Food Control 33 (2013) 254e261
Contents lists available at SciVerse ScienceDirect
Food Control
journal homepage: www.elsevier.com/locate/foodcont
Risk assessment of intervention strategies for fallen carcasses in beef
slaughter establishments
Vienna R. Brown a, *, Eric D. Ebel b, Michael S. Williams b
a
b
Colorado School of Public Health, Colorado State University, Fort Collins, CO 80526, USA
Office of Public Health Science, Risk Assessment Division, Food Safety and Inspection Service, United States Department of Agriculture, USA
a r t i c l e i n f o
a b s t r a c t
Article history:
Received 16 November 2012
Received in revised form
26 February 2013
Accepted 27 February 2013
In the slaughter establishment, cattle carcasses move along the line attached to a slaughter chain. Cattle
carcasses fall off the slaughter chain infrequently, but such an event results in carcasses potentially
contaminated with bacteria that exist on floors and equipment. Microbes in the feces and ingesta of
slaughtered livestock as well as microbes on the hide surfaces of those livestock contaminate the
slaughter environment. This environment often will include important foodborne pathogens, such as
Escherichia coli O157:H7. This analysis uses a risk assessment modeling approach to assess the potential
public health effects of standardizing treatments for carcasses that fall off the slaughter chain at
dehiding. This assessment examines combinations of six intervention options: 1) water rinse, 2) organic
acid rinse, 3) trim, 4) organic acid rinse and trim, 5) carcass trimming and cook, 6) condemn the carcass.
Potential improvement in public health results from progressive removal of the least effective of these
intervention options. The results of this analysis indicate that the number of annual human E. coli
O157:H7 illnesses avoided varies based on intervention typedorganic acid rinsing (281), carcass trimming (787), organic acid rinsing plus trimming (1533), trimming plus cooking (1539), and carcass
condemnation (1520). The model suggests that the numbers of illnesses prevented are largest and
similar when either the organic acid plus trim, trim plus cook, or condemn interventions are set as the
minimum. This conclusion was robust to sensitivity analysis of various uncertainties in the model.
Interestingly, it was found that a universal condemnation of fallen cattle was not a necessary intervention. Although it was assumed that most large slaughter establishments currently implement a
trimming plus cooking intervention for all fallen carcasses, the model suggests there is little difference
among the three best interventions.
Ó 2013 Elsevier Ltd. All rights reserved.
Keywords:
Escherichia coli O157:H7
Risk Assessment
HACCP
1. Introduction
During the slaughter process, cattle carcasses are shackled by
the rear legs and hung from a ceiling rail. Carcasses are intended to
remain connected to this overhead rail throughout the slaughter
process. At the end of this process, carcasses have been split
longitudinally and are moved into a chilling room to cool prior to
fabrication of the carcass into various beef products.
Carcasses suspended from an overhead rail can; 1) move
through the various slaughter processes rapidly, 2) be kept separate
from other carcasses, 3) afford ready access of slaughter personnel,
and 4) drain fluids efficiently. Hide removal occurs relatively early
in the slaughter process and is the first opportunity for microbial
pathogens to contaminate the underlying tissues.
* Corresponding author.
E-mail address: vrbrown@rams.colostate.edu (V.R. Brown).
0956-7135/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.foodcont.2013.02.037
Hide removal is completed in two phases. First, the hide is detached in the mid-back region, a metal bar is inserted through this
region, and the hide lifted to detach it from the caudal two-thirds of
the carcass. Following this phase, the hide is hanging over the head
and attached across the cranial one-third of the back, as well as the
neck, head, and forelimbs. The second phase uses a hydraulic system, and two conveyer belts running in opposite directions, to
catch the loose end of the hide and pull it off the remainder of the
carcass. Typically, the direction of the pull is toward the head and
away from the attachments of the carcass to the rail.
Given the force of the pulling in the second phase, there is a
chance that a carcass may be pulled off the rail. Although commercial slaughter establishments are engineered to keep carcasses
suspended on the rail while dehiding occurs, the process can fail
infrequently. Factors that increase the likelihood of a fallen carcass
during dehiding include; presence of horn stumps, ineffective
hooks, broken tendons, fast line speeds, and oversized carcasses.
V.R. Brown et al. / Food Control 33 (2013) 254e261
The U.S. Department of Agriculture, Food Safety and Inspection
Services (FSIS) is responsible for monitoring the sanitation and
hygiene of slaughter establishments. The requirement for FSIS
regulated products that fall on the floor is that the product
“.shall be cleaned (including trimming if necessary) or otherwise
handled in a manner approved by the inspector to assure that it
will not be adulterated or misbranded.” (9CFR Chapter 3(a)
318.2). FSIS regulations also address corrective actions, including
reconditioning of product, in 9CFR Chapter 3(e) 416.15. Nevertheless, it is the responsibility of each slaughter establishment to
develop its own standard operating procedure for carcasses that
contact the floor.
A carcass that falls from the rail becomes substantially
contaminated with bacteria that exist on floors and equipment in
the slaughter establishment environment. Microbes in the feces
and ingesta of slaughtered livestock, as well as on the hide surfaces
of those livestock, contaminate the environment; foodborne
pathogens, such as Escherichia coli O157:H7 and Salmonella spp. are
of particular concern. If not effectively removed from fallen carcasses, these pathogens represent foodborne hazards for consumers of beef products.
There is limited information and analysis regarding the topic of
carcasses that fall off the rail. A Best Practices for Beef Slaughter
Guideline, developed by the National Meat Association, Southwest
Meat Association, American Meat Institute, and the National Cattlemen’s Beef Association asserts, “Procedures should be in-place to
recondition any carcasses that fall.” (Harris & Savell, 2003).
Nevertheless, this publication does not provide any further details
or recommendations for managing or decontaminating these carcasses. Gill and Landers (2004) demonstrate that the numbers of
bacteria at visibly contaminated sites were reduced when the sites
were trimmed, such that highly contaminated carcasses that were
trimmed gained similar or superior condition to those carcasses
that were never contaminated.
Castillo, Lucia, Goodson, Savell, and Acuff (1998) report log reductions associated with various treatments to reduce bacteria of
fecal origin on beef carcasses. The most commonly used bacterial
reduction protocols involve water washes, organic acid rinses,
trimming external surfaces, cooking contaminated product, or
some combination of these options. Bacon et al. (2000) quantify the
change in the microbial populations on beef carcasses as they move
through different stages in the slaughter process.
Although carcasses that fall are exposed to substantial
contamination, carcasses that remain on the rail also can be
contaminated. This contamination may occur during dehiding or
evisceration, as well as through contact with contaminated cutting
utensils, contact surfaces, or nonpotable water (Smith, Fazil, &
Lammerding, 2012). Bosilevac et al. (2009) estimated prevalence
and levels of E. coli O157:H7 on hides and carcasses at two points in
the slaughter process in U.S. slaughter establishments and found
significant levels of contamination.
This analysis uses a risk assessment modeling approach to
assess the potential public health effects of proposing standardized treatments for carcasses that fall off the slaughter chain at
dehiding. Risk assessment is useful for examining the implications of different intervention strategies for reducing carcass
contamination and public health risks (Cassin, Lammerding,
Todd, Ross, & McColl, 1998; Ebel et al., 2004; Marks, Coleman,
Lin, & Roberts, 1998; Smith et al., 2012). Combinations of six
intervention options are assessed: 1) potable water rinse, 2)
organic acid rinse, 3) trim, 4) organic acid rinse and trim, 5) trim
and cook, and 6) condemn the carcass. The model output forecasts expected reductions in annual E. coli O157:H7 illnesses that
might result from universal application of the most effective
interventions.
255
2. Methods
This section explains the mathematical model developed to
examine the potential reduction in human illnesses that might
result from consistent performance of highly effective interventions
to treat fallen carcasses. This analysis focuses on E. coli O157:H7
illnesses that result from consumption of beef products that are
contaminated with this pathogen when carcasses fall off the rail at
dehiding. After outlining the mathematical model, this section explains the data and assumptions used to inform this model.
2.1. Model development
The output generated by the risk model is the change in annual
numbers of human E. coli O157:H7 illnesses (Dillnesses) that can be
attributed to the different intervention-scenarios that are applied
to carcasses that fall off the rail at dehiding. The output is generated
using numbers informed by current research and industry standards. The criteria used to compare the output will be a simple
ranking from lowest to highest illnesses avoided relative to a
baseline (status quo) estimate.
Based on arguments outlined in Williams, Ebel, and Vose
(2011a), we model the change in illnesses as proportional to the
change in contamination levels on carcasses between a baseline set
of interventions and an alternative (Alt) set of interventions. The
model for estimating the change in illnesses is
Dillnesses ¼
E½ZAlt E½ZBase lillnesses
E½ZBase where ZBase is the post-dehiding contamination per carcass in the
baseline scenario, ZAlt is the post-dehiding contamination per
carcass in an alternative scenario, and lillnesses is the current annual
rate of E. coli O157:H7 illnesses attributed to beef (i.e., the rate of
illnesses that correspond to the baseline contamination level). The
expectation operator (E[.]) indicates that we are using the expected
value (i.e., mean) of the contamination distribution estimated for
each scenario.
This simple approach to estimating the effect of alternative
policies hinges on two main assumptions: 1) the doses of the
pathogen consumed by humans are generally small and 2) the
doses consumed depend on the levels of the pathogen on carcasses
post-dehiding and the aggregation of effects associated with the
processes of converting carcasses to servings (i.e., partitioning,
mixing, amplification and attenuation), but these components are
independent of one another.1 The first assumption aligns with
previous risk assessments that have estimated average exposure
doses that are generally small (USDA, 2001). A small average dose
generally suggests that exposures occur in the linear part of a
typical doseeresponse function (Williams, Ebel, & Vose, 2011b).2
Because specific pathogen levels per carcass are generally unknown to slaughter establishment personnel, wholesalers, retailers
and consumers; it is reasonable to assume that carcasses and beef
products are handled independently of the levels of pathogens on
the products. Therefore, the second assumption also is reasonable.
We used a Monte Carlo simulation model to estimate ZBase and
ZAlt. The baseline distribution of contamination levels among
1
If D is the variable for dose per serving, Z is the variable describing postdehiding contamination per carcass, Y describes the conversion processes such
that D ¼ Z Y; then the simplification only requires that Z and Y are independent so
that E[D] ¼ E[Z] E[Y].
2
As explained in Williams et al. (2011b), a dose smaller than 7000 CFU is
generally in the “linear” part of the E. coli O157:H7 doseeresponse function.
Furthermore, the FSIS risk assessment of E. coli O157:H7 estimates most exposures
are well below this level (FSIS 2001).
256
V.R. Brown et al. / Food Control 33 (2013) 254e261
carcasses post-dehiding is a mixture of carcasses that do and do not
fall off the line at the dehiding stage. The contamination level on
carcasses that fall off the line also is a mixture of the eventual
contamination levels contributed by carcasses that are subjected to
the various interventions in use by slaughter establishments. Each
slaughter establishment decides the specific intervention strategy
used to recondition fallen carcasses. We start the estimation process by determining contamination levels according to six possible
intervention strategies.
A slaughter establishment class refers to the particular intervention strategy used to recondition fallen carcasses. Therefore, six
slaughter establishment classes characterize the entire cattle
slaughter industry. For a particular slaughter establishment class
that uses intervention s (where s ¼ 1, or 2,., or 6), the contamination level per carcass (Cs) is modeled as;
Cs ¼ ð1 pÞ Cbase þ p ðCbase þ AÞ Ms
Table 1
Summary information for all of model inputs. The parameter name, value, and information source are provided. Additional details are provided in the text.
Variable Variable name
symbol
Value/distribution
Reference
p
0.001; Bernoulli distribution
for each carcass; uncertainty
about p is modeled as
beta(5 þ 1, 5000e5þ1) in
sensitivity analysis
Poisson(lognormal(2.94,
1.07) 32,000)
Personal
communication
w1: water rinse only ¼ 0.025
w2: lactic acid only ¼ 0.015
w3: trim only ¼ 0.075
w4: lactic acid þ trim ¼ 0.075
w5: trim þ cook ¼ 0.80
w6: condemn ¼ 0.010
Fraction of carcass
contaminated if falls, beta
distribution (5.8, 3.87)
CFUs per gram of material
contacting carcass surface
if falls, lognormal (9.6, 1.8)
Adjustment variable for
converting grams to cm2,
gamma distribution (9.72,
94.74)
Pert distributions, see Table 2
for parameters
Kandel 2006
informs the
estimate for ws
Percentage of E. coli
0157 attributable to beef
Distribution;
17,587 to 149,631
Withee et al.,
2009
Scallan et al.,
2011
Cbase
ws
(1)
where p is the probability that a carcass falls off the line, Cbase is a
random variable of contamination levels on carcasses after dehiding (given that those carcasses do not fall off the line), A is the
contamination added to fallen carcasses, and Ms is the proportional
effectiveness of intervention s at removing contamination from
fallen carcasses.
To model the baseline post-dehiding carcass contamination
levels across the entire slaughter industry, we weight each class of
establishment by the fraction of carcasses it produces (ws) among
all carcasses produced by the industry per year
ZBase ¼ w1 C1 þ w2 C2 þ . þ w5 C5 þ w6 C6 :
An alternative scenario is distinguished from the baseline scenario with respect to the weighting of interventions applied across
the industry. To signify this difference, the weighting constants are
assigned a prime so
ZAlt ¼ w01 C1 þ w02 C2 þ . þ w05 C5 þ w06 C6 :
In the following discussion, we explain the data used to inform
the necessary inputs (p, Cbase, ws, A, Ms, lillnesses) to this model.
The model is solved using the R software (R Development Core
Team 2011). The target output to stabilize was Dillnesses; it further
depended on stabilized expected values for ZBase andZAlt. Because
ZBase and ZAlt were simply linear combinations of six random variables, however, any instability in model outputs occurred because
of instability in the simulated distributions for C1 through C6.
Through trial and error, we found that 50 million iterations
generated stable results for Dillnesses. Each iteration represented
an individual carcass that had fallen or not fallen. Fallen carcasses
were randomly treated with each of the possible interventions.
Therefore, the random variables, C1 through C6, were each estimated from 50 million simulated cattle. Each simulation scenario
was repeated three times to further stabilize the conclusions (i.e.,
150 million iterations per result) and illustrate the minor Monte
Carlo error across each simulation of 50 million iterations.
2.2. Data description
Table 1 briefly outlines each variable’s symbol, name, numerical
value or distribution parameters, and data used to inform the input.
2.3. Estimating the probability a carcass falls off the slaughter line
at dehiding, p
The probability that a carcass falls off the overhead rail at
dehiding was estimated to be approximately once per 1,000
A
Ms
lillnesses
Estimating the
probability a
carcass falls
of the slaughter
line at dehiding
Estimating the
contamination
on carcasses that
do not fall
Estimating the
fraction of total
carcasses subject
to different
fallen-carcass
interventions
Estimating the
amount of
contamination
added to carcasses
that fall
Estimating the
effectiveness
of different
fallen-carcass
nterventions
Estimating the
baseline number
of E. coli O157
illnesses per year
Bosilevac et al.,
2009, USDA 2001
Personal
communication
Cernicchiaro
et al., 2011
In-house
experiment
Castillo et al.,
1998
carcasses. We model carcass status as a Bernoulli process and assign
a value of 1 to carcasses that fall and a value of 0 to carcasses that do
not fall. This estimate was based on personal communication with
in-plant inspectors whose experience suggests that approximately
five carcasses fall in a slaughter establishment that harvests just
over 5000 animals daily (Gabel, 2011). Therefore, uncertainty about
this probability was modeled using a beta(5 þ 1, 5000 5þ1)
distribution (Vose, 2008) in the sensitivity analysis of the model.
2.4. Estimating the contamination on carcasses that do not fall,
Cbase
This input estimates the amount of contamination found on
individual carcasses that do not fall at dehiding. Bosilevac et al.
(2009) provided data on concentrations of E. coli O157:H7 on carcasses post-dehiding. The average concentration per carcass (colony forming units per cm2, CFU/cm2) was modeled by fitting a
lognormal(2.94, 1.07) distribution to these data using a maximum
likelihood technique. Uncertainty about the lognormal parameters
was not estimated because no measure of precision was provided
by Bosilevac et al. (2009).
The total carcass surface area of steers and heifers was assumed
as 32,000 cm2 (USDA, 2001). We further assumed that total CFUs
per carcass, Cbase, was the result of a Poisson process based on the
average concentration. In other words,
Cbase ðCFUsÞwPoisson 32; 000cm2 y ;
where y CFU=cm2 w lognormalð2:94; 1:07Þ:
V.R. Brown et al. / Food Control 33 (2013) 254e261
257
2.5. Estimating the amount of contamination added to carcasses
that fall, A
2.6. Estimating the effectiveness of different fallen-carcass
interventions, Ms
The amount of contamination added to a carcass that falls was
estimated by combining an estimate of the proportion of the
surface area contaminated (cm2) with an estimate of the concentration of contamination added to this area (CFU/cm2). The
proportion of the surface area contaminated due to a carcass
falling at dehiding was based on personal observation. This input
was modeled using a beta(5.8, 3.87)distribution such that its
approximate mean and standard deviation were 0.60 and 0.15,
respectively. This description of variability accounts for the
orientation of the fall and the process by which fallen carcasses
are recovered from the floor. Usually, carcasses fall on one side, but
it is sometimes the case that, in the process of retrieving these
carcasses, a larger area of contamination results. Alternatively,
some carcasses may not fall on a side but instead land on their
ventral or dorsal surfaces (possibly resulting in the contaminated
area being less than 50%). The contaminated surface area was
determined by multiplying the total surface area per carcass
(32,000 cm2) by the fraction of the carcass contaminated by the
fall.
The amount of contamination added to the contaminated surface area was developed by assuming the carcass fell onto material
that approximated cattle feces. The average concentration of
contamination (CFUs/g) in the material that contacts the fallen
carcass was assumed to be lognormal(9.6, 1.8)based on data from
Cernicchiaro, Pearl, McEwen, and LeJeune (2011). Nevertheless, a
gram of fecal material was expected to contaminate several square
centimeters of surface area because the force of a fall would
spread the material. We conducted a series of experimental trials
to approximate the degree of spread and enable a conversion of
CFU per gram to CFU per square centimeter.
Using refried beans as a surrogate for fecal material, we
modeled the spread of the surrogate by dropping a nearly 1 kg
textbook from a height of 107 cm (w41 inches) onto 1 g of the
material. The one gram of surrogate was previously placed on a
weighed piece of waxed paper that was then inverted and placed
on graph paper. Once the book was dropped on the surrogate, the
waxed paper was weighed to determine the amount of the surrogate that remained adherent to the graph paper. The surface
area of the graph paper contaminated with the surrogate was
measured by counting the squares (and parts of squares) with
adherent refried bean material. Fifteen such trials were repeated
to estimate the conversion from grams (amount that adhered to
the graph paper) to square centimeters (total surface of graph
paper contaminated) resulting in a mean of 0.103 g/cm2 and a
standard deviation of 0.032 g/cm2. A gamma(9.72,94.74)best fit
these data.
Therefore, the amount of contamination added to a carcass that
falls is modeled as
We assumed six interventions for fallen carcasses were operational in slaughter establishments across the country. The proportional reduction of contamination for each of the different
interventions was modeled as Ms where s ¼ 1 refers to a potable
water rinse, s ¼ 2 refers to a lactic acid rinse, s ¼ 3 refers to trimming of the carcass, s ¼ 4 refers to a lactic acid rinse plus trimming,
s ¼ 5 refers to trimming plus cooking, and s ¼ 6 refers to
condemnation of the carcass.
Previous research provides the mean log reduction resulting from
various carcass interventions (Castillo et al., 1998). Although intervention effectiveness varied somewhat between three regions of the
carcass (i.e., outside round, brisket, and clod) in that study, we used
the log reductions observed for contaminated briskets because these
provided the most conservative estimates of effectiveness. The paper
reported mean log reductions for potable water rinse (2.00 log), trim
(2.80 log), and lactic acid rinse plus trim (5.00 log) interventions.
Lactic acid rinse effectiveness was delineated by subtracting water
rinse only from the potable water rinse plus lactic acid rinse intervention. We confirmed this result by subtracting the trim only from
the trim plus lactic acid rinse intervention. Both calculations suggested that lactic acid rinsing results in a 2.20 log reduction.
Because Castillo et al. (1998) does not provide measures of
variability or uncertainty, we model variability about log effectiveness using Pert (min, mode, max) distributions. We assume the
maximum log reduction is the value reported by Castillo et al.
(1998) for each intervention because the laboratory environment
favors better performance of interventions than might be expected
in a slaughter establishment. The mode and minimum were
assumed as 0.5 logs and 1 log less than the maximum value,
respectively (Table 1).
To model the effectiveness of carcass cooking, the log reduction
is calculated using the following equation (USDA-FSIS 2001):
AðCFUÞwPoisson Surface Area Contaminated cm2
CFU added=cm2 ;
where Surface Area Contaminatedw32; 000 cm2
betað5:8; 3:87Þ;
CFU added=cm2 waðCFU=gÞ b g=cm2 ; and
aw lognormalð9:6; 1:8Þ; bwgammað9:72; 94:74Þ:
LR ¼ 6:6eð20:53e0:12 TempÞ:
The minimum value for cooking of 71 C (160 F) was based on
the minimum USDA recommendation for lethality of E. coli 0157:H7
(USDA 2011). The mode and maximum cooking effectiveness were
estimated using temperatures of 74 C (165 F) and 77 C (170 F,
respectively. Most commercial cooking of beef products intends to
reach temperatures greater than the minimum.
2.7. Estimating the fraction of total carcasses subject to different
fallen-carcass interventions, ws
Currently, slaughter establishments must have a plan for
reconditioning fallen carcasses, but there are no specific requirements for how to complete the reconditioning. There is evidence of between-establishment variability with regard to the
reconditioning process.
Personal experience at two large slaughter establishments indicates that both establishments trim all external surfaces of fallen
carcasses and route these carcasses for cooking only. This intervention is coded s ¼ 5 as outlined above. Current data shows that
four meatpacking companies slaughter 80% of all beef processed in
the United States (Kandel 2006), and both slaughter establishments
visited were among these companies used this intervention.
Therefore, we assume that all establishments operated by these
companies follow the same reconditioning protocol as observed in
the two establishments visited (i.e., w5 ¼ 0.80).
To inform assumptions for the other ws values, the following
factors were considered. First, it is unlikely that many slaughter
258
V.R. Brown et al. / Food Control 33 (2013) 254e261
establishments condemn an entire carcass for pet food following a
fall from the slaughter line during the dehiding process. Second,
establishments that either do not have the capacity to cook on-site
or do not sell their finished product to a cooking industry are likely
to either trim only or provide a lactic acid rinse and then trim all
external surfaces. Third, rinsing with potable water or lactic acid are
more unlikely scenarios as these do not provide a high degree of
decontamination. Nevertheless, we cannot rule these interventions
out because smaller establishments often do not have the design or
capital to use more thorough decontamination processes.
Based on the preceding factors and discussions with FSIS inspectors, we assumed the following ws values; w1 (potable water rinse
only) ¼ 0.025;w2 (lactic acid rinse only) ¼ 0.015; w3 (trim
only) ¼ 0.075; w4 (lactic acid þ trim) ¼ 0.075; w5 (trim þ cook) ¼ 0.80;
and w6 (condemn) ¼ 0.010. These assumptions apply to the baseline
scenario (Fig. 1). Alternative scenarios are explored to examine the
effect of promoting the more effective interventions.
We calculated each scenario by adding the frequency of the least
effective intervention to the frequency of the next least effective
intervention, such that the frequencies moved through each intervention from left to right on Fig. 1. The final alternative being a
mandatory carcass condemnation with frequency of 1.00. For
example, the first alternative is a lactic acid minimum scenario, such
that the frequency of potable water rinsing (the least effective intervention) is added to the frequency of acid rinsing (the next least
effective decontamination intervention). Therefore, w01 ¼ 0 and
w02 ¼ 0:04, while the remaining w0s values are the same as those in the
baseline scenario. Four other alternative scenarios are defined by the
least effective intervention still used (i.e., trim minimum, acid þ trim
minimum, trim þ cook minimum, and condemn minimum).
Given the limited evidence on which to base the values for ws,
the influence of these inputs on the analytic results is assessed via
sensitivity analysis.
2.8. Estimating the baseline number of E. coli O157 illnesses per
year, lillnesses
Scallan et al. (2011) estimate that 63,153 E. coli O157:H7 human
illnesses occur in the United States each year. The 90% credible
interval for this estimate is 17,587 to 149,631 cases per year. This
estimate pertains to all foodborne sources. Withee et al. (2009)
estimate that 33% of all E. coli O157:H7 illnesses are attributed to
consumption of beef. Therefore, we estimate lillnesses ¼ 3153 x
0.33 ¼ 20,840.
2.9. Sensitivity analysis
The results of this analysis depend on the assumptions used for
model inputs. Nevertheless, many of these inputs are based on
limited data or judgments. To explore the influence of alternative
input values, we generated results following changes to the
following model inputs:
p, the probability of a carcass falling
ws, the weighting fractions for the various intervention
strategies
Ms, the effectiveness of various intervention strategies
Other uncertain inputs were not examined at this time. For
example, the annual number of E. coli O157 illnesses and the beef
attribution fraction values that contribute to an estimate of lillnesses
are uncertain. Nevertheless, the effect of uncertainty in this case is
strictly linear, so a proportional increase or decrease in these values
will generate the same proportional change in the model’s estimates. The assumptions that contribute to fallen carcass (A)
contamination models are numerous and non-trivial. Nevertheless,
obvious alternative assumptions are not evident, and the effects of
these assumptions were not pursued at this time.
To assess the uncertainty, upper bound and lower bound scenarios were run individually after adjusting the input values to
determine the greatest number and smallest number of avoided
illnesses. All sensitivity scenarios were run with 30 million iterations.
3. Results
The mean and standard deviation of the number of E. coli O157
organisms per carcass after dehiding (given that carcasses do not fall
off the rail at dehiding), Cbase, are 2995 and 4,359, respectively. This
is a right-skewed distribution of O157 organisms per carcass with a
median (1697), 95th percentile (9793), and 99th percentile (20,196).
Intermediate model outputs include the distributions of
contamination levels on post-dehiding carcasses for each of the
individual intervention scenarios (i.e., C1,...,C6), as well as the
amount of contamination added to fallen carcasses (A). Accounting
for fallen carcasses during dehiding changes the distribution of
contamination per carcass depending on the particular intervention used (Table 3). Rinsing fallen carcasses with potable water
limits the average number of E. coli O157:H7 organisms on all carcasses to 7886 following dehiding. In this case, the standard deviation is 967,073. The use of acid rinse or trimming reduces both the
mean and standard deviation of these organisms per carcass
somewhat, but the combination of these two interventions essentially reduces the contamination toCbase levels.
Given the moments of the inputted distributions, it is possible to
use Eqn. (1) to approximate the results in Table 3. For example,
E C1;Water ¼ ð1 pÞ E½CBase þ p fE½CBase þ E½Ag E M1;Water
o
n
¼ ð0:999Þ ð2995Þ þ ð0:001Þ 2995 þ 1:4 108
3:5 102 z7892
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Std Dev C1;Water ¼ Var C1;Water
Var C1;Water ¼ Var½ð1 pÞ CBase þ Var½p H
Fig. 1. Baseline frequencies of the six interventions are shown.
H ¼ Var ðCBase þ AÞ M1;Water
¼ ðE½CBase þ E½AÞ2 Var M1;Water
2
þ E M1;Water ðVar½CBase þ Var½AÞ
þ ðVar½CBase þ Var½AÞ Var M1;Water
V.R. Brown et al. / Food Control 33 (2013) 254e261
Table 2
The assumed Pert distribution parameters for log reductions associated with each
intervention, as well as decimal reductions,a are shown.
Intervention
Water
Acid
Trim
Acid þ Trim
Cook
Trim þ Cook
a
Minimum
Mode
Maximum
1.00
1.50
2.00
1.20
1.70
2.20
1.80
2.30
2.80
4.00
4.50
5.00
5.27
5.87
6.47
Sum of trim and cook effects
Mean,
Standard deviation,
decimal
decimal reduction
reduction
3.5E-02
2.2E-02
7.5E-03
3.5E-05
1.5E-06
1.2E-08
1.5E-02
9.6E-03
5.4E-03
1.5E-05
8.2E-07
1.1E-08
259
Table 4
Model results for illnesses avoided based on five alternative scenarios are shown.
Each trial represents an estimate based on 50 million Monte Carlo iterations. The
average value is across the three trials.
Alternative scenarios
Trial 1
Trial 2
Trial 3
Average
Acid minimum
Trim minimum
Acid þ trim minimum
Trim þ cook minimum
Condemn minimum
277
795
1535
1541
1522
284
785
1551
1575
1538
282
782
1512
1518
1499
281
787
1533
1539
1520
eliminates lactic acid and potable water rinse interventions and
requires a trimming or more effective intervention (i.e.,
Decimal reductions are calculated as10LogReduction.
so Std Dev C1;Water z877; 000
For comparison, these calculations are essentially equivalent to
the Monte Carlo results for the other interventions.
The substantial and highly variable contamination added to
carcasses that fall at dehiding contributes to substantial variability
in contamination levels across all carcasses for less effective interventions (e.g., potable water, lactic acid rinse, or trimming). In
contrast, the variability was greatly reduced once the effectiveness
of the intervention is at least that inputted for the acid þ trim
scenario (i.e., a modal 4.5 log reduction).
The results in Table 3 convey the potential increased risk associated with any of the three least effective interventions. As long as
fallen carcasses are subjected to an acid wash plus trimming,
trimming plus cooking, or condemned outright, the apparent risk of
carcasses post-dehidingdas gauged by carcass contamination
levelsdis roughly equivalent to a situation where no fallen carcasses occur (i.e., Cbase).
Given an assumed baseline mixture of these interventions
across the beef slaughter industry, potential reductions in E. coli
O157:H7 human illnesses attributed to beef consumption accrue as
the less effective interventions are eliminated from practice across
the industry (Table 4).
Scenario 1
Potable water rinsing is the least effective decontamination
method; thus, the frequency of potable water was combined with
the acid only frequency. The model was run to determine the
number of illnesses prevented (i.e., w1 ¼ 0; w2 ¼ 0.04; w3 ¼ 0.075;
w40.075; w5 ¼ 0.80; and w6 ¼ 0.010). Our model estimates that
replacing the water wash intervention with a lactic acid rinse
industry-wide would prevent 281 illnesses annually attributable to
E. coli 0157:H7 in beef products.
Scenario 2
Lactic acid rinsing also is a minimally effective decontamination
method for heavily contaminated carcasses. This scenario
Table 3
Summary statistics of Monte Carlo simulations for post-dehiding numbers of E. coli
O157:H7 per carcass across different interventions applied to carcasses that fall off
the rail during dehiding.
w1 ¼ 0; w2 ¼ 0.0; w3 ¼ 0.115; w40.075; w5 ¼ 0.80; and
w6 ¼ 0.01). This trim minimum scenario is projected to prevent 787
illnesses annually that are attributable to E. coli 0157 in beef
products.
Scenarios 3e5
Requiring a minimum of lactic acid rinsing plus trimming, trimming plus cooking, or condemnation of carcasses that fall during
dehiding is projected to prevent more than 1500 illnesses annually
attributable to E. coli 0157:H7 in beef products.
The largest projected number of illnesses prevented (1539) is
associated with the combined “trim and cook” minimum scenario
in which w5 ¼ 0.99 and w6 ¼ 0.01. Fewer illnesses are prevented
(1520) if all fallen carcasses are condemned (i.e.,w6 ¼ 1.0). The fact
that trimming and cooking lowers risk below condemnation alone
(i.e., completely removing the carcass equating to zero risk from
that carcass) suggests that trimming and cooking decreases risk
below the non-fallen carcass baseline. This is because this intervention is often effective enough to eliminate, on average, all the
contamination added to the carcass following its fall as well as
some of the background contamination it acquired before the fall.
Table 3 illustrates that the average contamination for Ctrimþcook
(2992 E. coli O157:H7 per carcass) is less than the average for Cbase
and Ccondemn. Nevertheless, these findings assume that carcass interventions after dehiding are independent of the fallen carcass
intervention strategy employed such that the differences in
contamination levels translate proportionately to risk per serving.
3.1. Sensitivity analysis results
3.1.1. p, the probability of a carcass falling
Uncertainty about pis defined by a beta(5 þ 1, 5000e5þ1)
distribution with 5th and 95th percentiles of 0.0005 and 0.002,
respectively. We use these percentile values to generate upper and
lower bound scenarios (Table 5). Results of these alternative specifications of pgenerally imply that a doubling of this input will
result in approximately twice as many illnesses avoided, whereas
halving of this input will result in one-half as many illnesses
avoided.(Tables 6 and 7)
Table 5
Comparing illnesses avoided with alternative p input values.
Interventions
Mean
Standard deviation
Intervention
Lower bound scenarioa
Upper bound scenariob
Cwater
Cacid
Ctrim
Cacid&trim
Ctrim&cook
Ccondem
A, contamination added
7886
6121
4094
2997
2992
2995
1.4Eþ08
967,073
606,516
292,125
4566
4358
4359
7.2Eþ08
Acid minimum
Trim minimum
Acid, trim minimum
Trim, cook minimum
Condemn minimum
148
412
799
802
792
550
1517
2917
2928
2892
a
b
Lower bound scenario is defined by a p value of 0.0005.
Upper bound scenario is defined by a p value of 0.002.
260
V.R. Brown et al. / Food Control 33 (2013) 254e261
Table 6
Comparing illnesses avoided with alternative ws variables.
Intervention
Lower bound scenarioa
Upper bound scenariob
Acid minimum
Trim minimum
Acid, trim minimum
Trim, cook minimum
Condemn minimum
149
3912
755
757
737
1324
4464
6914
6929
6918
a
Lower bound scenario is bound by interventions (water, acid, trim, acid/trim,
trim/cook, and condemn) weighted at 0.0125, 0.0075, 0.0375, 0.0375, 0.90, and
0.005, respectively.
b
Upper bound scenario is bound by interventions equally weighted at 0.1667.
3.1.2. ws, the weighting fractions for the various intervention
strategies
We develop an upper bound scenario by assuming each intervention (potable water, acid, trim, acid þ trim, trim þ cook, and
condemn) is weighed equally across the industry. In the lower
bound scenario, we increase the fraction of production using the
trim þ cook intervention from 80% to 90% such that the weights of
the other interventions (potable water, acid, trim, acid þ trim, and
condemn) are adjusted proportionally (0.0125, 0.0075, 0.0375,
0.0375, and 0.005, respectively) to account for the remaining 10% of
total production.
Results of these alternative specifications of ws suggest, in the
upper bound, that illnesses avoided would increase 4.5 fold relative
to the default settings. In the lower bound scenario, illnesses are
approximately one-half those estimated using the default settings.
The substantial changes in illnesses avoided result from the
substantial changes in the initial mixture of interventions. For
example, the upper bound scenario assumes that 17% of fallen
carcasses are treated with the trim þ cook intervention before an
alternative mixture of interventions is implemented. Because the
alternative scenarios are changed to require effective minimum
interventions, the upper bound scenario provides more room for
improvement within the slaughter industry than the default scenario with respect to the ws inputs. The lower bound effect limits
the room for improvement within the industry such that avoidable
illnesses are reduced relative to the default settings.
The lower bound scenario is more plausible, because it is more
likely that we underestimate the fraction of production currently
subjected to the trim þ cook intervention by 10% than we have
over-estimated it by more than 60%. These results suggest that our
baseline estimates may be too large.
3.1.3. Ms, the effectiveness of various intervention strategies
The upper bound scenario reduces the efficacy of each intervention. Originally, the values reported in Castillo et al. (1998) were
conservatively set as the maximum value for the purposes of this
model to compensate for differing conditions between the
Table 7
Comparing illnesses avoided with alternative Ms values.
Intervention
Lower bound scenarioa
Upper bound scenariob
Acid minimum
Trim minimum
Acid, trim minimum
Trim, cook minimum
Condemn minimum
88
282
457
459
439
1038
2936
4792
4812
4796
a
Adjusts the efficacy of each intervention by assuming the Castillo et al. (1998)
values represent the modes of Pert distributions while the minimum and
maximum Pert parameters are derived by subtracting and adding one-half logs to
the mode, respectively.
b
Adjusts the efficacy of each intervention by deriving the mode parameter of the
Pert distribution as one log less than the maximum (established by the Castillo et al.
(1998) values) and deriving the minimum parameter of the Pert by subtracting two
logs from the maximum.
laboratory and the slaughterhouse. The mode and minimum values
were derived by reducing the efficacy of the treatment by a half log
from maximum to mode and another half log from mode to minimum, resulting in the maximum and minimum differing by one full
log. For the upper bound scenario, the maximum value remains
unchanged, but one log reductions are used to derive the mode and
minimum parameters for each Pert distribution. Nevertheless, a
minimum effectiveness of 0 logs was assumed if an adjustment
implied a negative value.
The lower bound scenario increases the efficacy of each intervention. In this scenario, the modes of the Pert distributions are the
values reported in Castillo et al. (1998). Maximum and minimum
Pert parameters are derived by adding and subtracting one-half
logs, respectively, from the mode.
The results of the upper bound scenario suggest that illnesses
avoided would increase roughly three-fold relative to the default
settings. In the lower bound scenario, illnesses avoided are roughly
one-third those estimated using the default settings.
The explanation of the changes observed in this analysis is
similar to the explanation for the ws inputs. Essentially, the upper
bound scenario resets the industry so that its baseline is less
effective in reducing contamination. As the alternative scenarios
are considered, however, there is more room for improvement as
the minimum allowed intervention becomes more effective. The
reverse effect applies to the lower bound scenario (i.e., less room for
improvement).
The increased variability of the upper bound scenario seems
more plausible than the increased effectiveness of the lower bound
scenario. Therefore, the effect of uncertainty about this input seems
to suggest that our baseline estimates may be too small.
4. Discussion
Approximately 35,000,000 cattle are slaughtered annually in
the United States; thus, we might estimate that nearly 35,000
carcasses fall off the rail during dehiding each year. Currently, there
is no specific requirement for how slaughter establishments should
recondition these carcasses. This risk assessment examines the
potential public health effects that might result from eliminating
less effective reconditioning interventions.
Depending on the share of beef production that might be
affected, this analysis suggests that eliminating the less effective
interventions for reconditioning fallen carcasses could improve the
safety of the beef supply. The results in Table 4 suggest that the
number of E. coli O157:H7 illnesses avoided doubles between the
trim minimum and acid þ trim minimum alternatives. Once a
minimum of acid þ trim has been met, the improvement in illnesses avoided is small.
Our results suggest there is limited improvement in illnesses
avoided from diverting fallen carcasses to cooking relative to
acid þ trimming. From our default inputs, the difference between
the acid þ trim minimum and the trim þ cook minimum is six
illnesses prevented. Cooking the entire carcass is both costly to
perform and is a lost opportunity to market products in the raw
form. When a carcass is diverted for cooking, it is ground and sold
to an industry that manufactures prepared meals; thus, the most
valuable cuts are ground. Condemning carcasses is not only an
expensive intervention, but also is responsible for fewer avoided
illnesses than both the acid þ trim minimum and trim þ cook
minimum alternatives.
Sensitivity analysis suggests that the true number of illnesses
avoided for the various alternative scenarios is uncertain. The
magnitudes of illnesses avoided may range from one-third to more
than four-times larger than our default predictions. The most
plausible sensitivity scenarios mostly balance each other such that
V.R. Brown et al. / Food Control 33 (2013) 254e261
our uncertainty may only range from one-half to three-times the
default predictions. Nevertheless, the conclusion about a minimum
intervention of acid þ trim seems to be robust to the model input
assumptions. In all of the uncertainty scenarios considered here,
the difference in illnesses avoided was generally substantial between the trim minimum and trim þ acid minimum scenarios
while this difference was trivial between trim þ acid, trim þ cook,
and condemn minimum scenarios.
This modeling approach is a simplification of reality with many
assumptions. Hence, additional uncertainty is associated with our
estimates that should be acknowledged. Nevertheless, the model
and its assumptions are reasonable approximations of reality and
based on the best available information. Although our findings are
meaningful, additional economic analysis will need to consider the
relative costs of the various interventions e as well as the possible
additional benefits from control of other pathogens (e.g., Salmonella
and other E. coli STEC bacteria) e before decision-makers might act
on these findings.
The outcome of the model indicates that acid þ trim is a
reasonable reconditioning intervention for carcasses that fall off the
overhead rail during dehiding that achieves nearly all the possible
benefits at less cost than more aggressive interventions. If all establishments considered this the minimum response to carcasses
that fall off the rail at dehiding, our analysis suggests a safer, more
wholesome beef supply would result.
References
Bacon, R., Belk, K., Sofos, J., Clayton, R., Reagan, J., & Smith, G. (2000). Microbial
populations on animal hides and beef carcasses at different stages of slaughter
in plants employing multiple-sequential interventions for decontamination.
Journal of Food Protection, 63(8), 1080e1086.
Bosilevac, J. M., Arthur, T. M., Bono, J. L., Brichta-Harhay, D. M., Kalchayanand, N.,
King, D. A., et al. (2009). Prevalence and enumeration of Escherichia Coli O157:
H7 and Salmonella in U.S. Abottoirs that process fewer than 1,000 head of cattle
per day. Journal of Food Protection, 72(6), 1272e1278.
Cassin, M. H., Lammerding, A. M., Todd, E. D., Ross, W., & McColl, R. S. (1998).
Quantitative risk assessment for Escherichia coli O157:H7 in ground beef hamburgers. International Journal of Food Microbiology, 41(1), 21e44.
261
Castillo, A., Lucia, L. M., Goodson, K. J., Savell, J. W., & Acuff, G. R. (1998). Comparison
of water wash, trimming, and combined hot water and lactic acid treatments
for reducing bacteria of fecal origin on beef carcasses. Journal of Food Protection,
61, 823e828.
Cernicchiaro, N., Pearl, D. L., McEwen, S. A., & LeJeune, J. T. (2011). Assessment of
diagnostic tools for identifying cattle shedding and super-shedding Escherichia
coli O157:H7 in a longitudinal study of naturally infected feedlot steers in Ohio.
Foodborne Pathogens, 8, 239e248.
Ebel, E., Schlosser, W., Krause, J., Orloski, K., Roberts, T., Narrod, C., et al. (2004). Draft
risk assessment of the public health impact of Escherichia coli O157:H7 in
ground beef. Journal of Food Protection, 67, 1991e1999.
Gabel, G. August 8, 2011. Veterinarian, FSIS. JBS Lamb Plant. Personal
communication.
Gill, C., & Landers, C. (2004). Microbiological conditions of detained beef carcasses
before and after removal of visible contamination. Meat Science, 66, 335e342.
Harris, K. B., & Savell, J. W. (2003, November). Best practices for beef slaughter. National Meat Association, Southwest Meat Association, American Meat Institute,
National Cattlemen’s Beef Association. Accessed February, 2012.
Kandel, W. (2006, June). AmberWaves e June 2006-meat-processing firms attract
hispanic workers to rural America. USDA Economic Research Service. Home Page.
Retrieved February, 2012, from. http://www.ers.usda.gov/AmberWaves/June06/
Features/MeatProcessing.htm.
Marks, H. M., Coleman, M. E., Lin, C. J., & Roberts, T. (1998). Topics in microbial risk
assessment: dynamic flow tree process. Risk Analysis, 18(3), 309e328.
R Development Core Team. (2011). R: A language and environment for statistical
computing(Online). Available at http://www.R-project.org.
Scallan, E., Hoekstra, R. M., Angulo, F. J., Tauxe, R. V., Widdowson, M. A., Roy, S. L.,
et al. (2011). Foodborne illness acquired in the United Statesemajor pathogens.
Emerging Infectious Disease, 17, 7e15.
Smith, B. A., Fazil, A., & Lammerding, A. M. (2012). A risk assessment model for
Escherichia coli O157:H7 in ground beef and beef cuts in Canada: evaluating the
effects of interventions. Food Control.
USDA. Food Safety and Inspection Service Home. Retrieved November 11, 2011,
from. http://www.fsis.usda.gov/News_.
USDA. (Sep 2001). Food Safety and Inspection Service. Risk assessment of E. coli O157:
H7 in ground beef. Accessed at. http://www.fsis.usda.gov/OPPDE/rdad/FRPubs/
00-023N/00-023NReport.pdf.
Vose, D. (2008). Risk analysis: A quantitative guide. Chichester, England: John Wiley
& Sons.
Williams, M., Ebel, E., & Vose, D. (2011a). Framework for microbial food-safety risk
assessments amenable to Bayesian modeling. Risk Analysis, 31(4), 548e565.
Williams, M. S., Ebel, E. D., & Vose, D. (2011b). Methodology for determining the
appropriateness of a linear dose-response function. Risk Analysis, 31(3), 345e
350.
Withee, J., Williams, M., Disney, T., Schlosser, W., Bauer, N., & Ebel, E. (2009).
Streamlined analysis for evaluating the use of preharvest interventions intended to prevent Escherichia coli 0157:H7 illness in humans. Foodborne Pathogens
and Disease, 6(7), 817e825.