Food Control 29 (2013) 394e400

Contents lists available at SciVerse ScienceDirect

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

Multiple regression model for thermal inactivation of Listeria

monocytogenes in liquid food products
J. Hein M. van Lieverloo a, *,1, Mattheus de Roode a, 2, Martijn B. Fox a, Marcel H. Zwietering b,
Marjon H.J. Wells-Bennik a
a
b

NIZO Food Research, Kernhemseweg 2, 6718 ZB Ede, The Netherlands

Laboratory of Food Microbiology, Wageningen University, PO Box 8129, 6700 EV Wageningen, The Netherlands

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 29 December 2011
Received in revised form
18 May 2012
Accepted 22 May 2012

A multiple regression model was constructed for thermal inactivation of Listeria monocytogenes in liquid
food products, based on 802 sets of data with 51 different strains and 6 cocktails of strains published
from 1984 to 2010. Signicant variables, other than inactivation temperature, were pH, sodium chloride
content, sugar content, the temperature of growth or storage before inactivation, in addition to a heat
shock before inactivation. The constructed model for thermal inactivation of L. monocytogenes has
a reduced variability as these variables are known to inuence the thermal resistance (and these are
known or controllable in practice). Mean simulation results of inactivation of L. monocytogenes during
pasteurisation (20 s, 76
C) of raw milk (calculated mean level after growth 14 cfu/l) were comparable
with results of a single regression model constructed from inactivation data found in experiments in milk
only (175 data sets, 18 strains/cocktails). Both models predicted a probability of survival of less than 1 in
a billion litres. The study shows that multiple regression modelling can be used to obtain a model from
all data available, with a limited and realistic uncertainty level, while retaining the variability of heat
resistance due to the 51 strains and 6 cocktails of strains (unknown and not controllable in practice).
2012 Elsevier Ltd. All rights reserved.

Keywords:
Pasteurisation
Sterilisation
pH
Sodium chloride
Sugar
Heat shock

1. Introduction
Listeria monocytogenes can cause listeriosis, a severe illness
with a high mortality rate, especially in vulnerable parts of populations, e.g. unborn children, elderly and immune-decient people
(World Health Organization, 2004, p. 269). L. monocytogenes is very
capable of growth in some food products, even when stored at
refrigeration temperatures. For such products, Listeria needs to be
inactivated, e.g. by thermal processing, and the probability of
survival of a single cell needs to be very low. The variability of
the efcacy of thermal inactivation of L. monocytogenes (e.g.
during pasteurisation) can be estimated using a single linear

* Corresponding author. Brabant Water, Magistratenlaan 200, 5223 MA s-Hertogenbosch, The Netherlands. Tel.: 31 73 6838888.
E-mail addresses: Hein.van.Lieverloo@brabantwater.nl, Hein.van.Lieverloo@
viaeterna.nl (J.H.M. van Lieverloo), matthew.de.roode@sensus.nl (M. de Roode),
Martijn.Fox@nizo.com (M.B. Fox), Marcel.Zwietering@wur.nl (M.H. Zwietering),
Marjon.Wells-Bennik@nizo.com (M.H.J. Wells-Bennik).
1
Present address: Viaeterna, Zandstraat 8, 5242 GR Rosmalen, The Netherlands.
Tel.: 31 6 24269886.
2
Present address: Sensus b.v., Borchwerf 3, 4704 RG Roosendaal, The
Netherlands.
0956-7135/$ e see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.foodcont.2012.05.078

regression model that is based on publicly available literature

data, not taking into account effects of variables other than inactivation temperature and holding time (Van Asselt & Zwietering,
2006).
When calculating inactivation of L. monocytogenes for a specic
food and specic processing conditions in Monte Carlo simulations,
this model is likely to overestimate the variability of the thermal
inactivation efcacy, as the model only takes into account the effect
of temperature and holding time. Factors that contribute to
increased variability include differences in food composition (pH,
NaCl etc.), processing conditions (e.g. those before heating),
bacterial strain characteristics (intra-species variability) and other
variables that have been shown to inuence thermotolerance of
L. monocytogenes (Doyle, Mazotta, Wang, Wiseman, & Scott, 2001).
L. monocytogenes has been shown to be more heat-sensitive in, for
example, apple juice (pH 4, Mazzotta, 2001) than in raw milk (pH
6.6, Doyle et al., 2001) and less sensitive in liquid egg products with
NaCl or sugar than in liquid egg products without these additives
(Palumbo, Beers, Bhaduri, & Palumbo, 1995). The variance of the
single, unadjusted predictive regression model based on
measurements in all sorts of media and food products is too high to
realistically predict variability of inactivation in a specic product
under specic processing conditions.

J.H.M. van Lieverloo et al. / Food Control 29 (2013) 394e400

However, restriction to a certain food group or product and xed

processing conditions may lead to underestimation of the variability of thermotolerance as inactivation data for single food
groups or products are available only for a limited number of
L. monocytogenes strains (Den Besten & Zwietering, 2012). The
objective of this research was to generate a multiple regression
model that can predict (variability of) thermal inactivation for
a maximum variety of strains from literature data while accounting
for effects of food composition and processing conditions. Thus, the
variance in the predictive inactivation model is limited to the
variance due to uncertainty about the strain present in the raw
product. As specic data on food composition are lacking in most
reported studies on heat inactivation in solids (sh, seafood, meat,
vegetables), the model was limited to uid foods and uid media. A
preliminary version of the model was presented at the Seventh
International Conference on Predictive Modelling in Foods in
Dublin (Van Lieverloo, De Roode, Fox, Zwietering, & Wells-Bennik,
2011).

395

 Heat shock (yes/no) and the temperature difference with last

phase before shock (n 54, 7e41
C);
 Adaptation to sodium chloride (n 6) or acid (n 56).
Missing data on pH and concentrations of fat, salt and sugars
in growth media, dairy, juices and egg (parts) were estimated
from other literature (pH of gravy: Huang, Yousef, Matthews, &
Marth, 1993) or the internet (see list of websites consulted,
following the list of literature references). Data sets (26) with
antimicrobials were not included: peroxide (Kamau, Doores, &
Pruitt, 1990; Lou & Yousef, 1996; Palumbo, Beers, Bhaduri, &
Palumbo, 1996), lactoperoxidase (Kamau et al., 1990), nisin
(Knight, Bartlett, McKellar, & Harris, 1999; Maisnier-Patin et al.,
1995) and ethanol (Lou & Yousef, 1996). In total, the 807 data
sets from 53 papers included 51 L. monocytogenes strains and 6
cocktails of strains. 279 data sets involved experiments with
strain Scott A, 118 data sets involved experiments with cocktails
of strains.
3. Theory and calculation

2. Materials and methods

3.1. Model construction
Heat inactivation data according to the Bigelow model
(Bigelow, 1921) and some condition variables were present in
a database constructed from literature as described in a metaanalysis by Van Asselt and Zwietering (2006). Data on more
variables were collected from papers cited in the latter paper
(Casadei, Esteves de Matos, Harrison, & Gaze, 1998; Chhabra,
Carter, Linton, & Cousin, 1999; Doyle et al., 2001; Holsinger,
Smith, Smith, & Palumbo, 1992; ICMSF, 1996; Mazzotta, 2001)
and 37 original papers from the reviews of ICMSF (1996) and
Doyle et al. (2001).
The database was further supplemented with other and more
recent literature on thermal inactivation in uid foods (EdelsonMammel, Whiting, Joseph, & Buchanan, 2005; Hassani, lvarez,
Raso, Condn, & Pagn, 2005; Hassani, Manas, Pagn, &
Condn, 2007; Hassani, Manas, Raso, Condn, & Pagn, 2005;
Huang, 2004; Ignatova, Leguerinel, Guilbot, Prvost, & Guillou,
2007; Juneja & Eblen, 1999; Maisnier-Patin, Tatini, & Richard,
1995; Van der Veen, Wagendorp, Abee, & Wells-Bennik, 2009).
Five groups of liquid heating menstrua were distinguished: Dairy
(n 260: milk, cream, butter and ice-cream), liquid eggs
(n 107: whole, white and yolk), gravy (n 87; beef and
chicken), fruit and vegetable juices (n 32: apple, orange, white
grape and cabbage) and liquid media (n 321: deionised water,
physiological saline, phosphate buffer, brain heart infusion,
tryptose phosphate broth and trypticase soy broth (with or
without yeast extract)). In total, 807 data sets were collected with
the following 13 variables:
 Inactivation temperature (50e80
C) and 10log of holding time
(9.59$104 e 356 min) needed for 10-fold inactivation (D). The
maximum coefcient of log-linear heat inactivation was
calculated from raw data in the case that only coefcients for
non-loglinear inactivation were provided;
 Heating method. Method 1 lab scale pasteuriser with ow
(n 93), method 2 low culture volume in large volume preheated menstruum (n 211), method 3 low volume in
submerged glass capillary tube or coil (n 366), method
4 large volume in glass vial in water bath (n 137).
 pH (n 807, 5.0e9.3), sodium chloride (NaCl, n 760,
0.003e20%), fat (n 518, 0.1e83%), sugars (n 456, 0.2e58%)
and lactate (n 33, 0.07e0.8%);
 Temperature (n 807, 0e43
C) and duration (n 807,
0.25e336 h) of last temperature phase (excluding heat shock);

Weight to volume fractions (wt/vol) of fat, sodium chloride,

sugars and lactate were 10logit transformed (10log(p/(1  p))) to
warrant linearity and a constant variance with a close to normal
distribution of the residuals of these regression submodels. Zero
values were transformed to the decimal below the lowest value,
i.e.5 for 10logit(wt/vol sodium chloride) and 4 for 10logit(wt/
vol fat, sugar and lactate). The duration of the last temperature
phase (0e336 h, culturing or storage, excluding heat shock) was
10
log transformed. Statistical analysis was performed using
GenStat 14.1 (VSN International Ltd.). The highest colinearities
that were found were: 1) between 10logit(sodium chloride) and
10
logit(fat), with a correlation coefcient (CC) of 0.38 and 2)
between the duration and temperature of the last temperature
phase excluding heat shock (CC 0.39). The all-subsets
regression procedure (selecting the best combination of signicant model variables) of GenStat was used to obtain the primary
model (without interaction terms) with signicant variables
(p < 0.05). Non-linear submodels (quadratic and cubic) were
tested for all continuous variables. As variables may interact (the
regression coefcient of one variable being dependent on the
value of another variable) the following interaction terms were
tested separately:
 Inactivation temperature (with all other variables, see
Section 2);
 All possible pairs of pH, 10logit(sugar), 10logit(sodium
chloride NaCl) and 10logit(fat);
 pH and acid adaptation;
 10logit(NaCl) and NaCl adaptation;
 Heat shock (yes/no), heat shock temperature difference and
duration and temperature of the last temperature phase.

3.2. Graphical presentation of the models

In the graphs with multiple regression models, the residual
variance of the adjusted 10logD is shown: Figs. 2, 3 and 5 for
instance demonstrate the effect of multiple adjustments on the
variance of 10logD versus inactivation temperature. The 10logD
values in graphs 2e5 were adjusted only with the sum of the
products of the regression coefcients and the mean value of
continuous variables, e.g. pH or 10logit(NaCl). No adjustments to

396

J.H.M. van Lieverloo et al. / Food Control 29 (2013) 394e400

3
Dairy (liquid, 260)

Adjusted 10 log D (in minutes)

log D (in minutes)

10

Dairy (liquid, 260)

Media (321)
Juice (32)
Gravy (87)

Liquid egg (107)

0
-1
-2
-3
-4

Media (321)

Juice (32)
Gravy (87)

Liquid egg (107)

0
-1
-2
-3
-4

45

50

55

60
65
70
75
Inactivation temperature (C)

80

85

45

50

55

60
65
70
75
Inactivation temperature (C)

80

85

Fig. 1. Variability of inactivation times (D time to 10-fold inactivation) of Listeria

monocytogenes as a function of the inactivation temperature. 10logD-values are not
adjusted for effects of any other variables; heating menstrua are not distinguished as
a variable in the regression analysis. Total number of data sets is 807. Lines indicate
limits of the 95% prediction interval.

Fig. 3. Variability of inactivation times (D time required for 10-fold inactivation) of

Listeria monocytogenes as a function of the inactivation temperature. 10logD-values are
adjusted for effects of variables other than temperature as described in model B
(Table 1). Total number of data sets is 807. Lines indicate limits of the 95% prediction
interval.

the 10logD-values were made in these graphs for binomial or

multinomial variables (e.g. heat shock or heating method). In
the predictions of heat inactivation in milk, adjustments for all
variables were applied.

Topt of 37
C, pHmin of 4.4, pHopt of 7, water activity (aw,min) of 0.92,

assuming an aw for milk of 0.993 (Te Giffel & Zwietering, 1999).
4. Results and discussion

3.3. Simulation of growth and heat inactivation in milk

4.1. Single regression model

The variability of heat inactivation of L. monocytogenes during

pasteurisation of milk was simulated by Monte Carlo analysis (Vose,
2008) using Microsoft Excel and the Palisade @Risk add-in (Palisade Inc.). Monte Carlo analysis was applied to calculate heat
inactivation kinetics with probability distribution functions (pdfs)
of data per variable instead of calculating with single data values.
The inactivation was calculated by multiplying the regression
coefcients per variable of the selected model with the uniform
pdfs of the corresponding variables with a given range (e.g. pH
6.5e6.7). For prediction of the 10logD for specic food products and
processing conditions, the standard error of the multiple regression
model of 10logD was used as the standard deviation of the normal
distribution of the model. To simulate growth during storage of raw
milk before pasteurisation, the Gamma model (Zwietering, Wit, &
Notermans, 1996) was used, with a mmax of 2 h1, Tmin of 1.5
C,

Fig. 1 shows the variability of the 10logD of the 807 data sets
plotted against inactivation temperature (T), showing the variability for 5 food groups. The explained variance (R2) of this single
regression model (logD 9.07  T/6.78, so logD76
C 2.14 and
z 6.78) is 71.8% and the standard error (s.e.) 0.471. Based on
Fig. 1 and warnings from the statistical software it was concluded
that the variance is not equal throughout the temperature range
(heteroscedasticity), indicating room for improvement.
4.2. Basic multiple regression model
The variance of thermotolerance of L. monocytogenes can be
limited by including the effect of variables other than temperature
in the inactivation model; this also contributes to a more constant
2.0
Dairy (liquid, 259)

Media (320)

1.5

Juice (32)

Media (321)

log D (in minutes)

Juice (32)
Gravy (87)

Liquid egg (107)

0
-1

10

Adjusted

10

log D (in minutes)

Dairy (liquid, 260)

Gravy (84)

1.0

Liquid egg (107)

0.5
0.0
-0.5

-2
-1.0

-3
-1.5
-5

-4
45

50

55

60
65
70
75
Inactivation temperature (C)

80

85

Fig. 2. Variability of inactivation times (D time required for 10-fold inactivation) of

Listeria monocytogenes as a function of the inactivation temperature. 10logD-values are
adjusted for effects of variables other than temperature as described in model A
(Table 1). Total number of data sets is 807. Lines indicate limits of the 95% prediction
interval.

-4

-3

-2
10

-1

logit(NaCl)

Fig. 4. Variability of inactivation times (D time required for 10-fold inactivation) of

Listeria monocytogenes as a function of the 10logit(sodium chloride). 10logD-values are
adjusted for effects of temperature and other variables as described in model C
(Table 1); heating menstrua are not distinguished as a variable in the regression
analysis. The line represents the model as discussed in the text. Total number of data
sets is 802.

J.H.M. van Lieverloo et al. / Food Control 29 (2013) 394e400

variables salt adaptation (p 0.46), acid adaptation (p 0.68), and

10
logit(lactate) (p 0.10) had no signicant effect below the
p 0.05 level, and therefore these variables were omitted. To limit
the complexity of the initial model, ve menstruum groups were
distinguished (dairy, liquid egg, meat gravy, fruit and cabbage juice,
and laboratory media), rather than 20 individual menstrua. Variability of 10logD was limited by accounting for effects of variables
other than temperature on inactivation. The variance at 60e70
C
remained too high compared with the variance at lower and
higher temperature, resulting in a model still showing heteroscedasticity, having 16 coefcients, with an R2 of 83.2% and a standard error of 0.364 (Fig. 2, model A in Table 1).

10

Adjusted log D (in minutes)

Dairy (liquid, 259)

Media (320)

Juice (32)
Gravy (84)

Liquid egg (107)

0
-1
-2
-3
-4
45

50

55

60
65
70
75
Inactivation temperature (C)

80

397

85

Fig. 5. Variability of inactivation times (D time required for 10-fold inactivation) of

Listeria monocytogenes as a function of the inactivation temperature. 10logD-values are
adjusted for effects of variables other than temperature as described in model C
(Table 1); heating menstrua are not distinguished as a variable in the regression
analysis. Total number of data sets is 802. Lines indicate limits of the 95% prediction
interval.

variance throughout the temperature range. To select variables for

the basic multiple regression model including processing conditions and menstruum composition, all possible combinations of the
14 available explanatory variables (including menstruum group)
were tested. This included leaving out one or more variables
(without non-linear submodels and interaction terms). The

4.3. Multiple regression model with constant variance of logD vs.

temperature
High 10logD values at the mid (60e70
C) temperature range were
linked to 5e20% sugar and/or 6e10% sodium chloride added to liquid
egg products. Low10logD values in this temperature range were
linked to long cold storage in chicken gravy. The presence of the
linear terms of 10logit(NaCl), 10logit(sugars) and 10log(duration of last
temperature phase) in model A could not reduce this high variance at
mid temperatures. The effects of variables may however not be linear
but quadratic or cubic. Including quadratic and cubic non-linear
submodels for all three variables in model A led to a considerable
improvement, also rendering the effect of the temperature of the last
temperature phase to be highly signicant again (p < 0.001). Variability of 10logD was limited by taking into account the effects of
variables other than temperature on inactivation. This model B had

Table 1
Coefcients (and standard error) of models of the effect of inactivation temperature and other variables on 10logD (D time in minutes for 10-fold inactivation). Signicance
levels are p < 0.001 unless indicated otherwise: **p < 0.01, *p < 0.05, & p > 0.1.
Variable

Intercept
Inactivation temperature (
C)
pH
logit(sugars wt/vol)a
logit(sugars wt/vol)2
10
logit(sugars wt/vol)3
10
logit(fat wt/vol)
10
logit(NaCl wt/vol)a
10
logit(NaCl wt/vol)2
10
logit(NaCl wt/vol)3
10
logit(NaCl wt/vol)4
Last temperature phase (
C)b
10
log (last temp. phase (h))b
10
log (last temp. phase (h))2
Heat shock difference (
C)c
Heat shock (yes/no)c
Heating method 2d
Heating method 3d
Heating method 4d
Liquid egge
Meat gravye
Cabbage / fruit juicee
Mediae
Estimated standard error
R2
Number of data sets
# of coefcients

Single regression models

Multiple regression models

All uids

Milk alone

Model A

Model B

Model C

9.07 (0.198)
0.148 (0.0033)
(z 6.78)

10.0 (0.225)
0.162 (0.0036)
(z 6.16)

9.56 (0.233)
0.156 (0.0029)
(z 6.40)
0.134 (0.014)
0.125 (0.018)

14.46 (0.321)
0.163 (0.0024)
(z 6.13)
1.30 (0.333)
1.231 (0.194)
0.381 (0.109)
0.038 (0.017)*

15.27 (0.48)
0.161 (0.0022)
(z 6.20)
0.143 (0.011)
0.918 (0.067)
0.163 (0.013)

5.67 (0.429)
2.11 (0.190)
0.237 (0.023)

7.59 (0.747)
3.70 (0.457)
0.754 (0.115)
0.0546 (0.0100)
7.08$103 (8.70$104)

10
10

0.077 (0.016)
0.104 (0.021)

7.45$103 (1.17$103)
0.252 (0.037)

6.18$103 (1.29$103)
0.293 (0.079)
0.130 (0.030)

0.0146 (0.0023)

0.471
71.8 %
807
2

0.277
92.3%
175
2

0.137 (0.054)*
0.103 (0.046)*
0.072 (0.057)&
0.201 (0.064)**
0.383 (0.074)
0.337 (0.094)
0.176 (0.058)**
0.364
83.2 %
807
16

0.494 (0.049)
0.193 (0.045)
0.196 (0.038)
0.130 (0.048)**
0.348 (0.092)
0.466 (0.116)
0.503 (0.077)
0.138 (0.091)&
0.294
89.1 %
807
20

0.483 (0.048)

0.302
88.2 %
802
11

As fractions.
Duration and temperature of last temperature phase, either during culturing or storage, not heat shock.
Heat shock (54 data sets), the temperature difference is with the last temperature phase (culturing or storage).
d
Heating method 1 lab scale pasteuriser with ow (n 93), 2 low culture volume in large volume pre-heated menstruum (n 211), 3 low volume in submerged glass
capillary tube or coil (n 366), 4 large volume in glass vial in water bath (n 137). Reference method is heating method 1.
e
Reference menstruum is dairy.
b
c

398

J.H.M. van Lieverloo et al. / Food Control 29 (2013) 394e400

an R2 of 89.1% and a standard error of 0.294 (having 20 coefcients)

and is presented in Table 1 and Fig. 3. Differences between heating
methods and between menstruum groups became more signicant
when changing from model A to model B.
Inclusion of quadratic and cubic terms for other variables and
interaction terms were investigated. In many cases submodels were
found that had signicant effects, however the increase of the R2
and the decrease of the s.e. were marginal in those cases.
4.4. Optimum model
An optimal model was considered to have less variables than
model B while retaining a high descriptive power, expressed as the
R2. The cubic submodel of 10logit(sugar) could be omitted without
changing the R2 and s.e. too much, leaving a model with all variables
and submodels highly signicant (p < 0.001, except the difference
between heating methods 1 and 4, p 0.008). The inclusion of
10
logit(NaCl) was considered necessary and a protective effect of
sodium is consistent with various individual reports (EdelsonMammel et al., 2005; Jorgensen, Stephens, & Knochel, 1995 and
Juneja & Eblen, 1999). The linear or linear quadratic submodels
alone were incapable of reducing variance. As the 4th power polynomial model of the adjusted 10logD versus 10logit(NaCl) followed
the experimental data at low and high NaCl concentrations much
better than the 3rd polynomial model, the 4th order submodel of
10
logit(NaCl) was included (see Fig. 4, the slight undulation of the line
at low NaCl concentrations follows measured effects but probably
does not reect actual effects). Preferably, inclusion of such complex
polynomials are avoided but were necessary to include inactivation
data at high levels (10%, 20%) of NaCl in liquid egg products. It is not
known whether similar effects occur with high levels of NaCl in other
liquids. To provide a model that is also applicable by manufacturers
for designing heat inactivation in liquid egg products, this
complexity was accepted. The effect of sugar concentrations was
described by Sumner, Sandros, Harmon, Scott, and Bernard (1991)
and the effect of pH by Juneja, Foglia, and Marmer (1998), Juneja
and Eblen (1999), Jorgensen, Hansen, and Knochel (1999), Chhabra
et al. (1999), Bayles (2004), Hassani, lvarez, et al. (2005), Hassani
et al. (2007) and Ignatova et al. (2007). The effect of the temperature before inactivation (excluding heat shock) was found by Knabel,
Walker, Hartman, and Mendonca (1990), Juneja et al. (1998), Rowan
and Anderson (1998) and Edelson-Mammel et al. (2005). The effect
of the heat shock was also considered necessary, considering
observations of Bunning, Crawford, Tierney, and Peeler (1990, 1992),
Linton, Pierson, and Bishop (1990), Linton, Webster, Pierson, Bishop,
and Hackney (1992), Jorgensen et al. (1999) and Hassani et al. (2007).
The effects of other variables (e.g. lactate, adaptation to salt, adaptation to acid) have also been described in literature (see review of
Doyle et al., 2001), but these effects were not signicant within the
total variability of the data used.
The simplest model with a relatively high R2 (88.7%) and low
standard error (0.299) was found by omitting 6 coefcients of 3
variables: the cubic submodel of 10logit(sugar), both submodels of
the duration of last temperature phase and the heating method
from model B. Furthermore, distinction of menstruum groups was
excluded to create a model for general application in uid food
products, resulting in an R2 of 87.5% and a standard error of 0.314.
After these exclusions, the model retained 11 coefcients and all
variables and submodels had a highly signicant effect (p < 0.001).
As there were 5 data sets with a high standardised residual (3.3
to 4.6), all leading to an underestimation of the logD, these 5 data
sets were removed: 1 milk (Donnelly & Briggs, 1986), 3 chicken
gravy (Huang, Yousef, Marth, & Matthews, 1992) and 1 trypticase
soy broth with yeast extract (Hassani, lvarez, et al., 2005). This
resulted in model C with an R2 of 88.2% and a standard error of

0.302 (Table 1 and Fig. 5). There were 12 warnings of data sets with
high leverage (0.045e0.150). These high leverages did not occur in
the linear regression between 10logD and temperature, but
between 10logD and other variables. Furthermore, the variance of
the regression between 10logD and some of the other variables was
not constant (heteroscedasticity). An example of non-constant
variance is shown for the regression of 10logD with 10logit(NaCl)
in Fig. 4. Considering the origin of the model data, which have been
published over a period of 25 years by different research groups
without a common experimental design that was focussed on
multiple regression analysis, these modelling uncertainties could
be expected. For the applicability of the model, this means that the
effect of some secondary variables may not be estimated correctly
for all possible uids. Therefore, a comparison of the results of the
multiple regression model with results from a single regression
model of specic uids and processing conditions remains necessary. Fig. 4 also shows that the level of variance of the effect of
10
logit(NaCl) on the adjusted 10logD is higher than the level of
variance of the effect of temperature (Fig. 5). This is the case for
other adjusting variables as well.
4.5. Comparison with single regression models
Model C was used to predict inactivation during pasteurisation
of milk (76
C, 20 s, no previous heat shock) and the results were
compared with predictions based on the single, unadjusted
regression model presented in section 4.1 and those based on
a single regression model of milk data only (175 data sets 0e4.5%
fat, with pH at 6.5 or 6.6, no previous heat shock, 18 strains or
cocktails of strains, 4 high standardised residuals, 1 warning of
high leverage, see Table 1). The assumed milk and processing
characteristics are presented in Table 2. The mean D during pasteurisation at 76
C was 0.222 s, resulting in a mean 10log inactivation of 90.2 at 20 s holding time, compared with a mean of 46.0
that was predicted using the single regression model of all uids,
and a mean of 74.4 that was predicted using the single regression
model of milk data alone (Table 2). The multiple regression model
C at 76
C nevertheless is more conservative in its prediction of
inactivation than the single regression model with milk data
alone, as model C has a larger standard error (0.302 versus 0.277).
The higher level of uncertainty in model C results in a lower 10log
inactivation during 20 s at the lower p 1$106 percentile of the
pdf (a probability of 1 in a million that 10logD 3.3 for model C
versus 3.8 for the single regression model, Table 2). The single,
unadjusted regression model of all uids has an even more
conservative mean and lower p 1$106 percentile (46.0 and 0.3
respectively), but is likely to underestimate the actual inactivation
due to the presence of unadjusted effects of secondary variables.
4.6. Simulation of heat inactivation in raw milk
Sanaa, Corroler, and Cerf (2004) estimated mean concentrations
of L. monocytogenes in raw milk from two areas in France at 0.3 and
0.8 cells/l, with their mean being 0.55 cells/l. Assuming a distribution of the concentration of Gamma (1; 0.55) cells/l in raw milk, the
mean concentration after growth during cold storage was estimated to be 14 cells/l. The P99.9999 in raw milk was approximately
13 cells/l, and after growth during cold storage approximately 3500
cells/l (16e80 h, 1 million iterations). The 10logD for inactivation of
L. monocytogenes at pasteurisation conditions was simulated in
a Monte Carlo analysis with the three models compared in Section
4.5. To include variability in milk composition, the following ranges
were assumed for the uniform distributions of the model variables:
pH 6.5e6.7; sugar 4.5e4.7%; NaCl 0.2e0.3%; temperature during
cold storage 5e7
C.

J.H.M. van Lieverloo et al. / Food Control 29 (2013) 394e400

399

Table 2
Prediction of thermal inactivation of Listeria monocytogenes during pasteurisation of milk (fat 0e4.5%, no previous heat shock) with model C from Table 1 on the mean heat
inactivation kinetics in the example of milk pasteurisation, compared with the single regression model from Section 4.1 and the single regression model of milk data alone.
Variable

Mean values
Milk pasteurisation

Intercept 10logD
Inactivation temperature (
C)
Secondary variables
pH
Sugars wt/vol
Sugars wt/vol e quadratic
NaCl wt/vol
NaCl wt/vol e quadratic
NaCl wt/vol e cubic
NaCl wt/vol e 4th power
Last temperature phase (
C)
Heart shock (yes/no)
Intercept 10log (D adjusted (min))
Predicted 10log (D (min))
Estimated standard error 10logD
Predicted D (s)
Inactivation time (s)
Predicted 10log inactivation (mean)
At lower p 0.001 of pdfa
At lower p 1$106 of pdfa
a

Predicted inactivation: milk pasteurisation

10

logit Single regression model milk Single regression model all uids Multiple regression model C all uids
10.0
12.4

76
6.6
0.046

1.317

0.0025

2.601

9.07
11.2

6
no
2.349
0.277
0.269

2.139
0.471
0.435

74.4
10.4
3.6

46.0
1.6
0.3

15.3
12.3
0.94
1.21
0.28
19.73
25.00
13.27
2.50
0.04
0
9.83
2.432
0.302
0.222

20
90.2
10.5
3.3

pdf Probability density function.

The Monte Carlo simulation with the single, unadjusted regression model of all uids resulted in a probability of 4.92$104 (SD
2.04$105 of 10 simulations of 1 million iterations each) of the
presence of a surviving L. monocytogenes cell in a litre of milk (which
is 1 in approximately every 2000 L). The results of this worst case
simulation are unrealistic. With the single regression model and the
multiple, adjusted regression model C of all uids, no survival was
found in 10 simulations of 1 million iterations each. The means of the
estimated lower p 1$106 percentile of the 10log inactivation during
these 10 simulations were 0.21 (SD 0.03) for the single regression
model of all uids, 3.49 (0.22) for the single regression model of milk
data alone and 3.00 (SD 0.20) for the multiple regression model C of
all uids. Therefore, the probability of survival of L. monocytogenes
during pasteurisation of raw milk at 76
C during 20 s can be estimated with model C at less than 1 in a billion (109) litres.
5. Conclusions
A multiple regression model was created from literature data to
predict heat inactivation of L. monocytogenes in uids such as dairy
(milk, cream, butter), meat gravy, fruit and vegetable juices and
liquid eggs with or without added sugar and sodium chloride. The
model includes the full variability of strain tolerance to heat and
limits the variability due to differences in food composition (pH,
sodium chloride, sugar) and processing conditions (storage
temperature, heat shock). The model shows the power of metaanalyses, the relative effects of variables other than temperature
and the uncertainties of predicting heat inactivation with data from
a limited number of strains.
As the model is constructed from data that have been published
in a period of more than 25 years without a common experimental
design, the inuence of some variables is not known for all conditions or tested with similar intensity. Therefore, the results of the
model are not certain enough to design heat inactivation processes
without comparing the mean results of the multiple regression
model with the mean results of the single regression models for the
specic liquid. We recommend to use the variability of the multiple
regression model (0.302 10logD) to calculate the variability due to
uncertainty about the L. monocytogenes strains present in the food.

Acknowledgements
This research was funded by the cooperative research
programme of the Dutch dairy industry. We thank M.Z. Tankerville
for critically reading the manuscript and providing valuable
comments.
Appendix A. Supplementary material
Supplementary data related to this article can be found online at
http://dx.doi.org/10.1016/j.foodcont.2012.05.078.
References
Bayles, D. O. (2004). Changes in heat resistance resulting from pH and nutritional
shifts of acid-adapted and non-acid-adapted Listeria monocytogenes Scott A.
Journal of Food Protection, 67, 316e321.
Bigelow, W. D. (1921). The logarithmic nature of thermal death time curves. Journal
of Infectious Diseases, 29, 528e536.
Bunning, V. K., Crawford, R. G., Tierney, J. T., & Peeler, J. T. (1990). Thermotolerance
of Listeria monocytogenes and Salmonella typhimurium after sublethal heat
shock. Applied and Environmental Microbiology, 56, 3216e3219.
Bunning, V. K., Crawford, R. G., Tierney, J. T., & Peeler, J. T. (1992). Thermotolerance
of heat-shocked Listeria monocytogenes in milk exposed to high-temperature,
short-time pasteurization. Applied and Environmental Microbiology, 58,
2096e2098.
Chhabra, A. T., Carter, W. H., Linton, R. H., & Cousin, M. A. (1999). A predictive model
to determine the effects of pH, milkfat, and temperature on thermal inactivation of Listeria monocytogenes. Journal of Food Protection, 62, 1143e1149.
Casadei, M. A., Esteves de Matos, R., Harrison, S. T., & Gaze, J. E. (1998). Heat
resistance of Listeria monocytogenes in dairy products as affected by the growth
medium. Journal of Applied Microbiology, 84, 234e239.
Den Besten, H. M. W., & Zwietering, M. H. (2012). Meta-analysis for quantitative
microbiological risk assessments and benchmarking data. Trends in Food Science
& Technology, 25, 34e39.
Donnelly, C. W., & Briggs, E. H. (1986). Psychrotrophic growth and thermal inactivation of Listeria monocytogenes as a function of milk composition. Journal of
Food Protection, 49, 994e998.
Doyle, M. E., Mazotta, S. A., Wang, T., Wiseman, D. W., & Scott, V. N. (2001). Heat
resistance of Listeria monocytogenes. Journal of Food Protection, 86, 410e429.
Edelson-Mammel, S. G., Whiting, R. C., Joseph, S. W., & Buchanan, R. L. (2005). Effect
of prior growth conditions on the thermal inactivation of 13 strains of Listeria
monocytogenes in two heating menstrua. Journal of Food Protection, 68, 168e172.
Hassani, M., lvarez, I., Raso, J., Condn, S., & Pagn, R. (2005). Comparing predicting
models for heat inactivation of Listeria monocytogenes and Pseudomonas aeruginosa at different pH. International Journal of Food Microbiology, 100, 213e222.

400

J.H.M. van Lieverloo et al. / Food Control 29 (2013) 394e400

Hassani, M., Manas, P., Raso, J., Condn, S., & Pagn, R. (2005). Predicting heat
inactivation of Listeria monocytogenes under nonisothermal treatments. Journal
of Food Protection, 68, 736e743.
Hassani, M., Manas, P., Pagn, R., & Condn, S. (2007). Effect of a previous heat shock
on the thermal resistance of Listeria monocytogenes and Pseudomonas aeruginosa at different pHs. International Journal of Food Microbiology, 116, 228e238.
Holsinger, V. H., Smith, P. W., Smith, J. L., & Palumbo, S. A. (1992). Thermal
destruction of Listeria monocytogenes in ice cream mix. Journal of Food Protection, 55, 234e237.
Huang, I.-P. D., Yousef, A. E., Marth, E. H., & Matthews, M. E. (1992). Thermal
inactivation of Listeria monocytogenes in chicken gravy. Journal of Food Protection, 55, 492e496.
Huang, I.-P. D., Yousef, A. E., Matthews, M. E., & Marth, E. H. (1993). Growth and
survival of Listeria monocytogenes in chicken gravy during cooling and refrigerated storage. Journal of Foodservice Systems, 7, 185e192.
Huang, L. (2004). Thermal resistance of Listeria monocytogenes, Salmonella heidelberg, and Escherichia coli 0157:H7 at elevated temperatures. Journal of Food
Protection, 67, 1666e1670.
ICMSF. (1996). Microorganisms in foods 5. Characteristics of microbial pathogens.
Blackie Academic & Professional London, ISBN 0412 47350 X.
Ignatova, M., Leguerinel, I., Guilbot, M., Prvost, H., & Guillou, S. (2007). Modelling
the effect of the redox potential and pH of heating media on Listeria monocytogenes heat resistance. Journal of Applied Microbiology, 105, 875e883.
Jorgensen, E., Stephens, P. J., & Knochel, S. (1995). The effect of osmotic shock and
subsequent adaptation on the thermotolerance and cell morphology of Listeria
monocytogenes. Journal of Applied Bacteriology, 79, 274e281.
Jorgensen, E., Hansen, T. B., & Knochel, S. (1999). Heat shock induced thermotolerance in Listeria monocytogenes 13e249 is dependent on growth phase, pH and
lactic acid. Food Microbiology, 16, 185e194.
Juneja, V. K., Foglia, T. A., & Marmer, B. S. (1998). Heat resistance and fatty acid
composition of Listeria monocytogenes: effect of pH, acidulant, and growth
temperature. Journal of Food Protection, 61, 683e687.
Juneja, V. K., & Eblen, B. S. (1999). Predictive thermal inactivation model for Listeria
monocytogenes with temperature, pH, NaCI, and sodium pyrophosphate as
controlling factors. Journal of Food Protection, 62, 986e993.
Kamau, D. N., Doores, S., & Pruitt, K. M. (1990). Enhanced thermal destruction of
Listeria monocytogenes and Staphylococcus aureus by the lactoperoxidase
system. Applied and Environmental Microbiology, 56, 2711e2716.
Knabel, S. J., Walker, H. W., Hartman, P. A., & Mendonca, A. E. (1990). Effects of
growth temperature and strictly anaerobic recovery on the survival of Listeria
monocytogenes during pasteurization. Applied and Environmental Microbiology,
56, 370e376.
Knight, K. P., Bartlett, F. M., McKellar, R. C., & Harris, L. J. (1999). Nisin reduces the
thermal resistance of Listeria monocytogenes Scott A in liquid whole egg. Journal
of Food Protection, 62, 999e1003.
Linton, R. H., Pierson, M. D., & Bishop, J. R. (1990). Increase in heat resistance of
Listeria monocytogenes Scott A by sublethal heat shock. Journal of Food Protection, 53, 924e927.
Linton, R. H., Webster, J. B., Pierson, M. D., Bishop, J. R., & Hackney, C. R. (1992). The
effect of sublethal heat shock and growth atmosphere on the heat resistance of
Listeria monocytogenes Scott A. Journal of Food Protection, 55, 84e87.
Lou, Y., & Yousef, A. E. (1996). Resistance of Listeria monocytogenes to heat
after adaptation to environmental stresses. Journal of Food Protection, 59,
465e471.

Maisnier-Patin, S., Tatini, S. R., & Richard, J. (1995). Combined effect of nisin and
moderate heat on destruction of Listeria monocytogenes in milk. Lait, 75, 81e91.
Mazzotta, A. S. (2001). Thermal inactivation of stationary phase and acid-adapted
Escherichia coli 0157:H7, Salmonella, and Listeria monocytogenes in fruit juices.
Journal of Food Protection, 64, 315e320.
Palumbo, M. S., Beers, S. M., Bhaduri, S., & Palumbo, S. A. (1995). Thermal resistance
of Salmonella spp. and Listeria monocytogenes in liquid egg yolk and egg yolk
products. Journal of Food Protection, 58, 960e966.
Palumbo, M. S., Beers, S. M., Bhaduri, S., & Palumbo, S. A. (1996). Thermal resistance
of Listeria monocytogenes and Salmonella spp. in liquid egg white. Journal of Food
Protection, 59, 1182e1186.
Rowan, N. J., & Anderson, J. G. (1998). Effects of above-optimum growth temperature and cell morphology on thermotolerance of Listeria monocytogenes cells
suspended in bovine milk. Applied and Environmental Microbiology, 64,
2065e2071.
Sanaa, M., Corroler, L., & Cerf, O. (2004). Risk assessment of listeriosis linked to the
consumption of two soft cheeses made from raw milk: Camembert of Normandy and Brie of Meaux. Risk Analysis, 24, 389e399.
Sumner, S. S., Sandros, T. M., Harmon, M. C., Scott, V. N., & Bernard, D. T. (1991). Heat
resistance of Salmonella typhimurium and Listeria monocytogenes in sucrose
solutions of various water activities. Journal of Food Science, 56, 1741e1743.
Te Giffel, M., & Zwietering, M. H. (1999). Validation of predictive models describing
the growth of Listeria monocytogenes. International Journal of Food Microbiology,
46, 135e149.
Van Asselt, E. D., & Zwietering, M. H. (2006). A systematic approach to determine
global thermal inactivation parameters for various food pathogens. International Journal of Food Microbiology, 107, 73e82.
Van der Veen, S., Wagendorp, A., Abee, T., & Wells-Bennik, M. H. J. (2009). Diversity
assessment of heat resistance of Listeria monocytogenes strains in a continuousow heating system. Journal of Food Protection, 72, 999e1004.
Van Lieverloo, J. H. M., de Roode, B. M., Fox, M. B., Zwietering, M. H., & WellsBennik, M. H. J. (2011). Multivariate regression model of thermal inactivation of
Listeria monocytogenes in liquid food products. In E. Cummins, J. M. Frias, &
V. P. Valdramidis (Eds.), Seventh International Conference on Predictive Modelling
in Foods e Conference Proceedings. Dublin, Ireland: (UCD, DIT, Teagasc).
Vose, D. (2008). Risk analysis: A quantitative guide (3rd ed.). Chisester: John Wiley &
Sons.
World Health Organization. (2004). Risk assessment of Listeria monocytogenes in
ready-to-eat foods. Technical report. Geneva, Switzerland, ISBN 92 4 156262 5.
Zwietering, M. H., De Wit, J. C., & Notermans, S. (1996). Application of predictive
microbiology to estimate the number of Bacillus cereus in pasteurised milk at
the point of consumption. International Journal of Food Microbiology, 30, 55e70.

Internet sites consulted (Last accessed on April 2, 2012)

http://www.nal.usda.gov/fnic/foodcomp/search/(several variables).
http://www.alsosalt.com/sodiumcontent.html (NaCl).
http://oto2.wustl.edu/men/sodium.htm (NaCl).
http://www.collectivewizdom.com/SodiumInCommonFoods.html (NaCl).
http://www.ntwrks.com/wmikev/chart2a.htm (fat, NaCl).
http://apjcn.nhri.org.tw/server/info/books-phds/books/foodfacts/html/data/data2g.
html (fat).
http://www.international-com.com/milk_powder_cream.html (fat, sugar).