A Quantitative Analysis of Cross-Contamination of Salmonella and Campylobacter spp. Via Domestic Kitchen Surfaces

Transcription

1 1892 Journal of Food Protection, Vol. 67, No. 9, 2004, Pages Copyright, International Association for Food Protection A Quantitative Analysis of Cross-Contamination of Salmonella and Campylobacter spp. Via Domestic Kitchen Surfaces H. D. KUSUMANINGRUM, E. D. VAN ASSELT, R. R. BEUMER,* AND M. H. ZWIETERING Laboratory of Food Microbiology, Department of Agrotechnology and Food Sciences, Wageningen University, P.O. Box 8129, 6700 EV Wageningen, The Netherlands MS : Received 29 September 2003/Accepted 16 March 2004 ABSTRACT Epidemiological data indicate that cross-contamination during food preparation in the home contributes noticeably to the occurrence of foodborne diseases. To help prevent such occurrences, the inclusion of a cross-contamination model in exposure assessments would aid in the development and evaluation of interventions used to control the spread of pathogenic bacteria. A quantitative analysis was carried out to estimate the probability of contamination and the levels of Salmonella and Campylobacter spp. on salads as a result of cross-contamination from contaminated chicken carcasses via kitchen surfaces. Data on the prevalence and numbers of these bacteria on retail chicken carcasses and the use of unwashed surfaces to prepare foods were collected from scientific literature. The rates of bacterial transfer were collected from laboratory experiments and literature. A deterministic approach and Monte Carlo simulations that incorporated input parameter distributions were used to estimate the contamination of the product. The results have shown that the probability of Campylobacter spp. contamination on salads is higher than that of Salmonella spp., since both the prevalence and levels of Campylobacter spp. on chicken carcasses are higher than those of Salmonella spp. It is realistic to expect that a fraction of the human exposure to Campylobacter spp., in particular, originates from cross-contamination in private kitchens during food handling. The number of human campylobacteriosis cases could be reduced either by reducing the degree of Campylobacter spp. contamination on chicken carcasses or by improving the hygiene in private kitchens. To eliminate the cross-contamination route, it is important to use separate surfaces or to properly wash the surfaces during the preparation of raw and cooked foods or ready-to-eat foods. Salmonella and Campylobacter spp. are zoonotic pathogens, with many animal species serving as reservoirs. These microorganisms have been the most common cause of human enteritis in western European countries (30) and the United States (21) during the past 5 years. Surveillance data often associate Salmonella- and Campylobacter-related illnesses with the consumption of raw or undercooked contaminated products and with cross-contamination during food preparation (30). To develop strategies for reducing the risk of infection with Salmonella or Campylobacter spp., a number of risk assessment studies have been carried out in poultry meat production chains (1, 6, 7, 20). Risk assessment includes hazard identification, hazard characterization, exposure assessment, and risk characterization. Exposure assessment is the estimation of how likely it is that an individual or a population will be exposed to a microbial hazard and what numbers of the microorganisms are likely to be ingested (16). In the later phase of a risk assessment, these variables are combined with the other estimates, including consumption patterns and dose-response relationships, to estimate the overall probability of illness per serving in a given population (16). Exposure assessment at a consumer level is of particular interest, because it is less controlled than during the other phases in food processing (19) and because, at this * Author for correspondence. Tel: 31 (317) ; Fax: 31 (317) ; rijkelt.beumer@wur.nl. point, the degree to which foods are contaminated is directly related to the public health. Since food handling is an important aspect in the preparation of safe foods, the incorporation of cross-contamination models in exposure assessments at this point will greatly help in the design of risk control interventions. Modeling can facilitate improved estimates and allow a quantification of food safety risks (4). Once the model has been developed, the impact of various control strategies and trends can be simulated (17). The objective of this study was to estimate the probability and level of contamination of Salmonella and Campylobacter spp. on foods as the result of cross-contamination. The probability of illness incurred by consuming the contaminated foods was also predicted. Cross-contamination is defined as the transmission of pathogens from naturally contaminated sources to the finished product. As an example, bacterial transfer from contaminated chicken carcasses via unwashed surfaces to salad vegetables was studied in this paper. Cucumber slices were used as a model for salad vegetables, since they have a flat surface, which increases the reproducibility of experiments. Assuming that the prevalence of Salmonella and Campylobacter spp. on raw products and their transfer rates do not have a single, constant value, they were described by probability density functions. Using these functions, both the variability (random effect of chance) and the uncertainty (lack of precise knowledge) of the variables can be incorporated into the results (16). Monte Carlo simulations were used to estimate

2 J. Food Prot., Vol. 67, No. 9 QUANTITATIVE ANALYSIS OF CROSS-CONTAMINATION 1893 FIGURE 1. Model of cross-contamination from contaminated retail chicken carcasses to salad vegetables via unwashed surfaces. the distribution of the prevalence of contamination and the levels of microorganisms on contaminated salads. This approach was compared to a deterministic calculation, since these calculations can be easily applied and can give a great deal of insight into the most important factors determining the probability of illness. The model obtained will be useful in developing and evaluating strategies for controlling the risks caused by cross-contamination during food preparation. MATERIALS AND METHODS Cross-contamination model. A model of bacterial crosscontamination from retail chicken carcasses to salad vegetables, e.g., cucumber slices, is shown in Figure 1. The prevalence of salad vegetable contamination (Pv) and an estimation of the level of that contamination (Cv) on salads in CFU per square centimeter are determined as follows: Pv P Fu (1) Cv N T 1/100 T 2/100 (2) where P the prevalence of contaminated chicken carcasses at retail ( ); Fu the prevalence of using unwashed surfaces ( ); N the level of microorganisms on contaminated chicken carcasses (CFU/cm 2 carcass); T 1 the transfer rates from chicken carcass to surfaces (CFU/cm 2 surface)/(cfu/cm 2 carcass) 100%; and T 2 the transfer rates from surfaces to salad vegetables (CFU/cm 2 salad vegetables)/(cfu/cm 2 surface) 100%. Data collection. Data on the prevalence and levels of Salmonella and Campylobacter spp. on raw chicken carcasses from retail stores, the prevalence of using unwashed surfaces, and the consumption of vegetables were collected from scientific literature. The transfer rates of Salmonella Enteritidis (phage type 4, chicken product isolate) and Campylobacter jejuni (NCTC 81116) from chicken carcasses to surfaces were collected by laboratory experiments. For this, 5 ml of bacterial cell suspension of approximately 10 6 CFU/ml was spread evenly with a pipette onto an approximately 150-g portion of raw chicken breast meat with skin and held at room temperature for 15 min to facilitate attachment. The levels of microorganisms on these artificially contaminated chicken breast meats were determined by sampling a 5- by 5-cm 2 area with cotton swabs, which thereafter were suspended in 10 ml of peptone saline solution and subsequently enumerated on tryptic soy agar and selective media: MLCB (CM 783, Oxoid, Basingstoke, UK) and Columbia-Blood Agar (CM 331, Oxoid) supplemented with modified Preston supplement (SR 117, Oxoid) as described by Kusumaningrum et al. (15). An unsampled part of the chicken portion was then put on a stainless steel surface (20 by 20 cm 2 ). After 5 min, the chicken portion was removed, and the surface area where the chicken portion had been placed was sampled using RODAC plates. The preparation of cell cultures and dilutions as well as the sampling and enumeration of microorganisms on surfaces were carried out following procedures described in previous work (15). The transfer rate data for Salmonella and Campylobacter spp. from stainless steel surfaces to cucumber slices were collected from previous work (15) and from additional experiments that were carried out using the same procedures. In total, the transfer experiments were performed 10 times with two parallel samples for each experiment. The cucumber slices were sampled by suspending them in 50 ml of sterile peptone saline solution and subsequently homogenizing them in a stomacher for 60 s at 260 rpm. The levels of pathogens were then determined using plate counting. In addition, data regarding the transfer of Enterobacter aerogenes from chicken breast meat via cutting boards to lettuce reported by Chen et al. (2) were included and compared with the results obtained in this study. Data analysis. Data from the literature were compiled and tabulated. Since bacterial contamination levels on chicken carcasses (N) are expressed as CFU per carcass, the numbers were converted to CFU per square centimeter using the following formula: total CFU/cm 2 carcass CFU on carcass/total carcass area (cm 2 ) (3) with CFU carcass CFU/ml ml used to rinse the carcass, and total carcass area (per square centimeter) 0.87w 635 (24), where w the weight of the carcass (grams), which was assumed to be 1,415 g (25). The total bacterial counts of each sample were determined from laboratory experiments, and the appropriate transfer rate was calculated as follows: transfer rate (percent) (CFU/cm 2 on destination)/(cfu/cm 2 on source) 100% (4) Dot plots were generated using Excel software (Microsoft, Redmond, Wash.). Both bacterial counts and percentages of transfer rates were log transformed to obtain normally distributed errors. Statistical analysis was performed using univariate analysis of variance in SPSS software with a significance level of 5% (SPSS Inc., Chicago, Ill.). Estimation of exposure levels per serving. The exposure level per serving of salad (Ce) (CFU) was calculated by multiplying the level of contamination on cucumber slices (Cv) (CFU per square centimeter) by the serving size (square centimeters). The serving size of salad vegetables was generated using the daily vegetable consumption in The Netherlands (8). Because the consumption size was expressed in a weight unit (grams) while the level of contamination on cucumber slices was expressed in CFU per square centimeter, the weight of the serving size was transformed to square centimeters using the following formula: cm 2 serving (w consumption /w slice ) d 2 /4 (5) where w consumption the weight of vegetable consumption (grams); w slice the weight of a cucumber slice with an approximate thickness of 0.3 cm (grams); and d the diameter of the cucumber slice (centimeters). The w slice and d values of the cucumber slices were measured experimentally.

3 1894 KUSUMANINGRUM ET AL. J. Food Prot., Vol. 67, No. 9 Probability distribution functions and Monte Carlo simulation. Literature and experimental data were transformed into appropriate probability distribution functions. Data sets for T 1 and T 2 were fitted to theoretical distributions using Bestfit (@Risk software version 4, Palisade Corporation, Newfield, N.Y.). The accuracy of fit of a distribution was ranked using the Anderson- Darling test, which emphasizes fitting a distribution at the tails as well as at the main body (29). A Monte Carlo simulation with Latin-Hypercube sampling was carried out to simulate the distribution of contamination probabilities and levels of Salmonella and Campylobacter spp. in salad vegetables as a result of cross-contamination (10,000 software, Palisade). Estimation of probability of illness per serving of contaminated salad. The probability of illness per serving of contaminated salad (P dr ) was estimated with two dose-response models. The first model used was the exponential dose-response model (18, 23): P dr 1 e (rce) (6) with r specific constant for the pathogen, and Ce level of exposure (CFU). The second model used was the Beta-Poisson dose-response model (18, 23): P dr 1 (1 Ce/ ) (7) with and dose-response parameters for the pathogen. The probability of illness per serving of salad (P ill ) was estimated by multiplying the P dr value by the probability that salad vegetables are contaminated (Pv). RESULTS Prevalence and levels of Salmonella and Campylobacter on retail chicken carcasses. The presence of Salmonella and Campylobacter spp. was qualitatively found in 4 to 53% and 26 to 83% of retail chicken carcasses, respectively (Table 1). These data were collected from recent studies carried out between 1999 and 2002 so that they would be as relevant as possible. Among these data sets, however, quantitative numbers of Salmonella spp. on chicken carcasses have been indicated in only one study in the UK, in which 0.8% of chicken carcasses were positive with direct counting, although 25% were positive when enrichment procedures were used (12). Countable levels of Campylobacter spp. have been indicated in 41 and 18% of chicken carcasses in the UK (12) and The Netherlands (3), respectively. For further assessment, levels of Salmonella and Campylobacter spp. on chicken carcasses were obtained from Jørgensen et al. (12), since the levels reported in that study were expressed as log CFU per carcass, not as most probable number (MPN) per carcass, as expressed by Dufrenne et al. (3). Although CFU data are preferable to MPN data, MPN data can also be used. Although different types of samples were used by Jørgensen et al. (12), including neck skin, carcass rinse without skin, carcass rinse with whole skin, and carcass rinse with neck skin, in the present study, only the data from their carcass rinse with whole-skin samples were used. For Campylobacter spp., the data set included 91 positives from 101 samples, of which 41 samples of the positives were countable by direct plating. Because the detection limit of the direct counting was 800 CFU per carcass, positive TABLE 1. Prevalence of Salmonella spp. and Campylobacter spp. on retail chicken carcasses Source Salmonella Zhao et al. Van der Zee et al. Dufrenne et al. Jørgensen et al. Uyttendaele et al. Harrison et al. No. of samples (n i ) No. of positives 9 b 34 b 19 c 60 b 45 d 50 b Prevalence Reference F(x) a Total 1, Campylobacter Uyttendaele et al. Van der Zee et al. Dufrenne et al. Zhao et al. Harrison et al. Jørgensen et al. Kramer et al d 81 b 56 c 130 b 73 b 199 b 165 b Total 1, a F(x) is the cumulative probability, with (n i )/(n 1), where n i is the number of the samples taken in each reference, and n is the total number of samples taken for all references. b Positive by enrichment. c Defined as positive if MPN per carcass 10. d Defined as positive if 1 CFU/100 cm 2 or 25 g. samples below this detection limit were assumed to have levels between 0 and 2.89 log CFU per carcass. The CFU values per carcass were transformed to log CFU per square centimeter with equation 3, using an average w value of 1,415 g (25), and are shown in Table 2. Prevalence of using unwashed surfaces. The prevalence of using unwashed surfaces during the preparation of raw and cooked foods or ready-to-eat (RTE) foods is shown in Table 3. On average, 26% of the consumers did not wash the surfaces during the preparation of raw and cooked foods or RTE foods. However, the same studies also indicated that only approximately 60% of the consumers always washed the surfaces during their preparation of raw and RTE foods (10, 11, 13, 28). This indicates that the prevalence of risky practices that could lead to cross-contamination is even higher than that shown by the data in Table 3. Cross-contamination from chicken to foods via surfaces. The transfer rates of Salmonella and Campylobacter spp. from artificially contaminated chicken carcasses to stainless steel surfaces and from these surfaces to cucumber slices are presented in Figures 2 and 3. In addition to these data, the transfer rates of E. aerogenes from artificially contaminated chicken carcasses to plastic cutting boards and subsequently to lettuce from Chen et al. (2) are included. Statistical analyses indicated that the transfer rates of E. aerogenes from chicken carcasses to plastic cutting boards were significantly different (P 0.05) from those of Salmonella and Campylobacter spp. from chicken carcasses to

4 J. Food Prot., Vol. 67, No. 9 QUANTITATIVE ANALYSIS OF CROSS-CONTAMINATION 1895 TABLE 2. Levels of Salmonella and Campylobacter spp. on retail Salmonella- and Campylobacter-positive chicken carcasses a Bacteria Salmonella d Levels (log CFU/ carcass) e Calculated levels b (log CFU/cm 2 ) n F(x) c ( 3.27) ( 0.38) e ( 0.37) Total 60 Campylobacter f e ( 3.27) ( 0.38) e ( 0.37) Total a Generated from Jørgensen et al. (12). b Calculated with formula CFU on carcass/(0.87w 635) (24), with w 1,415 g (25). c F(x) is the cumulative probability, with n i /(n 1), where n i is the number of the observed data point, and n is the sum of the number of data points (no. of samples). d Using 60 positives of 241 samples. e Positive by enrichment but below the detection limit by direct counting ( 800 CFU per carcass). f Using carcass rinse with whole-skin data, with 91 positives of 101 samples. stainless steel surfaces. Assuming that E. aerogenes is an indicator bacterium with attachment characteristics similar to those of Salmonella (2, 33) spp., it is likely that the transfer rates of bacteria from chicken carcasses to plastic cutting boards are higher than those observed for stainless steel surfaces. It was not possible to compare the bacterial transfer rates from stainless steel surfaces or from plastic cutting boards to salad vegetables, because different microorganisms, surfaces, and destinations were involved. Input distributions. Distributions of the prevalence of Salmonella and Campylobacter spp. on retail chicken carcasses were generated from the data sets in Table 1 and are given in Table 4. The prevalence was described by a cumulative distribution assuming a minimum and maximum prevalence of 1 and 60% for Salmonella spp. and a minimum and maximum prevalence of 1 and 90% for Campylobacter spp. Cumulative distributions were also used to describe the levels of bacteria on contaminated chicken carcasses, assuming a minimum and maximum probability of 2.0 and 2.0 log CFU/cm 2 for Salmonella spp. and a minimum and maximum probability of 2.0 and 7.0 log CFU/ cm 2 for Campylobacter spp. Since the data were given in log-level ranges, the mean log level of each range was used for the estimation of the contamination levels. The cumulative distribution was also chosen to describe the data obtained when using unwashed surfaces. Therefore, minimum and maximum prevalences of 1 and 40% were assumed. Normal distributions were selected to describe the logtransformed transfer rates from one surface to another for all test microorganisms because of their adequate goodness of fit and statistical convenience. Normal distributions were usually ranked first or second in goodness of fit using the Anderson-Darling test. When normal distributions ranked second, logistic distributions were the best fit. The normal distributions included in Table 4 indicate the mean values and standard deviations of the log transfer rates. Estimation of exposure levels per serving of salad. The probabilities of salad vegetables contaminated with Salmonella and Campylobacter spp. as the result of crosscontamination are shown in Figure 4. The mean value of the probability of contamination with Salmonella spp. was 4% with a 90% confidence interval of 0.3 to 10%. Contamination with Campylobacter spp. was estimated to occur at a higher percentage than contamination with Salmonella spp., with a mean value of 13% and a 90% confidence interval of 1 to 27%. The level of Salmonella spp. present on salad vegetables as the result of cross-contamination via unwashed surfaces was estimated to be 4.2 log CFU/cm 2 on average (Fig. 5). The level of contamination with Campylobacter spp. present on cucumber slices was estimated to have a mean value of 2.9 log CFU/cm 2 with a 90% confidence interval of 5.2 to 0.5 log CFU/cm 2. This means that 5% of the cucumber slices may be contaminated with Campylobacter spp. at a level of 1 CFU/10 5 cm 2 or less but also that 5% of the cucumber slices may be contaminated with 3.2 CFU/cm 2 or more. The prevalence of E. aerogenes TABLE 3. Prevalence of using unwashed cutting boards during the preparation of raw and cooked or ready-to-eat foods Source No. of samples (n) Unwashed use No. Prevalence ( ) F(x) a Remark Reference Voedingscentrum Worsfold and Griffith Ireland, Food Safety Authority Klontz et al. Jay et al ,000 1,620 1, Use as is/wiped with damp cloths Use dirty surfaces Rarely/never washed Use as is/wiped with damp cloths Use as is/wiped with damp cloths Total 4,076 1,071 a F(x) is the cumulative probability, with n i /(n 1), where n i is the number of the observed data point, and n is the sum of the number of data points (no. of samples).

5 1896 KUSUMANINGRUM ET AL. J. Food Prot., Vol. 67, No. 9 FIGURE 2. The transfer rates of (A) Salmonella spp. and (B) Campylobacter spp. from chicken carcasses to stainless steel surfaces and the transfer rates of (C) E. aerogenes from chicken carcasses to plastic cutting boards (2). contamination could not be estimated because of a lack of data regarding the frequency of contamination for these bacteria on chicken carcasses. Table 5 shows estimates of the probability of contaminated salads (Pv), the levels of pathogens on contaminated salads (Cv) due to cross-contamination, and the estimated dose (Ce) in a contaminated serving. These values are determined with a deterministic approach, using mean values and worst-case values for the various parameters, as well as by Monte Carlo simulation, in which the same equations were used that now included variability in the input parameters. The mean value of the prevalence of salad contamination (Pv) with Salmonella spp. is 4%, and the mean value with Campylobacter is 13% based on Monte Carlo simulation. The estimated level (Cv) for Salmonella spp. is approximately 1 CFU/10 4 cm 2, and that level for Campylobacter spp. is approximately 1 CFU/10 3 cm 2. The deterministic estimates of the prevalence of Salmonella and Campylobacter spp. contamination on salads were 6 and 16%, respectively, both also with levels of 1 CFU/cm 2. In a worst-case scenario, 20% of all preparation events result in the contamination of cucumber slices with Salmonella spp. at a level of 0.04 log CFU/cm 2 ( 1.0 CFU/cm 2 ). The prevalence of contamination with Campylobacter spp. in a worst-case scenario was estimated to be 32% at a level of CFU/cm 2. These results indicate a good comparison between deterministic calculations and Monte Carlo simulations. Slight variations were due to different sampling techniques. To estimate the dose per serving, serving sizes were derived from a Dutch national consumption survey (8). The average daily vegetable consumption was g/day based on a study of persons 43 years old. This consumption was transformed to square centimeters with equation 5. With an average weight of a cucumber slice of g and a diameter of cm, along with the assumption that only one side of the cucumber was in contact with the surface, the calculation resulted in a serving size of cm 2 /day. The mean exposures (Ce) with Salmonella and Campylobacter spp. per serving, both by deterministic calculation and Monte Carlo simulation, were estimated in several cases to be 1 CFU, as indicated by negative log values. These results suggest that, for example, when Pv (prevalence of contaminated salad) is and Ce (level of exposure) is 1.62 log CFU (0.024 CFU) per serving, the actual prevalence of contamination per serving is , meaning that 1 of 750 servings of the salads that are prepared will be contaminated with 1 CFU. Table 6 shows the same parameters (Pv, Cv, and Ce) as in Table 5, but they are now based on actual data (nonlog-transformed data). When data are log transformed, low bacterial counts have the same importance as high bacterial counts (e.g., the average of 5 and 7 log is 6 log). In practice, illnesses are usually caused by incidental high bacterial numbers in a product. Using actual data instead of logtransformed data gives a greater weight to these high bacterial numbers that approach real-life situations (e.g., the average of 10 5 and 10 7 is ). The average con- FIGURE 3. The transfer rates of (A) Salmonella spp. and (B) Campylobacter spp. from stainless steel surfaces to cucumber slices (15) and the transfer rates of (C) E. aerogenes from plastic cutting boards to lettuce (2).

6 J. Food Prot., Vol. 67, No. 9 QUANTITATIVE ANALYSIS OF CROSS-CONTAMINATION 1897 TABLE 4. Input distributions for respective parameters Bacteria Variable Distribution Rank a Salmonella P RiskCumul(0.01,0.60,{0.04,0.14,0.21,0.25,0.34,0.53},{0.21,0.45,0.54,0.77,0.91, 1.00}) N (log) RiskCumul( 2,2,{ 1.82,0.175,1.225},{0.95,0.97,0.98}) b T 1 (log %) RiskNormal(0.171,2) 1 T 2 (log %) RiskNormal(1.458,0.298) 2 Campylobacter P RiskCumul(0.01,0.90,{,0.34,0.63,0.71,0.77,0.83,0.83},{0.11,0.32,0.39,0.55, 0.63,0.83,1.00}) N (log) RiskCumul( 2,7,{ 1.82,0.175,1.225,2.225,3.225,6.225},{0.54,0.64,0.78,0.93, 0.98,0.99}) b T 1 (log %) RiskNormal(0.098,0.606) 2 T 2 (log %) RiskNormal(1.535,0.320) 2 Enterobacter aerogenes T 1 (log %) RiskNormal(1.146,0.452) 1 T 2 (log %) RiskNormal(0.769,0.626) 1 Fu RiskCumul(0.01,0.40,{0.03,0.08,0.18,,0.38},{0.04,0.06,0.31,0.70,1.00}) a Normal distribution rank by the Anderson-Darling test. b Using the mean of the range of log values. tamination level of Campylobacter spp. on chicken was high ( CFU/cm 2 ). This was caused by one sample containing 10 9 CFU per carcass. The other samples were in a range from 800 (detection limit) to 10 7 CFU per carcass. Neglecting the sample with the level 10 9 CFU per carcass, the contamination levels on chicken carcasses were estimated to be CFU/cm 2 (mean 2) and resulted in an estimation for exposure of CFU per serving of salad. Numbers based on actual data are higher than those based on log-transformed data, since the higher values are given more weight than those on the log scale. Estimation of probability of illness per serving of salad. The probability of illness per serving of contami- FIGURE 4. The probability density of the prevalence (Pv) that cucumber slices are contaminated with (A) Salmonella spp. and (B) Campylobacter spp. due to cross-contamination via unwashed surfaces. FIGURE 5. Probability distributions of estimated contamination levels (Cv) of (A) Salmonella spp. and (B) Campylobacter spp. on cucumber slices due to cross-contamination via unwashed surfaces.

7 1898 KUSUMANINGRUM ET AL. J. Food Prot., Vol. 67, No. 9 TABLE 5. Prevalence of contaminated salad (Pv), levels of contamination on salad (Cv), and dose of pathogen (Ce) per serving of salad due to cross-contamination, estimated by deterministic calculation and by mean Monte Carlo simulation using log-transformed data Campylobacter spp. Salmonella spp. Mean Monte Carlo simulation Worst case Mean deterministic Mean Monte Carlo simulation Worst case Mean deterministic Variable/process Unit/formula a a Log CFU/cm 2 Log % Log % Prevalence of contaminated chicken Prevalence using unwashed surfaces Contamination levels on chicken Transfer rates, chicken to surfaces Transfer rates, surfaces to salad Input P Fu Log N Log T 1 Log T c c c c P Fu ( ) Log(10 N 10 T 1/ T 2/100 ) Log(10 Ce serving) b Prevalence of contaminated salad Levels on contaminated salad (log CFU/cm 2 ) Dose of pathogen per serving of salad (log CFU) Output Pv Log Cv Log Ce a Calculated as ( (avg log leveli n i ))/n from Table 2. b Serving 306 cm 2, generated from Hulshof et al. (8) and transformed to square centimeters using equation 5, with the weight of a cucumber slice of g, and the diameter of cm measured experimentally. c Theoretical value. nated salad (P dr ) was estimated using the exponential and Beta-Poisson dose-response relationship, assuming that r (Salmonella) , r (Campylobacter) , (Salmonella) 0.428, (Salmonella) 8,524, (Campylobacter) 0.145, and (Campylobacter) (18, 23). The results of these calculations are given Table 7. The probability of illness per serving of salad (P ill ), estimated by multiplying the P dr value by the prevalence (Pv) of salads contaminated with Salmonella and Campylobacter spp., varies depending on the dose-response model and the data used. For the worst-case scenario, there is no difference between the logtransformed data and the actual data. For the mean case, particularly for the probability of illness caused by Campylobacter spp., there is a considerable difference between the log-transformed data and the actual data. As mentioned above, using the actual data, the numbers are higher, since the higher actual values have relatively more weight when compared with those on the log scale. Overall, the exponential model resulted in lower values of P ill than did the Beta-Poisson model. Using the exponential model in combination with the actual data, for example, the overall probability of illness (mean case) per serving of salad for Salmonella and Campylobacter spp. can occur with (1 of 10 million people) and (1 of 3,000 people), respectively. Using the Beta-Poisson model and the actual data, the proportion of illness caused by Salmonella and Campylobacter spp. is 1 of 300,000 people and 1 of 13 people, respectively. In Table 8, scenarios are presented to examine the effect of certain intervention strategies on the probability of illness per serving of salad due to cross-contamination with Campylobacter spp. Three distinct ways of reducing the probability of illness were analyzed: (i) reducing the prevalence of Campylobacter-positive retail chicken carcasses, (ii) reducing the concentration of Campylobacter spp. on the contaminated chickens, and (iii) improving the use of washed surfaces, clean surfaces, or both during food handling. In mean cases, to obtain a reduction in human cases by, for example, a factor of 10, the prevalence of contaminated chicken carcasses or the level of Campylobacter spp. contamination should be reduced by a factor of 10. A similar reduction in the number of human cases could be obtained by improving the level of food hygiene in private kitchens by a factor of 10. In the worst-case scenario, a 10- fold reduction in the number of human cases could be obtained by reducing the level of Campylobacter spp. on the contaminated chickens by a factor of 4,500 (3.65 log units), by reducing the prevalence of contaminated chicken carcasses, or by decreasing the use of unwashed surfaces by a factor of 10. A large reduction in the level of Campylobacter spp. on chickens is necessary to obtain a 10-factor reduction in the number of human cases in the worst-case scenario and is due to the very high dose (Ce) value, which does not correlate with the linear portion of the dose-response curve. DISCUSSION The development of a model, although simplified, can be helpful when evaluating the relationship between cross-

8 J. Food Prot., Vol. 67, No. 9 QUANTITATIVE ANALYSIS OF CROSS-CONTAMINATION 1899 TABLE 6. Prevalence of contaminated salad (Pv), levels of contamination on salad (Cv), and dose of pathogen (Ce) per serving salad due to cross-contamination, estimated by deterministic approach using actual data (nonlog-transformed data) Campylobacter spp. Salmonella spp. Mean 1 Mean 2 Worst case Mean Worst case Variable/process Unit/formula a,b a a CFU/cm 2 % % Prevalence of contaminated chicken Prevalence using unwashed surfaces Contamination levels on chicken Transfer rates, chicken to surfaces Transfer rates, surfaces to salad Input P Fu N T 1 T P Fu ( ) N T 1 /100 T 2 /100 (CFU/cm 2 ) Cv serving c (CFU) Prevalence of contaminated salad Levels on contaminated salad Dose of pathogen per serving of salad Output Pv Cv Ce a Calculated as ( (avg leveli n i ))/n from Table 2. b Without one sample with a contamination level of 10 9 to CFU per carcass. c Serving 306 cm 2, generated from Hulshof et al. (8) and transformed to square centimeters using equation 5, with the weight of a cucumber slice of g and the diameter of cm, measured experimentally. contamination and the probability of occurrence of related foodborne illness. The results of modeling, however, should be interpreted with care, with consideration given to the origin of the data and the assumptions made. During the development of this exposure assessment model, a number of data gaps were identified. The main data gap concerned the lack of quantitative numbers of contamination levels of Salmonella and Campylobacter spp. on chicken carcasses. Published data generally indicated the prevalence of these bacteria, i.e., the percentage of samples that were positive, but not the number of bacteria present. In a few studies, contamination levels were quantified, but it was evident that the Salmonella spp. numbers on retail chicken carcasses were very low, because the bacteria on the samples were not present in sufficient numbers to be detected by direct counting (3, 27). Furthermore, the distributions for prevalence and levels of contamination on chicken carcasses were based on the analysis of samples at retail points, neglecting transportation to the home and possible storage in home situations. This can possibly lead to an underestimation of the levels of Salmonella spp., especially because these bacteria can grow at room temperature. When data become available that incorporate the effect of transportation and possible storage in the kitchen before food preparation on the contamination levels of chicken carcasses, reality will be more accurately reflected. However, since Campylobacter spp. do not multiply at room temperature (9), the levels of Campylobacter spp. will not be considerably affected by ignoring these effects. In this study, stainless steel surfaces were used to prepare the salads. This could lead to an overestimation of the transfer rates to foods, as a preliminary study indicated that the transfer rate (percent) from stainless steel surfaces to a RODAC plate as a model of salad vegetables was higher than that from plastic cutting boards to a RODAC plate, with an average rate of 25 and 6.3%, respectively. However, the transfer rates of bacteria from chicken carcasses to stainless steel surfaces were likely lower than those from chicken carcasses to plastic boards. When Salmonella spp. and E. aerogenes data are compared (Table 4), the T 1 value for a plastic board is six times higher than that for stainless steel surfaces. These results indicate that in the event of a noncontinuous contact with bacteria (a single contamination), microorganisms are transferred to a stainless steel surface less readily than to a plastic cutting board, but once a stainless steel surface is contaminated, the bacteria are more readily transferred to foods. However, in the experiments with E. aerogenes, chicken was chopped on the cutting board, indicating that force was applied (2). A force of 500 g does not give an increase in transfer (15), but higher forces may. The transfer rates calculated in our study are fail-safe, since it is assumed that bacterial transfer takes place immediately until 15 min after contamination. Longer time intervals should result in reductions in the surviving numbers (15). This study indicates that the probability of illness caused by Campylobacter spp. per serving of cucumber salad due to cross-contamination via surfaces was higher than the probability of illness caused by Salmonella spp., since

9 1900 KUSUMANINGRUM ET AL. J. Food Prot., Vol. 67, No. 9 TABLE 7. Probability of illness (P ill ) per serving of salad due to cross-contamination during preparation, estimated by deterministic calculation using log-transformed and actual data Salmonella spp. Campylobacter spp. Variable/process Formula Mean Worst case Mean 1 Mean 2 Worst case Using log-transformed data Pv Prevalence of contaminated salad ( ) Log Ce Dose of pathogen per serving of salad P dr Probability of illness per serving of contaminated salad P ill Probability of illness per serving of salad (log CFU) 1 e (r 10 Ce ) a 1 ((1 10 Ce / ) ) b P dr Pv c P dr Pv d Using actual data (nonlog-transformed data) Pv Prevalence of contaminated salad ( ) Ce Dose of pathogen per serving of salad P dr Probability of illness per serving of contaminated salad P ill Probability of illness per serving of salad (CFU) 1 e (r Ce)a 1 ((1 10 Ce / ) ) b P dr Pv c P dr Pv d a Estimated using the exponential model, with r(salmonella) ; r (Campylobacter) (18, 23). b Estimated using the Beta-Poisson model, with (Salmonella) 0.428, (Salmonella) 8,524, (Campylobacter) 0.145, and (Campylobacter) (18, 23). c Estimated using the Pdr exponential model. d Estimated using the Pdr Beta-Poisson model.

10 J. Food Prot., Vol. 67, No. 9 QUANTITATIVE ANALYSIS OF CROSS-CONTAMINATION 1901 TABLE 8. Scenarios to examine the effect of intervention strategies on the probability of illness per serving of salad due to cross-contamination with Campylobacter spp. using the actual data Mean case Worst case Process/variable Unit/formula Current value Scenario 1 Scenario 2 Scenario 3 Current value Scenario 1 Scenario 2 Scenario 3 Input P Prevalence of contaminated chicken Fu Prevalence using unwashed surfaces N Contamination levels on chicken CFU/cm T 1 Transfer rates, chicken to surfaces % T 2 Transfer rates, surfaces to salad % Output Pv Prevalence of contaminated salad P Fu( ) Cv Estimated levels on contaminated N T 1 /100 T 2 /100 (CFU/cm 2 ) salad Ce Dose of pathogen per serving salad Cv serving a (CFU) P dr Probability of illness per serving of contaminated salad P ill Probability of illness per serving of salad 1 e (rce)b ( ) P dr Pv ( ) a Serving 306 cm 2, generated from Hulshof et al. (8) and transformed to square centimeters using equation 5, with the weight of a cucumber slice of g and the diameter of cm, measured by experiments. b Estimated using the exponential model with r(campylobacter) (18, 23). Scenario 1: reduction of prevalence of carcasses positive; scenario 2: reduction of contamination levels on carcasses; scenario 3: reduction of the use of unwashed surfaces.

11 1902 KUSUMANINGRUM ET AL. J. Food Prot., Vol. 67, No. 9 both the prevalence and contamination levels of Campylobacter spp. on chicken carcasses were higher. The use of washed surfaces, separate surfaces, or both will cut this cross-contamination route of bacterial contamination to RTE foods and avoid the occurrence of food infection. The two dose-response models that were used for estimating the probability of illness resulted in considerably different values. The Beta-Poisson model is believed to give a better fit with experimental data for infection with Campylobacter and Salmonella spp. in human volunteers than is the exponential model (18, 23). Using the Beta- Poisson model in this study, however, generally resulted in a higher probability of illness (e.g., 1:13 for Campylobacter spp.) when compared with the results of the exponential model (e.g., 1:3,000 for Campylobacter spp.), which seems to overestimate the occurrence of illness. The estimated probability of illness was also dependent on the parameters (r,, and ) used. Rose and Gerba (22) reported that the parameters of the Beta-Poisson dose-response relationship for Campylobacter spp. were and 55 and that the parameters for Salmonella spp. were 0.33 and Using these parameters, the estimated probability of illness caused by Campylobacter spp. is 10 times lower and that by Salmonella spp. is 50 times higher than the estimated probabilities that result when using the parameters of Teunis et al. (23). This result indicates that, in general, quantitative microbial risk assessment is hampered by different sources of uncertainty, making it difficult to estimate the absolute value of risk. In this case, the choice of the specific dose-response model or its parameters largely changes the magnitude of risk. The analysis does, however, allow the analyst to compare strategies that can be used to minimize the probability of illness, i.e., the relative change. Furthermore, it clearly shows the data that require more detailed collection in order to improve the prediction. This study illustrates how cross-contamination during food preparation in domestic kitchens can be modeled by linking currently available data with experimental data and shows that it is realistic to expect that a fraction of the human exposure to Campylobacter spp., in particular, originates from cross-contamination in private kitchens during food handling. Since consumers do not know the degree of contamination on the chicken carcasses that enter their homes, it is obviously important to use separate surfaces or to wash the surfaces during the preparation of raw and cooked foods or RTE foods. The model needs to be validated for strains and products other than those used in this study. The quantitative exposure assessment procedure is not static, since the data, assumptions, and models used may be changed when new information becomes available. Furthermore, the models can be used in a more extensive microbiological risk assessment to assess the influence of cross-contamination on foodborne disease. REFERENCES 1. Brown, M. H., K. W. Davies, C. M. P. Billon, C. Adair, and P. J. McClure Quantitative microbiological risk assessment: principles applied to determining the comparative risk of salmonellosis from chicken products. J. Food Prot. 61: Chen, Y. H., K. M. Jackson, F. P. Chea, and D. W. Schaffner Quantification and variability analysis of bacterial cross-contamination rates in common food service tasks. J. Food Prot. 64: Dufrenne, J., W. Ritmeester, E. Delfgou-van Asch, F. van Leusden, and R. de Jonge Quantification of the contamination of chicken and chicken products in the Netherlands with Salmonella and Campylobacter. J. Food Prot. 64: Foegeding, P. M Driving predictive modelling on a risk assessment path for enhanced food safety. Int. J. Food Microbiol. 36: Harrison, W. A., C. J. Griffith, D. Tennant, and A. C. Peters Incidence of Campylobacter and Salmonella isolated from retail chicken and associated packaging in South Wales. Lett. Appl. Microbiol. 33: Hartnett, E., L. Kelly, D. Newell, M. Wooldridge, and G. Gettinby A quantitative risk assessment for the occurrence of campylobacter in chickens at the point of slaughter. Epidemiol. Infect. 127: Hartnett, E., L. A. Kelly, G. Gettinby, and M. Wooldridge A quantitative risk assessment for campylobacters in broilers: work in progress. Int. Biodeterior. Biodegrad. 50: Hulshof, K., J. H. Brussaard, A. G. Kruizinga, J. Telman, and M. R. H. Lowik Socio-economic status, dietary intake and 10 y trends: the Dutch National Food Consumption Survey. Eur. J. Clin. Nutr. 57: International Commission on Microbiological Specifications for Foods Microorganisms in foods: microbiological specifications of food pathogens. Blackie A&P, London. 10. Ireland, Food Safety Authority Public knowledge and attitudes to food safety in Ireland. Report and Evaluation Services, Dublin. 11. Jay, L. S., D. Comar, and L. D. Govenlock A national Australian food safety telephone survey. J. Food Prot. 62: Jørgensen, F., R. Bailey, S. Williams, P. Henderson, D. R. A. Wareing, F. J. Bolton, J. A. Frost, L. Ward, and T. J. Humphrey Prevalence and numbers of Salmonella and Campylobacter spp. on raw, whole chickens in relation to sampling methods. Int. J. Food Microbiol. 76: Klontz, K. C., B. Timbo, S. Fein, and A. Levy Prevalence of selected food-consumption and preparation behaviors associated with increased risks of food-borne disease. J. Food Prot. 58: Kramer, J. M., J. A. Frost, F. J. Bolton, and D. R. A. Wareing Campylobacter contamination of raw meat and poultry at retail sale: identification of multiple types and comparison with isolates from human infection. J. Food Prot. 63: Kusumaningrum, H. D., G. Riboldi, W. C. Hazeleger, and R. R. Beumer Survival of foodborne pathogens on stainless surfaces and cross-contamination to foods. Int. J. Food Microbiol. 85: Lammerding, A. M., and A. Fazil Hazard identification and exposure assessment for microbial food safety risk assessment. Int. J. Food Microbiol. 58: Lammerding, A. M., and G. M. Paoli Quantitative risk assessment: an emerging tool for emerging foodborne pathogens. Emerg. Infect. Dis. 3: Medema, G. J., P. F. M. Teunis, A. H. Havelaar, and C. N. Haas Assessment of the dose-response relationship of Campylobacter jejuni. Int. J. Food Microbiol. 30: Nauta, M. J., S. Litman, G. C. Barker, and F. Carlin A retail and consumer phase model for exposure assessment of Bacillus cereus. Int. J. Food Microbiol. 83: Nauta, M. J., A. W. Van de Giessen, and A. M. Henken A model for evaluating intervention strategies to control Salmonella in the poultry meat production chain. Epidemiol. Infect. 124: Olsen, S. J., L. C. MacKinon, J. S. Goulding, N. H. Bean, and L. Slutsker. 17 March Surveillance for foodborne disease outbreaks United States, MMWR Surveillance Summaries

12 J. Food Prot., Vol. 67, No. 9 QUANTITATIVE ANALYSIS OF CROSS-CONTAMINATION (SS01). Available at: mmwrhtml/ss4901a1.htm. 22. Rose, J. B., and C. P. Gerba Use of risk assessment for development of microbial standards. Water Sci. Technol. 24: Teunis, P. F. M., O. G. Heijden, J. W. B. Van der Giessen, and A. H. Havelaar The dose-response relation in human volunteers for gastro-intestinal pathogens. RIVM report, National Institute of Public Health and the Environment, Bilthoven, The Netherlands. 24. U.S. Department of Agriculture Nationwide broiler chicken microbiological baseline data collection program, July 1994 June Science and Technology, Microbiological Division, U.S. Department of Agriculture, Food Safety and Inspection Service (USDA-FSIS), Washington, D.C. 25. U.S. Department of Agriculture AMS Poultry programs. Available at: chicken%20catalog.pdf. Accessed June Uyttendaele, M., P. De Troy, and J. Debevere Incidence of Salmonella, Campylobacter jejuni, Campylobacter coli, and Listeria monocytogenes in poultry carcasses and different types of poultry products for sale on the Belgian retail market. J. Food Prot. 62: Van der Zee, H., B. Wit, and E. De Boer Salmonella and Campylobacter in kip en kipproducten in De Ware(n)-Chemicus 1: Voedingscentrum Hygiene prive-huishouding. Research International, The Hague, The Netherlands. 29. Vose, D Risk analysis: a quantitative guide to Monte Carlo simulation modelling, 2nd ed. John Wiley & Sons, New York. 30. World Health Organization WHO surveillance programme for control of foodborne infections and intoxications in Europe, 7th report, WHO, Berlin. 31. Worsfold, D., and C. J. Griffith Assessment of the standard of consumer food safety behavior. J. Food Prot. 60: Zhao, C. W., B. L. Ge, J. De Villena, R. Studler, E. Yeh, S. H. Zhao, D. G. White, D. Wagner, and J. H. Meng Prevalence of Campylobacter spp., Escherichia coli, and Salmonella serovars in retail chicken, turkey, pork, and beef from the Greater Washington, DC, area. Appl. Environ. Microbiol. 67: Zhao, P., T. Zhao, M. P. Doyle, J. R. Rubino, and J. Meng Development of a model for evaluation of microbial cross-contamination in the kitchen. J. Food Prot. 61: