Designing Studies to Better
Understand Food Source
Attribution
Mike Hoekstra
National Center for Emerging and Zoonotic Infectious
Diseases
Division of Foodborne, Waterborne, and Environmental Diseases
Abstract
Attribution of illness to food commodity is a simple process of relating episodes of human illness through
consumption or handling of foods to instances of commodity contamination…except that the available data on
human illness, food consumption, and contamination are nowhere configured to make relating them simple.
The totality of agents that cause illness is not known. Surveillance for the agents that are known is not
complete. Surveillance reports rarely come with food specified as the cause, much less the commodity.
Outbreak investigations can produce cases of human illness that are tightly linked to specific food exposures,
but such tight links exist for only a fraction of reported outbreak cases, and outbreak cases are, in turn, only a
small fraction of all cases. Case control studies are typically aimed at attributing illness to causal food exposures
in the much larger population of sporadic illness. These studies link multiple food exposures to cases, but do so
in a very noisy fashion. The actual causal exposures are in turn inferred from control food exposures, also noisy
and with different potential biases. Consumption models, like that of Hald, link counts of human illness
aggregated by type to commodity contamination levels by type, through food consumption estimates, yielding
ecological associations. Further, commodity contamination levels can depend on the point in the food chain that
they are measured, creating potentially different attributions. Quantitative microbiological risk assessment
offers another route to attribution, building causal pathways from reservoir to consumption via probabilistic
models applied to the food chain. These are examples of existing ways to relate illness to contaminated food.
They are diverse, not exhaustive, and no single method can be deemed definitive given the large inherent
uncertainties in the data and in the model structures themselves. We present design considerations for each
these examples along with a paradigm for synthesizing an understanding of their collective food source
attribution outputs.
Outline
• Aim and Background
• Estimating the burden of foodborne illness
• Foodborne illness estimates
• Attribution and attributing
• Attributions
• Future directions
Aim
• Estimate the “burden” of human illness
caused by contaminated food
– at the individual pathogen/agent level and in
the aggregate
– where burden may be defined in terms of
severity (eg. illness vs. hospitalizations)
• Estimate the proportion of that burden
caused by specific food commodities
– where commodities are tied to regulation
– where burden may be specific to
subpopulation or illness outcome
Aim
• Intervene to reduce illness at point(s)
informed by estimated burden and
attribution
• Measure changes in amount of illness
– where power to detect change depends on
effect size and data stream
• Measure change in the proportion of
illness caused by specific food
commodities
Cycle of public health action
Burden Attribution
I
n
t
e
r
v
Attribution Trend e
n
t
i
o
n
Outline
• Aim and Background
• Estimating the burden of foodborne illness
• Foodborne illness estimates
• Attribution and attributing
• Attributions
• Future directions
Estimating illnesses
• Multiplicative models
• Data summarized with distributions
• Factors summarized with distributions
• Burden summarized with distributions
Estimates of US lab-confirmed Campylobacter
illnesses, based on data extrapolated
from each FoodNet site, by state
Multiplicative model
Multiplicative model
Estimated distribution of
Campylobacter Illness Burden
Outline
• Aim
• Estimating the burden of foodborne illness
• Foodborne illness estimates
• Attribution and attributing
• Attributions
• Future directions
Annual estimate of domestically acquired foodborne
illnesses, hospitalizations and deaths
31 Known Pathogens
Mean 90% credible interval
Illnesses (millions) 9.4 6.6 – 12.7
Hospitalizations 56,000 40,000 – 76,000
Deaths 1,350 700 – 2,250
Unspecified Agents
Mean 90% credible interval
Illnesses (millions) 38.4 19.8 – 61.2
Hospitalizations 72,000 10,000 – 157,000
Deaths 1,700 350 – 3,350
Summary of Results:
Domestically Acquired Foodborne illness
Summary of Results:
Domestically Acquired Foodborne illness
Deaths Hospitalizations
Illnesses Percent Foodborne
Links to additional information can
be found at…
www.cdc.gov/foodborneburden
Outline
• Aim
• Estimating the burden of foodborne illness
• Foodborne illness estimates
• Attribution and attributing
• Attributions
• Future directions
The Attribution Framework
Norovirus
Reservoir
Salmonella Production
Processing
E. Coli O157
Preparation
L. mono
Consumption
Norovirus Pathogen-Vehicle Plane
Salmonella
E. Coli O157
L. mono
Building Blocks in Framework
Outbreak Hypothetical Hypothetical Hypothetical Data Dom.
Based Validity? Validity? Validity?
Blending Hypothetical Data Dom.
Validity?
CaCo Hypothetical Data Dom. Data Dom.
Validity?
Consumption Data Dom. Data Dom. Data Dom. Hypothetical Hypothetical
Based Validity? Validity?
QMRA Model Dom Model Dom Model Dom Model Dom Model Dom
Expert Elic. Data wt’d Data wt’d Data wt’d Data wt’d Data wt’d
Opinion Opinion Opinion Opinion Opinion
Reservoir Production Processing Preparation Consumption
Outline
• Aim
• Estimating the burden of foodborne illness
• Foodborne illness estimates
• Attribution and attributing
• Attributions
• Future directions
Human Illness Data Sources and Related
Attribution Methodologies
Foodborne Human Illness
Sporadic Outbreak
Blending Simple Complex
Consumption-
CaCo Studies Sporadic and Commodity Commodity
based
Outbreak Data Attribution Attribution
Danish Model
Salmonella STEC 96 and Annual
Adaptation: Campylobacter Toxoplasma Listeria STEC Painter Model
Serotypes 99 MMWR
Salmonella
Food Commodity Hierarchy
All Food
Aquatic Land animals Plant
Fish Shellfish Dairy Eggs Meat-Poultry Grains-beans Oils-sugars Produce
Crustaceans Meat Fruits-nuts
Mollusks Beef Vegetables
Game Fungi
Pork Leafy
Yellow boxes identify 17 commodities Poultry Root
Sprout
Painter et al, J Food Protection 2009 Vine-stalk
Attributions
Illnesses (%)
Campylobacter
Grains-Beans
Crustaceans
Oils-Sugars
Fruits-Nuts
Vine-Stalk
Mollusks
Poultry
Sprout
Finfish
Game
Fungi
Dairy
Leafy
Total
Root
Pork
Beef
Eggs
Simple outbreak-
0 0 7 66 0 0 0 100%
• Missing values
• Incomplete classification
• Non-quantitative knowledge
• Weighting/combining information
Synthesis: Resolutions
• Expert elicitation
• EE/BMA hybrid
• Bayesian model averaging
• Integrated blending model (?)
Project 3
Project 7
Project 0
JAN 2013 JAN 2016
Analysis
Outbreak Theory
Attribution Data
Project 6
Analysis
Blended Theory
Attribution Data
Project 5
Analysis
Sporadic Theory
Attribution Data
Project 9
Project 4
Consumption-based Analysis
Theory
Models Data
Project 8
Analysis
Expert Theory
Elicitation Data
Project 2
Reporting
Synthesis
Theory
Project 10
Communication
Summary description Summary description
based on existing data based on revised data
and understanding and understanding
For more information please contact Centers for Disease Control and
Prevention
1600 Clifton Road NE, Atlanta, GA 30333
Telephone, 1-800-CDC-INFO (232-4636)/TTY: 1-888-232-6348
E-mail: cdcinfo@cdc.gov Web: www.cdc.gov
The findings and conclusions in this report are those of the authors and do not necessarily represent the official
position of the Centers for Disease Control and Prevention.
National Center for Emerging and Zoonotic Infectious Diseases
Division of Foodborne, Waterborne, and Environmental Diseases
In case you were thinking outbreaks can
solve all your problems…