Using meta-analysis to underpin a risk assessment model for the estimation of prevalence of Salmonella spp. on pork joints produced in Irish slaughterhouses Ursula Gonzales-Barron 1, Ilias Soumpasis 1, Francis Butler 1 & Geraldine Duffy 2 1 UCD School of Agriculture, Food Sci. & Vet. Med. 2 Ashtown Food Research Centre, Teagasc
Introduction Foodborne salmonellosis is a major public health issue and requires concerted efforts to control the pathogen in the food supply. Pork is one of the main sources for human salmonellosis (5-30% of human cases). The primary source of Salmonella in the whole pork production chain is the Salmonellainfected animal.
...Introduction In the slaughter process, contamination of 30% of Salmonella-positive carcasses arises from cross-contamination of other infected pigs in the slaughterhouse. Numerous researchers have observed that there is a strong association between the proportion of sub-clinically infected pigs entering the slaughter lines (carrying or excreting Salmonella) and the proportion of contaminated carcasses at the point of evisceration.
Objective To estimate the prevalence of Salmonella on pig carcasses and pork joints produced in Ireland using quantitative risk assessment techniques. To this effect, a stochastic relationship between Salmonella prevalence in pigs caeca and Salmonella prevalence on eviscerated carcasses was built taking into account sensitivities. Validation of results: Parallel study on the incidence of Salmonella on pork oyster cuts (n=720) produced in the boning halls of commercial pork abattoirs of Ireland.
Proportion of resulting Salmonellapositive eviscerated carcasses Methodology Prevalence in caecal contents Stunning, killing and bleeding Scalding Weighted regression 0.35 0.30 0.25 0.20 0.15 0.10 0.05 r 2 = 0.77 Dehairing, singe, and polishing 0.00 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Proportion of slaughter pigs carrying Salmonella in caeca Evisceration Splitting and trimming Final washing Chilling Meta-analysis Jointing
Proportion of slaughter pigs carrying Salmonella entering the slaughter lines (x ) - x = s C / n C = positive caecal samples / total number of samples - Correcting x for culture sensitivities (Se) of Salmonella x Source Proportion + caecal samples (x) Culture Culture Se Estimated protocol Se Corrected Prop+ caec samples (x ) Duggan (2008) Sorensen (2004) Kranker (2003) Quirke (2001) Davies (1999) Morgan (1987) Oosterom (1985) 87/193 216/1658 22/122 61/419 256/2205 71/149 35/145 28/151 44/220 PCR-MSRV NMKL NMKL RV+BGA DIASSALM RV+XLD TB+BGPRA 0.950 0.450 0.450 0.950 0.892 0.710 0.710 0.710 0.860 0.880 0.450 0.450 0.460 0.892 0.710 0.710 0.710 0.797 99/193 480/1658 31/122 133/419 287/2205 100/149 49/145 39/151 55/210
Proportion of slaughter pigs carrying Salmonella - Se values of culture protocols were relative to faecal contents and NOT to caecal contents in pigs. - x cannot be regarded as true prevalence. - Corrections provided a reliable metric to equalise the studies minimising the level of between-study heterogeneity.
Proportion of resulting Salmonella positive carcasses at the point of evisceration (x ) - y = s S / n S = positive carcass swabs / total number of carcass swabs. - Se of carcass swabs for every protocol could not be found in the literature. - Malorny et al. (2003): PCR method in relation to traditional culture methods for pig carcass swabs (Se=0.985). - Reasonable to correct s S in terms of PCR. Source Proportion + carcass (y) Relative Se Swab extent (cm 2 ) Estimated protocol Se Corrected Prop+ carcass (y ) Duggan (2008) Sorensen (2004) Kranker (2003) Quirke (2001) Davies (1999) Morgan (1987) 29/191 159/1665 6/117 42/419 155/2211 41/150 1 0.985 0.985 0.985 0.985 0.985 1000 1400 300 Entire 1000 1000 1 0.985 0.550 0.985 0.985 0.985 29/191 161/1665 11/117 43/419 157/2211 42/150 Oosterom (1985) 19/148 14/150 27/210 0.985 0.985 0.985 1000 0.985 0.985 0.985 19/148 14/150 27/210
Adding weights to individual studies - (x, y ) data pairs have been extracted from different studies presenting different degrees of precision. - In meta-analysis, a common method of weighting individual estimates is by their inverse variances. - Effect size parameter: RR se( RR ) i i 2 se 1 RR n i S ' ss 2 n s S ' S n C ' sc n s C ' C Source Duggan (2008) Sorensen (2004) Kranker (2003) Quirke (2001) Davies (1999) Morgan (1987) Oosterom (1985) Log-relative risk (log RR) -1.217-1.097-0.994-1.129-0.606-0.874-0.968-1.018-0.712 se(rr) 0.184 0.084 0.326 0.161 0.094 0.143 0.244 0.289 0.214 Weight (ω i ) 29.27 141.03 9.39 38.45 111.75 48.94 16.83 11.94 21.88 Fixed effects meta-analysis for RR, U test =368 P<0.001
Stochastic weighted linear regression - Non-parametric Bootstrap technique added uncertainty to m, c, and s. - B Bootstraps samples {(x 1, y 1)*, (x 2, y 2)*, (x 9, y 9)*} are created, where each (x i, y i)* is a random sample with replacement from {(x 1, y 1), (x 2, y 2), (x 9, y 9)}. - For every Bootstrap sample, a weighted linear regression was fitted to the data using ω i m B, c B and s B were calculated. - 50 000 Bootstrap samples were taken. - Distributions were fitted for m, c and s.
Stochastic weighted linear regression - m average 0.0193, standard deviation 0.0435 - c average 0.2909, standard deviation 0.1344 - s average 0.0502, standard deviation 0.0193 - Having defined an estimate of prevalence of Salmonella-carrier slaughter pigs (Pc), the Salmonella prevalence on eviscerated pig carcasses (Pev) is approximated as: Pev Normal m Pc c, s
Effect of splitting and trimming Alban and Stark (2005) Prev. increase of 16% Davies et al. (1999) Incidence increase in 50% Csp Pert 0,0.16,0.50 Psp 1 Csp Pev Effect of final rinsing Davies et al. (1999): Salmonella-positive results from two abattoirs before final rinsing (15/75) and after final rinsing (9/79). Rfr Beta 9 Beta 15 1, 79 1, 75 9 15 1 1 Pfr Pev Rfr
Effect of chilling A reduction effect of chilling on the recovery of Salmonella from pork carcasses has been observed by many researchers. Thus, data on Salmonella prevalence on pig carcasses before chilling and after chilling were taken from 9 published studies. Parametric meta-analysis was conducted Effect size measured was the relative risk of chilling. Study 1 Study 2 Study 3 Study 4 Study 5 Study 6 Study 7 Study 8 Study 9 Fixed -5-4 -3-2 -1 0 1 2 Log relative risk (log p T /p C )
Probability density...effect of chilling RR for chilling = Probability of encountering Salmonellapositive carcasses after chilling relative to the probability of encountering Salmonella-positive carcasses before chilling. The distribution of the reduction factor for the overall effect of the chilling operation (Rch) was therefore approximated by meta-analysis conducted on RR. 7 X <= 0.303 2.5% X <= 0.581 97.5% 6 5 Rch Normal e 0.868,0.166 4 3 2 1 Pch Rch Pfr 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Reduction in detected Salmonella prevalence due to chilling (Rch)
Increase in contamination in boning halls Berends et al. (1998): AP=0.67 of inadequate disinfection in cross contamination of pork with Salmonella spp. during the first 2 hours of production when the risk factor prevails, it provokes about two thirds of the total crosscontamination during the first production hours. Based on survey of Salmonella presence in Irish boning halls, the probability of inadequate cleaning and disinfection was defined as, Pert 0, 0.20, 0.30 p icd According to Berends et al. (1998), disinfection takes place 1-4 times a day (Np=Discrete (1,2,3,4)). The probability that disinfection is poorly performed at least once a day (p icd ) is ' Np p 1 1 Pert 0, 0.20, 0. 30 icd
Increase in contamination in boning halls Contribution of inadequate cleaning and disinfection (C icd ) on a particular day with respect to the cross-contamination that occurs during the first two production hours is C ' icd AP p icd With respect to all cross-contamination that occurs during a full working day of 8 h., the contribution of inadequate cleaning and disinfection (C icd ) is ' C icd C icd ~C icd% Thus, the prevalence of Salmonella in pork joints (Pj), which is the final model output, was estimated as 2 8 ~(1-C icd)% Pj Pch ' 1 Cicd
Model validation using Irish data The model s ability to produce accurate estimates, and intrinsically the effectiveness of the modeling capabilities of meta-analysis, were appraised using Irish data for the input parameter of prevalence of Salmonella-carrier slaughter pigs (Pc). Pc for Ireland was estimated in the following way: Source x=s C /n C Culture Culture protocol Se x =s C /n C Duggan et al. 87/193 PCR-MSRV 0.880 107/193 Quirke et al. 61/419 RV+BGA 0.460 133/419 UCD study 85/471 TB+BGPRA 0.797 99/471 POOLED DATA 339/1098 Pc=Beta(339+1,1098-339+1)
...Model validation using Irish data - The model s output (Pj) was compared to the results of an extensive survey of Salmonella incidence in pork joints produced in the boning halls of four representative Irish abattoirs. - This separate survey study of Salmonella prevalence and counts on pork joints covered an aspect within the same research project, and it is explained in detail in Prendergast et al. (2008). - The risk assessment model was developed in Microsoft Excel using the @Risk add-in (Industrial Edition 4.5.2, Palisade, NY), and run for 20 000 iterations using Latin Hypercube sampling. The weighted regression for 50 000 Bootstrap samples was performed in Matlab 7.1 (The Mathworks Inc, Natick, MA).
Proportion of resulting Salmonellapositive eviscerated carcasses Results and Discussion - The principal sources of microbial contamination during the processing of pigs: - Contamination from carrier animals (sub-clinically infected) - Cross-contamination from equipment, machinery inadequately disinfected from previous batches, staff practices. 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 r 2 = 0.77 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Proportion of slaughter pigs carrying Salmonella in caeca - Batches of Salmonella-positive slaughter pigs entering the abattoir impart a force of contamination during processing, that explains ~75-80% of the total contaminated carcasses at the beginning of the clean line.
Results and Discussion - The pool of Irish surveys on Salmonella prevalence in caecal contents of slaughter pigs (Pc) led to an estimate of 31.3% (95% CI: 28.6 34.1%) which is regarded to be a value close to the true prevalence. Observed incidence (underestimated values) - Cross-sectional study at four pig abattoirs in Northern Ireland, 31.4% (161/513) of slaughter pigs tested positive for Salmonella in caecal culture. - UK national survey 23% (578/2509) of the caecal samples were Salmonella positive in 34 pig abattoirs. - France national survey 24.8% (256/1030) of caecal samples tested positive for Salmonella in 18 pig abattoirs. - Denmark and The Netherlands have reported significantly lower incidence of Salmonella-carrier slaughter pigs (8.0% and 8.5%) severity of the application of national programmes for Salmonella control at farm level?
Cumulative probability Results and Discussion - Prevalence of Salmonella on eviscerated pig carcasses in Ireland. Pev=11.9%, 95%CI: 0.7-26.8% Observed incidence - Pearce et al. (2004) recovered 10% positive swabs after evisceration from an Irish abattoir. - In the Spanish region of Catalonia (Creus, 2007), Salmonella incidence found on pig carcasses after evisceration was 11.1% (40/360, 95%CI of 8.1-14.9%) sampled in three large slaughterhouses. 1.0 0.8 0.6 0.4 0.2 0.0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 Prevalence of Salmonella on eviscerated pig carcasses
Results and Discussion - After splitting and trimming, reduction in Salmonella levels may be related to the removal of residual contamination during the spraying of the carcasses, performed in order to remove bone and blood clots. - Thus, Salmonella prevalence on finished carcasses (after final rinsing and before chilling) was estimated to be Pfr=8.8%, 95%CI: 0.5-24.6% Observed incidence (underestimated values) - While figures cannot be directly comparable since they have not been corrected for culture protocol sensitivities nor are methods fully harmonised among countries, according to the last EFSA report, the average Salmonella incidence on carcass swabs detected among European slaughterhouses in 2006-2007 was 8.3% (95% CI: 6.3-11.0%).
Probability density Results and Discussion 20 X <= 0.003 2.5% X <= 0.115 97.5% 15 10 5 0 Mean = 0.040 Mean = 0.033 Simulation's output Fitted distribution Validation results 0.00 0.04 0.08 0.12 0.16 0.20 0.24 Prevalence of Salmonella in pork joints produced in Ireland Irish survey: Sampling pork cuts in boning halls Mean: 3.33% (24/720) 95% CI: 2.02-4.64% Output distribution of Salmonella prevalence in pork cuts produced in Ireland
Probability density Results and Discussion 35 30 25 20 15 10 5 0 Mean = 0.034 Mean = 0.040 Stage-by-stage model Validation results Regressional model 0.00 0.05 0.10 0.15 0.20 Prevalence of Salmonella in pork joints produced in Ireland Stage-by-stage model -No meta-analysis -No stochastic weighted regression - Contamination and decontamination factors were used for each stage from bleeding to jointing Distribution of Salmonella prevalence in pork joints in Ireland as an output of the regressional model in contrast to a previously-developed stage-by-stage model
Variables Results and Discussion Rfr Rch Csp Np p icd -0.382-0.221 0.106-0.030 0.019-0.6-0.4-0.2 0.0 0.2 0.4 0.6 Correlation Coefficients Key parameters having more influence on model s output. Reaffirmation that final rinsing and chilling are efficient at controlling carcass contamination CCP Expected that final rinsing with hot water would further decrease Salmonella prevalence
Conclusions A risk assessment model was built on the assumption that the occurrence of Salmonellainfected carcasses post-evisceration is proportional to the total contamination introduced by the carriers animals themselves entering the slaughter lines. The model output for the Salmonella prevalence on pork joints produced in boning halls was successfully validated with the results of an extensive survey carried out in four large Irish abattoirs.
...Conclusions The suitability of meta-analysis for integrating different findings and producing distributions for use in stochastic modelling was demonstrated. Final rinsing and chilling had strong impact on the prevalence of Salmonella on pork joints. According to this model, interim cleaning and disinfection in the boning halls has only a marginal effect on diminishing the amount of contaminated pork joints produced. as long as contaminated carcasses are being processed, about 90% of the cross-contamination is unavoidable.
Future Work This model only considers the contamination force inflicted by carriers animal. Future models should incorporate parameters for the cross-contamination from equipment, inadequate disinfection from previous batch, etc. Data gaps: Lack of research on presence of Salmonella on the skin of pigs entering the abattoir; Insufficient experiments for modelling the scalding process; Sensitivity values cannot be found for all culture protocols.
Acknowledgments SafeFood, QPorkChains and the Irish Department of Agriculture, Fisheries and Food
Thanks for your attention