International Journal of Food Microbiology 128 (2008) 10–15

Contents lists available at ScienceDirect

International Journal of Food Microbiology

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / i j f o o d m i c r o

Application of predictive modelling techniques in industry: From food design up to

risk assessment
Jeanne-Marie Membré a,⁎, Ronald J.W. Lambert b,c
a
Safety and Environmental Assurance Centre, Unilever, Colworth Park, Sharnbrook, Bedford MK44 1LQ, UK
b
Nestlé Research Centre Lausanne, Vers-Chez-Les-Blanc, 1000 Lausanne 26, Switzerland
c
Current address: Cranfield Health, Cranfield University, MK45 4DT, UK

A R T I C L E I N F O A B S T R A C T

Keywords: In this communication, examples of applications of predictive microbiology in industrial contexts (i.e. Nestlé
Industrial applications and Unilever) are presented which cover a range of applications in food safety from formulation and process
Product development design to consumer safety risk assessment. A tailor-made, private expert system, developed to support safe
Modelling tools
product/process design assessment is introduced as an example of how predictive models can be deployed
Risk assessment
for use by non-experts. Its use in conjunction with other tools and software available in the public domain is
discussed. Specific applications of predictive microbiology techniques are presented relating to investigations
of either growth or limits to growth with respect to product formulation or process conditions. An example
of a probabilistic exposure assessment model for chilled food application is provided and its potential added
value as a food safety management tool in an industrial context is weighed against its disadvantages. The role
of predictive microbiology in the suite of tools available to food industry and some of its advantages and
constraints are discussed.
© 2008 Elsevier B.V. All rights reserved.

1. Introduction strongly enabled the development of the mathematics and computer

Bridging the gap between scientific development and practical
implementation in industry has always been a challenge. Early
examples of the application of scientific principles to food preserva-
tion include Pasteur's work on the specificity of undesirable
fermentations in wine and the supply of lactic starter cultures by
Hansen in Denmark at the end of the 19th century. The Arrhenius
equation and the Bigelow model were used to mathematically
describe the effect of temperature on the heat resistance of micro-
organisms. The botulinum cook, developed some 90 years ago, may
have been established on the basis of the first predictive food model,
albeit only recently recognised as such, and has found widespread
utility in the food industry (McMeekin et al., 2002).
The 1980s saw a marked increase in interest in predictive micro-
biology as a result of some major food poisoning outbreaks caused by
both ‘traditional’ pathogens and foods (e.g. Salmonella in eggs) as well
as emerging pathogens with unusual characteristics (e.g. Listeria
monocytogenes able to grow at refrigeration temperatures). The
resulting public and political awareness contributed to the prioritisa-
tion of food safety research by governments in the USA, Europe,
Australia and New Zealand. The concomitant maturation of informa-
tion technology and step-change increase in computer technology
⁎ Corresponding author. Tel.: +44 1234 264881; fax: +44 1234 222632.
E-mail address: Jeanne-Marie.Membre@Unilever.com (J.-M. Membré).
power necessary for predictive modelling.
In the last two decades, predictive microbiology (PM) has
established itself as a scientific discipline in its own right, with
regular international multi-disciplinary conferences, scientific articles
in peer-reviewed journals, books and internet websites.
Current applications of PM in an industrial context are wide but
can be summarised into three groups of activities:
• Product innovation: Assessing speed of microbial proliferation,
growth limits, or inactivation rate associated with particular food
formulations and/or process conditions in order to develop new
products and processes, reformulate existing products, determine
storage conditions and shelf-life;
• Operational support: Supporting food safety decisions that need to
be made when implementing or running a food manufacturing
operation, such as designing in-factory heating regimes, setting
critical control points (CCPs) in HACCP, assessing impact of process
deviations on microbiological safety and quality of food products;
• Incident support: Estimating the impact on consumer safety or
product quality in case of problems with products on the market.
Beside this current use, PM might be also utilised to transfer new
risk management concepts into practical guidelines (Gorris, 2005;
Membré et al., 2007a).
In the following sections, examples of different types of applica-
tions of PM within Nestlé and Unilever are presented and some
lessons learnt regarding its utility in industrial contexts are discussed.

0168-1605/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijfoodmicro.2008.07.006
 J.-M. Membré, R.J.W. Lambert / International Journal of Food Microbiology 128 (2008) 10–15
2. Internal/external microbiological expert system, tools
and software
A classical application of PM is assessment of bacterial proliferation
or inactivation regarding particular intrinsic parameters (e.g. pH, water-
activity, acids, salt, preservatives), extrinsic parameters (e.g. chilling,
modified atmosphere) or processing (e.g. heat treatment, pressurisation,
irradiation). Many different researchers have developed models for
particular micro-organisms and made them available in the public
domain through peer-reviewed publications (McKellar and Lu, 2004).
From an industry perspective, the utility of these public-domain models
is often somewhat limited. In many cases, models have been developed
under laboratory conditions, are based on specific combinations of
parameters that might not be appropriate for the particular food
products of an industry, and have not always been validated or even used
in real food systems. Despite that, such models can be useful as long as
their limitations are recognised and considered in their application. To
overcome certain limitations, companies like Unilever have developed
and validated PM models tailored to their needs and the particular food
products in their portfolio. In general, these models are used by company
experts mostly to define windows-of-opportunity for safe product
innovation but can also be established for individual product innova-
tions. To make the deployment of these models effective and practical in
a large company, they would best be fit for use by non-experts which, in
the case of Unilever, are product developers residing in several centres
around the world. For this purpose, the company established a
microbiological expert system which is accessible by its staff through
the company's intranet. Users need to be trained before being able to
access this system. Part of the training is helping them to understand the
applicability and constraints of the expert system. One key applicability
aspect is that the system provides valuable information only within the
“window-of-opportunity” that the underlying models provide and that
innovations outside this window require the involvement of the
appropriate company experts. Typical constraints that staff need to
understand are that the system provides simulated information that
helps cut down on practical experimentation, but that does not fully
represent reality in terms of, for instance, the parameters considered or
the actual micro-organisms concerned. Given this understanding, the
expert system has been successfully used for many years in Unilever and
has led to speed up time-to-market in a responsible way. Fig. 1 shows a
screenshot of the expert system. Since its development, this tool has
been regularly updated using advances in the science of PM and
considering internal demand, in term of interface facilities and range of
applications. The fact that the system is held in a central location is a
significant advantage considering such updates.
In the absence of tailor-made PM models, as stand stand-alone tools
or in the form of an expert system, food companies may find generic PM
Fig. 1. Illustration of microbiological expert system accessible through Unilever intranet; the main users are non-modelling experts, e.g. product developers; it helps them to speed up
time-to-market in a responsible way; it is regularly updated using advances in science and also internal demand.
11
software such as PMP (http://www.arserrc.gov/mfs/PMP7_start.htm),
Growth Predictor (http://www.ifr.ac.uk/Safety/GrowthPredictor/) or
Sym'Previus (http://www.symprevius.net/, available through a yearly
subscription), very valuable resources. These pieces of software are easy-
to-use and allow experts and non-experts alike to obtain predictions on
microbial behaviour in three or four clicks. These systems are very user-
friendly and keep the underlying complexity from the user. Comparing
the different systems for the magnitude of difference in their predictions
in essence is a good practice, but in order to be able to decide which
system provides the most appropriate and useful results for the food
product at hand, some targeted experimentation will be necessary to
validate applicability in the particular product. The fact that different
systems provide different answers even when identical input para-
meters are used is not unexpected, but it may be difficult to appreciate
especially by non-trained, non-experts who typically may be overly
confident in predictions and may be faced with predictions that could be
significantly different requiring them to make a choice. For instance,
with L. monocytogenes, at temperature 10 °C, pH 6, aw 0.996, a 2 log-
increase is achieved within 48 h, 62 h or 82 h using Sym'Previus, PMP or
Growth Predictor, respectively. Of course, there are reasons behind these
model output differences, one being that different primary and
secondary models have been used to fit the data and then to generate
predictions, but these reasons are likely to be noticeable by PM
specialists only. A constraint of most PM software is that they cover
only a certain number of micro-organisms, often only pathogenic micro-
organisms, and have established the models most often in broth media.
As indicated before, this limits the utility for industrial application
regarding product development although work is ongoing for most
systems to extend the range of micro-organisms covered (including
spoilage micro-organisms) and to add models that have been validated
in food systems.
Despite these limitations, PM software and also PM databases such
as ComBase (http://www.combase.cc/links.html), can be of particular
value in situations where “instantaneous” information is required to
allow for a quick decision on product safety. This can be for instance
when a product on the market is believed to have manufacturing
defects and the magnitude of the impact of such defects on consumer
safety needs to be assessed on very short term.
Notably, PM models have found good application in quantitative
risk assessments. As illustrated further on (Section 4), external models
that have been implemented in PM software as well as published
stand-alone models may facilitate both hazard identification and
exposure assessment in risk assessments. While they sometimes are a
part of the exposure assessment model, they can more often even help
in screening the various hazards likely to be involved, establishing
which factors impact on microbiological behaviour and to what extent
and identifying suitable hazard mitigation strategies.
12 J.-M. Membré, R.J.W. Lambert / International Journal of Food Microbiology 128 (2008) 10–15

3. Rapid predictive microbiology using time to detection method hypothesis. The model can be used to quickly construct a challenge test,
by targeting areas specific to the developers needs, e.g. a given shelf-life.
It has been stated that a predictive model can give results at least Another example relates to Staphylococcus aureus, which remains a
1000 times faster than a traditional challenge test (Zwietering et al., major cause of food borne illness and is especially a concern in
1996). Indeed it was suggested that the challenge test be used to increase intermediate moisture foods with aw in the region 0.84–0.9. PM models
the confidence in the predictive model. In an industrial context, more for S. aureus are well established (Buchanan et al., 1993; McCann et al.,
often a predictive model arises from challenge test data, which is 2003). The Nestlé software model for S. aureus is based on the study of
particularly the case for growth/no-growth models. If a model is already ATCC standard 6538 (S. aureus subsp. aureus Rosenbach, also NCIB 9518,
in existence for a particular product design then it is a very valuable tool NCTC 10788). Application of the TTD modelling technique (Lambert and
for subsequent further refinements or alterations. Conversely, for new
product development, if a bespoke predictive model takes months to
obtain using traditional methods, this can be of little direct value if
product development has the same time scale. A model (e.g. a logistic
growth/no-growth model) produced for one product design may be of
no use for another design type where, for example, a different pre-
servation system has been applied. Often, what is required is knowledge
of the areas of growth/no-growth to design a more optimum challenge
study. Time to detection (TTD) can be used to quickly provide such
‘product-stability maps’ (Lambert and Bidlas, 2007a). These can then be
used to direct the expensive challenge tests, which ensure a focussing of
the limited resource available to do such studies.
TTD using optical density techniques is as valid as the traditional
plate techniques used for growth curves. The growth curve examines the
total change in microbial numbers with time whereas TTD focuses very
accurately on a smaller part of the growth curve, normally near the end
of exponential growth. Although parameters such as the specific growth
rate and lag can be obtained from TTD (Cuppers and Smelt, 1993;
Lindqvist, 2006) and attempts have been made to directly use optical
density data for the construction of growth curves (McClure et al., 1993)
it is the ability to produce vast amounts of accurate data on growth very
quickly that appeals to an industrial laboratory. One particular problem
with using TTD are the limited number of models available — much
published work involves fitting primary-type models to the optical
density curve and this can lead to erroneous conclusions, especially
where no calibration curve is used as pointed out e.g. by Dalgaard and
Koutsoumanis (2001).
By using TTD and through the assumption of the Gamma hypothesis
(Zwietering et al., 1992), predictive models can be produced within a few
days, which can be augmented or modified ensuring that PM becomes a
timely tool for product development. For example, following the TTD
approach, it took Nestlé a total of 10 days (Two people working a total of
60 hours and using six Bioscreen instruments (Labsystems, Helsinki,
Finland)) to develop a software model for Salmonella Poona that is able to
predict the independent effects of pH, salt, acetate and sorbate. Combined
studies of the various factors showed that the Gamma hypothesis was
valid, i.e. that each hurdle in combination can be treated independently
within the overall model (Lambert and Bidlas, 2007c). The calculated
minimum growth pH (3.81) and minimum growth aw (0.948) were
consistent with the range given in the literature for salmonellae. Fig. 2a
shows an example of the model output for combinations of pH and salt.
The outer contour of the model output is the absolute growth/no-growth
contour irrespective of initial inoculum or temperature. This is where the Fig. 2. a: An ‘Intervention Diagram’ for the combined hurdles of salt and pH against Sal-
product of the gamma factors results in no growth. monella Poona. This figure plots the relative effects of the combined gamma factors for pH
The plot is split into 4 so-called ‘intervention areas’ based on the and salt. At a value of 1, no effect on growth is observed and uninhibited growth with respect
to pH and salt will be observed, at a combined gamma=0.1, there is a ten fold increase in the
value of the product of the gamma factors. These areas are reflective of time to detection relative to the control environment. Contours for the combined gamma
the size of the hurdle requirement if a product has salt/pH attributes factors of 0.1, 0.01 and 0.001 are used in the figure to suggest the magnitude of the (extra)
within that area. These range from a major hurdle(s) requirement to hurdle application that should be applied. Key: Red = major hurdles required; Yellow =
none (outer contour and beyond). An applied hurdle can be a reduced moderate level of hurdles required; Green = minimum hurdle intervention required; Blue =
no intervention required to prevent growth. b: An ‘Intervention Diagram’ for the combined
temperature, an increase in hygiene (to reduce initial loading), a
hurdles of salt, pH, Na acetate (1500 mg l− 1) and K sorbate (250 mg l− 1) against Salmonella
shorter shelf-life or the addition of weak acid preservatives. Fig. 2b Poona. The addition of a specific concentration of Na acetate and K sorbate has a differential
shows the same system (in terms of pH and aw) but with the addition effect over the pH values given due to the pH dependent equilibrium of the acid and anion
of 1500 mg l− 1 Na acetate and 250 mg l− 1 of K sorbate. The minimum forms. At the lower pH there is a greater concentration of the antimicrobial acetic and sorbic
pH for growth has increased (there is a ‘hidden’ gamma factor from acids and this increases the effectiveness of the combined antimicrobial factors. With
respect to a plot of only the gamma factors for pH and salt, the influence of the gamma factor
the concentration of acetic acid, which has the effect of giving an for acetic and sorbic acids is seen as an increase in the minimum pH for growth. (For
apparent increase in the minimum pH for growth), whereas the interpretation of the references to colour in this figure legend, the reader is referred to the
minimum aw has remained constant — a requirement of the Gamma web version of this article.)
 J.-M. Membré, R.J.W. Lambert / International Journal of Food Microbiology 128 (2008) 10–15 13

Fig. 3. Effect of added salt on the loge time to detection of five strains of Staphylococcus aureus (left to right: ST121, ST55, ST68,ST84, ATCC 6538.; at NaCl = 12.2%, ST121 and ATCC 6538
failed to grow during the time of incubation of 5 days). Error bars are the standard deviation of the mean ln(TTD) value for 10 replicates per strain for each salt concentration.

Bidlas, 2007b) produced a multi-factor model including pH, aw (as salt),
various weak acids including acetate, sorbate, benzoate and nitrate, and
temperature. The model appeared to reproduce the experimental
observations well, especially in combinations. Product developers,
however, complained that the model appeared too conservative
especially with respect to pH and aw. Originally (in 2004) experiments
were done using a serial dilution of salt — usually a half-fold dilution
sequence from 20%. Since concentrations below 2% of salt have very little
effect on the time to growth of S. aureus, and since most data was
collected below this, there were few data in the region which enabled an
accurate modelling of the inhibition (i.e. at concentrations N5%).
Repeating the salt study using linear dilutions between 0.5 and 20% of
salt, returned the approximate literature values for pHmin (4.2) and a
calculated MIC of salt of 15.94% (aw = 0.88) for the standard strain used.
Assuming that there might be strong strain dependence, a selection
of factory isolates were examined in a similar way: pHmin ranged from
4.2–4.5, and MICsalt 14.6–16%. Fig. 3 shows the logarithmically
transformed TTD results for all strains examined. The diagram appears
to show that although there are small differences between strains,
there is an overall similar response to increasing salt concentration.
With increasing salt concentration the TTD increased as did the
variance in the data, this is a common observation as conditions
become more inimical (Ratkowsky et al., 1991; Ratkowsky et al., 1996;
Schaffner, 1998). The use of the log transformation to stabilise the
variance in order to conduct unbiased regression modelling is
commonplace, the use of the reciprocal transformation as used in
the model described by Lambert and Bidlas (2007a) also homogenises
the variance. The calculated MICsalt of the standard strain was found to

Fig. 4. Schematic representation of the exposure assessment model structure used to estimate probability of having 105 cfu g− 1 or more of Bacillus cereus in chilled foods, at the time
of consumption, under realistic market conditions. 'HT' stands for Heat-Treated or Heat Treatment.
14 J.-M. Membré, R.J.W. Lambert / International Journal of Food Microbiology 128 (2008) 10–15
be higher than the factory isolates, so is in effect conservative, but in a
positive/safe way.
4. Use of probabilistic exposure assessment in food
safety management
Since mid-90s, microbiological risk assessments have been devel-
oped in different parts of the world, early on mostly by academics, later
more and more by government experts or under their auspices. In all
quantitative risk assessments, some form of modelling is involved and in
many cases such models include PM models or combinations of PM
models with process models. The aim of such risk assessments is to
characterise and estimate the risk to consumers related to a certain
pathogen possibly present in a particular food product or product group.
While it is of interest to the industry to learn from those risk
assessments, there are also interesting industrial applications of some
of the techniques used in risk assessment (Lammerding, 2007). Unilever
staff have been engaged with some of those applications, e.g. linking
industrial food safety management to risk assessment or establishing
more sophisticated exposure assessments than traditionally were
possible (Brown et al., 1998; Van Gerwen and Gorris, 2004; Membré
et al., 2006). As an example of the latter application, Unilever developed
a probabilistic exposure assessment to assess the safety of REPFEDs
(refrigerated processed foods of extended durability) in relation to Ba-
cillus cereus. Fig. 4 illustrates the many factors considered in this
assessment, which included raw material contamination, heat treat-
ment, bacterial spore heat resistance, injured spore lag time and growth
potential, chilled supply-chain market (split into warehouse, retail,
consumers steps) and consumers' habits (e.g. time for shopping, time
before consuming). A sensitivity analysis, including uncertainty and
variability separation, was run to identify the impact of each factor on
the output (Membré et al., 2007b).
The exposure assessment model was structured to provide the
probability of having 105 cfu/g or more of B. cereus at the point of
consumption. Different scenarios of process, formulation and chilled
market were generated and outputs compared. In Fig. 5, the model
output is presented for two heat-treatment temperatures and three
heat-treatment times. The 95% confidence interval bars in this graph
capture the uncertainty in the model output. From these outputs, iso-
probability contour plots (Havelaar et al., 2004) can be generated to help
food developers to choose between various processes and formulations
that perform equivalently, this in analogy to the “window-of-opportu-
nity” approach followed with individual PM models or the Unilever
expert system.
Developing a probabilistic exposure assessment model, is of course
more time demanding, more resource and computer intensive than
building deterministic PM models involving two or three factors (e.g.
Fig. 5. Illustration of simulation results showing the probability of having 105 cfu g−1 or
more of Bacillus cereus when the product is consumed after various heat-treatment
conditions. Simulations were carried out for product with high raw material
contamination, pH of 6, aw of 0.996 and a shelf-life of 40 days.
temperature, aw and pH). Moreover, probabilistic models involving
sophisticated techniques such as 2D Monte Carlo decomposition, and,
uncertainty and variability separation, are difficult to set up, interpret
and communicate. They require skilled and experienced staff to be
deployed successfully. These constraints are out-balanced by the value
of running more realistic scenarios than deterministic assessments
would allow (e.g. considering variations of refrigerator temperatures),
being explicit and transparent in terms of the knowledge/data used in
the assessment as well as the uncertainty in the outputs, facilitation of
identification of data gaps and prioritising key risk contributing or
mitigating factors.
5. Conclusions
The predictive modelling history in Unilever and Nestlé shows
parallels with the rise in utility and applicability of PM within food
microbiology in general. The difference shows a strong lead by Unilever
microbiologists and managers to bring PM into their industry, whereas
Nestlé has preferred to continue with challenge test methods
augmented by published models or particular external studies where
appropriate. As product development continues to accelerate in nearly
every area of food manufacture, the use of PM as an accelerating tool is
recognised by all of industry. The degree to which it is applied in specific
areas is dependent on the understanding within the management
structures of the particular industries and the knowledge base that
underpins those decisions. However, its diligent application comes with
a number of pre-requisites amongst which are adequately increasing the
skill and training of non-expert users.
Many manufacturers have used PM to reduce the risk to consumer
associated with pathogen growth. The same is almost certainly true for
spoilage organisms, but tailor-made models for these are not or very
rarely published by industry for competitor advantage reasons. Product
developers often require models of spoilage organisms, for example in
low pH beverages yeast and mould spoilage is extremely important.
Only a few published mould and yeast models exist, however, and even
fewer for this particular food product. A recent review by (Dantigny et al.,
2005) emphasised the gap between the bacterial (especially pathogen)
models and those dealing with fungi: “In order to improve food quality
and safety, there is a need for a tool allowing prediction of fungal
development.”
Challenge studies require considerable labour, time, and materials,
and the number of parameters that can be tested is often limited. If a
predictive model is available (especially a validated model), then rapid
information about the microbial stability (with respect to spoilage) of
a product can be obtained and can itself be used in conjunction with
challenge studies to improve product stability and reduce costs. This
makes such work, when done appropriately and diligently, quite
valuable.
Within a safety frame, PM is recognised as a versatile and helpful (if
not essential) tool for risk assessment and the maturation of this area
of PM application is far from completed. Predictive modelling is a
continuously developing tool in industry, being driven by academics,
academic-industry joint ventures, governments, software and statis-
tical developments, legal and regulatory issues and to some extent the
industry's need to innovate and deliver food products that satisfy
consumer demands. Often, however, it is the need for ease of
communication to the internal end user, typically the product/process
developer that is the crux within industry. Application of predictive
microbiology within Unilever and Nestlé and probably in other
industries too, will continue along with further scientific develop-
ments in modelling techniques, microbiology and food technology.
Acknowledgements
The authors would like to thank Leon Gorris for critically reading
and improving the draft version of this manuscript.
 J.-M. Membré, R.J.W. Lambert / International Journal of Food Microbiology 128 (2008) 10–15 15

References McKellar, R.C., Lu, X., 2004. Modeling microbial responses in food. CRC Press, London.
McMeekin, T.A., Olley, J., Ratkowsky, D.A., Ross, T., 2002. Predictive microbiology:
towards the interface and beyond. International Journal of Food Microbiology 73,
Brown, M.H., Davies, K.W., Billon, C.M.P., Adair, C., McClure, P.J., 1998. Quantitative
395–407.
microbiological risk assessment: principles applied to determining the comparative
Membré, J.-M., Amézquita, A., Bassett, J., Giavedoni, P., Blackburn, C.d.W., Gorris, L.G.M.,
risk of salmonellosis from chicken products. Journal of Food Protection 61, 1446–1453.
2006. A probabilistic modeling approach in thermal inactivation: estimation of
Buchanan, R.L., Smith, J.L., McColgan, C., Marmer, B.S., Golden, M., Dell, B., 1993.
postprocess Bacillus cereus spore prevalence and concentration. Journal of Food
Response surface models for the effects of temperature, pH, sodium chloride, and
Protection 69, 118–129.
sodium nitrite on the aerobic and anaerobic growth of Staphylococcus aureus 1961.
Membré, J.-M., Bassett, J., Gorris, L.G.M., 2007a. Applying the food safety objective and
Journal of Food Safety 13, 159–175.
related standards to thermal inactivation of Salmonella in poultry meat. Journal of
Cuppers, H.G.A.M., Smelt, J.P.P.M., 1993. Time to turbidity measurement as a tool for
Food Protection 70, 2036–2044.
modelling spoilage by Lactobacillus. Journal of Industrial Microbiology 12, 168–171.
Membré, J.-M., Kan-King-Yu, D., Blackburn, C.d.W., 2007b. Development of a proba-
Dalgaard, P., Koutsoumanis, K., 2001. Comparison of maximum specific growth rates
bilistic lag model to predict the fate of Bacillus cereus spores in heat-treated
and lag times estimated from absorbance and viable count data by different
chilled foods (REPFEDs). In: Nychas, G.J.E., et al. (Ed.), 5th International Confer-
mathematical models. Journal of Microbiology Methods 43, 183–196.
ence of Predictive Modelling in Foods. Agricultural University of Athens, Greece,
Dantigny, P., Guilmart, A., Bensoussan, M., 2005. Basis of predictive mycology. International
pp. 261–264.
Journal of Food Microbiology 100, 187–195.
Ratkowsky, D.A., Ross, T., Macario, N., Dommett, T.W., Kamperman, L., 1996. Choosing
Gorris, L.G.M., 2005. Food safety objective: an integral part of food chain management.
probability distributions for modelling generation time variability. Journal of
Food Control 16, 801–809.
Applied Bacteriology 80, 131–137.
Havelaar, A.H., Nauta, M.J., Jansen, J.T., 2004. Fine-tuning food safety objectives and risk
Ratkowsky, D.A., Ross, T., McMeekin, T.A., Olley, J., 1991. Comparison of Arrhenius-type
assessment. International Journal of Food Microbiology 93, 11–29.
and Belehradek-type models for prediction of bacterial-growth in foods. Journal of
Lambert, R.J.W., Bidlas, E., 2007a. A study of the Gamma hypothesis: predictive
Applied Bacteriology 71, 452–459.
modelling of the growth and inhibition of Enterobacter sakazakii. International
Schaffner, D.W., 1998. Predictive food microbiology Gedanken experiment: why do
Journal of Food Microbiology 115, 204–213.
microbial growth data require a transformation? Food Microbiology 15, 185–189.
Lambert, R.J.W., Bidlas, E., 2007b. Gamma study of pH, nitrite and salt inhibition of
Van Gerwen, S.J.C., Gorris, L.G.M., 2004. Application of elements of microbiological risk
Aeromonas hydrophila. Applied and Environmental Microbiology 73, 2239–2246.
assessment in the food industry via a tiered approach. Journal of Food Protection 67,
Lambert, R.J.W., Bidlas, E., 2007c. Rapid predictive modelling for product development.
2033–2040.
Italian Journal of Food Science Special Issue: Shelf-life International Meeting 10–18.
Zwietering, M.H., deWit, J.C., Notermans, S., 1996. Application of predictive micro-
Lammerding, A.M., 2007. Using Microbiological Risk Assessment (MRA) in Food Safety
biology to estimate the number of Bacillus cereus in pasteurised milk at the point of
Management. ILSI Europe, Brussels. 90-78637-05-9.
consumption. International Journal of Food Microbiology 30, 55–70.
Lindqvist, R., 2006. Estimation of Staphylococcus aureus growth parameters from
Zwietering, M.H., Wijtzes, T., de Wit, J.C., Van't Riet, K., 1992. A decision support system
turbidity data: characterization of strain variation and comparison of methods.
for prediction of the microbial spoilage in foods. Journal of Food Protection 55,
Applied and Environmental Microbiology 72, 4862–4870.
973–979.
McCann, T.L., Eifert, J.D., Gennings, C., Schilling, M.W., Carter, W.H., 2003. A predictive
model with repeated measures analysis of Staphylococcus aureus growth data. Food
Microbiology 20, 139–147.
McClure, P.J., Cole, M.B., Davies, K.W., Anderson, W.A., 1993. The use of automated
turbidometric data for the construction of kinetic models. Journal of Industrial
Microbiology 12, 277–285.