Enhanced Bayesian Spline Regression Approach for Modelling Trends in Infections Caused by Pathogens Commonly Transmitted Through Food

Background: The Foodborne Diseases Active Surveillance Network (FoodNet) monitors illnesses caused by enteric pathogens at 10 U.S. sites to track progress toward federal disease reduction goals. The original trends frequentist model was developed to overcome statistical complications associated with FoodNet's catchment expansion during 1996-2004 but did not account for site-specific trends and treated year as a categorical variable. It was therefore sensitive to noise and single-year events, and shaped by trends in more populous sites. Methods: This paper describes an enhanced Bayesian model that overcomes these challenges by treating year as continuous, including an interaction between year and site, and using splines to better capture non-linear trends. Results: The enhanced model generates improved uncertainty estimates, can estimate incidence for years not represented by the surveillance dataset, and is dataset-agnostic. By publishing a pipeline for running the enhanced model, it can be readily adapted (i) for modeling trends in individual sites or among specific populations; (ii) for use with alternative datasets; and (iii) to include additional covariates. Conclusions: FoodNet generates the most up-to-date and highest-resolution enteric disease surveillance data in the United States. Publishing the pipeline for implementing this model improves methodologic transparency, and makes FoodNet data more accessible and available for guiding public health action in the near-real time.