Salmonella spp. in lymph nodes of fed and cull cattle: Relative assessment of risk to ground beef

Food Control 50 (2015) 423e434

Contents lists available at ScienceDirect

Food Control journal homepage: www.elsevier.com/locate/foodcont

Salmonella spp. in lymph nodes of fed and cull cattle: Relative assessment of risk to ground beef Min Li a, *, Sasidhar Malladi b, H. Scott Hurd a, 1, Timothy J. Goldsmith b, Dayna M. Brichta-Harhay c, Guy H. Loneragan d a

Department of Production Animal Medicine, College of Veterinary Medicine, Iowa State University, Ames, IA 50011, USA College of Veterinary Medicine, Center for Animal Health and Food Safety, University of Minnesota, St. Paul, MN 55108, USA U.S. Department of Agriculture, Agricultural Research Service, U.S. Meat Animal Research Center, Clay Center, NE 68933, USA d Department of Animal and Food Sciences, Texas Tech University, Lubbock, TX 79409, USA b c

a r t i c l e i n f o

a b s t r a c t

Article history: Received 3 April 2014 Received in revised form 8 September 2014 Accepted 9 September 2014 Available online 19 September 2014

Ground beef products have been implicated as the vehicle for the transmission of Salmonella in a number of outbreaks. Although carcass surface interventions have proven effective, Salmonella contamination in ground beef still occurs. Recent studies indicate that deep tissue lymph nodes (DTLNs) may be an important source of Salmonella contamination in ground beef products. We developed a stochastic simulation model covering the pre-to post-harvest stages to assess the relative contribution of DTLN as compared with carcass surface, to Salmonella in ground beef, and the impact of various pre- and postharvest interventions. The model addressed fed and cull cattle, and in high and low prevalence seasons. Contamination from carcass surfaces and DTLNs was simulated separately. Linear relationships were used to describe the changes of Salmonella surface concentration and prevalence at different processing stages. The baseline results indicate that on average over 90% of the Salmonella CFU load in a 2000 lb (907 kg) production lot originates from DTLN contamination as compared with carcass surface contamination. The relative contribution of DTLN contamination was fairly robust to changes in model parameters for ground beef from fed cattle, while it was comparatively more sensitive to changes in model parameters for cull cattle. The predicted mean Salmonella CFU load from DTLN contamination was considerably greater in ground beef production lots from fed cattle compared with cull cattle. Correspondingly, our scenario analysis suggested that generic pre-harvest interventions which can reduce Salmonella contamination in DTLNs would cause a greater total CFU load reduction in ground beef production lots from fed cattle compared with cull cattle. The study provides some valuable information for prioritizing control measures targeted at Salmonella contamination from the beef carcass surface or DTLNs based on the current knowledge. © 2014 Elsevier Ltd. All rights reserved.

Keywords: Salmonella Risk assessment Ground beef Lymph node Carcass surface Cattle

1. Introduction Ground beef products have been implicated in a number of salmonellosis outbreaks (Centers for Disease Control and Prevention [CDC], 2011, 2012, 2013). Cattle hide contamination is primarily responsible for the contamination of carcass surfaces, which may

* Corresponding author. Current address: S3-110 Schurman Hall, Department of Population Medicine and Diagnostic Sciences, College of Veterinary Medicine, Cornell University, Ithaca, NY 14853, USA. Tel.: þ1 607 253 4086; fax: þ1 607 253 3440. E-mail addresses: [email protected], [email protected] (M. Li). 1 Deceased. http://dx.doi.org/10.1016/j.foodcont.2014.09.011 0956-7135/© 2014 Elsevier Ltd. All rights reserved.

result in contaminated ground beef if the pathogens survive carcass interventions during processing (Arthur, Bosilevac, et al., 2008; Barkocy-Gallagher et al., 2003; Brichta-Harhay et al., 2008). Previous investigations indicate a fairly low level of Salmonella prevalence (on average less than 1%) on post-intervention carcasses due to the application of various carcass interventions in commercial beef processing plants (Brichta-Harhay et al., 2008; RiveraBetancourt et al., 2004). Despite the success of carcass interventions, however, testing results from the United States Department of Agriculture, Food Safety Inspection Service (USDA/ FSIS) indicate little reduction of Salmonella contamination in ground beef during the past decade, with generally over 2.0% prevalence in ground beef samples (25 g) during the past years

424

M. Li et al. / Food Control 50 (2015) 423e434

(USDA/FSIS, 2012). These discrepancies imply other potential sources for ground beef contamination, among which lymph nodes have received increasing attentions in recent years. A growing number of studies indicate that Salmonella may be harbored in bovine deep tissue lymph nodes (DTLNs), such as subiliac, iliofemoral, and superficial cervical LNs, which are embedded in adipose tissues that are frequently included in ground beef and thus could potentially contaminate the final product (Arthur, Brichta-Harhay, et al., 2008; Brichta-Harhay et al., 2012; Gragg, Loneragan, Brashears, et al., 2013, Gragg, Loneragan, Nightingale, 2013; Haneklaus et al., 2012; Koohmaraie et al., 2012). A two-year survey of Salmonella contamination in subiliac lymph nodes (n ¼ 3327) of cull and fed cattle across the nation indicated an average prevalence of 7.5%, and Salmonella harborage appeared to be affected by season, region, and animal type (Gragg, Loneragan, Brashears, et al., 2013). Since Salmonella harbored in DTLNs are likely to evade the antimicrobial carcass surface interventions currently implemented in the processing plants, other potential pre- and post-harvest interventions, such as vaccination, probiotics, lymph node removal, and final product treatment, could be explored to reduce or eliminate Salmonella originating from DTLNs. To harness limited resources to reduce the Salmonella burden, it is critical for the beef industry to employ a science-based, riskinformed approach to choose the most effective intervention strategies among a variety of control options. Therefore, the objective of this study was to develop a stochastic simulation model to assess the relative contribution of DTLNs to Salmonella contamination of ground beef products, and exercise the model with best available data to evaluate the relative impact of various pre- and postharvest intervention strategies. The following outcomes were evaluated: (i) relative contribution of DTLN contamination to the Salmonella CFU load in a 2000 lb (907 kg) production lot; (ii) total Salmonella CFU load in a production lot. The model could aid producers and beef packers in making informed decisions regarding the choice of effective control measures, and direct future research needs, especially for gathering appropriate data to improve the predictive ability of the model. 2. Materials and methods

Table 1 Scope of the risk assessment model. 1. The risk assessment model is designed for a typical high volume (16,000 production lots per year) beef processing plant that slaughters cattle, produces meat trims, and finally grinds the trims to ground beef. 2. It addresses fed cattle (steers and heifers) and cull cattle (cows and bulls) separately, based on their respective contamination data on hides and in DTLNs. 3. Generic, non-specific serovars of Salmonella enterica are considered. 4. Seasonal variations are modeled based on “high” (summer and fall) and “low” (winter and spring) prevalence times. 5. Plant data across different geographical locations are aggregated based on sample sizes and expressed as average values. 6. The surface contamination model starts from hide contamination at abattoir and ends at contamination in ground beef after grinding. 7. Internal model parameter estimates are based on averages from multiple published studies. 8. Data (publicly available) were collected from studies conducted for U.S. beef production and processing. 9. The prevalence and concentration (of positive samples) of Salmonella on carcasses and ground beef are estimated on a production lot basis, which is defined as a 2000 lb (907 kg) load of ground beef. 10. The risk reduction effect of post-grind interventions is not evaluated in the model. 11. No exported trims or trims from other packing plants are considered.

hide or airborne particulates created during the process. A linear transfer coefficient (P2) was assumed to describe this process and relate the Salmonella prevalence from 1000 cm2 hides to the prevalence on 8000 cm2 pre-evisceration carcasses (P3) (hide and carcass sampling areas in most data were 1000 and 8000 cm2, respectively). Similar linear regression models have been used previously to simulate the transfer of Escherichia coli O157:H7 prevalence from feces to pre-evisceration carcasses (Hurd & Malladi, 2012; USDA/FSIS, 2001). Studies that used tagged and matched hide and carcass samples were chosen to estimate the linear transfer coefficient (Arthur, Bosilevac, et al., 2008; BarkocyGallagher et al., 2003; Brichta-Harhay et al., 2007, 2008; RiveraBetancourt et al., 2004). Uncertainty associated with this parameter was estimated by linear regression parameter bootstrapping. To be more specific, the hide and carcass prevalence (denoted as independent and dependent variables, respectively) from

2.1. Model development A stochastic spreadsheet based simulation model covering the pre-harvest to post-harvest stages was utilized to predict the relative contribution of Salmonella from DTLNs. The model scope and key assumptions are summarized in Tables 1 and 2. The model is comprised of three modules as shown in Figs. 1e3: (1) Salmonella prevalence on carcass surfaces; (2) Salmonella concentration on carcass surfaces; and (3) Salmonella prevalence and concentration in DTLNs. The parameters and calculations for the three modules are listed in Table 3. 2.1.1. Modeling Salmonella prevalence on carcass surfaces The carcass surface prevalence module starts with post-stun hide prevalence (Fig. 1). The transfer of Salmonella from live animals' feces to hide surfaces was not attempted because previous studies have showed that the transport and lairage stages after harvest and before slaughtering have a commingling effect which can increase the magnitude of contamination and mess up the genetic association between pre- and post-harvest bacterial isolates (Arthur et al., 2007, Arthur, Bosilevac, et al., 2008). Hide prevalence (P1) for fed and cull cattle, in high and low prevalence seasons, was estimated from multiple studies across different geographical regions (Table 3). During dehiding, Salmonella can be transferred to the carcass surface through direct contact with the

Table 2 Key assumptions of the risk assessment model. 1. Contamination of a production lot is independent of previous lots processed, i.e., lot-to-lot contamination is negligible. 2. Pre-evisceration carcass prevalence is a linear function of in-plant hide prevalence. 3. No mixing of ground beef from cows and bulls with that from steers and heifers is assumed. 4. Simple linear fits were assumed to be appropriate for relating Salmonella prevalence and concentration at various production and processing stages. 5. Salmonella is assumed to be randomly distributed in combo bins of ground beef (although clusters of bacteria may exist) and modeled with Poisson distribution. 6. All contamination is assumed to be on the external surface of the carcass and in DTLNs. 7. The Salmonella prevalence and concentration on all 8000 cm2 segments of carcass surface is assumed to be similar to the values estimated from the 8000 cm2 sampling area in the Brichta-Harhay et al. (2007, 2008) studies. 8. The Salmonella prevalence and concentration on all 1000 cm2 segments of hide surface is assumed to be similar to the values estimated from 1000 cm2 sampling area in the Brichta-Harhay et al. (2007, 2008) studies. 9. External carcass contamination is assumed to come mostly from direct contact with contaminated hide during dehiding. 10. Internal carcass contamination from GI (gastro-intestinal) content was not considered in the model. 11. During fabricating, trimming, and grinding, cross-contamination from hands, equipment, and contaminated carcasses or trims is assumed not to differentially add new bacteria into the grinding load.

M. Li et al. / Food Control 50 (2015) 423e434

425

Fig. 3. Model pathway of the Salmonella DTLN contamination module.

Fig. 1. Model pathway of the carcass surface prevalence module.

individual studies were first described by Binomial (n, p)/n distributions, respectively, where n is the sample size, p is the percent of positive samples, and Binomial (n, p) is the number of positive samples. Using the bootstrapping technique, both of the hide and carcass prevalence distributions from all studies were resampled together in pairs for a number of 1000 times. Each paired bootstrap replicates in one iteration were used to calculate a linear regression coefficient and hence 1000 linear coefficients were calculated after the iterations, which were then best fitted with a Normal (0.44, 0.045) distribution to represent the uncertainty of the linear transfer coefficient. Following hide removal, carcasses are subject to multiple carcass intervention steps to minimize bacterial contamination on the surface. Because of data limitations, specific intervention steps were modeled as generic interventions and the impact on prevalence reduction (P4) was evaluated using a linear model linking the

change of prevalence from pre-evisceration to post-intervention carcasses on 8000 cm2 external carcass surface area segments. The number of contaminated 8000 cm2 external surface area segments (P11) from cattle contributing ground beef for a production lot was simulated with a Binomial distribution with parameters as, the number of 8000 cm2 carcass segments added to a production lot (P10), and the Salmonella prevalence on 8000 cm2 postintervention carcass segments (P5). 2.1.2. Modeling Salmonella concentration on carcass surfaces The carcass surface concentration module starts with Salmonella concentration on pre-evisceration carcasses (Fig. 2). The transfer of bacterial concentration from hide to carcass was not included because the linear regression fit linking hide and carcass concentrations seemed to underestimate the Salmonella pre-evisceration carcass surface concentration compared with observed data (Bosilevac et al., 2009b; Brichta-Harhay et al., 2008). The Salmonella concentration on pre-evisceration carcasses (C1) was fitted with an empirical cumulative distribution based on data obtained from

Fig. 2. Model pathway of the carcass surface concentration module.

426

Table 3 Summary of parameter estimates for modeling Salmonella contamination on carcass surfaces and in DTLNs and baseline model simulation results. Cattle-season

Data source

Distribution/@RISK or Excel formula

Estimated valuea

Salmonella prevalence on carcass surfaces Salmonella prevalence on 1000 cm2hide samples at slaughter (P1)

Fed-High

Weighted sampling of Beta distributionb estimated from individual studies

Mean: 70.4% 95% PI: 49.7%e92.3%

Cull-High

Barkocy-Gallagher et al. (2003); Brichta-Harhay et al. (2008); Kalchayanand et al. (2009); RiveraBetancourt et al. (2004) Barkocy-Gallagher et al. (2003); Rivera-Betancourt et al. (2004) Brichta-Harhay et al. (2007, 2008)

Cull-Low

Brichta-Harhay et al. (2007, 2008)

All

Arthur, Bosilevac, et al. (2008); Barkocy-Gallagher et al. (2003); Rivera-Betancourt et al. (2004); Brichta-Harhay et al. (2007, 2008) e

Weighted sampling of Beta distributionb estimated from individual studies RiskExtvalueMin (0.960, 0.068, RiskTruncate (0, 1))c RiskExtvalueMin (0.927, 0.074, RiskTruncate(0, 1))c Normal (0.44, 0.045)

Mean: 36.7% 95% PI: 28.2%e95.9% Mean: 90.2% 95% PI: 74.6%e99.1% Mean: 87.5% 95% PI: 70.2%e98.4% Mean: 0.441 95% PI: 0.366e0.515

P1*P2

Mean: 31.0% 95% PI: 19.8%e44.4% Mean: 16.2% 95% PI: 11.3%e41.2% Mean: 39.7% 95% PI: 30.5%e41.8% Mean: 38.5% 95% PI: 29.1%e47.3% Mean: 0.018 95% PI: 0.0072e0.033 Mean: 0.57% 95% PI: 0.19%e1.2% Mean: 0.3% 95% PI:0.1%e0.7% Mean: 0.73% 95% PI: 0.28%e1.4% Mean: 0.71% 95% PI: 0.27%e1.3% Weighted weights: Fed cattle: 734 lb (333 kg) Cull cattle: 630 lb (286 kg) Fed cattle: 18% Cull cattle: Uniform (75%, 90%) 15 for fed cattle 4 for cull cattle 32,000 cm2 for fed cattle; 24,000 cm2 for cull cattle 60 for fed cattle; 12 for cull cattle Mean: 0.34 95% PI: 0.0e2.0 Mean: 0.18 95% PI: 0e1.0 Mean: 0.089 95% PI: 0e1.0 Mean: 0.085 95% PI: 0e1.0

Fed-Low

Linear transfer coefficient for hide prevalence to prevalence in 8000 cm2 pre-evisceration samples (P2) Pre-evisceration carcass prevalence on 8000 cm2 carcass surface segments (P3)

Fed-High Fed-Low Cull-High Cull-Low

Effectiveness of generic interventions in reducing prevalence (linear coefficient) (P4) Post-intervention carcass prevalence on 8000 cm2 surface segments (P5)

All Fed-High

Rivera-Betancourt et al. (2004); Brichta-Harhay et al. (2007, 2008) e

Gamma (3.79, 0.0042) P3*P4

Fed-Low Cull-High Cull-Low Average carcass weight (P6)

e

FSIS, 2001

Proportion of carcass weight contributing to trim (P7)

e

No. of carcasses in a production lot (P8)

e

FSIS, 2001; Industrial expert estimate e

Carcass surface area (P9)

e

FSIS, 2001

No. of 8000 cm2 carcass segments per lot (P10) No. of contaminated 8000 cm2 carcass segments per lot (P11)

e Fed-High

e

P8*(P9/8000 cm2) Binomial (P10, P5)d

All

Brichta-Harhay et al. (2007, 2008)

Cumulative distribution

All

Brichta-Harhay et al. (2007, 2008)

Normal (3.56, 0.95)

764, 703, 539, 851 lb (347, 319, 244, 386 kg) for steers, heifers, cows, and bulls, respectively

2000 lb/P6*P7

Fed-Low Cull-High Cull-Low Salmonella concentration on carcass surfaces Concentration of Salmonella on pre-evisceration carcasses (C1) (CFU/100 cm2) Generic intervention effect on the log reduction of Salmonella concentration on carcasses based on indicator bacteria data (C2) (log CFU/100 cm2)

Mean: 2.21 95% PI:0.3e2.1 Mean: 3.56 95% PI: 2.00e5.12

M. Li et al. / Food Control 50 (2015) 423e434

Parameter description

Mean: 2.51 95% PI: 0.99e4.61

MAX(0.4, 10^(LOG(C1)-C2))e

0.4

All

e

MAX(0.4, 10^(LOG(C1)-C3))e

0.4

Fed-High

FSIS, 2001 e

C7 

e

C6 

Fed-High

Gragg, Loneragan, Brashears, et al. (2013)

Fed-Low

Arthur, Brichta-Harhay, et al. (2008); Gragg, Loneragan, Brashears, et al. (2013) Gragg, Loneragan, Brashears, et al. (2013)

All

Arthur, Brichta-Harhay, et al. (2008); Gragg, Loneragan, Brashears, et al. (2013) Assumption

RiskDuniform distributionf aggregated Beta (86, 347) and Beta (133, 487) RiskDuniform distributionf aggregated Beta (9, 445) and Beta (5, 567) RiskDuniform distributionf aggregated Beta (5, 605) and Beta (24, 588) RiskDuniform distributionf aggregated Beta (7, 603) and Beta (15, 557) Round (Pert (2, 4, 6),0)

Fed-High

e

Binomial (L2*P8, L1)d

Fed

Gragg (2012)

RiskExtvalue (13.9, 5.45, RiskTruncate (4, 40))g

Cull

Unpublished data

Fed

Gragg (2012)

RiskLognorm (33.17, 32.72,RiskShift(1.92), RiskTruncate(3, 445)) RiskBeta (115, 31)

Cull

Unpublished data

RiskBeta (12, 20)

Fed

Gragg (2012)

Cull

Unpublished data

RiskWeibull(1.88, 2.21,RiskShift(-0.21), Risktruncate (0, 4)) RiskUniform (0.15, 2.2)

All

Salmonella concentration on 8000 cm2 postintervention carcass segments based on indicator bacteria data (C4) (CFU/100 cm2) Salmonella concentration on 8000 cm2 postintervention carcass segments (C5) based on lab data (CFU/100 cm2) Percent of carcass surface area represented in trims (C6) Total Salmonella CFU added to a production lot from carcass surface contamination based on indicator bacteria data (C7)

PP11 i¼1

75% for fed cattle; 90% for cull cattle Mean: 12.43 95% PI: 0e48 Mean: 6.83 95% PI: 0e24 Mean: 3.67 95% PI: 0e29 Mean: 3.58 95% PI: 0e29 Mean: 8.88 95% PI: 0e48 Mean: 4.66 95% PI: 0e24 Mean: 2.70 95% PI: 0e29 Mean: 2.60 95% PI: 0e29

C4  80

Fed-Low Cull-High Cull-Low

Total Salmonella CFU added to a production lot from carcass surface contamination based on lab data (C8)

Fed-High

PP11

C i¼1 5

 80

Fed-Low Cull-High Cull-Low Salmonella prevalence and concentration in DTLNs Salmonella prevalence in DTLNs (L1)

Cull-High Cull-Low Number of DTLNs per carcass added to a production lot (L2) Number of contaminated DTLNs added to a production lot (L3)

from from from from

Fed-Low Cull-High Cull-Low DTLN weight (L4)

Proportion of enumerated positive DTLNs (L5)

Salmonella concentration in enumerable contaminated DTLNs (log CFU/g) (L6)

Mean: 20.7% 95% PI: 17.4%e23.8% Mean: 1.4% 95% PI: 0.43%e2.9% Mean: 2.4% 95% PI: 0.4%e4.9% Mean: 1.8% 95% PI: 0.64%e3.4% Mean: 4 95% PI: 2.8e5.2 Mean: 12.4 95% PI: 6e20 Mean: 0.86 95% PI: 0e3 Mean: 0.38 95% PI: 0e2 Mean: 0.30 95% PI: 0e1 Mean: 16.83 95% PI: 7.98e29.96 Mean: 35.18 95% PI: 8.3e93.6 Mean: 0.79 95% PI: 0.73e0.84 Mean: 0.37 95% PI: 0.24e0.52 Mean: 1.62 95% PI:0.23e3.28 Mean: 1.18 95% PI: 0.25e2.1

427

(continued on next page)

M. Li et al. / Food Control 50 (2015) 423e434

Gamma (4.96, 0.51) þ Lognormal (2.18, 1.07)

All

Arthur, Bosilevac, et al. (2008); Castillo, Lucia, Goodson, Savell, and Acuff (1998); Cutter and Rivera-Betancourt (2000); Nutsch et al. (1997); Retzlaff et al. (2004) e

Generic intervention effect on the log reduction of Salmonella concentration on carcasses based on lab data (C3) (log CFU/100 cm2)

Parameter description Salmonella CFUs in a contaminated DTLN (L7) (CFU/DTLN)

428

Table 3 (continued ) Cattle-season

Data source

Distribution/@RISK or Excel formula h

Fed

L4*If(RiskBinomial(1, L5),10^ L6,1)

Cull Total Salmonella CFU from DTLNs in a production lot (L8) (CFU)

Fed-High

e

RiskCompound (L3, L7)i

Fed-Low Cull-High Cull-Low Baseline model simulation results Total Salmonella CFU load in a production lot (S1) Relative contribution of Salmonella from DTLNs to the total CFU load in a production lot based on indicator bacteria or lab data (S2)

All All

C7þL8 or C8þL8

RiskMeanðL8 Þ RiskMeanðL8 ÞþRiskMeanðC7 Þ

Estimated valuea Mean: 4331 95% PI: 11e24,086 Mean: 442 95% PI: 3e2425 Mean: 53,513 95% PI: 3610e155,217 Mean: 3743 95% PI: 0e20,885 Mean: 169 95% PI: 0e773 Mean: 129 95% PI: 0e474 See Table 5 See Table 5

or

RiskMeanðL7 Þ RiskMeanðL8 ÞþRiskMeanðC8 Þ

a

95%PI: 95% probability interval. Beta distribution: A two parameter continuous probability distribution often used to model the uncertainty associated with the probability of an event. The Beta (Sþ1, NSþ1) distribution is used in Bayesian analysis to model the uncertainty in the probability of success of a Binomial process when s successes are observed in N trials. c RiskExtValueMin (a, b) specifies an extreme value min distribution with location parameter a and shape parameter b. Risktruncate (minimum, maximum) defines the minimumemaximum range allowable for samples drawn for the distribution in which the RiskTruncate function is included. d Binomial distribution: A two parameter discrete probability distribution estimating the number of successes in a sequence of N independent trials and each trial yields success with a probability of p. e The detection limit for Salmonella on carcass surface was 0.4 CFU/100 cm2 in Brichta-Harhay et al. (2007, 2008). f RiskDuniform ({X1,X2, … Xn}) is a discrete uniform distribution with n outcomes valued at X1 through Xn. g RiskExtvalue (a, b) describes an extreme value distribution with location parameter a and shape parameter b. h Salmonella level for enrichment positive and enumeration negative samples is 1 CFU/g for both fed and cull cattle. i RiskCompound (dist#1,dist#2) calculates the sum of a number of samples from dist#2 where the number of samples drawn from dist#2 is given by the value sampled from dist#1. b

M. Li et al. / Food Control 50 (2015) 423e434

M. Li et al. / Food Control 50 (2015) 423e434

Brichta-Harhay et al. (2007, 2008). This function was used because the best fit parametric distributions were not able to adequately incorporate the extreme variability (tail probability or very high carcass surface concentrations in a few of the samples) in the reported data. Data on the effectiveness of carcass surface interventions in reducing Salmonella carcass concentrations in commercial beef processing plants are scarce. Therefore, this parameter was estimated based on a couple of indirect data sources: 1) commercial plant sampling data for reduction of indicator microorganisms (aerobic plate count) before evisceration and after chilling (C2); 2) laboratory data for Salmonella reduction on inoculated carcass samples after application of certain carcass interventions (C3). Data from the laboratory tests involved two consecutive carcass interventions: organic acid wash and steam pasteurization. Their reduction effects measured in log CFU/100 cm2 were fitted with Lognormal (2.18, 1.07) and Gamma (4.96, 0.51) distributions, respectively, and then summed up to estimate the total bacterial reduction. The amount of Salmonella from individually contaminated 8000 cm2 post-intervention carcass surface segments (C4 and C5) were summed up and multiplied by the percent of carcass surface area represented in beef trims (C6), to estimate the total Salmonella CFU load from carcass surface contamination in a production lot (C7 and C8) (Fig. 2). 2.1.3. Modeling Salmonella prevalence and concentration in DTLNs The DTLN contamination module involved modeling the prevalence and concentration of Salmonella in DTLNs (Fig. 3). The DTLN prevalence data from individual studies were first fitted with Beta distributions, which were then aggregated with equal probability using a discrete uniform distribution for each category of animaltype and seasonal combinations (L1) (Arthur, Brichta-Harhay, et al., 2008; Gragg, Loneragan, Brashears, et al., 2013) (Table 3). The number of contaminated DTLNs added to a production lot (L3) was estimated by the Binomial distribution by accounting for DTLN prevalence (L1) and the number of major DTLNs (subiliac, superficial cervical, and popliteal lymph nodes) or equivalent DTLN tissues added to a production lot, which was the product of the number of DTLNs per carcass added to a production lot (L2) (assumed to be a minimum of 2, maximum of 6 and most likely value of 4, due to data gaps) and the number of carcasses in a production lot (P8). The proportion of enumerated positive DTLNs (L5) and Salmonella concentration in enumerable contaminated DTLNs (L6) for fed cattle were obtained from Gragg (2012) and Gragg, Loneragan, Brashears, et al. (2013), which tested 1501 DTLN samples of fed cattle across three geographical regions in the U.S. in a two-year span. In a subset of 618 samples, 67% (n ¼ 144) were positive and 114 enumerable. The enumeration data were fitted with a Weibull (1.88, 2.21, RiskShift (0.21)) log CFU/g distribution and truncated at a minimum of 0 log CFU/g and a maximum of 4 log CFU/g based

429

on the raw data. These two parameters (L5 and L6) for cull cattle were based on an unpublished study through personal communication with Dr. Guy Loneragan at Texas Tech University and Dr. Dayna Brichta-Harhay at the U.S. Meat Animal Research Center. Similar to the fed cattle study, the cull cattle study collected 1655 cull cattle DTLNs across different geographical regions in different seasons and detected Salmonella prevalence and concentration. Only 1.8% samples were found contaminated with Salmonella and a small proportion of the positive samples were enumerable, which was fitted with a Uniform (0.15, 2.2) log CFU/g distribution. Thereafter, the Salmonella CFUs in a contaminated DTLN (L7) was calculated via multiplying the weight of DTLNs (L4) by the concentration of Salmonella in contaminated DTLNs (CFU/g), and the latter was calculated to be 10L6 CFU/g if the contaminated DTLNs were enumerable and a minimum of 1 CFU/g if non-enumerable, using the Binomial (1, L5) distribution to assign the state of either enumerable or non-enumerable for prevalence positive samples. Finally, the total Salmonella CFU load from DTLNs in a production lot (L8) was calculated by summing up Salmonella CFUs in all contaminated DTLNs. Very limited data were available to quantify the effect of pre-harvest interventions. Therefore, the model assumed multiple scenarios to study the impact of potential pre-harvest interventions in reducing lymph node prevalence and concentration, which is to be explained in Section 2.3. Finally, the relative contribution of DTLNs was defined as the percentage of the total Salmonella CFU load from DTLNs versus the total CFU load contributed from both DTLNs and carcass surfaces in a production lot (Table 3). 2.2. Evidence collection for model parameters A systematic search of scientific literature was conducted to identify all relevant publications regarding Salmonella contamination and interventions during pre- and post-harvest stages. Three databases, including Agricola, PubMed, and Food Science and Technology Abstract (FSTA) were searched from 1996 to 2013 (HACCP was implemented in 1996 in the USA). Data sources encompassed journal papers, conference proceedings, thesis/dissertations, and annual reports, and were restricted to North America (USA, Canada, and Mexico) studies. The overall search term was “Pathogen AND Population AND Outcome”, which was adapted from Sargeant, Amezcua, Rajic, and Waddell (2007).The specific search terms are shown in Table 4. A total of 3814 references were obtained after the search, 2168 references passed duplication deletion and relevance screening, and ultimately 42 references remained after quality assessment. Criteria for quality assessment were comprised of sound study designs and experimental procedures, adequate statistical analysis, and sufficient description of intervention protocols, etc. In case of key data deficiencies from the literature, external expert opinions were elicited and applicable unpublished studies used.

Table 4 Search terms for systematic reviews of literature data. Search terms

Bacterial prevalence and concentration

Pathogen Population Outcome

Salmonella OR Salmonellae cattle OR cow* OR steer* OR heifer* OR bull* OR calf OR calves OR beef vaccin* OR immuniz* OR immunis* OR SRP OR “porin protein*” fecal OR faecal OR feces OR faeces OR manure OR hide OR carcass* OR shed* OR feedlot* OR feedyard* OR feeding OR farm* OR harvest OR siderophore* OR bacteriocin* OR antimicrobial* OR antibiotic* OR antibacterial* OR phage* OR bacteriophage* OR diet* OR preharvest OR pre-harvest OR postharvest OR post-harvest OR “sodium chlorate” OR probiotic* OR biocontrol OR “competitive exclusion” OR transport* OR lairage* OR “holding pen*” OR trailer* OR truck* OR “direct fed microbial*” OR lactob* OR bifidobac* OR “lactic acid bacteria” OR abattoir* OR slaughter* OR kill* OR plant* OR processing OR dehid* OR dehair* OR skin* OR dress* OR vaccum* OR eviserat* OR wash* OR rins* OR steam OR pasteuriz* OR spray* OR acid* OR “hot water” OR chlorine OR chill* OR cool* OR debon* OR boning OR cut* OR fabricat* OR trim* OR grind* OR “lymph node*” OR intervention* OR mitigat* OR “control measure*” OR decontaminat* OR disinfect*

Pre- and post-harvest intervention

430

M. Li et al. / Food Control 50 (2015) 423e434

Table 5 Total Salmonella CFU load and relative contribution of DTLNs to the total CFU load in a ground beef production lot. Parameter

Relative contribution of Salmonella from DTLNs to the total Salmonella CFU load in a production Total Salmonella CFU load in a production lot a b

Data sourcea

Indicator bacteria Lab test Indicator bacteria Lab test

Fed cattle

Cull cattle

High season

Low season

High season

Low season

99.97% 99.98% 53,526 (3625e155,235)b 53,522 (3624e155,235)

99.82% 99.87% 3750 (0e20,886) 3748 (0e20,886)

97.56% 98.36% 170 (0e787) 169 (0e772)

96.83% 97.78% 135 (0e492) 134 (0e484)

Two types of post-harvest carcass intervention data, in-plant indicator bacteria data and Salmonella lab test data, were used to calculate model outcomes, respectively. Data in the parentheses represent 95% probability intervals.

Evidences were extracted from appropriate studies passing quality assessment and tabulated for corresponding model parameters comprising a variety of categories, such as cattle type, season, geographical location, prevalence, concentration, detection method, and intervention effect. To account for variability in the complex beef system, relevant data from multiple studies were compiled to define parameter estimates, and appropriate weights were assigned to each study proportional to their sample sizes.

season, and 66.9%e99.8% in low season) with the increase of the effectiveness of generic post-harvest interventions in reducing Salmonella prevalence and/or concentration. However, for cull cattle (Table 7), the relative contribution of DTLNs to the total CFU load was fairly sensitive to the effectiveness of the post-harvest interventions, and varied from 9.0% to 97.6% in high season, and 7.6%e96.8% in low season, across different effectiveness levels.

2.3. Scenario analysis

3.4. Impact of generic pre-harvest carcass interventions on the total Salmonella CFU load in a production lot

Multiple scenario analyses with respect to key parameters were conducted to estimate how changes in their values impact model outcomes. These scenarios included: multiplying the effectiveness of generic post-harvest interventions in reducing carcass surface prevalence and concentration by multipliers from 20 to 100%; reducing DTLN prevalence by 0e50% and concentration by 0e1 log CFU/g owing to generic pre-harvest interventions; and changing the number of major DTLNs (or equivalent lymph node tissues) per carcass added to a production lot from 1 to 8. The @RISK software (version 6.2, Palisade, Newfield, NY), an add-in the Excel spreadsheet, was used to develop and simulate the model with 100,000 iterations for the baseline model and each scenario analysis. 3. Results 3.1. Relative contribution of Salmonella from DTLNs to the total CFU load in a production lot Table 5 indicates that for each category of animal type and season combinations, most of the Salmonella contamination originates from DTLNs, which ranged from 99.8% to 99.9% for fed cattle and from 96.8% to 98.4% for cull cattle, respectively. These results indicate that the contribution of DTLNs to a production lot contamination is relatively robust to animal type and seasonal difference. However, since the DTLN concentration data for cull cattle were based on enumeration data for a small sample size of DTLNs, there is a greater uncertainty associated with the cull cattle results compared with the fed cattle. 3.2. Total Salmonella CFU load in a production lot

As illustrated in Table 8, for fed cattle in high season, increasing the effectiveness of generic pre-harvest interventions in reducing the Salmonella prevalence and concentration in DTLNs reduced the total Salmonella CFU load considerably from 16,985 to 855 while not having a significant impact on the relative contribution from DTLNs. In comparison, for cull cattle in high season (Table 9), a small change of the total CFU load from 58.2 to 6.9 owing to the increased effectiveness of generic pre-harvest interventions caused a significant reduction in the DTLN contribution from 92.9% to 40.2%. Similar trends were observed in low season and in high season for fed and cull cattle, respectively (data not shown). 3.5. Impact of the number of major DTLNs on the total Salmonella CFU load in a production lot As shown in Fig. 4, the total Salmonella CFU load in a production lot for fed cattle in high season increased dramatically when the number of major DTLNs per animal added to a ground beef Table 6 Impact of generic post-harvest carcass interventions on the relative contribution of DTLNs to the total CFU load in a production lot for fed cattle. Multiplier for effectiveness Multiplier for effectiveness of post-harvest of post-harvest interventions in interventions in reducing carcass prevalenceb a reducing carcass concentration 20% 40% 80% 100% 20% 40% 80%

Table 5 shows that the total Salmonella CFU load in a production lot has seasonal and animal type variations. The predicted CFU load was greater, for fed cattle than for cull cattle, and in high season than in low season. 3.3. Impact of generic post-harvest carcass interventions on the relative contribution of DTLNs to Salmonella CFU load in a production lot As indicated in Table 6, the relative contribution of DTLNs to the total CFU load for fed cattle remained high (96.5%e99.9% in high

100%

96.5%c (66.9%)d 98.6% (86.3%) 99.3% (94.2%) 99.3% (94.7%)

96.6% (67.7%) 98.7% (87.4%) 99.4% (95.3%) 99.5% (95.8%)

97.1% (73.1%) 99.1% (90.8%) 99.8% (97.8%) 99.8% (98.3%)

99.2% (94.2%) 99.8% (98.4%) 99.9% (99.7%) 99.9% (99.8%)

a The effect of post-harvest interventions in reducing carcass concentration was estimated by multiplying the distribution of intervention effect with fractional multiplier from 20 to 100%. b The effect of post-harvest interventions in reducing carcass prevalence was estimated by multiplying the distribution of intervention effect with fractional multiplier from 20 to 100%. c Data represent the relative contribution of DTLNs to the total CFU load in a production lot for fed cattle in high season. d Data in the parentheses represent the relative contribution of DTLNs to the total CFU load in a production lot for fed cattle in low season.

M. Li et al. / Food Control 50 (2015) 423e434

431

Table 7 Impact of generic post-harvest carcass interventions on the relative contribution of DTLNs to the total CFU load in a production lot for cull cattle. Multiplier for effectiveness of post-harvest interventions in reducing carcass concentrationa

Multiplier for effectiveness of post-harvest interventions in reducing carcass prevalenceb 20% c

20%

9.0% (7.6%)d 26.1% (21.9%) 54.2% (48.0%) 56.8% (52.7%)

40% 80% 100%

40%

80%

100%

9.7% (8.1%) 28.1% (22.6%) 58.5% (53.7%) 63.3% (58.4%)

15.3% (12.2%) 38.0% (33.8%) 75.8% (71.9%) 80.9% (77.1%)

58.6% (53.9%) 84.2% (80.9%) 96.7% (95.5%) 97.6% (96.8%)

a The effect of post-harvest interventions on reducing carcass concentration was estimated by multiplying the distribution of intervention effect with fractional multiplier from 20 to 100%. b The effect of post-harvest interventions on reducing carcass prevalence was estimated by multiplying the distribution of intervention effect with fractional multiplier from 20 to 100%. c Data represent the relative contribution of DTLNs to the total CFU load in a production lot for cull cattle in high season. d Data in the parentheses represent the relative contribution of DTLNs to the total CFU load in a production lot for cull cattle in low season.

Table 8 Impact of generic pre-harvest interventions on the relative contribution of DTLNs and total CFU load in a production lot for fed cattle in high season. Reduction in Salmonella DTLN concentration due to generic pre-harvest intervention (log CFU/g)

Effectiveness of generic pre-harvest intervention in reducing Salmonella DTLN prevalence 0%

10%

30%

50%

0.5

16,985a (99.9%)b 5379 (99.8%) 1713 (99.3%)

15,269 (99.9%) 4836 (99.7%) 1531 (99.1%)

11,855 (99.9%) 3757 (99.6%) 1195 (98.9%)

8473 (99.8%) 2674 (99.5%) 855 (98.4%)

1 1.5 a

Data represent the total Salmonella CFU load in a production lot. Data in the parentheses represent relative contribution of DTLNs to the total CFU load in a production lot. b

production lot was increased from 1 to 8. However, the relative contribution of DTLN was not significantly affected, which remained over 99%. For cull cattle in the high prevalence season, increasing the number of major DTLNs resulted in an increase in the average CFU load per production lot from 47 to 341 CFUs, which corresponded to an increase in the DTLN contribution from 91.2% to 98.8%. 4. Discussion

DTLNs, with relatively little contribution from the carcass surface contamination. An explanation for these results is that a single contaminated DTLN (L7) (on average, 103.64 CFU for fed cattle and 102.65 CFU for cull cattle) can often add considerably more CFUs than does a contaminated 8000 cm2 post-intervention carcass segment (less than 101.5 CFU for both fed and cull cattle). Furthermore, a greater number of contaminated DTLNs (L3) than that of 8000 cm2 contaminated carcass surface segments (P11) were predicted to be included in a production lot, particularly for fed cattle with higher DTLN prevalence (Table 3). These results also indicate that the carcass surface interventions are quite effective and very few Salmonella CFUs would be added to a production lot from the surface. These results are consistent with previous USDA/ARS studies (Barkocy-Gallagher et al., 2003; Brichta-Harhay et al., 2008), where very few post-intervention carcasses were enrichment positive for Salmonella and an even smaller proportion of those enrichment positive samples were enumerable at very low concentrations. The scenario analyses indicated that, for fed cattle, the relative contribution of DTLN contamination is fairly robust to changes in model parameters such as the effectiveness of post-harvest interventions. There is a greater uncertainty regarding the relative contribution of DTLN contamination for cull cattle and the results were more sensitive to changes in model parameters in this case. The model results also indicate that ground beef from cull cattle would have a lower Salmonella load in orders of magnitude relative

The results from the baseline scenario indicate that most of the Salmonella load in a ground beef production lot originates from

Table 9 Sensitivity analysis of the impact of generic pre-harvest interventions on the relative contribution of DTLNs and total CFU load in a production lot for cull cattle in high season. Reduction in Salmonella DTLN concentration due to generic pre-harvest intervention (log CFU/g) 0.5 1 1.5 a

Effectiveness of generic pre-harvest intervention in reducing Salmonella DTLN prevalence 0% a

58.2 (92.9%)b 21.4 (80.1%) 9.5 (56.6%)

10%

30%

50%

53.3 (91.4%) 18.7 (78.7%) 8.3 (59.0%)

42.2 (89.1%) 15.8 (76.2%) 7.3 (48.3%)

32.3 (87.9%) 11.6 (69.2%) 6.9 (40.2%)

Data represent the total Salmonella CFU load in a production lot. Data in the parentheses represent relative contribution of DTLNs to the total CFU load in a production lot. b

Fig. 4. Sensitivity analysis of the impact of the number of major DTLNs (or equivalent lymphatic tissues) per carcass added to a production lot on (A) the total Salmonella CFU load in a production lot and (B) the relative contribution of Salmonella from DTLNs for fed cattle in high season.

432

M. Li et al. / Food Control 50 (2015) 423e434

to fed cattle. This is because compared with fed cattle, cull cattle contributed less carcasses per production lot (P8), and had a lower DTLN Salmonella prevalence (L1), resulting in fewer contaminated DTLNs being included in a production lot (L3). Moreover, the proportion of enumerated positive DTLNs (L5) and Salmonella concentration in enumerable contaminated DTLNs (L6) of cull cattle were lower, leading to less CFU load per contaminated DTLN (L7) in cull cattle. The model reflects the observed difference in Salmonella DTLN prevalence and concentration between fed and cull cattle, and indicates that cull cattle DTLNs are not likely a great contributor to Salmonella contamination in ground beef. This may reflect discrepancies in hygiene or mitigation practices prior to harvest, as well as differences in Salmonella prevalence in cattle environment (Gragg, Loneragan, Brashears, et al., 2013). Several categories of pre-harvest interventions, such as vaccines, are being evaluated for reducing Salmonella in DTLNs of beef cattle (Edrington, Loneragan, Hill, Genovese, Brichta-Harhay et al., 2013, Edrington, Loneragan, Hill, Genovese, He, 2013). We evaluated the impact of generic pre-harvest interventions which could reduce the prevalence and/or concentration of Salmonella in DTLNs on the model outcomes. Our scenario analysis indicated that generic pre-harvest interventions that could reduce Salmonella prevalence and concentration in DTLNs could also result in a significant reduction in the total Salmonella CFU load in a ground beef production lot for fed cattle. This is not unexpected since most of the Salmonella CFUs of fed cattle could be attributed to DTLN contamination. However, owing to the lower Salmonella CFUs from DTLNs of cull cattle, the generic pre-harvest interventions were predicted to cause a smaller reduction in the total CFU load in a production lot. Because there is a paucity of studies on the efficacies of specific interventions in reducing Salmonella in DTLNs or on carcass surfaces, we evaluated the impacts of all the applicable generic pre- or post-harvest interventions instead. This simplified approach is feasible since the ultimate purpose of this study is to compare the overall relative impacts of pre-and post-harvest interventions rather than specific intervention effects. For this purpose and from the perspective of model parsimony, although the specific approach has merits in providing better precision, it would increase model complexity and introduce additional parameters with large uncertainties (due to data gaps), which might increase the overall uncertainty and reduce the accuracy of the model (Zwietering, 2009). Similar modeling approaches have been utilized in some previous risk assessment studies where the effectiveness of carcass interventions on the change of E. coli O157:H7 prevalence and concentration were the aggregate effect of specific decontamination events (Cassin, Lammerding, Todd, Ross, & McColl, 1998; Hurd & Malladi, 2012; USDA/FSIS, 2001). In addition, for post-harvest interventions, the generic intervention effect is a compilation or summation of the effects of different specific interventions regarding the sequences of their applications between evisceration and chilling and thus generally reflects the actual decontamination effects in post-harvest processing. Therefore, the generic approach was considered sufficient to answer the risk management question in this study. As for the pre-harvest interventions, the hypothetical

interventions of different effectiveness levels were used to compare its relative impact with the post-harvest interventions. This is a strategy commonly employed in risk assessment studies to cope with data limitations when evaluating the effectiveness of different interventions (Cassin et al., 1998; Delhalle et al., 2009; Dodd, Sanderson, Jacob, & Renter, 2011). Although DTLNs have been found a major contributor to ground beef Salmonella contamination, there may be potentially alternate explanations for the results that need to be further explored before recommending appropriate intervention strategies. For example:  There may be sources of Salmonella other than the external carcass surface and DTLNs which were not considered here. Cross-contamination of freshly-cut meat surfaces from environmental sources may occur during the fabrication stage, such as equipment surfaces, conveyor belts, cutting surfaces, knives, gloves, and aprons (Gill, Badoni, & McGinnis, 1999; Gill, Bryant, & Landers, 2003; Youssef, Badoni, Yang, & Gill, 2013). However, because of the scarcity of empirical data, we were not able to include these potential contamination sources in the model. Probably for the similar reason, cross-contamination was not modeled in some other risk assessment studies for E. coli during beef processing (Cummins, Nally, Butler, Duffy, & O'Brien, 2008; Dodd et al., 2011; Signorini & Tarabla, 2009). Another potential source is the internal surface contamination owing to gut rupture during evisceration. However, since the incidence of gut rupture is rare, it was omitted in several risk assessment studies (Cassin et al., 1998; Signorini & Tarabla, 2009). Moreover, deep cuts in the carcass which may occasionally occur in the harvest process may protect Salmonella from some of the post-harvest interventions (Koohmaraie et al., 2012). Salmonella harbored in such cuts may not be detectable by the surface sponge sampling method used in most studies.  The effectiveness of carcass interventions was based on Salmonella lab test data and in-plant indicator bacteria data. Bacterial reduction estimated from these data might be greater than the actual Salmonella reduction caused by in-plant interventions in commercial operations, thus leading to underestimation of Salmonella concentration on post-intervention carcasses. However, the available carcass sampling data does indicate a very low prevalence and concentration of Salmonella at this stage (Barkocy-Gallagher et al., 2003; Brichta-Harhay et al., 2008). The validity of the model should be further tested against reported data. As data are lacking for both the total Salmonella CFUs in a production lot and for the relative impact of lymph node and carcass surface contaminations, the model validation was carried out in an indirect way, which involves calculating the total CFUs in a 2000 lb production lot based on the Salmonella detection data reported by Bosilevac et al. (2009a). The study sampled 4136 ground beef portions of 65 g each and the detection results could be categorized into 4 groups according to the levels of contamination, i.e., enrichment negative, enrichment positive and enumeration negative, 2e4 CFU/g, and 40 CFU/g. Table 10 presents the input parameters for estimating the total Salmonella CFUs in a production

Table 10 Input parameters for estimating the Salmonella CFU load in a production lot from empirical data presented in Bosilevac et al. (2009a, 2009b). Input parameter

Salmonella enrichment and enumeration results for 65 g ground beef samples Enrichment negative

Enrichment positive, enumeration negative

2e4 CFU/g

40 CFU/g

Sample size Expected proportion of ground beef samples Salmonella CFUs per 65 g ground beefa

3964 95.78% 0

162 3.94% 65*Uniform (0e2)

9 0.24% 65*Uniform (2e4)

1 0.048% 40*65

a

The limit of detection was 2 CFU/g of ground beef.

M. Li et al. / Food Control 50 (2015) 423e434

lot, including the number of samples, the proportion of 65 g ground beef samples (reported as mean values), and the Salmonella CFUs per 65 g ground beef samples at each of the contamination levels. The uncertainty about the proportion of samples was simulated using the Dirichlet (3965, 163, 10, 2) distribution, which corresponded with the number of samples in each group. The number of Salmonella CFUs per 65 g was simulated using Uniform distributions. Summing up the Salmonella CFUs in each contamination level (calculated by multiplying the proportion of samples and Salmonella CFUs/65 g sample and by the number of 65 g portions in a 2000 lb production lot, i.e., 2000 lb/65 g) resulted in the total CFUs in a production lot, which was estimated to be in the range of 3393 to173,713 CFUs with a mean of 59,824 CFUs after simulation. The ground beef collected in Bosilevac et al. (2009a) had various percentages of lean meat (73%e90%) from cull cattle and fat meat from fed cattle. Therefore, assuming the same ratios of lean meat from cull cattle to fat meat from fed cattle in this study, the resulting total CFU load in a production lot would generally fall in that calculated range and lean toward the lower end. There might be greater uncertainty about the cull cattle results due to the limited enumerable sample size of DTLNs as mentioned earlier. However, the model prediction does reflect the low level of cull cattle concentration data in the sampling surveys. In addition, the total CFU load (minimum of 3393 CFUs) was substantially greater than the predicted CFU load originating from carcass surface per production lot, suggesting that another contamination source apart from carcass surfaces (i.e., lymph nodes) may be contributing significantly to the Salmonella load in a production lot. Although the model prediction seems reasonable, further improvements are warranted in several fields. To begin with, the Salmonella prevalence at different processing points was related using linear regression models (such as the change of Salmonella prevalence from hide to carcass during dehiding and on carcasses due to post-harvest interventions). While the linear fit is simpler, nonlinear models may provide a better fit to understand the bacterial transfer dynamics influenced by the complex beef processing system (Havelaar, Evers, & Nauta, 2008). Evaluating alternate nonlinear relationships (such as logistic regressions, nonlinear transfer ratios, and other nonlinear models) with more empirical data can improve model accuracy and predictive ability (Dodd et al., 2011; Smith, Fazil, & Lammerding, 2013). Furthermore, we assumed that contamination in DTLNs and on carcass surfaces were independent. Some previous serotyping investigations and challenge tests on the sources of DTLN contamination revealed that Salmonella are most likely introduced into DTLNs by crossing the host's integument transdermally through various pathways, such as transfer from cattle hides or environment via skin abrasions, biting flies, or diseases of the integument (Edrington, Loneragan, Hill, Genovese, Brichta-Harhay et al., 2013, Edrington, Loneragan, Hill, Genovese, He, 2013; Gragg, Loneragan, Brashears, et al., 2013, Gragg, Loneragan, Nightingale, 2013). There has been observed correlations between Salmonella prevalence on cattle hides, in cattle environment and in DTLNs (Gragg, Loneragan, Brashears, et al., 2013). Considering that cattle hides are a common reservoir for Salmonella, it is not surprising that hides could be a common source of Salmonella to both DTLNs and carcass surfaces. However, the transmission of Salmonella from hides to DTLNs and to carcass surfaces are generally two independent events that occur at different stages of the cattle chain, i.e. the former mainly occurs during pre-harvest while the latter takes place during slaughter). Although we cannot completely rule out the possibility that Salmonella may enter DTLNs from environment during transport and lairage through similar transdermal pathways, the duration of infection is relatively very short as compared with that of preharvest (less than 24 h during lairage and a few days at the max during

433

transport and marketing (for cull cattle), versus several months during preharvest), and thus the infection during transport and lairage is not likely to have very much contribution to the prevalence and level change in DTLNs as compared with the preharvest infection. Therefore, we assumed the independence of contamination events on carcass surfaces and in DTLNs. Nevertheless, since Salmonella in DTLNs has recently emerged as an important food safety concern in the beef industry, this assumption is derived from published studies at the time of this analysis. As evidence accumulates that further elucidates the contamination mechanisms involved, the relationship between DTLNs and carcass surface contamination may be better characterized and quantified in the future. It is also worth noting that the model estimated contribution of Salmonella contamination to ground beef from trims originating from either all from fed cattle or all from cull cattle. In reality, notwithstanding, a good portion of ground beef in the U.S. is made by combining lean trims from cull cattle with fat trims from beef cattle to get the right lean: fat ratio desired (80:20; 90:10, etc). This is because cull cows (in particular dairy cows) do not have much fat and hence fat needs to be added from other sources. For future modeling work it would be meaningful to include mixing proportions of lean and fat trim in the model to get a better understanding of the downstream impact on Salmonella contamination of ground beef by the upstream trim sources. We acknowledge some other limitations of the study. The analysis focused on estimating the mean Salmonella load and relative contribution of DTLN or surface contamination. Some of the sources of variability in the Salmonella load in a production lot, such as differences in abattoir characteristics, cross contamination rates and effectiveness of antimicrobial interventions across different carcasses were not explicitly modeled. Studies have also shown a considerable variability in lymph node prevalence across different feedlots (Haneklaus et al., 2012). Occasionally, such variability can result in extremely contaminated production lots and ground beef samples (Bosilevac et al., 2009a). Therefore, the variability associated with model parameters will need to be exploited in future modeling effort with available data to improve model accuracy. Explicit modeling of all the sources of variability may also be beneficial for developing models to predict Salmonella prevalence in test portions of ground beef given the estimated CFU load in a production lot. Another limitation is that there were sparse or no published studies to inform some of the model parameters. For example, Salmonella concentration on pre-evisceration carcasses and the prevalence in DTLNs was estimated from a small number of studies. Furthermore, the model is representative of average values from all published studies and does not explicitly account for regional variations in prevalence. Given this, the average results from the model would not directly be applicable for predicting the relative contribution of DTLNs in specific beef production regions. With these caveats in mind, this study nevertheless provides valuable information for prioritizing control measures targeted at Salmonella contamination in ground beef, originating from beef carcass surfaces or DTLNs as modeled on currently available data. Acknowledgments This project is funded in part by the The Beef Checkoff and the USDA National Institutes of Food and Agriculture (Award No. 201151110-31081). The authors are grateful to Drs. Terrance M. Arthur and Joseph M. Bosilevac from USDA/ARS for kindly providing scientific advice and experimental data to define key model parameters, and Drs. Darby Murphy and Sarah E. Strayer from University of Minnesota for contributing to literature search and screening.

434

M. Li et al. / Food Control 50 (2015) 423e434

The author Dr. H. Scott Hurd passed away before the manuscript submission. The rest of the authors acknowledge his significant contributions in designing, conducting, and preparing manuscript of this project. References Arthur, T. M., Bosilevac, J. M., Brichta-Harhay, D. M., Guerini, M. N., Kalchayanand, N., Shackelford, S. D., et al. (2007). Tranportation and lairage envrionment effects on prevalence, numbers, and diverstiy of Escherichia coli O157:H7 on hides and carcasses of beef cattle at processing. Journal of Food Protection, 70, 280e286. Arthur, T. M., Bosilevac, J. M., Brichta-Harhay, D. M., Kalchayanand, N., King, D. A., Shackelford, S. D., et al. (2008). Source tracking of Escherichia coli O157:H7 and Salmonella contamination in the lairage environment at commercial US beef processing plants and identification of an effective intervention. Journal of Food Protection, 71, 1752e1760. Arthur, T. M., Brichta-Harhay, D. M., Bosilevac, J. M., Guerini, M. N., Kalchayanand, N., Wells, J. E., et al. (2008). Prevalence and characterization of Salmonella in bovine lymph nodes potentially destined for use in ground beef. Journal of Food Protection, 71, 1685e1688. Barkocy-Gallagher, G. A., Arthur, T. M., Rivera-Betancourt, M., Nou, X., Shackelford, S. D., Wheeler, T. L., et al. (2003). Seasonal prevalence of shiga toxin-producing Escherichia coli, including O157:H7 and non-O157 serotypes, and Salmonella in commercial beef processing plants. Journal of Food Protection, 66, 1978e1986. Bosilevac, J. M., Arthur, T. M., Bono, J. L., Brichta-Harhay, D. M., Kalchayanand, N., King, D. A., et al. (2009a). Prevalence and characterization of Salmonella in commercial ground beef in the United States. Applied and Environmental Microbiology, 75, 1892e1900. Bosilevac, J. M., Arthur, T. M., Bono, J. L., Brichta-Harhay, D. M., Kalchayanand, N., King, D. A., et al. (2009b). Prevalence and enumeration of Escherichia coli O157: H7 and Salmonella in U.S. abattoirs that process fewer than 1,000 head of cattle per day. Journal of Food Protection, 72, 1272e1278. Brichta-Harhay, D. M., Arthur, T. M., Bosilevac, J. M., Guerini, M. N., Kalchayanand, N., & Koohmaraie, M. (2007). Enumeration of Salmonella and Escherichia coli O157:H7 in ground beef, cattle carcass, hide and faecal samples using direct plating methods. Journal of Applied Microbiolology, 103, 1657e1668. Brichta-Harhay, D. M., Arthur, T. M., Bosilevac, J. M., Kalchayanand, N., Schmidt, J. W., Wang, R., et al. (2012). Microbiological analysis of bovine lymph nodes for the detection of Salmonella enterica. Journal of Food Protection, 75, 854e858. Brichta-Harhay, D. M., Guerini, M. N., Arthur, T. M., Bosilevac, J. M., Kalchayanand, N., Shackelford, S. D., et al. (2008). Salmonella and Escherichia coli O157:H7 contamination on hides and carcasses of cull cattle presented for slaughter in the United States: an evaluation of prevalence and bacterial load by immunomagnetic separation and direct plating methods. Applied and Environmental Microbiology, 74, 6289e6297. Cassin, M. H., Lammerding, A. M., Todd, E. C. 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, 21e44. 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. Centers for Disease Control and Prevention (CDC). (2011). Multistate outbreaks of Salmonella Typhimurium infections linked to ground beef. Retrieved from http:// www.cdc.gov/salmonella/typhimurium-groundbeef/index.html. Centers for Disease Control and Prevention (CDC). (2012). Multistate outbreaks of Salmonella Enteritidis infections linked to ground beef. Retrieved from http:// www.cdc.gov/salmonella/enteritidis-07-12/index.html. Centers for Disease Control and Prevention (CDC). (2013). Multistate outbreaks of Salmonella Typhimurium infections linked to ground beef. Retrieved from http:// www.cdc.gov/salmonella/typhimurium-01-13/. Cummins, E., Nally, P., Butler, F., Duffy, G., & O'Brien,, S. (2008). Development and validation of a probabilistic second-order exposure assessment model for Escherichia coli O157:H7 contamination of beef trimmings from Irish meat plants. Meat Science, 79, 139e154. Cutter, C. N., & Rivera-Betancourt, M. (2000). Interventions for the reduction of Salmonella Typhimurium DT 104 and Non-O157:H7 Enterohemorrhagic Escherichia coli on beef surfaces. Journal of Food Protection, 63, 1326e1332. Delhalle, L., Saegerman, C., Messens, W., Farnir, F., Korsak, N., van der Stede, Y., et al. (2009). Assessing interventions by quantitative risk assessment tools to reduce the risk of human salmonellosis from fresh minced pork meat in Belgium. Journal of Food Protection, 72, 2252e2263. Dodd, C. C., Sanderson, M. W., Jacob, M. E., & Renter, D. G. (2011). Modeling preharvest and harvest interventions for Escherichia coli O157 contamination of beef cattle carcasses. Journal of Food Protection, 74, 1422e1433.

Edrington, T. S., Loneragan, G. H., Hill, J., Genovese, K. J., Brichta-Harhay, D. M., Farrow, R. L., et al. (2013). Development of challenge models to evaluate the efficacy of a vaccine to reduce carriage of Salmonella in peripheral lymph nodes of cattle. Journal of Food Protection, 76, 1259e1263. Edrington, T. S., Loneragan, G. H., Hill, J., Genovese, K. J., He, H., Callaway, T. R., et al. (2013). Development of a transdermal Salmonella challenge model in calves. Journal of Food Protection, 76, 1255e1258. Gill, C. O., Badoni, M., & McGinnis, J. (1999). Assessment of the adequacy of cleaning of equipment used for breaking beef carcasses. International Journal of Food Microbiology, 46, 1e8. Gill, C. O., Bryant, J., & Landers, C. (2003). Identification of critical control points for control of microbiological contamination in processes leading to the production of ground beef at a packing plant. Food Microbiology, 20, 641e650. Gragg, S. E. (2012). Salmonella in the lymph nodes of cattle presented for harvest (Doctoral dissertation). Texas Tech University. Retrieved from http://repositorie s.tdl.org/ttu-ir/handle/2346/50733. Gragg, S. E., Loneragan, G. H., Brashears, M. M., Arthur, T. M., Bosilevac, J. M., Kalchayanand, N., et al. (2013). Cross-sectional study examining Salmonella enterica carriage in subiliac lymph nodes of cull and feedlot cattle at harvest. Foodborne Pathogen and Disease, 10, 368e374. Gragg, S. E., Loneragan, G. H., Nightingale, K. K., Brichta-Harhay, D. M., Ruiz, H., Elder, J. R., et al. (2013). Substantial within-animal diversity of Salmonella isolates from lymph nodes, feces, and hides of cattle at slaughter. Applied and Environmental Microbiology, 79, 4744e4750. Haneklaus, A. N., Harris, K. B., Griffin, D. B., Edrington, T. S., Lucia, L. M., & Savell, J. W. (2012). Salmonella prevalence in bovine lymph nodes differs among feedyards. Journal of Food Protection, 75, 1131e1133. Havelaar, A. H., Evers, E. G., & Nauta, M. J. (2008). Challenges of quantitative microbial risk assessment at EU level. Trends in Food Science & Technology, 19, S26eS33. Hurd, H. S., & Malladi, S. (2012). An outcome model to evaluate risks and benefits of Escherichia coli vaccination in beef cattle. Foodborne Pathogen and Disease, 10, 952e961. Kalchayanand, N., Brichta-Harhay, D. M., Arthur, T. M., Bosilevac, J. M., Guerini, M. N., Wheeler, T. L., et al. (2009). Prevalence rates of Escherichia coli O157:H7 and Salmonella at different sampling sites on cattle hides at a feedlot and processing plant. Journal of Food Protection, 72, 1267e1271. Koohmaraie, M., Scanga, J. A., De la Zerda, M. J., Koohmarie, B., Tapay, L., Beskhlebnaya, V., et al. (2012). Tracking the sources of Salmonella in ground beef produced from nonfed cattle. Journal of Food Protection, 75, 1464e1468. Nutsch, A., Phebus, R., Riemann, M. J., Schafer, D. E., Boyer, J. E., Wilson, R. C., Jr., et al. (1997). Evaluation of a steam pasteurization process in a commercial beef processing facility. Journal of Food Protection, 60, 485e492. Retzlaff, D., Phebus, R., Nutsch, A., Riemann, J., Kastner, C., & Marsden, J. (2004). Effectiveness of a laboratory-scale vertical tower static chamber steam pasteurization unit against Escherichia coli O157:H7 and Salmonella Typhimurium, and Listeria innocua on prerigor beef tissue. Journal of Food Protection, 67, 1630e1633. Rivera-Betancourt, M., Shackelford, S. D., Arthur, T. M., Westmoreland, K. E., Bellinger, G., Rossman, M., et al. (2004). Prevalence of Escherichia coli O157:H7, Listeria monocytogenes, and Salmonella in two geographically distant commercial beef processing plants in the United States. Journal of Food Protection, 67, 295e302. Sargeant, J. M., Amezcua, M. R., Rajic, A., & Waddell, L. (2007). Pre-harvest interventions to reduce the shedding of E. coli O157 in the faeces of weaned domestic ruminants: a systematic review. Zoonoses Public Health, 54, 260e277. Signorini, M., & Tarabla, H. (2009). Quantitative risk assessment for verocytotoxigenic Escherichia coli O157:H7 in ground beef hamburgers in Argentina. International Journal of Food Microbiology, 132, 153e161. Smith, B. A., Fazil, A., & Lammerding, A. M. (2013). A risk assessment model for Escherichia coli O157:H7 in ground beef and beef cuts in Canada: evaluating the effects of interventions. Food Control, 29, 364e381. United States Department of Agriculture, Food Safety Inspection Service (USDA/ FSIS). (2001). Risk assessment of the public health impact of Escherichia coli O157: H7 in ground beef: technical report. Retrieved from http://www.fsis.usda.gov/ OPPDE/rdad/FRPubs/00-023N/00-023NReport.pdf. United States Department of Agriculture, Food Safety Inspection Service (USDA/ FSIS). (2012). Progress report on Salmonella and Campylobacter testing of raw meat and poultry products, 1998e2012. Retrieved from http://www.fsis.usda.go v/wps/wcm/connect/8d792eef-f44d-4ccb-8e25-ef5bdb4c1dc8/Progress-Repo rt-Salmonella-Campylobacter-CY2012.pdf?MOD¼AJPERES. Youssef, M. K., Badoni, M., Yang, X., & Gill, C. O. (2013). Sources of Escherichia coli deposited on beef during breaking of carcasses carrying few E. coli at two packing plants. Food Control, 31, 166e171. Zwietering, M. H. (2009). Quantitative risk assessment: is more complex always better? Simple is not stupid and complex is not always more correct. International Journal of Food Microbiology, 134, 57e62.