International Journal of Food Microbiology 130 (2009) 7787

Contents lists available at ScienceDirect

International Journal of Food Microbiology

j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / i j f o o d m i c r o

Analysis of foodborne outbreak data reported internationally for source attribution

J.D. Greig a,, A. Ravel b
a

Laboratory for Foodborne Zoonoses, Public Health Agency of Canada, 160 Research Lane, Guelph, Ontario, Canada N1G 5B2
Laboratoire de lutte contre les zoonoses d'origine alimentaire, Agence de la sant publique du Canada, Facult de mdecine vtrinaire, 3200, rue Sicotte, C.P. 5000, Saint-Hyacinthe,
Qubec, Canada J2S 7C6

a r t i c l e

i n f o

Article history:
Received 15 April 2008
Received in revised form 29 October 2008
Accepted 20 December 2008
Keywords:
Food attribution
Outbreak
Foodborne pathogen
Food vehicle
Enteric disease
Zoonoses

a b s t r a c t
Analysis of foodborne outbreak data is one approach to estimate the proportion of human cases of specic
enteric diseases attributable to a specic food item (food attribution). Although we recognize that for a
variety of reasons reported outbreaks represent only a small portion of all actual outbreaks, using outbreak
data for food attribution is the only methodological approach where, theoretically, there is an actual direct
link between the pathogen, its source and each infected person. The purpose of this study was to explore the
usefulness of foodborne outbreak data extracted from publicly available international electronic reports and
publications to provide estimates of food attribution, to derive and compare these estimates between
regions, while improving the understanding of the pathogen/food vehicle combination. Electronic reports
and publications of foodborne outbreaks that occurred globally since the 1980s were systematically scanned
and their data were extracted and compiled in a database. A system of food categorization was developed and
food vehicles assigned accordingly. The association between the aetiology and the food source was
statistically described for outbreaks with both reported aetiology and incriminated food vehicle. Differences
in associations between Australia and New Zealand, Canada, the European Union (EU) and the United States
(US) were explored using multiple correspondence analysis and were formally tested between the EU and
the US for selected pathogens and food sources. As a result, the food and aetiology cross tabulation of 4093
foodborne outbreaks that occurred globally between 1988 and 2007 is presented and discussed. For a few
aetiologies and some foods the association is very specic. The lack of a specic association between the
other foods and aetiologies highlights the potential roles of cross-contamination, environmental
contamination and the role of the infected foodhandler along the food chain from farm to fork. Detailed
analysis of the four regions highlighted some specic associations: Salmonella Enteritidis outbreaks occurred
relatively often in the EU states with eggs as the most common source; Campylobacter associated outbreaks
were mainly related to poultry products in the EU and to dairy products in the US; there was an association
between Escherichia coli outbreaks and beef in Canada; and while Salmonella Typhiumurium outbreaks were
relatively common in Australia and New Zealand, across all regions, Salmonella was associated with a variety
of food groups. The value and limitations of the study are discussed, as well as the extrapolation of the food
attribution estimates beyond their outbreak context.
Crown Copyright 2008 Published by Elsevier B.V. All rights reserved.

1. Introduction
Food attribution for enteric disease is dened as the estimation of
the proportion of human cases for each disease that can be attributed
to a specic food item or commodity. Attribution has recently been
highlighted as an important tool currently lacking, but of great interest
in food safety risk analysis (Rocourt et al., 2003; Batz et al., 2004;
ICMSF, 2006; Doyle and Erickson, 2006; Havelaar et al., 2007). It
provides useful information for risk proles, risk assessments, risk
management and risk communication regarding food safety. More

Corresponding author. Laboratory for Foodborne Zoonoses, Public Health Agency of

Canada, 160 Research Lane, Unit 106, Guelph, Ontario, Canada N1G 5B2. Tel.: +1 519 826
2377; fax: +1 519 826 2367.
E-mail address: judy_greig@phac-aspc.gc.ca (J.D. Greig).

specically, food attribution allows the estimation of the number of

human cases, hospitalizations, deaths and their related costs for each
pathogen-food vehicle combination (Adak et al., 2005; Batz et al.,
2004). Ultimately, food attribution informs the decisions of food safety
risk managers for effective resource allocation to improve food safety
and help prevent enteric disease in the community.
Potential methodological approaches for quantifying food attribution estimates were recently reviewed and few approaches were
judged interesting and worthwhile (Batz et al., 2005; Havelaar et al.,
2007; Anonymous, 2008). Two methods rely on epidemiological
principles, i.e. the analysis of outbreak data and casecontrol studies.
Two methods are based on microbiological approaches: quantitative
microbial risk assessment and microbial subtyping combined with
microbial source tracking. A model based on Bayesian inference
applied to Salmonella enterica serotyping and phagetyping was

0168-1605/$ see front matter. Crown Copyright 2008 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.ijfoodmicro.2008.12.031

78

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

Table 1
A description of the hierarchical system used for classifying food vehicles
Main category

Subcategories (number of foods

listed)

Examples

Beef

Beef dishes (10)

Ground beef (2)
Other beef (12)
Whole muscle beef (5)
Ham (5)
Other pork (7)
Pork dishes (1)
Chicken dishes (11)
Ground chicken (1)
Other chicken (5)
Whole chicken (6)
Ground turkey (1)
Other turkey (2)

Stew, beef curry

Ground beef, hamburgers
Liver, heart
Roasts, steaks
Ham, ham dishes
Chitterlings, pork chops
Pork, mixed dishes
Chicken curry, chicken pot pie
Ground chicken
Nuggets, popcorn
Baked, BBQ
Turkey meatballs
Turkey & gravy, turkey deli
meat
Turkey salad, mixed dishes
Roasted, smoked
Other ground poultry
Other poultry
Other poultry dishes
Duck, quail
Rabbit, moose
Salami, deli meats

Pork

Chicken

Turkey

Other poultry

Other meats

Eggs
Dairy products

Turkey dishes (2)

Whole turkey (3)
Other ground poultry (1)
Other poultry (1)
Other poultry dishes (1)
Other whole poultry (2)
Game (15)
Luncheon (4) made of different
meats
Meat dishes (4)
Other meats (4)
Egg dishes (20)
Eggs (8)
Cheese (12)
Dairy substitute (3)
Ice cream (5)

Milk (8)
Other dairy (11)
Seafood
Finsh (21)
Molluscan shellsh (7)
Other seafood (10)
Seafood dishes (14)
Produce
Fruits (20)
Produce dishes (12)
Vegetables (35)
Nuts
Almonds
Halva
Nuts/dry fruits
Peanut butter
Peanuts
Sesame seeds
Tahini
Cereals
Cooked cereals (3)
Dry cereals (3)
Bakery items
Bakery (10)
Breads (6)
Multi-ingredient Ethnic foods (16)
foods
Other multi-ingredient foods (13)
Rice/beans/stufng/pasta
dishes (18)
Salads (8)
Sandwiches (9)
Sauces (7)
Home canned
Home canned goods
goods
Beverages
Juices (11) pasteurized and
unpasteurized
Other beverages (9)
Other foods
Other foods (16)

Haggis, meat sauce for pasta

Lamb, porcupine
Custards, French toast
Scrambled, boiled
Cottage cheese, raw-milk
cheese
Soy milk, rice milk
Ice cream bar, home made ice
cream
Goat, powdered milk
Butter, yogurt
Salmon, cod
Clams, oysters
Imitation crab, crustaceans
Chowders, crab cakes
Apples, blueberries
Fruit salad, salsa
Lettuce, sprouts

Porridge, cassava
Infant cereal, other cereals
Frosting/icing, cakes
Tortilla, bread
Chinese food, enchilada
Pureed diet, soup
Macaroni and cheese, fried rice
Pasta salad, potato salad
Submarine sandwich, chicken
Salad dressing, gravy

Historically, various authors in the United Kingdom (Panisello

et al., 2000; Adak et al., 2005), Europe (Rocourt et al., 2003),
Australia (Dalton et al., 2004) and the USA (Smith DeWaal and
Barlow, 2002; Hoffmann et al., 2006; Painter, 2006) have analyzed
foodborne disease outbreak data by pathogen, food, or the food/
pathogen combination, indirectly providing food attribution estimates. Methodology varied between studies resulting in a range of
results and presentation, making comparison over time or between
regions difcult. The purpose of this study was to explore the
usefulness of foodborne outbreak data extracted from publicly
available international reports to provide estimates of food attribution, and to derive and compare these estimates between countries,
as well as improve the understanding of the pathogen/food vehicle
combination.
2. Materials and methods
2.1. Data sources and identication of outbreaks
To identify outbreaks, information was systematically collected
from peer-reviewed journals, listservs, press releases, health units
and government websites worldwide. Weekly searches of Current
ContentsAgriculture, Biology & Environmental Sciences and Current ContentsLife Sciences, the Institute for Scientic Information,
were conducted using the keywords diarrhea, outbreak, foodborne,
waterborne, zoonoses, case control, food, and the specic pathogens
(Escherichia coli, Shigella, Bacillus cereus, Salmonella, Campylobacter,
Staphylococcus aureus, Vibrio, Listeria monocytogenes, Streptococcus,
Clostridium, Toxoplasma, Cryptosporidium, Giardia, and norovirus).
Line listings of outbreaks from the Centers for Disease Control and
Prevention (CDC) for the years 19952005 were reviewed as well as
the European Union Food Safety Authority report of zoonotic
outbreaks for the years 2004 and 2005, the rst years these line
lists were available. The information reviewed in this study includes
foodborne outbreaks that occurred during the last 20 years (1988 to
2007).
2.2. Denition of foodborne outbreak useful for food attribution
A foodborne outbreak was dened as an event in which two or
more people became ill after consuming a common food or meal. An
outbreak useful for food attribution was dened as a foodborne
outbreak for which the available report identied both the aetiology
and the food vehicle, although the level of evidence varied. The
purpose of this study was to identify specic foods and pathogens
responsible for illness; therefore, outbreaks caused by pathogens
such as Shigella, Giardia, and viruses (including norovirus) commonly transmitted person-to-person were only included if (a) there
was microbiological evidence of food or waterborne transmission,
(b) the investigation demonstrated the source, method of contamination and survival of agent, or (c) statistical evidence strongly
associated a specic food or water source in the outbreak (Bryan
et al., 1997).

Apple, orange

2.3. Classication of food vehicle

Alcohol, tea
Chocolate, spices

When the identication and compilation of the outbreak reports

began there were no ofcially recognized methods of food categorization. A hierarchical classication system was therefore developed
to group the food into categories (Table 1). The multi-ingredient
foods group includes foods made from a variety of ingredients (e.g.
casseroles and stews) from which the precise source of infection
cannot be identied. If, during an investigation, an item in a multiingredient food was clearly identied as the source of the outbreak
(e.g. raw egg in Caesar salad dressing) then the outbreak was placed in
that category (Eggs).

developed and used in Denmark for human salmonellosis attribution

(Hald et al., 2004). The nal approach that was identied relied on
expert opinion elicitation (Batz et al., 2005; Hoffmann et al., 2006).
Each method has its own strengths and drawbacks and none seems to
be exactly and universally appropriate for food attribution for all
enteric diseases (Batz et al., 2005; Havelaar et al., 2007; Anonymous,
2008).

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

79

Table 2
Sources of publicly available foodborne outbreak reports identied between 1988 and 2007 useful for food attribution (i.e. includes the aetiological agent and the food vehicle)
Region

United States of America

European Unionb
Australia and New Zealand
Canada
Other countriesc
Total

Number of outbreaks
useful for food
attribution (%)

Source of reports
Governmenta

Peer reviewed journals

Non-peer reviewed journals

List-servs

2168
(100%)
1287
(100%)
246
(100%)
208
(100%)
184
(100%)
4093
(100%)

1995
(92.0%)
1075
(83.5%)
220
(89.4%)
164
(78.8%)
71
(38.6%)
3525
(86.1%)

107
(4.9%)
182
(14.1%)
10
(4.1%)
26
(12.5%)
54
(29.3%)
379
(9.3%)

17
(0.8%)
4
(0.3%)
2
(0.8%)
8
(3.8%)
5
(2.7%)
36
(0.9%)

49
(2.3%)
26
(2.0%)
14
(5.7%)
10
(4.8%)
54
(29.3%))
153
(3.7%)

The label Government refers to line lists, state and provincial departments of health and government publications.
Includes Austria, Belgium, Czech Republic, Cyprus, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta,
Netherlands, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden and the United Kingdom.
c
Includes Argentina, Bahrain, BosniaHerzegovina, Brazil, Chile, China, Croatia, Egypt, Guatemala, Honduras, Hong Kong, Iceland, India, Iran, Israel, Japan, Jordan, Kazakhstan,
Kyrgyzstan, North Korea, Mexico, Morocco, Norway, Peru, Philippines, Russia, Saudi Arabia, Serbia, Singapore, South Africa, South Korea, Switzerland, Taiwan, Tajikistan, Thailand,
Turkey, Ukraine, Uruguay, Venezuela, Vietnam, Zambia.
b

2.4. Data management

When a public report of a foodborne outbreak was identied, it
was examined to ensure inclusion criteria were met (aetiology and
vehicle) and duplication prevented. If multiple sources of the same
information existed, the published manuscript was selected as the
reference source. Outbreak details were entered into a Microsoft
Access 2000 database. Conrmed cases were identied by laboratory analysis while presumptive cases exhibited symptoms compatible with the disease and were epidemiologically linked by a common
meal, food or water source.
2.5. Data analysis
All recorded foodborne outbreaks with known aetiology and food
vehicle were statistically described by agent, food category and
country separately, and then further cross tabulated by agent and food
vehicle based on the number of outbreaks. At that stage, pathogens
responsible for less than ten outbreaks were grouped into an Other
category reecting their bacterial, parasitic or viral nature. The
Salmonella enterica category was further divided into Salmonella
Enteritidis, Salmonella Typhimurium and Other Salmonella enterica
to group all other serotypes and outbreaks where the serotype was not
reported. Some initial food categories were also merged due to a
limited number of outbreaks in their respective category: Turkey and
Other poultry were merged into a Turkey + other poultry category,
Nuts were placed in the Produce category, and Home canned goods
in the Other foods category. From that cross tabulated table,
proportions of outbreaks by row and by column were calculated to
provide the food source/pathogen prole (namely food attribution)
and the pathogen/food source prole.
To compare source attribution between countries, the initial
dataset was restricted by geographical scope to include data from
four regions due to the quantity of available data. The regions were
dened as Australia and New Zealand, Canada, the US and the EU
(2007 members inclusive: Austria, Belgium, Bulgaria, Cyprus, Czech
Republic, Denmark, Estonia, Finland, France, Germany, Greece,
Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden
and the United Kingdom). The regions were then compared by
similarity in population and socioeconomic development: Australia
and New Zealand vs. Canada, and the US vs. the EU. The most recent
decade with the most complete data was chosen because it has been
illustrated that food attribution based on outbreak data changes over

time with new methodologies introduced, new pathogens and new

food vehicles (e.g. Cyclospora spp. and fruits) (Ravel et al., 2008,
unpublished results). Possible changes over time in sources attribution estimates have also been highlighted and it has been recommended to update such estimates (Anonymous, 2008). The decade
19962005 was chosen for analysis because line lists were not
available from the US and EU beyond 2005.
The three dimensional (micro organism, region, food vehicle) table
was built and analyzed visually. To further explore and underscore the
signicant structure in this table, a multiple correspondence analysis
(MCA) of the outbreaks based on their region, food vehicle and
aetiology was conducted. Finally, the test of homogeneity was used to
formally test the differences between the US and the EU food
attribution values for the most frequently reported agents after
regrouping some food categories.
MCA is a descriptive statistical technique designed to explore and
visualize the relationships between three or more categorical
variables (Greenacre, 1984). It is based on a contingency table
displaying some measure of correspondence between the various
categories of each variable (micro organism, food vehicle and region).
MCA computes the inertia, which is the equivalent of the variance for
quantitative variables, and breakdowns the total inertia in axes that
gradually explain less of the inertia. Beyond the intensive mathematical computation, the most interesting output of MCA is the
representation of the multidimensional dataset on a two-dimensional
map that minimizes the deformation and underscores the relationships between all categories. The map is interpreted based on the
location of the points only, the two axes and the distances between
points having no straightforward interpretation. Points in the middle
of the gure, close to the axis origin, reveal micro organisms, main
food vehicles or regions that have no remarkable association with any
other categories. On the other hand, points away from the origin and
found in approximately the same direction from the origin highlight
some notable associations between the categories they represent. To
help interpret the dimensions, MCA computes the contribution of
every category to each dimension. A contribution is considered
signicant on one dimension when its value is greater than the
relative weight of the category, i.e. the number of observations for this
category, divided by the total number of observations (outbreaks).
MCA also computes the squared cosines of the angle between the axis
and the direction of each category from the origin. Larger square
cosine values indicate a stronger representation of the category in the
particular dimension before the multi dimensional space is projected
to the two dimension map.

80

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

Table 3
Micro organism-specic food source prole, by proportion (%) for foodborne outbreaks reported internationally between 1988 and 2007 based on publicly available reports
Micro organism

Produce

Multi-ingredient
foods

Seafood

Beef

Pork

Dairy
products

Chicken

Bacterial
Bacillus cereus
Campylobacter spp.
Clostridium botulinum
Clostridium perfringens
Escherichia coli
Listeria monocytogenes
Salmonella Enteritidis
Salmonella Typhimurium
Other Salmonella enterica
Shigella spp.
Staphylococcus aureus
Vibrio spp.
Other bacteria

74
191
108
248
389
53
991
270
657
83
182
54
62

8.1
4.7
14.8
2.8
19.5
1.9
3.6
3.6
21.0
28.9
3.3
3.7
9.7

56.8
14.1
0.9
20.2
11.8
5.7
10.1
10.7
13.9
30.1
22.0
1.9
6.5

4.1
2.6
25.0
2.0
0.5
11.3
4.2
4.8
2.6
9.6
3.3
90.7
3.2

6.7
4.7
7.4
39.1
44.2
5.7
5.1
8.5
9.6
6.0
13.7

2.7
0.5
4.6
6.5
0.5
11.3
1.5
6.7
5.6
2.4
21.4

4.1
34.6
3.7
0.4
9.8
41.5
6.4
11.9
6.2
14.5
11.0

13.5
29.3
0.9
14.5
1.0
1.9
9.9
10.4
13.6
6.0
8.2
1.9
4.8

35
47
23

91.4

8.6

26.1

55
552
19

40.0
16.5
21.1

Parasitic
Cyclospora spp.
Trichinella spp.
Other parasites
Viral
Hepatitis A virus
Norovirus
Other viruses
a

11.3

19.4

10.6

36.2
8.7

17.4

8.7

1.8
2.0

2.2
2.2

2.9

13.0

14.6
40.2
26.3

20.0
13.0
31.6

1.8
4.4
5.3

27.4

Other
meats

Bakery
items
1.4
0.5
0.9

Beverages

2.1
13.9
4.8
6.9
13.2
0.7
6.7
3.5

1.0

4.4

12.1
7.4
3.4

0.2
1.1
1.7
2.4

4.4

5.0

Eggs

Other
foods

Totala

1.4
1.6
0.9

1.4
24.1

9.7
0.3
7.6
1.7
2.2
4.6

43.4
18.2
13.6

1.1
1.5
0.9

3.3

3.9

0.6

1.6

3.2

100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%

Turkey + other
poultry

5.2
2.8

1.9
11.3

1.6

100%
100%
100%

53.2
26.1

7.3
1.8

1.8
8.7
15.8

9.1
5.8

1.1

0.4

1.1

100%
100%
100%

The sum over all food vehicles may not equal 100% due to rounding.

All statistical analyses were conducted using XLStat 2007.5 software

(Addinsoft, Paris, France) as an add-in to Microsoft Excel 2003.

listed in the US CDC line lists for bacteria/viruses/parasites and

multiple pathogens were useful for food attribution while only 28% of
the EU line list data for 20042005 was useful.

3. Results
3.2. Overall results for food attribution
3.1. Usefulness of outbreak data
For all regions, except the Other countries category, government
sources provided the majority of foodborne outbreak data useful for
food attribution (86.1% of total) (Table 2). The second most important
source for the US, EU and Canada were peer reviewed journals. Major
sources for the Other countries category included government
sources (38.6%), peer reviewed journals (29.3%) and List-servs (29.3%).
The usefulness of foodborne outbreak reports was only quantied for
the US and the EU. For the years 19952005, 49% of the outbreaks

Among the 4093 foodborne outbreaks for which a report was

identied, accessible and useful, almost 70% were attributable to
Salmonella, Norovirus and E. coli (46.9, 13.5 and 9.5% respectively).
Salmonella Enteritidis was the most frequent Salmonella serotype
(991 foodborne outbreaks or 24.1% of total), followed by Typhimurium
(270 outbreaks or 6.6%). The most frequently reported food categories
were Multi-ingredient foods, Eggs, and Produce and Beef ex aequo
at the third rank (17.0, 14.3, 12.2 and 12.2% of all outbreaks,
respectively).

Table 4
Food source-specic micro organism prole (%) for foodborne outbreaks reported internationally between 1988 and 2007 based on publicly available reports (source ranked by
decreasing number of micro organisms)
Micro organism

Bacillus cereus
Campylobacter spp.
Clostridium botulinum
Clostridium perfringens
Escherichia coli
Listeria monocytogenes
Salmonella Enteritidis
Salmonella Typhimurium
Other Salmonella enterica
Shigella spp.
Staphylococcus aureus
Vibrio spp.
Other bacteria
Cyclospora spp.
Trichinella spp.
Other parasites
Hepatitis A virus
Norovirus
Other viruses
Totala
a

Produce

Multi-ingredient
foods

Seafood

Beef

Pork

Dairy
products

Chicken

Other
meats

Bakery
items

Beverages

Turkey + other
poultry

Eggs

Other
foods

N = 498
1.3
1.9
3.3
1.5
15.3
0.3
7.3
3.3
27.8
4.9
1.3
0.5
1.3
6.5

N = 697
6.1
3.9
0.2
7.2
6.6
0.5
14.4
4.2
13.1
3.6
5.8
0.2
0.6
0.5

N = 277
1.1
1.9
9.8
1.9
0.8
2.2
15.2
4.7
6.2
2.9
2.2
17.7
0.8

N = 498
1.1
1.9
1.7
19.5
34.6
0.7
10.1
4.7
12.7
1.1
5.1

N = 197
1.1
0.6
2.6
8.2
1.1
3.1
7.7
14.8
18.8
1.1
19.8

N = 337
0.9
19.6
1.2
0.3
11.3
6.6
18.7
9.5
12.2
3.6
6.0

N = 365
2.8
15.4
0.3
9.9
1.1
0.3
26.9
7.7
24.4
1.4
4.2
0.3
0.9

N = 167

N = 231
0.5
0.5
0.5

N = 82

N = 105

N = 584
0.2
0.6
0.2

N = 55
1.9

22.9
1.0
3.9
16.2
5.8
28.6

73.7
8.4
15.3

20.0
7.3
11

5.8

1.2

1.9

1.0

0.4

7.4
6.1
39.1

5.8

0.4

11.0

100%

100%

100%

100%

1.3
4.5
18.3
0.9
100%

1.2
31.9
0.8
100%

1.1
4.0
26.0
2.2
100%

2.4
9.0
7.2
16.2
4.2
4.2
10.8
13.8
4.8

6.1

5.1

1.1

8.7
1.1
0.6
5.6

1.2
0.6
3.6

0.6
4.4

2.4
6.0

100%

100%

100%

100%

The sum over all micro organisms may not equal 100 due to rounding.

1.8

20.8

52.0
8.7
9.6

2.5
3.7
13.5
2.5

3.9

47.3

1.3

1.5

0.3
4.9
0.3
100%

9.6
3.7

4.2

0.5

15.0
0.5
20.8
1.3
100%

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

81

Table 5
Pathogen and main food vehicle cross-tabulation for foodborne outbreaks reported between 1996 and 2005 in four regions based on publicly available reports (restricted to the most
frequently reported pathogens)
Micro organism

Region

Beef

Campylobacter spp.

Aus-NZ
Canada
EU
US
Total
Aus-NZ
Canada
EU
US
Total
Aus-NZ
Canada
EU
US
Total
Aus-NZ
Canada
EU
US
Total
Aus-NZ
Canada
EU
US
Total
Aus-NZ
Canada
EU
US
Total
Aus-NZ
Canada
EU
US
Total
Aus-NZ
Canada
EU
US
Total

5
1
7
4
5
14
64
87

30
15
52

4
3
8

6
29
35

22
22

Clostridium perfringens

Escherichia coli

Salmonella Enteritidis

Salmonella Typhimurium

Other Salmonella Enterica

S. aureus

Norovirus

Total
a

Pork

2
13
15

34
12
79
125

37
9
46
3
3
6
7
19
1
5
9
39
54

12
3
15
7

8
24
34

2
21
23

2
37
39

24
24
385

1
10
11
140

10
7
24
2

Chicken

Turkey + other
poultry

1
3
3

75
16
91
14
1
2
11
28
4
3
16
58
81
1

1
1
9
7
17
1

Other
meatsa

4
2

Eggs

3
3

Dairy
products
1
1
9
37
48

3
5
10
6
4
5
15

5
17
8
30
5
326
65
396
31
1
12
3
47
6
2
39
23
70

1
35
21
57
3
2
2
17
24
1
2
12
14
29

4
1
6

3
4
7
6
2
2
2
12

3
3
22
28

2
7
10
19

11
12

4
4

1
7
8

2
4
6
1

12
3
15

1
15
16
318

1
5
6
92

10
10
85

1
2
524

12
12
215

Seafood

Produce

4
4
2

2
7
9
1

2
4

5
6

2
2

2
3
51
56

21
18
39
8

6
8
21
35
3

4
12
1
2
5
7
15
1
2
3
6
9
3
6
45
63
145

4
4
11
10
10
3
79
102
1
2
2
5
1
2
7
76
86
310

Bakery
items

1
1

Multi-ingredient
foods

Beverages

Other
foods

4
1
6
13
24
9

16
2
62
80
160
18
5
32
167
222

7
27
43

1
2
3

92
16
108
14
2
2
1
19
2
4
13
19
1
3
4
8
3
6
38
47
205

3
3
28
34
1
42
44
87
17
2
10
29
2
1
12
55
70
4
3
27
34
6
3
4
197
210
531

Total

2
1
11
14

55
41
189
285

11
2
2
2

1
3
1
2
2
5

11
4

4
1
1
1
3

1
1

32
32
56

6
6
25

14
671
226
911
113
11
46
68
238
30
33
119
347
529
7
1
29
124
161
20
8
26
471
525
3031

Meat products where type of meat was not clearly stated.

Table 3 provides the source prole of each micro organism, i.e. the
distribution (or the proportion) of the food vehicles that were
incriminated according to the outbreak investigation and the
categorization method used in this study, resulting in food attribution
estimates. It shows a variable degree of specicity for the association
between the micro organism and food categories. On one hand,
foodborne outbreaks due to Cyclospora spp., Trichinella spp., Vibrio
spp. and viruses other than Norovirus and Hepatitis A (HAV), were
linked to foods from only a very limited number of categories. On the
other hand, Salmonella related foodborne outbreaks and those due to
Norovirus were associated with foods from all 13 food categories, with
none being signicantly predominant, with the exception of the Eggs
category for S. Enteritidis (43.4% of all S. Enteritidis outbreaks) and the
Multi-ingredient food category for Norovirus. For the other pathogens, the outbreaks were associated with more than seven food
categories (Nthan half the number of food categories used in this
study), with one or two categories being predominant, for example
E. coli and Beef (44.2%), L. monocytogenes and Dairy products (41.5%),
HAV and Produce (40.0%) or Clostridium perfringens and Beef (39.1%).
Table 4 describes the micro organism-specic prole of each food
category based on the number of foodborne outbreaks associated with
it. It clearly highlights the specicity or lack thereof in the relationship
between a food category and a micro organism in a complementary
way compared to Table 3. On one hand, the Eggs and the Other foods
categories were linked to outbreaks due to less than 10 (half) micro

organisms among those considered and grouped in this study (19 in

total), with two pathogens largely predominant. On the other hand,
the Produce, Multi-ingredient foods and Seafood categories were
the vehicles for outbreaks due to all but one or two micro organisms
(parasitic pathogens). Generally, one or two pathogens were predominant within each food category in terms of the number of
outbreaks. The ve strongest associations were: S. Enteritidis in nearly
three quarters (73.7%) of egg-related outbreaks, and in half (52%) of
bakery associated outbreaks, Clostridium botulinum in nearly half
(47.4%) of the outbreaks related to Other foods (essentially home
canned food), Norovirus in 39% of outbreaks linked to the Beverages
category, and E. coli in one third (34.6%) of the beef-related outbreaks.
The precise beverages associated with Norovirus outbreaks were
water (37 outbreaks), ice likely contaminated by an infected
foodhandler (16), and other beverages (16).
3.3. Multivariate descriptive analysis of regions, micro organisms and
food vehicles
The dataset was restricted to foodborne outbreaks occurring
between 1996 and 2005, in Australia and New Zealand, Canada, the
EU and the US, and to the most common pathogens, in an effort to
compare the source attribution value between regions, but the resulting
number was rather small and dominated by the US data (Table 5),
making the visual appraisal of differences between the regions for

82

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

Fig. 1. Map resulting from the multiple correspondence analysis of the micro organism and main food vehicle cross-tabulation of foodborne outbreaks that occurred between 1996
and 2005 in four regions (Australia and New Zealand, Canada, the European Union and the United States of America) based on publicly available reports (restricted to the most
frequent pathogenic micro organisms). It shows the relative positions of the regions (in brown), food vehicles (in blue) and micro organisms (in yellow) on the rst 2 dimensions
following the multiple correspondence analysis. The rst and the second dimension encompass 52% and 15% of the total variance in the multidimensional data set, respectively. The
size of the bubbles is proportional to the number of outbreaks for the category it represents (Total number of outbreaks = 3031). The map is interpreted based on the location of the
points only, the two axes and the distances between points having no straightforward interpretation. Points in the middle of the gure, closed to the axis origin, reveal micro
organisms, main food vehicles or regions that have not remarkable association with any other categories. On the other hand, points away from the origin and found in approximately
the same direction from the origin highlight some remarkable associations between the categories they represent seen in the corresponding contingency table, which are three here:
EU, S. Enteritidis and Eggs; S. Typhimurium and Australia and New Zealand; and E. coli, Canada and Beef.

distribution of pathogen-specic outbreaks by food vehicle very

difcult. The EU had more reported and linked outbreaks associated
with S. Enteritidis than the US (three times more). The relatively small
differences between Campylobacter spp. outbreaks in the EU and the US
identied Dairy products, Chicken and Multi-ingredient foods as the
most frequently associated food categories in both regions.
The outputs from the MCA showed that the rst two dimensions
successfully encompass approximately 52% and 15% of the total inertia,
or in other words the total variance, present in the original three
dimensional table (i.e. Table 5). S. Enteritidis, the EU and the Eggs
category were the variable modalities contributing the most to
Dimension 1, while the Australia and New Zealand region, S.
Typhimurium, E. coli and the Beef category were the ones that
contributed most to Dimension 2 (numerical results not shown). With
all squared cosine values far below 1 (numerical results not shown),
there were no categories located along any of those dimensions,
indicating a general lack of meaning of the dimensions. Finally, the
projection of the four regions, food vehicles and pathogen categories on
the rst two dimensions produced a map most representative of the
specic structure included in Table 5 (Fig. 1). In particular, it showed
three interesting subsets of categories pointing along the same direction
from the origin: EU, S. Enteritidis and Eggs along the rst dimension far

away from the origin; S. Typhimurium and the Australia and New
Zealand region on the second dimension; and E. coli, Canada and Beef in
the fourth quadrant, relatively far from the origin on the second
dimension. Each of those subsets highlights a specic association
between the categories it includes compared to all other categories.
The formal statistical comparisons between the US and EU for the
distribution of outbreaks by food vehicle showed signicant differences for Campylobacter spp., S. Enteritidis and Salmonella other than
Enteritidis or Typhimurium, but not for C. perfringens and S.
Typhimurium (Table 6). In particular, Campylobacter associated outbreaks were mainly related to poultry products in the EU and to dairy
products in the US. Outbreaks related to S. Enteritidis were more
specically associated to egg products in the EU compared to the US,
while the most frequent vehicle for outbreaks under the Other
S. enterica category was Eggs in the EU and poultry products in the
US.
4. Discussion
The association between pathogens causing gastrointestinal illness
in the population and the specic food that serves as a vehicle for
these infections is the essence of food attribution. Over the years,

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

83

Table 6
Test for homogeneity between the European Union and the United States for the distribution of foodborne outbreaks by food vehicle for the most frequently reported pathogens
publicly available between 1996 and 2005
Micro organism

Region

Beef

Pork + Other
meats a

Campylobacter spp.

EU
US
Total
EU
US
Total
EU
US
Total
EU
US
Total
EU
US
Total

14
64
78
37
9
46
6
7
13
9
39
48

5
18
23
15
7
22
12
9
21
15
34
49

Clostridium perfringens

Salmonella Enteritidis

Salmonella Typhimurium

Other Salmonella enterica

a
b

Chicken + turkey +
other poultry
34
18
52
6
51
57
84
23
107
6
12
18
19
80
99

Eggs

Dairy
products

Seafood

Bakery
items

Produce

Multi-ingredient
foods

326
65
391

39
23
62

9
37
46

35
21
56

12
14
26

21
18
39

5
7
12

92
16
108

8
21
29

6
13
19
7
27
34
42
44
86

12
55
67

Total
49
68
117
32
160
192
660
224
884
24
28
52
111
252
363

Observed Chi2
value
22.01

2.44

101.75

2.21

48.46

p-value
b0.0001

0.485

b 0.0001

0.331

b 0.0001

Meat products where type of meat was not clearly stated.

Numbers of outbreaks not considered for the test because at least one cell for the food category was below ve.

many reports from various countries have been published summarizing foodborne outbreak data. Some tabulated the aetiological agents
and food vehicles separately (see for examples Adams and Moss,
2000; Anonymous, 2002, 2004; Ray, 2004; Smith Dewaal et al., 2006),
whereas others also provided the aetiology-food cross-tabulation (see
for examples Rocourt et al., 2003; Lynch et al., 2006). The scope of
those reports varied not only geographically but chronologically.
There was also variation in the way aetiological agents and foods were
dened and categorized, even within a country, impeding direct
comparisons of food attribution across countries for the same time
period. This study is specically focused on food attribution and only
the association between food and micro-organisms is presented and
discussed. In addition, its scope was global in an effort to assess the
similarities of associations between food and aetiologies between
countries, to provide more reliable food attribution estimates.
The general advantages and disadvantages of using foodborne
outbreak data to derive source attribution gures have been reviewed
(Batz et al., 2005; Anonymous, 2008). The major strength is that the food
attribution estimates are based on actual observations linking cases,
causes and vehicles. Practically, this approach is straightforward and easy
to implement using existing datasets. Such foodborne outbreak datasets
generally provide good temporal and geographical coverage and allow
dealing with many pathogens and food vehicles at the same time. The
major weakness of this approach is the representativeness of outbreaks
compared to the more numerous sporadic cases. Many biases can possibly
impede the extrapolation of ndings from outbreak cases to all cases:
detection bias, investigation bias, and reporting or publication bias. As a
result, some pathogens and some food vehicles may be articially
highlighted and others underestimated. In addition, there is no
standardized categorization of food and the a posteriori categorization of
vehicles is limited by the content of the outbreak dataset, which likely has
not been uniformly populated over time and space. Those general
limitations inherent to the use of foodborne outbreak data for food
attribution will not be further discussed here. However, this study has
some specic features that require discussion before interpreting its
ndings: the data source; the unit of analysis; the food and aetiology
categorization used; the comparison between countries; the time frame
considered; and the use of MCA.
Data sources included electronically accessible public reports of
foodborne outbreaks from government sources, peer-reviewed publications, non-peer reviewed journals, and list-servs. The majority of
data (86.1%) were from government sources; 9.3% were from peerreviewed journals. Publication bias has been demonstrated for
outbreaks published through peer-reviewed journals (O'Brien et al.,
2006). These authors quantied the publication bias for the aetiologies,

food sources, and modes of transmission separately by comparing

the proportions derived from the peer-reviewed papers and from the
national foodborne outbreak database in England and Wales;
however, they did not compare the two data sources for the
association between foods and aetiology. Moreover, they stated that
the food source was available in 98% (51/55) of published outbreaks
and in 66% (1169/1763) of recorded outbreaks, indicating a tendency
to publish outbreaks for which both the aetiologic cause and the
implicated food vehicle were reported. Since the food source and the
causative agent are both required for any work on food attribution
based on outbreak data, this methodological approach implies this
restriction beyond the publication bias. The impact of this restriction
on the extrapolation of the results will be discussed at the end of this
section. More importantly, the use of data reported electronically
favoured certain countries, not only in terms of public health
capacity but also in informatics technology infrastructure and use.
This resulted in over representation of some countries, especially the
US and the EU state members and required further renement of the
analysis describing individual data. Only outbreaks reporting an
aetiological agent and an implicated food vehicle were included in
the analysis with the assumption that these variables were correct
based on reliable epidemiological or microbiological evidence during
the outbreak investigation. This information is not systematically
documented in outbreak reports, impeding stratication and further
analysis based on level of evidence. This weakness associated with
outbreak data could be reduced by systematic and standardized
documentation of the evidence implicating the causative agent and
vehicle during outbreak investigations.
The unit of interest and analysis for this source attribution study was
the outbreak. No data concerning the number of cases or deaths are
presented compared to previous outbreak data synthesis. The unit was
chosen because outbreak cases are non-independent subjects by
denition and the issue explored relates to food consumption resulting
in illness. The actual number of cases was considered to be more related
to the causative pathogen, the setting and contributing factors that lead
to the consumption of unsafe food rather than to the nature of the
vehicle per se. Therefore, a family outbreak caused by undercooked
ground beef patties and a country-wide outbreak linked to undercooked
beef patties served at a fast-food chain have the same food source and
each counts for one in the association between beef and the causative
agent. To support this, Rangel et al. (2005) reported that E. coli associated
foodborne outbreaks in the US were signicantly larger when related to
produce compared to ground beef (median number of cases: 20 vs. 8)
and the distribution of settings varied. In addition, when used for
ranking source and illness, food specic attribution values are multiplied

84

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

by the number of estimated cases of the disease, the latter derived by

multiplying the actual number of reported cases by the underreporting
rate (Batz et al., 2004; Adak et al., 2005). Adak et al. (2005) used the
number of outbreaks in their estimation of food attribution. In contrast,
Batz et al. (2004) provided little rationale for the use of outbreak cases to
estimate food attribution while building their risk ranking model. In
brief, we decided it was preferable not to consider the size of the
outbreaks when generating the food attribution values, understanding
that the impact or burden of the food-pathogen combination on
population health is highly relevant for risk assessment and policymaking and this impact will be dealt with at a subsequent stage.
The analysis of foodborne outbreak data for the purpose of source
attribution requires categorization of the reported food vehicle;
previously various food category denitions or hierarchical structures
were used. The lack of a classication standard has been a major
problem when comparing summary reports of foodborne outbreak
data. Recently the EU proposed standards to uniformly compile
information from its state members (Anonymous, 2007). For the
purposes of this work, the outbreak was attributed to the most specic
food vehicle, even if contained as an ingredient in a complex food or
dish, providing the best information available from the outbreak
reports (eggs in a homemade Caesar salad dressing). Compared to
other summary reports, the Multi-ingredient food may be less
specic representing complex foods when no specic item was
identied as the vehicle in the report, however, reciprocally, complex
food from previous similar works may be hidden under a more
specic food category in our results. Painter (2006) have recently
attempted to attribute an outbreak related to a complex food to a
specic item using a method that considers average recipes for the
complex food and the risk prole of each specic ingredient. This
method was not applied in this study because it is relatively new and
there were doubts concerning its application on global outbreak data.
Aetiology was classied based on biological taxonomy and is less
prone to misclassication, although some outbreaks could have been
misclassied for S. enterica if the outbreak report did not include a
precise enough serotype compared to our grouping. The Other Salmonella enterica (not Enteritidis and Typhimurium) category may
indeed include these serotypes if not completely typed during the
investigation, thus reducing our ability to demonstrate and test the
actual specic food sources for these two common Salmonella
serotypes.
This study was designed with the widest geographical scope in an
effort to compare source attribution between countries. It was
assumed that consistency of food attribution estimates across
countries for a given micro organism would strengthen our condence
in their validity but the method of data collection lead to over
representation of a few countries, limiting formal comparison
between countries. The study did show signicant differences
between countries as discussed below, and MCA was an innovative
feature of this work supporting such differences. To overcome
problems in the formal testing of differences between countries
concerning the association between food sources and aetiology, MCA
was used to describe and, more importantly, visualize the relationship
between these three categorical variables. MCA visually complements
the quantitative table it analyzes (Table 5). We used outbreaks that
occurred during the last two decades, with more in-depth analysis on
the last decade of available data (19962005). Considering the natural
dynamics in the epidemiology of infectious gastrointestinal diseases
combined with the dynamics of the anthropological activities related
to food production, processing, distribution and consumption and
human behaviour, the food attribution results may appear dated. The
time period was chosen a priori to have the largest number of
outbreaks to increase statistical power to illustrate true differences
between countries. That being said, food attribution certainly changes
over time and estimates should be revisited regularly as recommended by others (Anonymous, 2008).

This study provides the most comprehensive set of food attribution

estimates using outbreak data to date, based on more than 4000
outbreaks that occurred globally over the last two decades. Although it
encompasses the most commonly known pathogens responsible for
infectious enteric diseases, it mainly represents the situation in the US and
other occidental countries (Australia, New Zealand, Canada and the EU
state members). The tabulation of foodborne outbreaks by food vehicle,
pathogen and region showed variable specicities in association between
those categories. Outbreak vehicles for some micro organisms seem very
specic across all regions. This was clear for parasitic outbreaks: 91% of
cyclosporiasis outbreaks were linked to Produce, and 36% and 53% of
trichinellosis outbreaks were linked to Pork and Other meat (a category
that may include wild game), respectively. The biological cycle and
characteristics of these parasites prevent cross-contamination (see Adams
et al., 2000; Ray, 2004). Similarly, seafood was the most frequent vehicle
(91%) for outbreaks due to Vibrio spp., a known aquatic genus (Adams et
al., 2000; Ray, 2004).
Many pathogens were associated with several or all food categories,
while others were associated with one particular food category as
illustrated by: C. perfringens mainly associated with Beef, L. monocytogenes with Dairy products, HAV with Produce, B. cereus with Multiingredient food, E. coli with Beef and Norovirus with Multi-ingredient
food. Other pathogens such as C. botulinum, S. aureus, and S. enterica
Typhimurium had no clear more frequent food vehicles. Outbreaks due to
Campylobacter spp., S. enterica Enteritidis or to S. enterica other than
Enteritidis or Typhimurium were associated with many food vehicles
although there were some differences between countries, as further
discussed below. The relative lack of specicity in association between
foods and pathogens highlighted by this study is intriguing given that
most zoonotic pathogens have at least one main reservoir, usually food
animals, and a few main routes of transmission, usually food vehicles of
animal origin. Direct human-to-human transmission is usually considered
nonexistent or marginal for the most common infectious gastrointestinal
diseases such as non typhi salmonellosis and campylobacteriosis. If so,
then at any point in the food chain, cross-contamination, environmental
contamination or contamination by food handlers could explain our
ndings. Environmental or anthropological recontamination of processed
foods after nal preparation has been recently reviewed and highlighted
as a source of pathogens (Reij, Den Aantrekker, and ILSI Europe Risk
Analysis in Microbiology Task Force, 2004). A recent review of 816
foodborne outbreaks associated with infected food workers showed 33%
were associated with norovirus, 18% with Salmonella, 10% with HAV, 6.5%
with S. aureus and 4% with Shigella (Greig et al., 2007), thus demonstrating
the signicant role of food handlers in transmission of zoonotic
pathogens. The current study showed the following pathogens were
associated with few (other viruses: 5), several (Shigella: 8; HAV: 9) or all
(norovirus: 13) main food categories demonstrating a non-food specic,
foodborne transmission of non-zoonotic enteropathogens from a human
reservoir (symptomatic or asymptomatic) to other individuals. This could
also be true for zoonotic pathogens, partially explaining the relatively
wide range of food vehicles associated with outbreaks.
Salmonellosis outbreaks were the most numerous over the four
regions (Australia and New Zealand, Canada, EU, US) with little variation
between them. First, outbreaks related to S. enterica Enteritidis,
Typhimurium, and other (including non-specied) seemed to each be
linked to a wide range of food items, with no predominant source except
the strong association between S. Enteritidis and the Eggs category. This
is consistent with other reports; for example, a study in England and
Wales between 1992 and 1996 provided a cross tabulation of 350
foodborne outbreaks by aetiology and vehicle of transmission (Panisello
et al., 2000). Of them, 291 were Salmonella outbreaks linked to all vehicles
considered: dessert, sauce, sandwich and miscellaneous food (37%);
poultry (23%); eggs (14%); red meat (13%); seafood (5%); vegetables (4%);
and seafood (3%). In the USA, Hoffmann et al. (2006) elicited food source
attribution values from experts and compared them to values derived
from outbreak data (source of outbreak data not cited but presumably

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

from the US). The outbreak-based estimates included the following food
vehicles: eggs (36.5% of the outbreaks), poultry (18.3%), produce (16.6%),
dairy (7.2%), beef (6.3%) and pork (2.9%); whereas the mean estimates over
the 44 experts were 21.8, 35.1, 11.7, 7.3, 10.9 and 5.7%, respectively, with
those numbers statistically different from the outbreak estimates for all
commodities except dairy. Dalton et al. (2004) analyzed 60 Salmonella
foodborne outbreaks that occurred in Australia between 1995 and 2000.
Salmonellosis was linked to 14 of 20 listed food categories with
miscellaneous (27%), chicken (17%) and sh (9%) the three most frequent
vehicles. Secondly, salmonellosis food attribution is different between
countries. In particular, Australia and New Zealand experienced a
relatively higher number of outbreaks due to S. Typhimurium compared
to Canada, the EU and the US, and no outbreaks due to S. Enteritidis. In
addition, relatively more S. Enteritidis outbreaks appeared to have
occurred in the EU with a stronger association with eggs and egg products
compared to the US. MCA clearly underscores these differences. Enteric
infection by S. Enteritidis due to consumption of eggs has been repeatedly
demonstrated in Europe and has been a major food safety issue of concern
for more than a decade (see for example: Rabsch et al., 2001; De Jong and
Ekdahl, 2006; Little et al., 2006; Domnguez et al., 2007). In Denmark, it
was estimated that 45 to 50% of all human salmonellosis cases were due to
eggs in 1998 (Anonymous, 1998). In the US, there was an increase in S.
Enteritidis cases in the 80s and mid 90s followed by a decrease due to the
implementation of quality assurance programs for eggs (Mumma et al.,
2004; Patrick et al., 2004; Lynch et al., 2006). In Australia, S. Enteritidis
infection seems to be mainly a travel-associated disease, encompassing at
least 69 and 84% of all cases in 2004 and 2005, respectively (The
OzFoodNet Working Group, 2005, 2006).
Another difference highlighted by MCA between the regions was
the clustering of Canada, E. coli and Beef. The association between
beef and E. coli was expected since ruminants, especially cattle, are the
most important reservoir of human pathogenic E. coli strains. Their
stronger association with Canada compared to the EU and more
particularly to the US reects the actual numbers: the three regions
showed the same six most frequent food vehicles, but the relative
proportion of the Beef vehicle is much higher for Canada (34/55
outbreaks or 62%) compared to EU (12/41 or 29%) and the US (79/189
or 42%). In their review of E. coli associated outbreaks in the US, Rangel
et al. (2005) showed that ground beef was associated with 41% of
foodborne outbreaks, followed by unknown food vehicles (23%) and
produce (21%). No biological or epidemiological explanations have
been found to support a true difference between Canada and other
countries in relation to vehicles of foodborne outbreaks.
Results for the pathogen/food source prole demonstrated an
almost common lack of specicity in the relationship between food
categories and agents. Exceptions included the Eggs category which
was associated with nine aetiological categories out of the 19 used,
with the three Salmonella categories encompassing 92% of the eggassociated outbreaks, and Beverages associated with 10 agents out of
the 19 possible, mostly Norovirus and E. coli (39 and 21% of the
beverage-associated outbreaks, respectively). It appears that almost
all foods can serve as the vehicle for a wide range of pathogens
resulting in an outbreak, even when there is no theoretical link
between an animal reservoir and the food item itself. Contamination
of food products can occur at any point in the food chain from primary
production in the eld to the point of consumption by the consumer.
This also supports possible cross-contamination between foods, and
recontamination during processing or preparation, including contamination by food handlers. Among the non-zoonotic diseases, the three
predominant vehicles included the Produce, Multi-ingredient food,
and Seafood categories for the viral enteric outbreaks, and Produce,
Multi-ingredient food and Dairy products for shigellosis outbreaks.
The predominance of these four food categories could be explained as
follows. Foods within these categories are among those that could be
minimally processed before consumption and therefore are more
likely to transmit infection if contaminated (e.g. fruits and vegetables;

85

raw milk or cheese; and raw or undercooked seafood). Preparation of

multi-ingredient foods and some dairy products may require
considerable manipulation resulting in greater opportunities for
cross-contamination, environmental contamination or contamination
by food handlers. It can alternatively be argued that these categories
are broad and may encompass higher volumes of food compared to
more specic categories, such as the various meat commodities. This
would lead to more frequent associations with outbreaks even with a
uniform food-specic risk of disease transmission. The extension of
the actual vehicle for foodborne outbreaks from the traditional list of
foods of animal origin to include produce and complex or multiingredient foods has already been noted (Hughes, Gillepsie, O'Brien,
and the Breakdowns in Food Safety Group, 2007).
An analysis of foodborne outbreak data for estimation of food
attribution requires data from foodborne outbreaks for which both the
causative agent and the food vehicle are reported. Those outbreaks are
a fraction of all declared outbreaks because aetiology and vehicle are
rarely reported for a variety of reasons. In addition, and irrespective of
the results of the outbreak investigation, outbreaks and their related
cases represent a very small portion of the actual cases of foodborne
acute gastrointestinal infections in the human population in any area
for reasons summarized by Lynch et al. (2006). As a result, outbreak
data are considered to introduce a bias when estimating the disease
specic occurrence (e.g. cases, hospitalized, deaths) or when estimating the source specic burden (Batz et al., 2005). At rst glance, using
data in this study from foodborne outbreaks with both reported
aetiology and food vehicle could lead to an even greater bias. However,
when considering the extrapolation of the present ndings beyond
the outbreak context, we must remember the topic is food attribution,
the relationship between the causative pathogen and the food vehicle.
Does the detection of an outbreak inuence the food-aetiology
combination? At the beginning of an investigation due to a priori
opinions, bias may exist in the choice of the pathogen or the food
tested. Especially after identication of the pathogen, bias may occur
in the choice of foods tested and negative results may not be reported.
We recognize that knowledge about one piece of the combination may
inuence the outbreak investigation and its results (for example, a
Salmonella-related outbreak may trigger more extensive egg products
or poultry meat testing when the actual food vehicle is peanut butter).
From experience it would be surprising if outbreak investigations
generally did not target and test the most common food vehicles
associated with a particular pathogen and, consequently, the most
frequently reported foods acting as vehicles for acute infectious
enteric diseases detected during outbreak may not in reality be the
most frequent. We believe that the documented reports of food/
pathogen combinations from outbreaks are indicative of the most
common food/pathogen combinations (among the known pathogens); while the absence of such ndings could hide only less frequent
food vehicles at a given point in time. Therefore, we conclude that the
analysis of foodborne outbreak data by vehicle and aetiology provides
a relatively reliable order of magnitude when estimating food
attribution. Although there is uncertainty around these estimates, it
may be greater for diseases rarely reported as outbreaks compared to
their incidence in the community, such as campylobacteriosis. When
building their risk ranking model, Batz et al. (2004) used US foodborne
outbreak data based on case numbers while eliciting opinions on food
attribution from 101 experts. They reported the comparison of food
attribution between outbreak data results and expert judgement for
campylobacteriosis and listerioisis. The authors underlined the minor
differences for listeriosis, but some major differences for campylobacteriosis explained by the fact that Campylobacter associated
outbreaks are relatively rare compared to its sporadic occurrence. In
a separate paper on the same work, Hoffmann et al. (2008) detailed
their estimates of the burden of foodborne illnesses (annual
occurrence, hospitalization and deaths) for various foodborne diseases using expert-based and outbreak-based food attribution values.

86

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787

They showed differences, sometime large, between the two methods

for deriving food attribution values. They advocated the use of expert
opinion because outbreak data generally would be incomplete,
whereas using a structured method of eliciting expert opinion,
allowing for assessment of the reliability of the expert-based data,
would cover all food-pathogen combinations. The reasons for the
discrepancies between expert-based and outbreak-based estimates of
burden of illness were not discussed in the paper, although they
highlighted the consequences of using inaccurate estimates when
making decisions on the control of foodborne diseases. Finally we
should keep in mind that using outbreak data for food attribution is
the only methodological approach where there is an actual, direct link
between the pathogen, its source and each infected person, compared
to the other approaches that were judged relevant.
5. Conclusion
This study highlights the usefulness of foodborne outbreak data
obtained from publicly available reports to estimate food attribution,
when both the aetiology and the incriminated food are reported. It
allows generation of food attribution estimates for the most commonly
reported enteropathogens based on the largest international foodborne outbreak database to date where the food/pathogen combination is known. This study highlights a common lack of high specicity
in the association between aetiology and food vehicle for both zoonotic
and non-zoonotic diseases, suggesting that cross-contamination,
environmental contamination and food handler contamination may
be common along the food chain. Some differences between countries
appeared, especially for S. enterica and Campylobacter suggesting
differences in the epidemiology of these important human enteropathogens. Despite some limitations in the methodology, especially
concerning food categorization, the food attribution estimates provided by this study should give an initial point for estimating the
health and economic burden of some food-pathogen combinations,
at least for the occidental countries. Next steps should include
validation of these results from other relevant approaches for food
attribution.
Acknowledgments
We would like to acknowledge the constructive comments
provided by Dr. Andrea Ellis, Dr. Shannon Majowizc, Dr. Frank
Pollari and Dr. Paul Sockett of the Center for Foodborne, Environmental and Zoonotic Infectious Disease, Public Health Agency of
Canada.
References
Adak, G.K., Meakins, S.M., Yip, H., Lopman, B.A., O'Brien, S.J., 2005. Disease risks from
foods, England and Wales, 19962000. Emerging Infectious Diseases 11,
365372.
Adams, M.R., Moss, M.O., 2000. Food Microbiology, 2nd ed. Royal Society of Chemistry,
Cambridge, England, p. 479.
Anonymous, 1998. Annual Report on Zoonoses in Denmark 1998. http://zoonyt.dzc.dk/
annualreport1998/index.html. (Accessed 15 April 2008).
Anonymous, 2002. Outbreak alert. Closing the Gaps in Our Federal Food-Safety Net.
Updated and Revised September 2002. Center for Science in the Public Interest,
Washington, pp. 158.
Anonymous, 2004. Outbreak alert. Closing the Gaps in Our Federal Food-Safety Net.
Updated and Revised March 2004. Center for Science in the Public Interest,
Washington, pp. 121.
Anonymous, 2007. Report of the Task Force on Zoonoses Data Collection on
harmonising the reporting of food-borne outbreaks through the Community
reporting system in accordance with Directive 2003/99/EC. The European Food
Safety Authority (EFSA) Journal 123, 116.
Anonymous, 2008. Scientic Opinion of the Panel on Biological Hazards on a request
from EFSA on Overview of methods for source attribution for human illness from
food borne microbiological hazards. The European Food Safety Authority (EFSA)
Journal 764, 143.
Batz, M.B., Hoffmann, S.A., Krupnick, A.J., Morris, J.G., Sherman, D.M., Taylor, M.R., Tick, J.S.,
2004. Identifying the most signicant microbiological foodborne hazards to public
health: a new risk ranking model. Food Safety Research Consortium Discussion Paper

Series Number 1 (September 2004). http://www.rff.org/documents/FRSC-DP-01.pdf.

(Accessed 14 February 2008).
Batz, M.B., Doyle, M.P., Morris, J.G., Painter, J., Singh, R., Tauxe, R.V., Taylor, M.R.,
Lo Fo Wong, D.M., 2005. Attributing illness to food. Emerging Infectious
Diseases 11, 993999.
Bryan, F.L., Guzewich, J.J., Todd, E.C.D., 1997. Surveillance of foodborne disease III.
Summary and presentation of data on vehicles and contributory factors; their value
and limitations. Journal of Food Protection 60, 701714.
Dalton, C.B., Gregory, J., Kirk, M.D., Stafford, R.J., Givney, R., Kraa, E., Gould, D., 2004.
Foodborne disease outbreaks in Australia, 1995 to 2000. Communicable Diseases
Intelligence 28, 211224.
De Jong, B., Ekdahl, K., 2006. The comparative burden of salmonellosis in the European
Union member states, associated and candidate countries. BioMed Central (BMC)
Public Health 6, 4.
Domnguez, A., Torner, N., Ruiz, L., Martnez, A., Bartolom, R., Sulleiro, E., Teixid, A.,
Plasencia, A., 2007. Foodborne Salmonella-caused outbreaks in Catalonia (Spain),
1990 to 2003. Journal of Food Protection 70, 209213.
Doyle, M.P., Erickson, M.C., 2006. Reducing the carriage of foodborne pathogens in
livestock and poultry. Poultry Science 85, 960973.
Greenacre, M.J., 1984. Theory and Applications of Correspondence Analysis. Academic
Press, London.
Greig, J.D., Todd, E.C., Bartleson, C.A., Michaels, B.S., 2007. Outbreaks where food
workers have been implicated in the spread of foodborne disease. Part 1.
Description of the problem, methods, and agents involved. Journal of Food
Protection 70, 17521761.
Hald, T., Vose, D., Wegener, H.C., Koupeev, T., 2004. A Bayesian approach to quantify the
contribution of animal-food sources to human salmonellosis. Risk Analysis 24,
255269.
Havelaar, A.H., Braunig, J., Christiansen, K., Cornu, M., Hald, T., Mangen, M.J., Molbak, K.,
Pielaat, A., Snary, E., Van Pelt, W., Velthuis, A., Wahlstrom, H., 2007. Towards an
integrated approach in supporting microbiological food safety decisions. Zoonoses
Public Health 54, 103117.
Hoffmann, S., Fischenbeck, P., Krupnick, A., McWilliams, M., 2006. Eliciting information
on uncertainty from heterogeneous expert panels attributing U.S. foodborne
pathogen illness to food consumption. Discussion Paper 06-17-REV, Resources for
the Future. http://ageconsearch.umn.edu/bitstream/123456789/25872/1/
dp060017.pdf. (Accessed 23 January 2008).
Hoffmann, S., Fishbeck, P., Krupnick, A., McWilliams, M., 2008. Informing riskmitigation priorities using uncertainty measures derived from heterogeneous
expert panels: a demonstration using foodborne pathogens. Reliability Engineering
& System Safety 63, 687698.
Hughes, C., Gillepsie, I.A., O'Brien, S.J., the Breakdowns in Food Safety Group, 2007.
Foodborne transmission of infectious intestinal disease in England and Wales,
19922003. Food Control 18, 766772.
ICMSF (International Commission on Microbiological Specications for Foods), 2006.
Use of epidemiologic data to measure the impact of food safety control programs.
Food Control 17, 825837.
Little, C.L., Walsh, S., Hucklesby, L., Surman-Lee, S., Pathak, K., Hall, Y., de Pinna, E.,
Threlfall, E.J., Maund, A., Chan, C.H., 2006. Salmonella contamination in non-UK
produced shell eggs on retail sale in some regions of England. Eurosurveillance 11.
http://www.eurosurveillance.org/ew/2006/061123.asp#4. (Accessed 22 February
2008).
Lynch, M., Painter, J., Woodruff, R., Braden, C., 2006. Surveillance for FoodborneDisease Outbreaks United States, 19982002. Morbidity and Mortality Weekly
Report Centers for Disease Control and Prevention Surveillance Summaries 55,
1 3 4 htt p://ww w.cdc.gov/m mw r/preview/mmwrhtml/s s5 510 a1.h tm .
(Accessed 22 February 2008).
Mumma, G., Grifn, P., Meltzer, M., Braden, C., Tauxe, R.V., 2004. Egg quality assurance
programs and egg-associated Salmonella Enteritidis infections, United States.
Emerging Infectious Diseases 10, 17821789 http://www.cdc.gov/ncidod/EID/
vol10no10/pdfs/04-0189.pdf. (Accessed 22 February 2008).
O'Brien, S.J., Gillepsie, I.A., Sivanesan, M.A., Elson, R., Hughes, C., Adak, G.K., 2006.
Publication bias in foodborne outbreaks of infectious intestinal disease and its
implication for evidence-based food policy. England and Wales 19922003.
Epidemiology and Infection 134, 667674.
OzFoodNet Working Group, 2005. Reported foodborne illness and gastroenteritis in
Australia: annual report of the OzFoodNet network, 2004. Communicable
Diseases Intelligence 29, 165192 http://www.health.gov.au/internet/wcms/
Publishing.nsf/Content/cda-cdi2902-pdf-cnt.htm/$FILE/cdi2902f.pdf. (Accessed
22 February 2008).
OzFoodNet Working Group, 2006. Burden and causes of foodborne disease in
Australia: annual report of the OzFoodNet network, 2005. Communicable
Diseases Intelligence 30, 278300 http://www.health.gov.au/internet/wcms/
Publishing.nsf/
Content/cda-cdi3003b.htm. (Accessed 22 February 2008).
Painter, J., 2006. Estimating attribution of illnesses to food vehicle from reports of
foodborne outbreak investigations. Abstract. Society for Risk Analysis. 2006 Annual
Meeting. Baltimore, MD. December 36, 2006.
Panisello, P.J., Roney, R., Quantick, P.C., Stanwell-Smith, R., 2000. Application of
foodborne disease outbreak data in the development and maintenance of HACCP
systems. International Journal of Food Microbiology 59, 221234.
Patrick, M.E., Adcock, P.M., Gomez, T.M., Altekruse, S.F., Holand, B.H., Tauxe, R.V.,
Swerdlow, D.L., 2004. Salmonella Enteritidis infections, United States, 19851999.
Emerging Infectious Diseases 10, 1710.
Rabsch, W., Tschpe, H., Bumler, A.J., 2001. Non-typhoidal salmonellosis: emerging
problems. Microbes and Infection 3, 237247.

J.D. Greig, A. Ravel / International Journal of Food Microbiology 130 (2009) 7787
Rangel, J.M., Sparling, P.H., Crowe, C., Grifn, P.M., Swerdlow, D.L., 2005. Epidemiology of
Escherichia coli O157:H7 outbreaks, United States, 19822002. Emerging Infectious
Diseases 11, 603609.
Ravel, A., 2008. Analysis of Canadian foodborne outbreak data for food attribution. In
preparation for submission in 2008 (Exact reference to come in the near future).
Ray, B., 2004. Food Microbiology, 3rd ed. CRC Press, Boca Raton, Florida, p. 608.
Reij, M.W., Den Aantrekker, E.D., ILSI Europe Risk Analysis in Microbiology Task Force,
2004. Recontamination as a source of pathogens in processed foods. International
Journal of Food Microbiology 91, 111.

87

Rocourt, J., Moy, G., Vierk, K., Schlundt, J., 2003. Present State of Foodborne Disease in
OECD Countries. Food Safety Department. World Health Organization, Geneva
Switzerland, pp. 139.
Smith De Waal, C., Barlow, K., 2002. Outbreak Alert. Center for Science in the Public
Interest. http://cspinet.org/reports/outbreak_report.pdf. (Accessed 23 January
2008).
Smith Dewaal, C., Hicks, G., Barlow, K., Alderton, L., Vegosen, L., 2006. Foods associated
with food-borne illness outbreaks from 1990 through 2003. Food Protection Trends
26, 466473.