- Email: [email protected]
A process risk model for the shelf life of Atlantic salmon fillets
International Journal of Food Microbiology 73 (2002) 47 – 60 www.elsevier.com/locate/ijfoodmicro
A process risk model for the shelf life of Atlantic salmon fillets S.K.J. Rasmussen *, T. Ross, J. Olley, T. McMeekin School of Agricultural Science, University of Tasmania, GPO Box 252-54, Hobart Tasmania 7000, Australia Received 15 January 2001; received in revised form 23 May 2001; accepted 20 September 2001
Abstract The shelf life of Atlantic salmon (Salmo salar) portions produced for retail distribution is examined and the dominant aerobic spoilage organism is identified. Characterization of the harvesting and processing operations allow the development of a stochastic mathematical model, a process risk model (PRM), which predicts the range of the possible shelf life for the portions under normal retail and distribution. The considered risk is the failure to achieve the nominal ‘use by’ date. Bacterial counts from surface swabs, water, ice, and fish samples, collected over a period of 9 months, are fitted to distribution functions for use within the model. Comparisons are made between the distributions fitted to the observed bacterial levels and the predicted levels for the slurry water, initial surface contamination on the fish, and for the predicted and observed shelf life. Storage temperature of the packaged salmon portions has the greatest influence on shelf life, with contamination from contact surfaces and other sources being the next most important. The range of bacterial counts on the portions was between 0.6 and 5 log10 cfu/cm2. The model predicts bacterial counts in the slurry water to have an average value of 3.36 log10 cfu/ml, whereas the observed slurry water bacterial counts were 3.35 log10 cfu/ml. The predicted average initial bacterial contamination is 3.31 log10 cfu/cm2 on the fish surface and 3.23 log10 cfu/cm2 on the observed. The average predicted shelf life is 6.5 days, compared to an observed value of 6.2 days at 4 °C. D 2002 Elsevier Science B.V. All rights reserved. Keywords: Process risk model; Shelf life; Atlantic salmon
1. Introduction Fish quality, microbiology, and shelf life have been studied extensively for the last seven decades. This is the first reported effort to use risk assessment techniques to predict the shelf life of any seafood product from catching to retail from one factory. Process risk models (PRM) have been applied in risk assessments concerned with public health outcomes (Cassin et al.,
*
Corresponding author. Tel.: +61-3-62-26-1831, +61414693459 (mobile); fax: +61-3-62-26-2642. E-mail address: [email protected] (S.K.J. Rasmussen).
1998). Fish are recognised as being highly perishable, having a relatively short shelf life (Fey and Regenstein, 1982). It is recommended that fish should be chilled as quickly as possible after harvest and kept on ice until consumed (Liston, 1992; Sumner et al., 1984). The concept of temperature control as a means to prolong shelf life is well understood, and the response of bacteria to temperature has been widely studied (Olley and Ratkowsky, 1973; Ratkowsky et al., 1982, 1983; Bremner et al., 1987; McMeekin et al., 1991). Accurately defining shelf life is important to both the consumer and the producer but with any processing, storage, and biological system, there is variability.
0168-1605/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 8 - 1 6 0 5 ( 0 1 ) 0 0 6 8 7 - 0
48
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
This variability can have a large influence on the shelf life of the product. Rather than the shelf life being a single point value, there is a range of possible values that depends on variations in the production process, initial bacterial contamination, and the storage conditions. Storage trials allow the shelf life of a product to be determined reasonably accurately (McMeekin and Olley, 1986). With the concurrent use of stochastic modelling, the range of possible shelf life estimates of the product can be elucidated. Production improvements may then be defined to reduce contamination loads on the salmon fillets and increase the shelf life of the product. The application of stochastic modelling techniques has not been previously used for evaluation and prediction of shelf life. This paper describes the development of a process risk model, stating assumptions and explaining the modelling process. Validation of the model is also described. A background to the philosophy of Monte Carlo risk assessment modelling may be found in Cassin et al. (1998), and further use of computer-aided techniques that may be used for the seafood may be found in Nielsen et al. (1991).
2. Materials and methods 2.1. Sample collection Water and fish: Atlantic salmon (Salmo salar) samples were collected from the harvesting and the processing factories. The fish samples collected from the harvesting factory were whole, head on, gilled and gutted (HOGG) salmon, and were kept on ice until processing at the laboratory. At the harvesting factory, the salmon are weighed, sorted, and placed into large insulated plastic containers (slurry bins). Slurry bins have a holding volume of 1000 l, and are filled with water from bulk water storage tanks at the harvesting factory, ice, and fish. Fresh water was used throughout the processing of the salmon. On two occasions during the processing, the slurry bins are spiked with chlorine, i.e. a solution of sodium hypochlorite is added, with an effective dose of 30 mg/l of chlorine. Samples taken at the harvesting factory included water, fish, ice, and surface swabs from the processing line. In situations where samples were taken from ‘‘working surfaces’’, this would refer to sampling
done at the harvesting factory, and when ‘‘production line’’ is used, this would refer to sampling from the processing factory. Water samples collected from the slurry bins after they had been spiked had the active chlorine neutralised by the addition of a fixed amount of sodium thiosulphate (Na2S2O3, Analar, Poole, England), following the standard method (Clesceri et al., 1989). 2.2. Bacterial population development on salmon portions Retail packages of Atlantic salmon portions obtained from a local producer were incubated at 4 °C for up to 11 days. The portions were incubated aerobically in stomacher bags (Bio-Service, Huntingdale, Victoria) after aseptic transfer. Microbiological sampling and enumeration was undertaken throughout the storage period to determine the lag phase and the growth rate of the spoilage microbiota. Shelf life was defined as the time taken for the bacterial population to reach 107 cfu/g. Total viable counts on the portions were determined by aseptically excising a 16cm2 area of flesh from the surface of the portion, using a sterile 4 4-cm stainless steel template. Alternatively, 10 g of flesh was added to 90 ml of diluent (0.9% NaCl with 0.1% bacteriological peptone, Oxoid L37), and homogenised using a Colworth Stomacher 400 (A.J. Seward, London, UK). A 1.0-ml aliquot of the homogenate was serially 10-fold diluted, and 0.1-ml aliquots of appropriate dilutions was spread onto duplicate plates of Standard Plate Count Agar (PCA, Oxoid CM463) and Pseudomonas-selective agar (Oxoid CM559 with CFC supplement Oxoid SR103). All plates were incubated at 25 °C for 2 –3 days (previous studies had shown that colony numbers do not increase further after 48 h). After incubation, colonies were counted and total numbers were expressed in terms of log10 cfu/g or log10 cfu/cm2. 2.3. Identification of the specific spoilage organism The fastest growing organisms on the salmon portions that became numerically dominant were designated as the specific spoilage organism (Dalgaard et al., 1993). They were identified as follows. Isolated colonies on PCA were sub-cultured onto Nutrient Agar (NA, Oxoid CM3), and identified to a
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
genus level using a dichotomous tree (Molin et al., 1983). The growth characteristics of the spoilage organism were monitored in various environments as described below. 2.3.1. Bacterial growth rates in slurry water Water samples collected at the processing factory were transported to the laboratory in polyfoam containers with reusable ice packs and incubated at 5 F 0.1 and 10 F 0.1 °C in shaking waterbaths (Ratek Instruments, Boronia, Victoria). Samples were periodically removed for microbiological enumeration over a period of 4 days. Slurry water aliquots of 1.0 ml were serially diluted, and 0.1 ml of the appropriate dilutions was spread onto duplicate plates of PCA and CFC agar. Plates were incubated at 25 °C for 2 –3 days. Generation times were calculated by linear regression from the slope of the growth curve during the exponential phase (approximately six data points) for both total viable counts and CFC plates. 2.3.2. Effect of chlorine on the lag phase of bacteria present in the slurry water Water samples collected from the harvesting operation were divided into two approximately equal volumes. Both volumes were inoculated with a Pseudomonas sp., in stationary phase, to an initial load of 104 cfu/ml, and incubated in a shaking water bath at 10 F 0.1 °C. The inoculum was isolated from the slurry water and was grown in nutrient broth (Oxoid CM1). One water sample was treated with hospital grade sodium hypochlorite (NaClO, White King, Kiwi Brands) to a final concentration of 30 mg/l active chlorine. The other was used as a control. Samples were removed periodically over 3 days for microbiological enumeration as described in Section 2.3.1. 2.4. Microbiological quality of water samples Water samples were obtained at the processing factory from the slurry bins after they had been ‘‘spiked’’ with chlorine (30 mg/l) and mixed with an aerator. Samples were taken through a bunghole at the bottom of the bins and collected in sterile 500-ml Schott bottles. Samples were transported to the laboratory in polyfoam boxes containing reusable ice packs, or examined at the processing factory. The
49
bacterial load was determined as described in Section 2.2 using both PCA and CFC agar plates. Other water samples were taken from the slurry bins at the harvesting factory and at the processing factory prior to the addition of chlorine. In some cases, it was possible to collect samples from the same slurry bin. The samples were processed within 8 h of collection. There were approximately 60 water samples taken from the processing factory and 30 from the harvesting factory. 2.5. Water held as surface film on salmon portions Five preweighed filter paper discs (28-mm2 area, Whatman Filter papers, grade 42) were placed on the skin-side surface of the salmon portions for approximately 25 s. The weight of the absorbed water was determined by reweighing the disc. The weight of water collected by the filter disc was converted to volume by weighing 1.0 ml of the slurry water from which the fish came and using this value as a conversion factor. Where the weight of the 1-ml sample was greater than 1 g, the inverse of this weight was used as the conversion factor. Five additional preweighed discs were placed on the same place as the first, for the same time, and the amount of absorbed liquid was determined by weight difference. 2.6. Temperature history of the slurry bins Tinytag and Tiny Talk (Gemini Data Loggers, West Sussex, UK) and Delphi (Delphi Industries, Auckland, New Zealand) electronic temperature data-loggers were placed in the slurry bins at the harvesting factory as the bins were being filled with salmon. The loggers were removed between 2 and 4 days later, just prior to the bins being emptied at the processing factory. Tinytag and Tiny Talk loggers were set to record the temperature every 10 min, and Delphi loggers every 15 min. The temperatures recorded by the loggers were offloaded onto a personal computer using OTLM v1.4 (Gemini Data Loggers) and MIRINZ logger interrogation Program v7.01 (Delphi Industries) software as appropriate. From the recorded temperatures, the maximum possible cell density of pseudomonads was predicted using Food Spoilage Predictor software v2.1 (Gemini Data Loggers, Port Macquarie, NSW), based on the model of Neumeyer
50
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
et al. (1997), or by integrating the time and average temperature for the Gemini and Delphi loggers, respectively. Comparisons between the predicted cell density and the observed bacterial numbers (cf. Section 2.3) were made. Growth of microorganisms in the slurry water between the harvest and the processing factories was assumed to be predominantly that of pseudomonads. For inclusion of this growth prediction into the PRM, the proportion of pseudomonads in the slurry water was determined by differential plate counts (see Section 2.2). Spot temperature checks were routinely performed at the processing factory to determine the temperature of the slurry bin water as it reached this location. Approximately 1100 samples of this type were obtained for inclusion into the model. 2.7. Distribution temperatures The temperatures to which the packaged salmon portions were exposed between the processing factory and retail outlets were monitored using temperature loggers as described in Section 2.6. This information was used to predict the effect of the distribution time and temperatures on the shelf life of the portions. The information from the temperature plots was used in the production of the PRM. 2.8. Modelling the harvesting and processing operations A mathematical model of the harvest to packaging operation was prepared using Analytica 2.0 for the PC (Lumina Decision Systems, Los Gatos, CA, USA). The flow model that was developed in Analytica is shown in Fig. 1, the numbers that are shown at each step indicate the relevant calculation used, as described in Table 1. The PRM was developed to predict the distribution of shelf life of the salmon portions based on the experimental data for:
Shelf life; Growth rate; Initial bacterial counts in the slurry water—both total counts and those for Pseudomonas; Final bacterial counts in slurry water—both total counts and those for Pseudomonas; Average temperature of the slurry water;
Time between harvesting and processing; Maximum possible growth in the slurry for Pseudomonas; Effect of chlorine—induction of bacterial population into a lag phase; Initial bacterial counts on the salmon portions—both total counts and those for Pseudomonas; Percentage pseudomonads on the fish surface; Surface counts on the processing line; Volume of water film on the fillet surfaces; and Distribution temperatures and times.
Shelf life estimates were based on the predicted time for Pseudomonas to reach densities of 107 cfu/g or cm2. Analytica was chosen for the PRM for its well-designed interface, its ready availability, and its facility for clear presentation of data. The microbial profiles assembled for each part of the process were fitted to distribution functions (BestFitv2.0c, Palisade, Newfield, NY, USA) allowing comparisons of the observed and the predicted levels. From the methods described in Sections 2.1 –2.6, three microbial profiles were developed to characterise the production process. Each profile was used for parts of the PRM, implementing the use of stochastic modelling to describe the production process. 2.8.1. Bacterial levels in the slurry bin water The final bacterial load in the slurry bin water was predicted using the Pseudomonas growth rate model based on the modified square root equation (Ratkowsky et al., 1983) refined by Neumeyer et al. (1997). sqrtðgrowth rateÞ ¼ bðT Tmin Þf1 exp½cðT Tmax g
ð1Þ
where: Tmin = 6.1, Tmax = 41, b = 0.1673, c = 0.192. To predict cell density, the lag induced by addition of chlorine was subtracted from time between harvesting and processing, and then divided by the growth rate (in hours). The growth predicted was then added to the initial bacterial levels to produce a final value for the predicted bacterial load in the slurry water (cfu/ ml). This gave a value for the total bacterial count in the slurry water, and is not an indication of only the pseudomonads present.
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
51
Fig. 1. Model flow chart.
To assess the accuracy of the model, a comparison was made between the predicted and the experimentally obtained values, which were fitted to a distribution function. The comparison (comparison 1, from Fig. 1) was made by overlaying the probability distributions for both the predicted and the observed data, comparing the mean and minimum and maximum values, and by comparing the values on probability bands (5%, 25%, 50%, 75% and 95%). 2.8.2. Initial contamination on salmon portions The bacterial contamination on the salmon portions was estimated using the data described in Section 2.4, and the predicted value as described in Section 2.8.1.
The range of values for bacterial contamination on the salmon portions was fitted to a distribution function and compared with the predicted values as described in Section 2.8.1. 2.8.3. Shelf life of salmon portions Harvested salmon are processed and packaged in vacuum packages prior to transport to retail destinations. The packages are opened and the salmon portions are displayed for the sale. The shelf life of the salmon portions was governed by the growth of Pseudomonas, the specific spoilage organism. The time the portions spend in the vacuum pack were treated as part of the lag phase because pseudomonads
52
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
Table 1 Detail of the model Number
Step
Distribution
Parameters
Comment
1
Temperature distribution (log)
Gamma (1.54, 1.33) 0.53
a, b
2
Time (log)
Gamma (16.94, 4.53)
a, b
3
Chlorine spike
(27/6.76) generation time
4
Generation time
5
Initial water counts
1/((0.1673 sqrt(0.995 0.947) (temperature distribution (log) + 6.1) (1 exp(0.192 (temperature distribution (log) 41))))^2) Normal (1.45, 0.86)
From the collected temperature logger (log) data; the 0.53 shifts the distribution so that the outputs may contain values below 0 °C From collected temperature logger (log) data; time is in hours The lag time for Pseudomonas in hours; from collected data Pseudomonas growth rate equation from Neumeyer et al. (1997)
6
%Psd in water
Triang (0.01, 0.1, 0.5)
7
Growth
8
Rationalised growth Predicted water counts Normalised predicted water counts Surface film
(Time (log) chlorine spike)/ (generation time 0.30103)) + (log10 (%Psd in water 10^initial water counts)))) If growth < 0 then 0 else growth
9 10
11 12 13
14 15 16 17
Surface contamination Combination
Start surface contamination Mid surface contamination End surface contamination Combined surface contamination
18
Fish contamination
19
%Psd on fish
Mean, standard deviation Min, most likely, max
Assumed distribution for the proportion of pseudomonads in the bacteria present in the slurry water Growth of Pseudomonas in the slurry water
Removes negative growth values
log10 ((10^initial water counts + 10^rationalized growth) If predicted water counts > 7 then 7 else predicted water counts Normal (0.00823, 0.00169) + normal (0.0041, 0.00124) Gamma (25.53, 0.12)
From collected data, in log10 cfu/ml
Predicted final level in slurry water Removes unrealistic values above 107
Mean, standard deviation a, b
From collected data; volume of water on fillet surface (ml) From collected data; contamination from water surface film, log10 cfu/ml Calculates bacterial contamination from the surface water film
(Gamma (2., 0.92)) 0.0324
Mean, standard deviation a, b
Gamma (5.46, 0.37)
a, b
Contamination level at the beginning of the production line, log10 cfu/cm2 Contamination level in the middle of the production line, log10 cfu/cm2 Contamination level at the end of the production line, log10 cfu/cm2 Calculates the average contamination from contact surfaces (steps 14, 15, 16).
log10 (10^(surface contamination + (10^normalised predicted water counts surface film)) Normal (1.67, 0.97)
log10 (10^end surface contamination + 10^mid surface contamination + 10^start surface contamination)/3 log10 (10^combined surface contamination + 10^combination) Normal (0.52, 0.31)
Mean, standard deviation
Predicts the bacterial contamination on the fillets From collected data
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
53
Table 1 (continued ) Number
Step
Distribution
20
Normalise %Psd on fish
21
22
Psd contamination level Normalised Psd
If %Psd on fish >1 then 1 else %Psd on fish, and if normalise %Psd < 0 then 0 else normalise %Psd log10 (normalise %Psd on fish (10^fish contamination)
23
Temperature
24
Aerobic storage
25 26
Lag phase Shelf life
27
Storage time in Vac pack Anaerobic storage
28
29
Anaerobic shelf life
If Psd contamination level > 7 then 7 else Psd contamination level Triang (0, 4, 10)
Parameters
Comment Removes values below 0% and above 100%
Calculates the number of Pseudomonads on the fish surface
Min, most likely, max
1/((0.1673 sqrt(0.995 0.947) (temperature + 6.1) (1 exp(0.192 (temperature 41))))^2) (27/6.76) aerobic storage (((7 normalise Psd) (10/3)) aerobic storage + lag phase ))/24 ((Triang (1, 4, 4)) 24) + lag phase
Removes values above 107, i.e. samples with a shelf life of 0 days The storage temperature of the fillets; estimate of possible storage temperatures Generation time for Pseudomonas
Relative lag time Predicts the shelf life (days) Estimate of days between packaging and retail
1/((0.1673 sqrt(0.995 0.947) (temperature + 6.1) (1 exp(0.192 (temperature 41))))^2) (((7 normalise Psd) (10/3)) (anaerobic storage) + lag phase))/24
Growth rate equation for Pseudomonas
Shelf life with retail distribution included
Numbers indicate the location of the step in the model influence diagram (Fig. 1).
are strictly aerobic. Consequently, two shelf life predictions were modelled. The first assumed that the portions were stored aerobically for the entire time, i.e. the packs are opened immediately after the fish has been packaged. The second considers the time the portions spend in vacuum packs during distribution before opening at the retail outlet. This is a more realistic estimate of the shelf life. The shelf life was predicted by subtracting the number of pseudomonads on the portions at the defined spoilage level (107 cfu/cm2), then dividing it by the growth rate at a given temperature and adding the lag phase observed during the shelf life studies (as described in Section 2.2). The temperature of distribution to retail outlets was obtained through temperature logging (as described in Section 2.7), and allowed the prediction of the remaining shelf life from the time the portions reached the retail outlets and were opened. A range of possible shelf life values were obtained from the model and comparisons (as described in Section 2.8.1) were made between the predicted shelf life and the observed shelf life for the salmon portions.
A sensitivity analysis was performed on the PRM, which ranked the input variables of time, initial counts, surface contamination, transport temperature and distribution temperature, according to their influence on the final outcome of the model, using the ‘Importance’ function in Analytica. The ‘Importance’ of a variable is based on the changes in that particular variable’s value, and the subsequent effect on the final output of the model. Only independent variables may be selected for this analysis.
3. Results 3.1. Bacterial population development One hundred three samples of salmon portions were examined for bacterial contamination level and composition of the spoilage microbiota. Initial counts (day 0) ranged from 2 to 5 log10 cfu/g, with an average count of 3.23 log10 cfu/g s.d. F 0.72 log10. A lag phase that consistently lasted 27 h at 4 °C was observed. This was followed by exponential
54
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
growth that slowed to a stationary phase after 6 days, by which time the salmon samples had spoiled. The average generation time during the exponential growth phase was 6.76 h at 4 °C. The shelf life, defined as the time to reach 107 cfu/g, was between 6 and 7 days, and the average for all samples was 6.23 days. Variations in shelf life correlated well with variations in the initial counts on the salmon portions; an initial count of 4.1 log cfu/g corresponded to a shelf life of 6 days whereas an initial count of 3 log cfu/g corresponded to a shelf life of nearly 8 days. 3.2. Identification of spoilage bacteria The composition of the microbiota present on the salmon portions was monitored during the storage period. On the aerobically stored portions, a diverse range of bacteria was initially present. As the portions reached spoilage, only two genera were usually isolated, with fluorescent and nonfluorescent Pseudomonas dominant, and occasionally Moraxella. When the samples were overtly spoiled, nonfluorescent Pseudomonas was the only bacteria isolated; however, this does not mean that there were no other bacterial genera present. From the growth rate experiments conducted at 2, 4, 5 and 10 °C, and the shelf life trials, conducted at 4 °C, nonfluorescent Pseudomonas was identified as the specific spoilage organism for the aerobically stored salmon portions (within a spoilage domain of 2– 10 °C). Anaerobically stored portions were predominantly colonised by Gram-positive bacteria (lactic acid bacteria and coryneforms) present during the storage period. Pseudomonas was almost entirely absent from the final microbiota of the vacuum-packaged portions. However, replicate plates of samples from the vacuum-packaged portions, incubated aerobically at 25 °C, contained Pseudomonas in low numbers ( < 1% of the colonies present). The increasing dominance of Pseudomonas over the aerobic storage period results from their faster growth rate relative to the other bacteria on the salmon portions (Gill and Newton, 1977) in the considered temperature range. 3.2.1. Bacterial growth rates in slurry water Pseudomonads present in the water have a lag phase of between 20 and 24 h. Following resolution of the lag phase, generation times for pseudomonads
at 10 °C were between 3.5 and 4 h. For those at 5 °C, the generation time was between 5.7 and 9.2 h. 3.2.2. Microbiological quality of water from the harvesting factory Water at the harvesting factory is chlorinated to 30 mg/l and bacterial counts are less than 1 cfu/100 ml. A test for chlorine in the slurry water at the harvesting factory, prior to addition of fish or ice, showed < 0.2 mg/l of chlorine present, suggesting that even with the initial high level of chlorine present in the bulk storage water used to fill the slurry bins, it is quickly bound to free organic matter in the slurry bins and neutralised. Bacterial counts for the slurry water at the harvesting factory ranged from undetectable ( < 1 cfu/ ml) to 2.5 log10 cfu/ml. 3.2.3. Effect of chlorine on the lag phase of bacteria added to slurry water Addition of the inoculum caused the bacterial load in the water to rise to 104 cfu/ml. Chlorine was added immediately after the addition of the inoculum, with further sampling starting at 24 h. The addition of chlorine did not greatly reduce the number of bacteria present in the water sample. The difference between the chlorinated and nonchlorinated water samples was 0.1 log cfu/ml. Counts did not begin to rise until after 30 h. 3.3. Microbiological quality of the slurry water samples Fifty-five slurry water samples were examined for bacterial contamination and diversity. The bacterial counts ranged between 1.1 and 5.3 log10 cfu/ml, with an average count of 3.3 log10 cfu/ml. The organisms most commonly found in the slurry water were nonfluorescent Pseudomonas (29.4% of the identified bacteria), Moraxella (17.6%), coryneforms (17.6%), and Aeromonas/Vibrio (13.7%). 3.4. Water held as surface film on salmon portions The average volume of water held as a surface film on the salmon portions was 0.0130 F 0.0043 ml/cm2. This volume of water, in conjunction with the slurry water bacterial load, was used to estimate the initial contamination load on the salmon portions. Samples
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
that were taken just after the backbones were removed from the fish showed surface counts between 1.9 and 4.4 log10 cfu/cm2. Duplicate samples were taken at the end of the processing line, prior to the fish being packaged. These second bacterial counts indicated an increase in surface contamination of between 0.2 and 1.1 log10 cfu/cm2. 3.5. Temperature history of the slurry bins Thirty-three temperature histories were recorded for different slurry bins. The time between filling and emptying the slurry bins was between 44 and 100 h. The time varied due to the level of activity at both the harvesting and processing factories. Temperatures within the slurry bins also varied, with averages over the storage temperature between 0.5 and + 6.9 °C. The longest time which was observed was close to 100 h ( 4 days), with the average time taken between filling and processing being 70 h (2.9 days).
55
Taking a lag phase for the bacteria in the slurry bin water into account, growth was possible under the temperature/time conditions in the slurry bins. 3.6. Modelling the growth of Pseudomonas throughout the harvesting and processing operations Each of the experimentally gathered data sets was fitted to a distribution function that most closely matched the possible range of values. In some cases, the best fitting distribution was not supported in Analytica so the next highest ranked function was used instead. A summary of each of the variables and their corresponding fitted distribution is given (Table 1). Three main phases in the production/distribution process that were modelled could be directly compared with the experimentally obtained values. These were slurry water counts, initial fish contamination, and shelf life. A summary flow-chart of the model is given in Fig. 1, and a mathematical description of the model is given in Table 1.
Fig. 2. Comparison between the distribution function fitted to the observed data and the predicted slurry water counts.
56
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
3.6.1. Modelling bacterial levels in slurry water Several inputs were used to predict the growth in the slurry water (as described in Section 2.8). The output from this part of the PRM was then rationalised, i.e. values for potential negative growth were disregarded. The final output, predicted water counts, was obtained by adding the initial counts to the growth. A comparison between the predicted water counts and the measured counts is shown (Fig. 2). The range of predicted values for the slurry water bacterial counts ranged from 0.6 to 1012 cfu/ml. The predicted upper value was considered unrealistic due to the effect of nutrient depletion at cell densities above 109 cfu/ml. Values for the predicted water counts above 107 were excluded from further calculations within the model as the standard operating procedure (SOP) at the processing factory would eliminate slurry bins entering the processing line with high bacterial counts in the slurry bin water.
3.6.2. Modelling initial bacterial counts on salmon portions The level of initial surface contamination on the salmon portions was modelled using three input variables: predicted water counts (cf. Section 3.6.1), surface water film, and contact surface contamination. The amount of water contaminating the fish surface was defined as the sum of the two distribution functions fitting the filter disc data. The water in the surface film was assumed to come from the slurry water, thus the slurry water contamination on the fish surface was the product of the volume of water on the fish surfaces and the bacterial load (cfu/ml) in the slurry water. Further contamination occurred along the processing line. Data for the level of contact surface contamination was obtained from routine environmental swabs taken at the processing factory. Three areas along the processing line were considered to make important contributions to the contamination of fish
Fig. 3. Comparison between the distribution function fitted to the observed and the predicted bacterial counts on salmon portions.
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
57
Fig. 4. Comparison between the range of predicted shelf life of aerobically and anaerobically stored portions.
surfaces. For each of these areas, a distribution function was fitted to the data and the results were averaged. Bacterial counts from the processing line varied between 0 and 3 log10 cfu/ml. Addition of the range of contact surface contamination with the contamination level from the slurry water produced a value that was close to the observed contamination levels found on the salmon portions. The predicted values were compared to the observed surface contamination levels (Fig. 3).
3.6.3. Modelling the shelf life of salmon portions The final stage of the model predicts the range of shelf lives possible for the salmon portions. The proportion of pseudomonads, in the initial contamination on the portions, ranged between 40% and 100%. These percentage values were converted to log10 cfu for shelf life predictions. For the simplicity of the modelling, the time that the salmon portions spend in vacuum packs was included as part of the lag phase of Pseudomonas.
Fig. 5. Relative importance of the input variables on the shelf life of salmon portions.
58
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
The predicted mean shelf life at 4 °C under aerobic conditions is 6.46 days, which is comparable with the average of 6.23 days obtained from the storage trials. The predicted shelf life of the aerobically stored portions and the portions that spend time in the vacuum packages are compared in Fig. 4. Also shown (Fig. 4) is a comparison of the cumulative probabilities for the predicted shelf life of the salmon portions stored aerobically and anaerobically. 3.6.4. Sensitivity analysis of input variables Sensitivity analysis performed on the input variables for the model revealed that the distribution/ storage temperatures and the time under which the salmon portions were stored after packaging had the greatest influence on the shelf life. The next most important variable was contact surface contamination. Fig. 5 shows the relative importance of each of the input variables. These were in increasing order of importance: initial water counts in the slurry bins, transport temperature and time, surface film contamination, contact surface contamination, and distribution time/temperature. Since temperature almost completely determines shelf life, attempts at improving the shelf life of the salmon portions should be directed at storing them at a lower temperature rather than attempting to reduce surface contamination or any of the other variables affecting the shelf life.
4. Discussion Process risk models involve situations in which uncertainty and variability restrict a definitive outcome due to limited data on the prevalence or density of the organism of concern, and/or absence of information on important elements of the risk assessment, for example the infective dose of a pathogen. Developing a process risk model to characterise the shelf life of a food product has (to our knowledge) not been reported in the scientific literature. However, it is a situation where a stochastic approach is likely to yield meaningful estimates of risk, as large amounts of relevant information on all parts of the harvesting, processing, and distribution chain can be collected in a longitudinal study. In this study, the production of salmon fillet portions for retail sale was examined for the PRM.
Stochastic modelling allows the identification of the variables with greatest influence in the production process and suggests means to implement improvements for extending shelf life. In this situation, it was the storage temperature that had the greatest influence on the final shelf life, which has been demonstrated previously (McMeekin and Olley, 1986; Ratkowsky et al., 1982, 1983; Huss et al., 2000). The other useful application of the model could be to explain the relatively short shelf life of the salmon portions, if this was due to the high numbers of spoilage pseudomonads. The effect of the proportion of the number of pseudomonads present, as part of the initial contamination on the salmon portions, can be seen. However, reductions in the number of pseudomonads will have little overall effect on the shelf life of the salmon portions. To extend the shelf life by any substantial amount, the growth rate of the spoilage pseudomonads would have to be limited. The normal handling procedure for the vacuumpackaged salmon portions which were studied is the transport to their retail destination, generally within 2 days of being packaged, at which time the vacuum packages are opened for display and retail sale. Growth of aerobic spoilage bacteria does not occur during the time between packaging and when the packages are opened, as these are in the lag phase, and the SSO (Pseudomonas) survives but is unable to grow in the anaerobic conditions prevailing during transport. To produce a PRM that predicts the shelf life of the portions, the lag phase and growth rate of the SSO were determined by storage trials. The growth of bacteria in the slurry water was calculated from the distribution of the average time/temperature conditions. The inclusion of actual temperature plots into the model was not practical due to the limitations of the modelling program used. However, the use of average time and temperatures was a good estimate of the conditions in the slurry bins. Ross (1999) demonstrated that in similar situations, the difference between the growth predicted from the average time/ temperature and the integrated time/temperature history was insignificantly small. Initial storage trials were conducted at 4 °C, with subsequent trials at higher and lower temperatures, to check the accuracy of the model outside of the original range. Several other salmon portions were stored at 1, 2 and 10 °C to
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
validate the model. The observed shelf life changed in proportion to the temperature, with the difference between the observed and predicted shelf lives being within 0.5 days of each other, for each experimental set. That is, the difference between the predicted and the observed shelf life at 1 °C was less than half a day. This indicated that the model was accurate at these temperatures, and that it is capable of predicting the shelf life of salmon portions between 1 and 10 °C. Overall good agreement was found between the predicted range of values and the fitted distributions. The predicted and observed values were almost identical for the bacterial loads in the slurry water (Fig. 2). This demonstrates clearly that where sufficient information is collected and uncertainty is minimised, a stochastic model can provide an accurate microbial profile of a specific processing operation. Conversely, where less data is available, the distributions may not overlap as closely, as can be seen with the comparison between the predicted bacterial contamination load on the portions and the distributions fitted to the observed data (Fig. 3). This comparison was further complicated by a change in operating procedures on the filleting line during the course of the 9-month study. Using the sensitivity analysis capabilities of Analytica, it was possible to determine the most influential factor on shelf life, and to rank strategies to extend the shelf life of the product. The model predicted that reduction of contamination in the slurry bin water by a factor of 10 would result in an extension of shelf life of 0.4 days, reducing contamination on the filleting line by a factor of 10 would give a predicted extension in shelf life of 1 day, whereas reduction in storage temperature from 6 to 3 °C would result in an increase in shelf life of 5 days. This indicates clearly that temperature is the most influential factor affecting the shelf life, a well-recognised phenomenon (McMeekin and Olley, 1986; Ratkowsky et al., 1982, 1983; Huss et al., 2000). In this context, the value of a PRM is not limited specifically to the prediction of shelf life of a specific food product. It can become an important element of a quality assurance program and will find utility in correcting potential microbiological problems by incorporation into an HACCP plan. While many authors restrict the HACCP concept to food safety issues, the principles are equally applicable to the
59
assurance of microbial food quality. In the specific example described, the hazard is the failure to reach a specified ‘use by’ date with the quality assurance points identified by the PRM. ‘‘What-if scenarios’’ using the PRM provide the processor with the most effective means to attain with certainty the required shelf life which may vary for different markets. A further advantage is that the PRM is not static and additional data can be added to a current model to further reduce uncertainty or to evaluate the effect of changes in processing protocols. We conclude that the application of stochastic approaches through development of a PRM constitutes a powerful tool to promote continual improvement and to attain microbial food quality objectives.
Acknowledgements The financial support and cooperation of a Tasmanian salmon producer and the University of Tasmania are gratefully acknowledged.
References Bremner, H.A., Olley, J., Vail, A.M.A., 1987. Estimating time – temperature effects by a rapid systematic sensory method. In: Kramer, D.E., Liston, J. (Eds.), Seafood Quality Determination. Elsevier Science, Amsterdam, pp. 413 – 435. Cassin, M.H., Lammerding, A.M., Todd, E.C.D., Ross, W., McColl, S.R., 1998. Quantitative risk assessment for Escherichia coli O157:H7 in ground beef hamburgers. International Journal of Food Microbiology 41, 21 – 44. Clesceri, L.S., Greenberg, A.E., Trussell, R.R., 1989. Standard Methods for the Examination of Water and Wastewater, 17th edn. American Public Health Association, Washington, DC, USA, pp. 9-30 – 9-35. Dalgaard, P., Gram, L., Huss, H.H., 1993. Spoilage and shelf life of cod fillets packed in vacuum or modified atmospheres. International Journal of Food Microbiology 19, 283 – 294. Fey, M.S., Regenstein, J.M., 1982. Extending shelf life of fresh wet red hake and salmon using CO2 – O2 modified atmosphere and potassium sorbate ice at 1 °C. Journal of Food Science 47, 1048 – 1054. Gill, C.O., Newton, K.G., 1977. The development of aerobic spoilage flora on meat stored at chill temperatures. Journal of Applied Bacteriology 43, 189 – 195. Huss, H.H., Reilly, A., Embark, K.B., 2000. Prevention and control of hazards in seafood. Food Control 11, 149 – 156. Liston, J., 1992. Bacterial spoilage of food. In: Huss, H.H., Jakobsen, M., Liston, J. (Eds.), Developments in Food Science 30:
60
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
Quality Assurance in the Fish Industry. Elsevier, Holland, pp. 93 – 105. McMeekin, T.A., Olley, J., 1986. Predictive microbiology. Food Technology in Australia 38, 331 – 334. McMeekin, T.A., Olley, J., Ross, T., 1991. Application of predictive microbiology to assure the quality and safety of fish and fish products. In: Huss, H.H., Jakobsen, M., Liston, J. (Eds.), Developments in Food Science 30: Quality Assurance in the Fish Industry. Elsevier, Holland, pp. 459 – 478. Molin, G., Stenstro¨m, I.-M., Ternstro¨m, A., 1983. The microbial flora of herring fillets after storage in carbon dioxide, nitrogen or air at 2 °C. Journal of Applied Bacteriology 55, 49 – 56. Neumeyer, K., Ross, T., McMeekin, T.A., 1997. Development of a predictive model to describe the effects of temperature and water activity on the growth of spoilage pseudomonads. International Journal of Food Microbiology 38, 45 – 54. Olley, J., Ratkowsky, D.A., 1973. Temperature function integration and its importance in the storage and distribution of flesh foods
above the freezing point. Food Technology in Australia 25, 66 – 73. Nielsen, J., Reines, J.H., Jespersen, C.M., 1991. Quality assurance in the fishing industry with emphasis on the future use of vision techniques. In: Pau, L.F., Olafsson, R. (Eds.), Fish Quality Control by Computer Vision. Marcel Dekker, New York, pp. 3 – 20. Ratkowsky, D.A., Olley, J., McMeekin, T.A., Ball, A., 1982. Relationship between temperature and growth rate of bacterial cultures. Journal of Bacteriology 149, 1 – 5. Ratkowsky, D.A., Lowry, R.K., McMeekin, T.A., Stokes, A.N., Chandler, R.E., 1983. Model for bacterial culture growth rate throughout the entire biokinetic temperature range. Journal of Bacteriology 154, 1222 – 1226. Ross, T., 1999. Predictive microbiology for the meat industry. Meat and Livestock Australia, Sydney, 196 pp. Sumner, J.L., Gorczyca, E., Cohen, D., Brady, P., 1984. Do fish from tropical waters spoil less rapidly in ice than fish from temperate waters? Food Technology in Australia 36, 328 – 329.
A process risk model for the shelf life of Atlantic salmon fillets S.K.J. Rasmussen *, T. Ross, J. Olley, T. McMeekin School of Agricultural Science, University of Tasmania, GPO Box 252-54, Hobart Tasmania 7000, Australia Received 15 January 2001; received in revised form 23 May 2001; accepted 20 September 2001
Abstract The shelf life of Atlantic salmon (Salmo salar) portions produced for retail distribution is examined and the dominant aerobic spoilage organism is identified. Characterization of the harvesting and processing operations allow the development of a stochastic mathematical model, a process risk model (PRM), which predicts the range of the possible shelf life for the portions under normal retail and distribution. The considered risk is the failure to achieve the nominal ‘use by’ date. Bacterial counts from surface swabs, water, ice, and fish samples, collected over a period of 9 months, are fitted to distribution functions for use within the model. Comparisons are made between the distributions fitted to the observed bacterial levels and the predicted levels for the slurry water, initial surface contamination on the fish, and for the predicted and observed shelf life. Storage temperature of the packaged salmon portions has the greatest influence on shelf life, with contamination from contact surfaces and other sources being the next most important. The range of bacterial counts on the portions was between 0.6 and 5 log10 cfu/cm2. The model predicts bacterial counts in the slurry water to have an average value of 3.36 log10 cfu/ml, whereas the observed slurry water bacterial counts were 3.35 log10 cfu/ml. The predicted average initial bacterial contamination is 3.31 log10 cfu/cm2 on the fish surface and 3.23 log10 cfu/cm2 on the observed. The average predicted shelf life is 6.5 days, compared to an observed value of 6.2 days at 4 °C. D 2002 Elsevier Science B.V. All rights reserved. Keywords: Process risk model; Shelf life; Atlantic salmon
1. Introduction Fish quality, microbiology, and shelf life have been studied extensively for the last seven decades. This is the first reported effort to use risk assessment techniques to predict the shelf life of any seafood product from catching to retail from one factory. Process risk models (PRM) have been applied in risk assessments concerned with public health outcomes (Cassin et al.,
*
Corresponding author. Tel.: +61-3-62-26-1831, +61414693459 (mobile); fax: +61-3-62-26-2642. E-mail address: [email protected] (S.K.J. Rasmussen).
1998). Fish are recognised as being highly perishable, having a relatively short shelf life (Fey and Regenstein, 1982). It is recommended that fish should be chilled as quickly as possible after harvest and kept on ice until consumed (Liston, 1992; Sumner et al., 1984). The concept of temperature control as a means to prolong shelf life is well understood, and the response of bacteria to temperature has been widely studied (Olley and Ratkowsky, 1973; Ratkowsky et al., 1982, 1983; Bremner et al., 1987; McMeekin et al., 1991). Accurately defining shelf life is important to both the consumer and the producer but with any processing, storage, and biological system, there is variability.
0168-1605/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 8 - 1 6 0 5 ( 0 1 ) 0 0 6 8 7 - 0
48
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
This variability can have a large influence on the shelf life of the product. Rather than the shelf life being a single point value, there is a range of possible values that depends on variations in the production process, initial bacterial contamination, and the storage conditions. Storage trials allow the shelf life of a product to be determined reasonably accurately (McMeekin and Olley, 1986). With the concurrent use of stochastic modelling, the range of possible shelf life estimates of the product can be elucidated. Production improvements may then be defined to reduce contamination loads on the salmon fillets and increase the shelf life of the product. The application of stochastic modelling techniques has not been previously used for evaluation and prediction of shelf life. This paper describes the development of a process risk model, stating assumptions and explaining the modelling process. Validation of the model is also described. A background to the philosophy of Monte Carlo risk assessment modelling may be found in Cassin et al. (1998), and further use of computer-aided techniques that may be used for the seafood may be found in Nielsen et al. (1991).
2. Materials and methods 2.1. Sample collection Water and fish: Atlantic salmon (Salmo salar) samples were collected from the harvesting and the processing factories. The fish samples collected from the harvesting factory were whole, head on, gilled and gutted (HOGG) salmon, and were kept on ice until processing at the laboratory. At the harvesting factory, the salmon are weighed, sorted, and placed into large insulated plastic containers (slurry bins). Slurry bins have a holding volume of 1000 l, and are filled with water from bulk water storage tanks at the harvesting factory, ice, and fish. Fresh water was used throughout the processing of the salmon. On two occasions during the processing, the slurry bins are spiked with chlorine, i.e. a solution of sodium hypochlorite is added, with an effective dose of 30 mg/l of chlorine. Samples taken at the harvesting factory included water, fish, ice, and surface swabs from the processing line. In situations where samples were taken from ‘‘working surfaces’’, this would refer to sampling
done at the harvesting factory, and when ‘‘production line’’ is used, this would refer to sampling from the processing factory. Water samples collected from the slurry bins after they had been spiked had the active chlorine neutralised by the addition of a fixed amount of sodium thiosulphate (Na2S2O3, Analar, Poole, England), following the standard method (Clesceri et al., 1989). 2.2. Bacterial population development on salmon portions Retail packages of Atlantic salmon portions obtained from a local producer were incubated at 4 °C for up to 11 days. The portions were incubated aerobically in stomacher bags (Bio-Service, Huntingdale, Victoria) after aseptic transfer. Microbiological sampling and enumeration was undertaken throughout the storage period to determine the lag phase and the growth rate of the spoilage microbiota. Shelf life was defined as the time taken for the bacterial population to reach 107 cfu/g. Total viable counts on the portions were determined by aseptically excising a 16cm2 area of flesh from the surface of the portion, using a sterile 4 4-cm stainless steel template. Alternatively, 10 g of flesh was added to 90 ml of diluent (0.9% NaCl with 0.1% bacteriological peptone, Oxoid L37), and homogenised using a Colworth Stomacher 400 (A.J. Seward, London, UK). A 1.0-ml aliquot of the homogenate was serially 10-fold diluted, and 0.1-ml aliquots of appropriate dilutions was spread onto duplicate plates of Standard Plate Count Agar (PCA, Oxoid CM463) and Pseudomonas-selective agar (Oxoid CM559 with CFC supplement Oxoid SR103). All plates were incubated at 25 °C for 2 –3 days (previous studies had shown that colony numbers do not increase further after 48 h). After incubation, colonies were counted and total numbers were expressed in terms of log10 cfu/g or log10 cfu/cm2. 2.3. Identification of the specific spoilage organism The fastest growing organisms on the salmon portions that became numerically dominant were designated as the specific spoilage organism (Dalgaard et al., 1993). They were identified as follows. Isolated colonies on PCA were sub-cultured onto Nutrient Agar (NA, Oxoid CM3), and identified to a
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
genus level using a dichotomous tree (Molin et al., 1983). The growth characteristics of the spoilage organism were monitored in various environments as described below. 2.3.1. Bacterial growth rates in slurry water Water samples collected at the processing factory were transported to the laboratory in polyfoam containers with reusable ice packs and incubated at 5 F 0.1 and 10 F 0.1 °C in shaking waterbaths (Ratek Instruments, Boronia, Victoria). Samples were periodically removed for microbiological enumeration over a period of 4 days. Slurry water aliquots of 1.0 ml were serially diluted, and 0.1 ml of the appropriate dilutions was spread onto duplicate plates of PCA and CFC agar. Plates were incubated at 25 °C for 2 –3 days. Generation times were calculated by linear regression from the slope of the growth curve during the exponential phase (approximately six data points) for both total viable counts and CFC plates. 2.3.2. Effect of chlorine on the lag phase of bacteria present in the slurry water Water samples collected from the harvesting operation were divided into two approximately equal volumes. Both volumes were inoculated with a Pseudomonas sp., in stationary phase, to an initial load of 104 cfu/ml, and incubated in a shaking water bath at 10 F 0.1 °C. The inoculum was isolated from the slurry water and was grown in nutrient broth (Oxoid CM1). One water sample was treated with hospital grade sodium hypochlorite (NaClO, White King, Kiwi Brands) to a final concentration of 30 mg/l active chlorine. The other was used as a control. Samples were removed periodically over 3 days for microbiological enumeration as described in Section 2.3.1. 2.4. Microbiological quality of water samples Water samples were obtained at the processing factory from the slurry bins after they had been ‘‘spiked’’ with chlorine (30 mg/l) and mixed with an aerator. Samples were taken through a bunghole at the bottom of the bins and collected in sterile 500-ml Schott bottles. Samples were transported to the laboratory in polyfoam boxes containing reusable ice packs, or examined at the processing factory. The
49
bacterial load was determined as described in Section 2.2 using both PCA and CFC agar plates. Other water samples were taken from the slurry bins at the harvesting factory and at the processing factory prior to the addition of chlorine. In some cases, it was possible to collect samples from the same slurry bin. The samples were processed within 8 h of collection. There were approximately 60 water samples taken from the processing factory and 30 from the harvesting factory. 2.5. Water held as surface film on salmon portions Five preweighed filter paper discs (28-mm2 area, Whatman Filter papers, grade 42) were placed on the skin-side surface of the salmon portions for approximately 25 s. The weight of the absorbed water was determined by reweighing the disc. The weight of water collected by the filter disc was converted to volume by weighing 1.0 ml of the slurry water from which the fish came and using this value as a conversion factor. Where the weight of the 1-ml sample was greater than 1 g, the inverse of this weight was used as the conversion factor. Five additional preweighed discs were placed on the same place as the first, for the same time, and the amount of absorbed liquid was determined by weight difference. 2.6. Temperature history of the slurry bins Tinytag and Tiny Talk (Gemini Data Loggers, West Sussex, UK) and Delphi (Delphi Industries, Auckland, New Zealand) electronic temperature data-loggers were placed in the slurry bins at the harvesting factory as the bins were being filled with salmon. The loggers were removed between 2 and 4 days later, just prior to the bins being emptied at the processing factory. Tinytag and Tiny Talk loggers were set to record the temperature every 10 min, and Delphi loggers every 15 min. The temperatures recorded by the loggers were offloaded onto a personal computer using OTLM v1.4 (Gemini Data Loggers) and MIRINZ logger interrogation Program v7.01 (Delphi Industries) software as appropriate. From the recorded temperatures, the maximum possible cell density of pseudomonads was predicted using Food Spoilage Predictor software v2.1 (Gemini Data Loggers, Port Macquarie, NSW), based on the model of Neumeyer
50
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
et al. (1997), or by integrating the time and average temperature for the Gemini and Delphi loggers, respectively. Comparisons between the predicted cell density and the observed bacterial numbers (cf. Section 2.3) were made. Growth of microorganisms in the slurry water between the harvest and the processing factories was assumed to be predominantly that of pseudomonads. For inclusion of this growth prediction into the PRM, the proportion of pseudomonads in the slurry water was determined by differential plate counts (see Section 2.2). Spot temperature checks were routinely performed at the processing factory to determine the temperature of the slurry bin water as it reached this location. Approximately 1100 samples of this type were obtained for inclusion into the model. 2.7. Distribution temperatures The temperatures to which the packaged salmon portions were exposed between the processing factory and retail outlets were monitored using temperature loggers as described in Section 2.6. This information was used to predict the effect of the distribution time and temperatures on the shelf life of the portions. The information from the temperature plots was used in the production of the PRM. 2.8. Modelling the harvesting and processing operations A mathematical model of the harvest to packaging operation was prepared using Analytica 2.0 for the PC (Lumina Decision Systems, Los Gatos, CA, USA). The flow model that was developed in Analytica is shown in Fig. 1, the numbers that are shown at each step indicate the relevant calculation used, as described in Table 1. The PRM was developed to predict the distribution of shelf life of the salmon portions based on the experimental data for:
Shelf life; Growth rate; Initial bacterial counts in the slurry water—both total counts and those for Pseudomonas; Final bacterial counts in slurry water—both total counts and those for Pseudomonas; Average temperature of the slurry water;
Time between harvesting and processing; Maximum possible growth in the slurry for Pseudomonas; Effect of chlorine—induction of bacterial population into a lag phase; Initial bacterial counts on the salmon portions—both total counts and those for Pseudomonas; Percentage pseudomonads on the fish surface; Surface counts on the processing line; Volume of water film on the fillet surfaces; and Distribution temperatures and times.
Shelf life estimates were based on the predicted time for Pseudomonas to reach densities of 107 cfu/g or cm2. Analytica was chosen for the PRM for its well-designed interface, its ready availability, and its facility for clear presentation of data. The microbial profiles assembled for each part of the process were fitted to distribution functions (BestFitv2.0c, Palisade, Newfield, NY, USA) allowing comparisons of the observed and the predicted levels. From the methods described in Sections 2.1 –2.6, three microbial profiles were developed to characterise the production process. Each profile was used for parts of the PRM, implementing the use of stochastic modelling to describe the production process. 2.8.1. Bacterial levels in the slurry bin water The final bacterial load in the slurry bin water was predicted using the Pseudomonas growth rate model based on the modified square root equation (Ratkowsky et al., 1983) refined by Neumeyer et al. (1997). sqrtðgrowth rateÞ ¼ bðT Tmin Þf1 exp½cðT Tmax g
ð1Þ
where: Tmin = 6.1, Tmax = 41, b = 0.1673, c = 0.192. To predict cell density, the lag induced by addition of chlorine was subtracted from time between harvesting and processing, and then divided by the growth rate (in hours). The growth predicted was then added to the initial bacterial levels to produce a final value for the predicted bacterial load in the slurry water (cfu/ ml). This gave a value for the total bacterial count in the slurry water, and is not an indication of only the pseudomonads present.
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
51
Fig. 1. Model flow chart.
To assess the accuracy of the model, a comparison was made between the predicted and the experimentally obtained values, which were fitted to a distribution function. The comparison (comparison 1, from Fig. 1) was made by overlaying the probability distributions for both the predicted and the observed data, comparing the mean and minimum and maximum values, and by comparing the values on probability bands (5%, 25%, 50%, 75% and 95%). 2.8.2. Initial contamination on salmon portions The bacterial contamination on the salmon portions was estimated using the data described in Section 2.4, and the predicted value as described in Section 2.8.1.
The range of values for bacterial contamination on the salmon portions was fitted to a distribution function and compared with the predicted values as described in Section 2.8.1. 2.8.3. Shelf life of salmon portions Harvested salmon are processed and packaged in vacuum packages prior to transport to retail destinations. The packages are opened and the salmon portions are displayed for the sale. The shelf life of the salmon portions was governed by the growth of Pseudomonas, the specific spoilage organism. The time the portions spend in the vacuum pack were treated as part of the lag phase because pseudomonads
52
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
Table 1 Detail of the model Number
Step
Distribution
Parameters
Comment
1
Temperature distribution (log)
Gamma (1.54, 1.33) 0.53
a, b
2
Time (log)
Gamma (16.94, 4.53)
a, b
3
Chlorine spike
(27/6.76) generation time
4
Generation time
5
Initial water counts
1/((0.1673 sqrt(0.995 0.947) (temperature distribution (log) + 6.1) (1 exp(0.192 (temperature distribution (log) 41))))^2) Normal (1.45, 0.86)
From the collected temperature logger (log) data; the 0.53 shifts the distribution so that the outputs may contain values below 0 °C From collected temperature logger (log) data; time is in hours The lag time for Pseudomonas in hours; from collected data Pseudomonas growth rate equation from Neumeyer et al. (1997)
6
%Psd in water
Triang (0.01, 0.1, 0.5)
7
Growth
8
Rationalised growth Predicted water counts Normalised predicted water counts Surface film
(Time (log) chlorine spike)/ (generation time 0.30103)) + (log10 (%Psd in water 10^initial water counts)))) If growth < 0 then 0 else growth
9 10
11 12 13
14 15 16 17
Surface contamination Combination
Start surface contamination Mid surface contamination End surface contamination Combined surface contamination
18
Fish contamination
19
%Psd on fish
Mean, standard deviation Min, most likely, max
Assumed distribution for the proportion of pseudomonads in the bacteria present in the slurry water Growth of Pseudomonas in the slurry water
Removes negative growth values
log10 ((10^initial water counts + 10^rationalized growth) If predicted water counts > 7 then 7 else predicted water counts Normal (0.00823, 0.00169) + normal (0.0041, 0.00124) Gamma (25.53, 0.12)
From collected data, in log10 cfu/ml
Predicted final level in slurry water Removes unrealistic values above 107
Mean, standard deviation a, b
From collected data; volume of water on fillet surface (ml) From collected data; contamination from water surface film, log10 cfu/ml Calculates bacterial contamination from the surface water film
(Gamma (2., 0.92)) 0.0324
Mean, standard deviation a, b
Gamma (5.46, 0.37)
a, b
Contamination level at the beginning of the production line, log10 cfu/cm2 Contamination level in the middle of the production line, log10 cfu/cm2 Contamination level at the end of the production line, log10 cfu/cm2 Calculates the average contamination from contact surfaces (steps 14, 15, 16).
log10 (10^(surface contamination + (10^normalised predicted water counts surface film)) Normal (1.67, 0.97)
log10 (10^end surface contamination + 10^mid surface contamination + 10^start surface contamination)/3 log10 (10^combined surface contamination + 10^combination) Normal (0.52, 0.31)
Mean, standard deviation
Predicts the bacterial contamination on the fillets From collected data
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
53
Table 1 (continued ) Number
Step
Distribution
20
Normalise %Psd on fish
21
22
Psd contamination level Normalised Psd
If %Psd on fish >1 then 1 else %Psd on fish, and if normalise %Psd < 0 then 0 else normalise %Psd log10 (normalise %Psd on fish (10^fish contamination)
23
Temperature
24
Aerobic storage
25 26
Lag phase Shelf life
27
Storage time in Vac pack Anaerobic storage
28
29
Anaerobic shelf life
If Psd contamination level > 7 then 7 else Psd contamination level Triang (0, 4, 10)
Parameters
Comment Removes values below 0% and above 100%
Calculates the number of Pseudomonads on the fish surface
Min, most likely, max
1/((0.1673 sqrt(0.995 0.947) (temperature + 6.1) (1 exp(0.192 (temperature 41))))^2) (27/6.76) aerobic storage (((7 normalise Psd) (10/3)) aerobic storage + lag phase ))/24 ((Triang (1, 4, 4)) 24) + lag phase
Removes values above 107, i.e. samples with a shelf life of 0 days The storage temperature of the fillets; estimate of possible storage temperatures Generation time for Pseudomonas
Relative lag time Predicts the shelf life (days) Estimate of days between packaging and retail
1/((0.1673 sqrt(0.995 0.947) (temperature + 6.1) (1 exp(0.192 (temperature 41))))^2) (((7 normalise Psd) (10/3)) (anaerobic storage) + lag phase))/24
Growth rate equation for Pseudomonas
Shelf life with retail distribution included
Numbers indicate the location of the step in the model influence diagram (Fig. 1).
are strictly aerobic. Consequently, two shelf life predictions were modelled. The first assumed that the portions were stored aerobically for the entire time, i.e. the packs are opened immediately after the fish has been packaged. The second considers the time the portions spend in vacuum packs during distribution before opening at the retail outlet. This is a more realistic estimate of the shelf life. The shelf life was predicted by subtracting the number of pseudomonads on the portions at the defined spoilage level (107 cfu/cm2), then dividing it by the growth rate at a given temperature and adding the lag phase observed during the shelf life studies (as described in Section 2.2). The temperature of distribution to retail outlets was obtained through temperature logging (as described in Section 2.7), and allowed the prediction of the remaining shelf life from the time the portions reached the retail outlets and were opened. A range of possible shelf life values were obtained from the model and comparisons (as described in Section 2.8.1) were made between the predicted shelf life and the observed shelf life for the salmon portions.
A sensitivity analysis was performed on the PRM, which ranked the input variables of time, initial counts, surface contamination, transport temperature and distribution temperature, according to their influence on the final outcome of the model, using the ‘Importance’ function in Analytica. The ‘Importance’ of a variable is based on the changes in that particular variable’s value, and the subsequent effect on the final output of the model. Only independent variables may be selected for this analysis.
3. Results 3.1. Bacterial population development One hundred three samples of salmon portions were examined for bacterial contamination level and composition of the spoilage microbiota. Initial counts (day 0) ranged from 2 to 5 log10 cfu/g, with an average count of 3.23 log10 cfu/g s.d. F 0.72 log10. A lag phase that consistently lasted 27 h at 4 °C was observed. This was followed by exponential
54
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
growth that slowed to a stationary phase after 6 days, by which time the salmon samples had spoiled. The average generation time during the exponential growth phase was 6.76 h at 4 °C. The shelf life, defined as the time to reach 107 cfu/g, was between 6 and 7 days, and the average for all samples was 6.23 days. Variations in shelf life correlated well with variations in the initial counts on the salmon portions; an initial count of 4.1 log cfu/g corresponded to a shelf life of 6 days whereas an initial count of 3 log cfu/g corresponded to a shelf life of nearly 8 days. 3.2. Identification of spoilage bacteria The composition of the microbiota present on the salmon portions was monitored during the storage period. On the aerobically stored portions, a diverse range of bacteria was initially present. As the portions reached spoilage, only two genera were usually isolated, with fluorescent and nonfluorescent Pseudomonas dominant, and occasionally Moraxella. When the samples were overtly spoiled, nonfluorescent Pseudomonas was the only bacteria isolated; however, this does not mean that there were no other bacterial genera present. From the growth rate experiments conducted at 2, 4, 5 and 10 °C, and the shelf life trials, conducted at 4 °C, nonfluorescent Pseudomonas was identified as the specific spoilage organism for the aerobically stored salmon portions (within a spoilage domain of 2– 10 °C). Anaerobically stored portions were predominantly colonised by Gram-positive bacteria (lactic acid bacteria and coryneforms) present during the storage period. Pseudomonas was almost entirely absent from the final microbiota of the vacuum-packaged portions. However, replicate plates of samples from the vacuum-packaged portions, incubated aerobically at 25 °C, contained Pseudomonas in low numbers ( < 1% of the colonies present). The increasing dominance of Pseudomonas over the aerobic storage period results from their faster growth rate relative to the other bacteria on the salmon portions (Gill and Newton, 1977) in the considered temperature range. 3.2.1. Bacterial growth rates in slurry water Pseudomonads present in the water have a lag phase of between 20 and 24 h. Following resolution of the lag phase, generation times for pseudomonads
at 10 °C were between 3.5 and 4 h. For those at 5 °C, the generation time was between 5.7 and 9.2 h. 3.2.2. Microbiological quality of water from the harvesting factory Water at the harvesting factory is chlorinated to 30 mg/l and bacterial counts are less than 1 cfu/100 ml. A test for chlorine in the slurry water at the harvesting factory, prior to addition of fish or ice, showed < 0.2 mg/l of chlorine present, suggesting that even with the initial high level of chlorine present in the bulk storage water used to fill the slurry bins, it is quickly bound to free organic matter in the slurry bins and neutralised. Bacterial counts for the slurry water at the harvesting factory ranged from undetectable ( < 1 cfu/ ml) to 2.5 log10 cfu/ml. 3.2.3. Effect of chlorine on the lag phase of bacteria added to slurry water Addition of the inoculum caused the bacterial load in the water to rise to 104 cfu/ml. Chlorine was added immediately after the addition of the inoculum, with further sampling starting at 24 h. The addition of chlorine did not greatly reduce the number of bacteria present in the water sample. The difference between the chlorinated and nonchlorinated water samples was 0.1 log cfu/ml. Counts did not begin to rise until after 30 h. 3.3. Microbiological quality of the slurry water samples Fifty-five slurry water samples were examined for bacterial contamination and diversity. The bacterial counts ranged between 1.1 and 5.3 log10 cfu/ml, with an average count of 3.3 log10 cfu/ml. The organisms most commonly found in the slurry water were nonfluorescent Pseudomonas (29.4% of the identified bacteria), Moraxella (17.6%), coryneforms (17.6%), and Aeromonas/Vibrio (13.7%). 3.4. Water held as surface film on salmon portions The average volume of water held as a surface film on the salmon portions was 0.0130 F 0.0043 ml/cm2. This volume of water, in conjunction with the slurry water bacterial load, was used to estimate the initial contamination load on the salmon portions. Samples
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
that were taken just after the backbones were removed from the fish showed surface counts between 1.9 and 4.4 log10 cfu/cm2. Duplicate samples were taken at the end of the processing line, prior to the fish being packaged. These second bacterial counts indicated an increase in surface contamination of between 0.2 and 1.1 log10 cfu/cm2. 3.5. Temperature history of the slurry bins Thirty-three temperature histories were recorded for different slurry bins. The time between filling and emptying the slurry bins was between 44 and 100 h. The time varied due to the level of activity at both the harvesting and processing factories. Temperatures within the slurry bins also varied, with averages over the storage temperature between 0.5 and + 6.9 °C. The longest time which was observed was close to 100 h ( 4 days), with the average time taken between filling and processing being 70 h (2.9 days).
55
Taking a lag phase for the bacteria in the slurry bin water into account, growth was possible under the temperature/time conditions in the slurry bins. 3.6. Modelling the growth of Pseudomonas throughout the harvesting and processing operations Each of the experimentally gathered data sets was fitted to a distribution function that most closely matched the possible range of values. In some cases, the best fitting distribution was not supported in Analytica so the next highest ranked function was used instead. A summary of each of the variables and their corresponding fitted distribution is given (Table 1). Three main phases in the production/distribution process that were modelled could be directly compared with the experimentally obtained values. These were slurry water counts, initial fish contamination, and shelf life. A summary flow-chart of the model is given in Fig. 1, and a mathematical description of the model is given in Table 1.
Fig. 2. Comparison between the distribution function fitted to the observed data and the predicted slurry water counts.
56
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
3.6.1. Modelling bacterial levels in slurry water Several inputs were used to predict the growth in the slurry water (as described in Section 2.8). The output from this part of the PRM was then rationalised, i.e. values for potential negative growth were disregarded. The final output, predicted water counts, was obtained by adding the initial counts to the growth. A comparison between the predicted water counts and the measured counts is shown (Fig. 2). The range of predicted values for the slurry water bacterial counts ranged from 0.6 to 1012 cfu/ml. The predicted upper value was considered unrealistic due to the effect of nutrient depletion at cell densities above 109 cfu/ml. Values for the predicted water counts above 107 were excluded from further calculations within the model as the standard operating procedure (SOP) at the processing factory would eliminate slurry bins entering the processing line with high bacterial counts in the slurry bin water.
3.6.2. Modelling initial bacterial counts on salmon portions The level of initial surface contamination on the salmon portions was modelled using three input variables: predicted water counts (cf. Section 3.6.1), surface water film, and contact surface contamination. The amount of water contaminating the fish surface was defined as the sum of the two distribution functions fitting the filter disc data. The water in the surface film was assumed to come from the slurry water, thus the slurry water contamination on the fish surface was the product of the volume of water on the fish surfaces and the bacterial load (cfu/ml) in the slurry water. Further contamination occurred along the processing line. Data for the level of contact surface contamination was obtained from routine environmental swabs taken at the processing factory. Three areas along the processing line were considered to make important contributions to the contamination of fish
Fig. 3. Comparison between the distribution function fitted to the observed and the predicted bacterial counts on salmon portions.
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
57
Fig. 4. Comparison between the range of predicted shelf life of aerobically and anaerobically stored portions.
surfaces. For each of these areas, a distribution function was fitted to the data and the results were averaged. Bacterial counts from the processing line varied between 0 and 3 log10 cfu/ml. Addition of the range of contact surface contamination with the contamination level from the slurry water produced a value that was close to the observed contamination levels found on the salmon portions. The predicted values were compared to the observed surface contamination levels (Fig. 3).
3.6.3. Modelling the shelf life of salmon portions The final stage of the model predicts the range of shelf lives possible for the salmon portions. The proportion of pseudomonads, in the initial contamination on the portions, ranged between 40% and 100%. These percentage values were converted to log10 cfu for shelf life predictions. For the simplicity of the modelling, the time that the salmon portions spend in vacuum packs was included as part of the lag phase of Pseudomonas.
Fig. 5. Relative importance of the input variables on the shelf life of salmon portions.
58
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
The predicted mean shelf life at 4 °C under aerobic conditions is 6.46 days, which is comparable with the average of 6.23 days obtained from the storage trials. The predicted shelf life of the aerobically stored portions and the portions that spend time in the vacuum packages are compared in Fig. 4. Also shown (Fig. 4) is a comparison of the cumulative probabilities for the predicted shelf life of the salmon portions stored aerobically and anaerobically. 3.6.4. Sensitivity analysis of input variables Sensitivity analysis performed on the input variables for the model revealed that the distribution/ storage temperatures and the time under which the salmon portions were stored after packaging had the greatest influence on the shelf life. The next most important variable was contact surface contamination. Fig. 5 shows the relative importance of each of the input variables. These were in increasing order of importance: initial water counts in the slurry bins, transport temperature and time, surface film contamination, contact surface contamination, and distribution time/temperature. Since temperature almost completely determines shelf life, attempts at improving the shelf life of the salmon portions should be directed at storing them at a lower temperature rather than attempting to reduce surface contamination or any of the other variables affecting the shelf life.
4. Discussion Process risk models involve situations in which uncertainty and variability restrict a definitive outcome due to limited data on the prevalence or density of the organism of concern, and/or absence of information on important elements of the risk assessment, for example the infective dose of a pathogen. Developing a process risk model to characterise the shelf life of a food product has (to our knowledge) not been reported in the scientific literature. However, it is a situation where a stochastic approach is likely to yield meaningful estimates of risk, as large amounts of relevant information on all parts of the harvesting, processing, and distribution chain can be collected in a longitudinal study. In this study, the production of salmon fillet portions for retail sale was examined for the PRM.
Stochastic modelling allows the identification of the variables with greatest influence in the production process and suggests means to implement improvements for extending shelf life. In this situation, it was the storage temperature that had the greatest influence on the final shelf life, which has been demonstrated previously (McMeekin and Olley, 1986; Ratkowsky et al., 1982, 1983; Huss et al., 2000). The other useful application of the model could be to explain the relatively short shelf life of the salmon portions, if this was due to the high numbers of spoilage pseudomonads. The effect of the proportion of the number of pseudomonads present, as part of the initial contamination on the salmon portions, can be seen. However, reductions in the number of pseudomonads will have little overall effect on the shelf life of the salmon portions. To extend the shelf life by any substantial amount, the growth rate of the spoilage pseudomonads would have to be limited. The normal handling procedure for the vacuumpackaged salmon portions which were studied is the transport to their retail destination, generally within 2 days of being packaged, at which time the vacuum packages are opened for display and retail sale. Growth of aerobic spoilage bacteria does not occur during the time between packaging and when the packages are opened, as these are in the lag phase, and the SSO (Pseudomonas) survives but is unable to grow in the anaerobic conditions prevailing during transport. To produce a PRM that predicts the shelf life of the portions, the lag phase and growth rate of the SSO were determined by storage trials. The growth of bacteria in the slurry water was calculated from the distribution of the average time/temperature conditions. The inclusion of actual temperature plots into the model was not practical due to the limitations of the modelling program used. However, the use of average time and temperatures was a good estimate of the conditions in the slurry bins. Ross (1999) demonstrated that in similar situations, the difference between the growth predicted from the average time/ temperature and the integrated time/temperature history was insignificantly small. Initial storage trials were conducted at 4 °C, with subsequent trials at higher and lower temperatures, to check the accuracy of the model outside of the original range. Several other salmon portions were stored at 1, 2 and 10 °C to
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
validate the model. The observed shelf life changed in proportion to the temperature, with the difference between the observed and predicted shelf lives being within 0.5 days of each other, for each experimental set. That is, the difference between the predicted and the observed shelf life at 1 °C was less than half a day. This indicated that the model was accurate at these temperatures, and that it is capable of predicting the shelf life of salmon portions between 1 and 10 °C. Overall good agreement was found between the predicted range of values and the fitted distributions. The predicted and observed values were almost identical for the bacterial loads in the slurry water (Fig. 2). This demonstrates clearly that where sufficient information is collected and uncertainty is minimised, a stochastic model can provide an accurate microbial profile of a specific processing operation. Conversely, where less data is available, the distributions may not overlap as closely, as can be seen with the comparison between the predicted bacterial contamination load on the portions and the distributions fitted to the observed data (Fig. 3). This comparison was further complicated by a change in operating procedures on the filleting line during the course of the 9-month study. Using the sensitivity analysis capabilities of Analytica, it was possible to determine the most influential factor on shelf life, and to rank strategies to extend the shelf life of the product. The model predicted that reduction of contamination in the slurry bin water by a factor of 10 would result in an extension of shelf life of 0.4 days, reducing contamination on the filleting line by a factor of 10 would give a predicted extension in shelf life of 1 day, whereas reduction in storage temperature from 6 to 3 °C would result in an increase in shelf life of 5 days. This indicates clearly that temperature is the most influential factor affecting the shelf life, a well-recognised phenomenon (McMeekin and Olley, 1986; Ratkowsky et al., 1982, 1983; Huss et al., 2000). In this context, the value of a PRM is not limited specifically to the prediction of shelf life of a specific food product. It can become an important element of a quality assurance program and will find utility in correcting potential microbiological problems by incorporation into an HACCP plan. While many authors restrict the HACCP concept to food safety issues, the principles are equally applicable to the
59
assurance of microbial food quality. In the specific example described, the hazard is the failure to reach a specified ‘use by’ date with the quality assurance points identified by the PRM. ‘‘What-if scenarios’’ using the PRM provide the processor with the most effective means to attain with certainty the required shelf life which may vary for different markets. A further advantage is that the PRM is not static and additional data can be added to a current model to further reduce uncertainty or to evaluate the effect of changes in processing protocols. We conclude that the application of stochastic approaches through development of a PRM constitutes a powerful tool to promote continual improvement and to attain microbial food quality objectives.
Acknowledgements The financial support and cooperation of a Tasmanian salmon producer and the University of Tasmania are gratefully acknowledged.
References Bremner, H.A., Olley, J., Vail, A.M.A., 1987. Estimating time – temperature effects by a rapid systematic sensory method. In: Kramer, D.E., Liston, J. (Eds.), Seafood Quality Determination. Elsevier Science, Amsterdam, pp. 413 – 435. Cassin, M.H., Lammerding, A.M., Todd, E.C.D., Ross, W., McColl, S.R., 1998. Quantitative risk assessment for Escherichia coli O157:H7 in ground beef hamburgers. International Journal of Food Microbiology 41, 21 – 44. Clesceri, L.S., Greenberg, A.E., Trussell, R.R., 1989. Standard Methods for the Examination of Water and Wastewater, 17th edn. American Public Health Association, Washington, DC, USA, pp. 9-30 – 9-35. Dalgaard, P., Gram, L., Huss, H.H., 1993. Spoilage and shelf life of cod fillets packed in vacuum or modified atmospheres. International Journal of Food Microbiology 19, 283 – 294. Fey, M.S., Regenstein, J.M., 1982. Extending shelf life of fresh wet red hake and salmon using CO2 – O2 modified atmosphere and potassium sorbate ice at 1 °C. Journal of Food Science 47, 1048 – 1054. Gill, C.O., Newton, K.G., 1977. The development of aerobic spoilage flora on meat stored at chill temperatures. Journal of Applied Bacteriology 43, 189 – 195. Huss, H.H., Reilly, A., Embark, K.B., 2000. Prevention and control of hazards in seafood. Food Control 11, 149 – 156. Liston, J., 1992. Bacterial spoilage of food. In: Huss, H.H., Jakobsen, M., Liston, J. (Eds.), Developments in Food Science 30:
60
S.K.J. Rasmussen et al. / International Journal of Food Microbiology 73 (2002) 47–60
Quality Assurance in the Fish Industry. Elsevier, Holland, pp. 93 – 105. McMeekin, T.A., Olley, J., 1986. Predictive microbiology. Food Technology in Australia 38, 331 – 334. McMeekin, T.A., Olley, J., Ross, T., 1991. Application of predictive microbiology to assure the quality and safety of fish and fish products. In: Huss, H.H., Jakobsen, M., Liston, J. (Eds.), Developments in Food Science 30: Quality Assurance in the Fish Industry. Elsevier, Holland, pp. 459 – 478. Molin, G., Stenstro¨m, I.-M., Ternstro¨m, A., 1983. The microbial flora of herring fillets after storage in carbon dioxide, nitrogen or air at 2 °C. Journal of Applied Bacteriology 55, 49 – 56. Neumeyer, K., Ross, T., McMeekin, T.A., 1997. Development of a predictive model to describe the effects of temperature and water activity on the growth of spoilage pseudomonads. International Journal of Food Microbiology 38, 45 – 54. Olley, J., Ratkowsky, D.A., 1973. Temperature function integration and its importance in the storage and distribution of flesh foods
above the freezing point. Food Technology in Australia 25, 66 – 73. Nielsen, J., Reines, J.H., Jespersen, C.M., 1991. Quality assurance in the fishing industry with emphasis on the future use of vision techniques. In: Pau, L.F., Olafsson, R. (Eds.), Fish Quality Control by Computer Vision. Marcel Dekker, New York, pp. 3 – 20. Ratkowsky, D.A., Olley, J., McMeekin, T.A., Ball, A., 1982. Relationship between temperature and growth rate of bacterial cultures. Journal of Bacteriology 149, 1 – 5. Ratkowsky, D.A., Lowry, R.K., McMeekin, T.A., Stokes, A.N., Chandler, R.E., 1983. Model for bacterial culture growth rate throughout the entire biokinetic temperature range. Journal of Bacteriology 154, 1222 – 1226. Ross, T., 1999. Predictive microbiology for the meat industry. Meat and Livestock Australia, Sydney, 196 pp. Sumner, J.L., Gorczyca, E., Cohen, D., Brady, P., 1984. Do fish from tropical waters spoil less rapidly in ice than fish from temperate waters? Food Technology in Australia 36, 328 – 329.