Purpose: We evaluated multiple approaches to down-weighting older outbreak data used to estimate foodborne illness source attribution. Whereas there is no objective notion of a ‘median’ down-weighting, we sought a robust compromise between using all data and using only recent data.
Methods: We identified 952 outbreaks in CDC’s Foodborne Disease Outbreak Surveillance System (FDOSS) caused by four priority pathogens (Salmonella, Escherichia coli O157, Campylobacter, and Listeria monocytogenes) from 1998-2012 in which a food was implicated. We explored an array of approaches for down-weighting older outbreaks in a model for estimating attribution fractions.
Results: Excluding older outbreaks resulted in estimates of no attributable risk for some food categories. Including older outbreaks without applying down-weighting resulted in higher estimated attributions for some food categories with markedly decreased outbreaks over time. We chose a compromise from the array of down-weighting schemes based on an ad hoc 50% minimum weight for the most recent 5 years and 10% maximum for older information: down-weight data from outbreaks older than 5 years using an exponential decay function of 0.71. Using this approach, 67% of the information used to estimate attribution came from 2008-2012, 28% from 2003-2007, and 5% from 1998-2002.
Significance: This study identified an approach to limiting the influence of older outbreaks in attribution estimates without losing valuable information for pathogens and food categories with sparse data. Improved attribution estimates can provide essential information to regulators and public health officials for targeting risk interventions.