Table. Current approaches to food attribution.

Approach	Primary advantages	Primary limitations	Refs
Denmark Salmonella Accounts	Microbial subtyping provides direct link between public health endpoint and animal	Difficult to expand to other pathogens; requires distinctive subtypes across reservoirs	5,6
	High reporting of illnesses (social health care)	Focus on animals ignores nonanimal sources
	National, temporal coverage for both illnesses and animal/product monitoring	Focus on reservoirs, not food products at point of consumption
UK outbreak data	Large dataset: national, temporal coverage	May not correlate with sporadic case data	8
	Results correlate with local epidemiologic findings	Not all pathogens well represented
		Dependence on general practitioners
US outbreak data	National and temporal coverage	May not correlate with sporadic case data	11,12
	Large common dataset	Geographic and temporal inconsistencies (local reporting) and biases towards certain foods
	Straightforward, uses existing data	Not all pathogens well represented
	Outbreaks and outbreak cases can be aggregated into food categories
Case-control studies	Population-based studies	Survey format has recall bias and other limits	17,20–24
	Captures risk factors not included in most surveillance data (travel, food preparation questions)	Long exposure windows (problems with common exposures)
	Can implicate risks missed by laboratory testing	Durable immunity in population can impede associating exposures with illnesses
		No laboratory verification
Microbial subtyping	Subtyping of illnesses and foods can provide direct link between public health endpoint and source of infection	For animal sourcing, subtypes must be distinctive across species (see Danish Salmonella Accounts)	25–28,5,6
	Can be used to identify specific foods (outbreak investigations) or animal reservoirs (source tracking by species)	Utility may be limited to certain pathogens
	Many different techniques, growing fast	Resource intensive; requires human surveillance, extensive monitoring of food and animals, plus laboratory testing, data storage, analysis
Risk assessments	Can estimate cases not captured by surveillance methods (not limited by underreporting or biases in epidemiologic methods)	Predictive; cannot be verified	29–33
	Uses consumption and contamination data ignored by surveillance-based approaches	Large uncertainties in dose-response models and exposure estimates
		Resource- and time-intensive (each pathogen-food combination requires its own exhaustive study)
Food monitoring data	Captures upstream contamination (avoids environmental and cross-contamination after purchase)	Not usable for food attribution unless made compatible (through subtyping or other means) with public health data	34–36
Expert elicitation/judgment	Useful when data are sparse or conflicting	Respondents can be similarly biased	15,37–39
	Formal methods increase utility	Requires some level of consensus for reasonable error bounds
		Based on perception, not data