Challenges in data needs for assessment of food product risk and attribution of foodborne illnesses to food products in the United States Chuanfa Guo, Carl Schroeder, and Janell Kause Office of Public Health Science Food Safety and Inspection Service United States Department of Agriculture Fourth International Conference on Agriculture Statistics October 22-24, 2007 Beijing, China Better information. Better decisions. Risk assessment.

2 Food Safety and Attribution Food safety risk assessment requires a farm-to- table approach Food safety risk assessment requires a farm-to- table approach Farm  Processing  Retail  Consumer The same concept applies to attribution of foodborne illnesses to food products The same concept applies to attribution of foodborne illnesses to food products  Where does contamination occur?  How does contamination occur?  How can it be prevented? Better information. Better decisions. Risk assessment.

3 Attribution of Foodborne Illnesses to Food Products Better information. Better decisions. Risk assessment.

4 FoodNet Attribution Activities Foodborne Diseases Active Surveillance Network (FoodNet) attribution working group and modeling subgroup  Centers for Disease Control and Prevention (CDC)  Food Safety and Inspection Service (FSIS)  Food and Drug Administration (FDA)  State health departments Better information. Better decisions. Risk assessment.

5 Use of Expert Elicitation for Food Safety and Attribution FSIS and RTI International conducted expert elicitations in 2005 and 2007  Rank the public health risks posed by bacterial hazards in processed meat and poultry products  Attribution of foodborne illnesses to specific pathogens as a result of consuming or handling processed meat and poultry products Better information. Better decisions. Risk assessment.

6 Salmonella Attribution Model Hald et al. A Bayesian approach to quantify the contribution of animal-food sources to human salmonellosis. Risk Anal (1): Hald et al. A Bayesian approach to quantify the contribution of animal-food sources to human salmonellosis. Risk Anal (1): Microbial subtyping provides link between public health endpoint and source of infection. Microbial subtyping provides link between public health endpoint and source of infection. Bayesian framework uses Markov Chain Monte Carlo simulation to estimate number of human salmonellosis. Bayesian framework uses Markov Chain Monte Carlo simulation to estimate number of human salmonellosis. The approach quantifies the contribution of each of the major animal-food sources to human salmonellosis. The approach quantifies the contribution of each of the major animal-food sources to human salmonellosis. Better information. Better decisions. Risk assessment.

7 Adaptation of the Salmonella Attribution Model to U.S. Data Joint effort by FSIS, CDC, FDA, and state partners under the FoodNet Attribution Working Group and Modeling Subgroup. Objectives: Joint effort by FSIS, CDC, FDA, and state partners under the FoodNet Attribution Working Group and Modeling Subgroup. Objectives:  Estimate the number of cases of human salmonellosis attributable to various food sources  Support risk managers and regulators when deciding how to allocate resources  Identify data needs and gaps for future attribution studies Better information. Better decisions. Risk assessment.

8 Salmonella Attribution Model Parameters Salmonella prevalence by serotype in a food source (p) Salmonella prevalence by serotype in a food source (p) Amount of particular food consumed (M) Amount of particular food consumed (M) Food source dependent factor (a) Food source dependent factor (a) Serotype dependent factor (q) Serotype dependent factor (q) Better information. Better decisions. Risk assessment.  Expected number of salmonellosis cases ( )

9 Attribution Data Sources Human salmonellosis cases, by serotype Human salmonellosis cases, by serotype  Public Health Laboratory Information System (PHLIS), Foods - Salmonella prevalence, by serotype Foods - Salmonella prevalence, by serotype  Beef, ground beef, chicken, turkey, pork, and processed egg products, FSIS in-plant samples,  Shell eggs, Pennsylvania SE Pilot Project, Food consumption data Food consumption data  USDA/Economic Research Service, Outbreak and travel information Outbreak and travel information  Salmonellosis cases reported to FoodNet, 2004 Better information. Better decisions. Risk assessment.

10 Preliminary Model Results Estimated Percentage Distributions of Human Salmonellosis Cases, Better information. Better decisions. Risk assessment. * Shell egg data from the Pennsylvania Pilot Project,

11 Data Gaps and Limitations of Salmonella Attribution Model The stochastic model does not attribute all observed salmonellosis cases to food sources. The stochastic model does not attribute all observed salmonellosis cases to food sources. Model does not address other foodborne sources (produce, dairy, etc) of Salmonella. Model does not address other foodborne sources (produce, dairy, etc) of Salmonella. Model does not attribute any salmonellosis cases to non-food source, environmental exposures, pets, farm animals, water, etc. Model does not attribute any salmonellosis cases to non-food source, environmental exposures, pets, farm animals, water, etc. Egg data are limited ( data available) Egg data are limited ( data available) Better information. Better decisions. Risk assessment.

12 Challenges in Data Needs For Food Attribution To obtain data concerning pathogen prevalence and distribution in a wide variety of potential food vehicles and other for other important sources of human exposure, such as indirect sources of contamination and non-food sources To obtain data concerning pathogen prevalence and distribution in a wide variety of potential food vehicles and other for other important sources of human exposure, such as indirect sources of contamination and non-food sources To ensure that existing data sources continue to adequately represent the burden of foodborne illnesses in the U.S. population and the distribution of the associated pathogen in food vehicles and exposure sources of interest To ensure that existing data sources continue to adequately represent the burden of foodborne illnesses in the U.S. population and the distribution of the associated pathogen in food vehicles and exposure sources of interest To refine existing data so that the comparisons between data from various sources are based on similar units of observation at the necessary levels of discrimination for defined points along the farm-to-table continuum To refine existing data so that the comparisons between data from various sources are based on similar units of observation at the necessary levels of discrimination for defined points along the farm-to-table continuum Better information. Better decisions. Risk assessment.

13 Results of 2007 Expert Elicitation Results of 2007 Expert Elicitation Rank the public health risks posed by bacterial hazards in each of 25 categories of processed meat and poultry products Rank the public health risks posed by bacterial hazards in each of 25 categories of processed meat and poultry products Score of 1 to 10 for likelihood of illness from consuming or handling meat and poultry products among healthy adults and vulnerable consumers Score of 1 to 10 for likelihood of illness from consuming or handling meat and poultry products among healthy adults and vulnerable consumers  1 – least likelihood  10 – greatest likelihood  Attribute foodborne illnesses of specific pathogens to consuming or handling processed meat and poultry products Better information. Better decisions. Risk assessment.

14 Likelihood of Illness Among Healthy Adults Finished Product Type Median Score (1-10) Level of Confidence (1-3) Raw ground or otherwise non-intact chicken Raw ground or otherwise non-intact turkey 92.3 Raw ground or otherwise non-intact poultry – no chicken or turkey Raw intact chicken 82.6 Raw intact turkey 82.5 Raw intact poultry – other than chicken or turkey 81.9 Raw ground or otherwise non-intact beef 82.5

15 Likelihood of Illness Among Vulnerable Consumers Finished Product Type Median Score (1-10) Level of Confidence (1-3) Raw ground or otherwise non-intact chicken Raw ground or otherwise non-intact beef Raw ground or otherwise non-intact turkey 92.5 Raw ground or otherwise non-intact poultry – no chicken or turkey 92.0 Raw intact chicken Raw intact turkey 82.6 Raw intact poultry – other than chicken or turkey 82.1

16 Attribution of Foodborne Illness of Salmonella (Non-Typhi) to Meat and Poultry Products Better information. Better decisions. Risk assessment.

17 Attribution of Foodborne Illness of Salmonella (Multidrug Resistant) to Meat and Poultry Products Better information. Better decisions. Risk assessment.

18 Closing Remarks  Assessment of food product safety and attribution of foodborne illnesses require extensive data originating from various sources  Available data sources often suffer from methodological limitations and the unavailability of certain types of data often result in critical data gaps  Expert elicitation is useful when epidemiologic data are lacking, are sparse, or are highly uncertain to fill the critical gaps in food safety studies  Ensure and maximize the quality, objectivity, utility, and integrity of the data Better information. Better decisions. Risk assessment.

19 Acknowledgements FoodNet Attribution Modeling Subgroup FoodNet Attribution Modeling Subgroup CDC: Fred Angulo, Mike Hoekstra, Elaine Scallan, Xin Tong CDC: Fred Angulo, Mike Hoekstra, Elaine Scallan, Xin Tong FSIS: Carl Schroeder, Chuanfa Guo, Liane Ong, Kristin Holt, Patty Bennett, Bonnie Kissler, Evelyne Mbandi, Reza Roodsari, Jane Harman, Alecia Naugle, Bonnie Rose FSIS: Carl Schroeder, Chuanfa Guo, Liane Ong, Kristin Holt, Patty Bennett, Bonnie Kissler, Evelyne Mbandi, Reza Roodsari, Jane Harman, Alecia Naugle, Bonnie Rose Oregon Health Division : Paul Cieslak Oregon Health Division : Paul Cieslak Georgia Division of Public Health: Dana Cole Georgia Division of Public Health: Dana Cole Decisionalysis Risk Consultants, Inc.: Emma Hartnett Decisionalysis Risk Consultants, Inc.: Emma Hartnett Better information. Better decisions. Risk assessment. Expert Elicitation Activities Expert Elicitation Activities  RTI International: Shawn Karns, Mary Muth, Michaela Coglaiti  FSIS: Janell Kause, Chuanfa Guo, Matthew Michael, Cynthia Williams, Don Anderson

20 THANK YOU Better information. Better decisions. Risk assessment.