 FoodNet Population Survey estimates. We ran prospective simulations using our Bayesian approach at three time points during the investigation. The posterior probability that leafy green consumption prevalence was higher than the general population prevalence increased as additional case-patients were interviewed. Probabilities were >0.70 for multiple leafy green items 2 weeks before the exact binomial p-value was statistically significant. A Bayesian approach to assessing exposure prevalence among cases could be superior to the one-sample binomial test typically used during foodborne outbreak investigations.