food additives & contaminants -...

60
Published by TAYLOR & FRANCIS LTD Food Additives & Contaminants EDITORS: Prof John Gilbert Prof Timothy Phillips Department for Environment, Food and Rural Affairs Texas A&M University Central Science Laboratory College of Veterinary Medicine Sand Hutton Department VAPH York YO41 1LZ Mail stop: 4458 College Station UK TX 77843-4458 USA FAX: +44 (0)1904 462426 e-mail: [email protected] [email protected] Dr O Vitrac UMR Fractionnement des Agro-Ressources et Emballage INRA 614 Moulin de la Housse 51687 Reims cedex2 FRANCE 7 July 2005 Dear Dr Vitrac Exposure of consumers to substances from plastic packaging materials: assessment of the contribution of styrene from yoghurt pots – Vitrac et al (Manuscript Ref: 04-JG/FAC/090) Thank you for returning your revised manuscript and for your responses to the referees comments. I am pleased to inform you that your paper is now accepted for publication. Nevertheless, I still feel that there is too much specialised mathematics in the paper which will be a ’turn-off’ for most FAC readers but will not ask you to make any further revisions. I have forwarded your paper to the Publishers from who you will receive proofs (sent electronically as a PDF File) in due course, which you will need to check carefully. You will also need to look critically at the Figures when you get the proofs and ensure they are sized appropriately to maintain legibility. Once typesetting and corrections to your paper has been completed it will be available in electronic form as PReVIEW on the Taylor and Francis web-site http://www.tandf.co.uk/journals/titles/0265203X.asp prior to publication in the Journal. Kind regards Yours sincerely Prof John Gilbert Editor

Upload: others

Post on 21-Jul-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Published by TAYLOR & FRANCIS LTD

Food Additives & ContaminantsEDITORS: Prof John Gilbert Prof Timothy Phillips

Department for Environment, Food and Rural Affairs Texas A&M UniversityCentral Science Laboratory College of Veterinary MedicineSand Hutton Department VAPHYork YO41 1LZ Mail stop: 4458 College StationUK TX 77843-4458

USAFAX: +44 (0)1904 462426e-mail: [email protected] [email protected]

Dr O VitracUMR Fractionnement des Agro-Ressources et EmballageINRA 614Moulin de la Housse51687 Reims cedex2FRANCE 7 July 2005

Dear Dr Vitrac

Exposure of consumers to substances from plastic packaging materials: assessment of the contribution of styrene from yoghurt pots – Vitrac et al(Manuscript Ref: 04-JG/FAC/090)

Thank you for returning your revised manuscript and for your responses to the referees comments. I am pleased to inform you that your paper is now accepted for publication. Nevertheless, I still feel that there is too much specialised mathematics in the paper which will be a ’turn-off’ for most FAC readers but will not ask you to make any further revisions.

I have forwarded your paper to the Publishers from who you will receive proofs (sent electronically as a PDF File) in due course, which you will need to check carefully. You will also need to look critically at the Figures when you get the proofs and ensure they are sized appropriately to maintain legibility.

Once typesetting and corrections to your paper has been completed it will be available in electronic form as PReVIEW on the Taylor and Francis web-site

http://www.tandf.co.uk/journals/titles/0265203X.asp

prior to publication in the Journal.

Kind regards

Yours sincerely

Prof John GilbertEditor

Page 2: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Exposure of consumers to substances from plastic packaging materials. 1.

Assessment of the contribution of styrene from yogurt pots.

Olivier Vitrac1, Jean-Charles Leblanc2

1UMR Fractionnement des Agro-Ressources et Emballage, INRA 614, Moulin de la Housse. 51687 Reims cedex

2. France.

Fax. 33-3-26-91-39-16

E-mail: [email protected]

2UR Méthodologies d’analyse du risque alimentaire , INRA, INAP-G, 16 rue Claude Bernard, 75005 Paris,

France

Abstract

A generic methodology for the assessment of consumer exposure to substances migrating

from packaging materials into foodstuffs during storage is presented. Consumer exposure at

household scale is derived from the efficient probabilistic modeling of the contamination of

all packed food product units (e.g. a yogurt pot, a milk bottle…) that are consumed by a given

household during one year. Exposure of given population is finally estimated by gathering the

exposure distributions of individual households respectively to suitable weights (conveniently

household sizes). Tractable calculations are achieved by combining both i) an efficient

resolution of migration models and ii) a methodology combining different sources of

uncertainty and variability. The full procedure is applied on the assessment of consumer

exposure to styrene from yogurt pots based on yearly purchases data of more than 5400

households in France (about 2 millions of yogurt pots) and an initial concentration of styrene

1/42

5

10

15

20

25

Page 3: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

in yogurt wall pots, which is assumed to be normally distributed with an average value of 500

ppm and a standard deviation of 150 ppm. Results are discussed regarding both the sensitivity

of the migration model to boundary conditions and household practices. By assuming a

partition coefficient of 1 and a Biot number of 100, the estimated median household exposure

to styrene calculated is ranged between 1 and 35 µg⋅day-1⋅person-1 (5th and 95th percentiles)

with a likely value of 12 µg⋅day-1⋅person-1 (50th percentile). It is found that exposure does not

vary independently with the average consumption rate and contact times. Thus, falsely

assuming an uniform contact time equal to the sale-by-date for all yogurts overestimates

significantly the daily exposure (5th and 95th percentiles of 2 and 110 µg⋅day-1⋅person-1

respectively) since high consumers showed quicker turnovers of their stock.

Key words: consumer exposure, packaging, styrene, risk assessment

2/42

30

35

40

45

Page 4: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Notations

Roman symbols

Bi dimensionless mass Biot number [-]

eqFC concentration in food at thermodynamic equilibrium [mg⋅kg-1]

D diffusion coefficient [m⋅s-2]

kE individual (per capita) consumer exposure [mg⋅day-1⋅pers-1]

2i

it DFol×= dimensionless Fourier number related to the ith yogurt [-]

kN number of consumed yogurts for the kth household [-]

K partition coefficient [-]

F

P

V

V

F

PLρρ= dilution factor [-]

kP size of the kth household [-]

SD sale-by-date [days]

PV volume of the packaging material [m3]

FV yogurt volume [m3]

0c initial concentration in the packaging material [mg⋅kg-1]

i yogurt pot index [-]

k household index [-]

l yogurt pot thickness [m]

Sr safety ratio in assessed exposure (see 3.6) [-]

Vr variability ratio in exposure assessment (see 3.6) [-]

k household index [-]

0cs normalized standard deviation of 0c [-]

Ds normalized standard deviation of ( )log D [-]

it storage time of the ith yogurt [s]

*iv dimensionless migration rate [-]

3/42

50

55

60

65

70

Page 5: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Greek symbols

Pr packaging density [kg⋅m-3]

Fr yogurt density [kg⋅m-3]

1. Introduction

Motivation

Over the last decade, chronic exposure to chemicals ingested through diet has exerted an

increasing impact on technical, political and legislative decisions (e.g. regulations, industrial

practices,…). Despite considerable improvements in quantification limits for measuring

substances in food, it is nevertheless currently not possible to identify all potential sources of

intake of a given chemical over lifetime (Lau and Wong 2000). A major difficulty is that large

variations of contaminant concentrations in foodstuffs may occur according to industrial and

household practices (Svensson, 2002).

The contamination of food products through packaging materials illustrates the stakes for both

contamination risk and exposure assessments. The accumulation of contaminants from food

packaging materials in packed food products occurs through a migration process during all

steps of the product life including: during process and distribution , during storage by both

retailers and households. As a result, the final concentration in a packed food product will

depend on factors that are specific to the product (composition, rheology, volume ratio

between food and packaging) but also on its overall history (time, temperature, vibrations,

possible stirring). For products having shelf-lives longer than a few days, the storage time by

consumers may be larger than the time for processing and distribution, so that the

contamination of food products – when they are consumed – may be strongly influenced by

consumer behaviors and practices (Vitrac et al. 2005).

4/42

75

80

85

90

95

Page 6: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Typical accumulation kinetics for non equilibrated systems are related to an effect of storage

time ( st ) proportional to stb with 0.5 1b£ £ . When the process responsible for the release of a

given contaminant is controlled by internal diffusion (without limiting effect of the packaging

volume), one gets 12

b = at constant temperature (Cranck 1975). When the limiting process is

either a reaction or an interfacial transport, b is expected to be close to 1 (Treybal 1980,

Vergnaud 1991). These kinetic considerations illustrate that the distribution of storage times

at household scale may modify drastically the range of the contamination of packed food

products. Indeed for a same overall food consumption, it is expected that a household

including high consumers or doing regular purchases consumes less contaminated packed

food than a household doing infrequent purchases. For products regularly consumed, such a

scatter between household practices can be responsible for an inter-household variability in

food contamination, which may be higher than the intra-household variability. Possible

correlations between consumption patterns and storage times, and consequently between

consumption and contamination, imply that a probabilistic approach at house scale should be

preferred to a coarse grained assessment of individual risk assuming that modeling quantities

– contamination and consumption levels – are independent. An additional argument in

supporting detailed description of contamination at the moment of consumption is to analyze

separately the effects of variability and uncertainty on quantitative risk estimates.

The current paper proposes a modular risk assessment model dedicated to the quantitative

assessment of exposure to substances originating from the plastic layer in contact with food

items, which may obey to general principles defined by Huggett et al. (1998). For this

purpose, a full probabilistic modeling driven by the knowledge of both the migration of

assessed substances and of the consumption is devised. Unlike other empirical methods based

on Monte Carlo simulations (Hamey and Harris 1999, Gauchi and Leblanc 2002, Lambe

5/42

100

105

110

115

120

5

Page 7: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

2002, Gibney and van der Voet 2003), such an approach would be not as limited by the

availability of contamination data and by strategies used to handle undetectable levels, false-

negative, false-positive results and outliers.

Background

The interest of combining the knowledge of both physical processes and food pathway

conditions to predict the migration of packaging substances was early discussed by Chatwin

and Katan (1989). They suggested that such an approach could partly replace long and

prohibitive surveys of contamination in real food products if the degree of variability in the

migration occurring in real life should be undertaken. The predictability of the contamination

of foodstuffs by substances migrating from plastics in contact is achievable through a physical

formulation of the migration process between and food items (European Commission 2003).

The use of physical models to replace contamination data in exposure assessment was first

proposed by Vermeire et al. (1993) for both occupational and consumer exposure and

specifically by Lickly et al. (1995) for food contact materials. It has been recently proposed to

handle conveniently inherent sources of uncertainty on initial concentration in plastic

materials, physiochemical properties (diffusion and partition coefficients), geometry factors,

contact times, food texture via a stochastic resolution of transport equations (Vitrac and

Hayert 2005). It is worth to notice that this formulation strives for likely or “best” estimators

of modeling quantities (e.g. diffusion coefficients) rather than overestimates as detailed by

Begley et al. (2005). Indeed, the concept of conservatism as it used either in point exposure

estimates or for compliance determination introduces a priori assumptions and choices that

deliberately yield overestimates of chronic exposure and that do not separate the effects of

both variability and uncertainty (Leber 2001, Vitrac 2003).

6/42

125

130

135

140

145

Page 8: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

The main restrictions for the application of full probabilistic approaches are the lack of food

consumption data including correspondence with packaging formulations (starting substances,

typical concentrations…) and uses (food matrices, geometry…). The proposed dimensionless

formulation detailed in Vitrac and Hayert (2005) provides reliable strategies to handle

incomplete information on packaging usage and food storage.

The first strategy relies on the assumption of the invariance of some distribution shape

parameters related to geometric quantities, contact times, transport properties. As a result,

only likelihood values the main scale parameters are required with enough accuracy.

Important distribution parameters such as contact times can be either measured or

reconstructed. This methodology was applied to estimate the possible risk of contamination of

12 real packed food products by a phenolic antioxidant starting from the yearly purchases of

6,422 households in France (Vitrac et al. 2005).

In this work, we refine the approach by calculating the probability density function (pdf) of

the contamination of all food units that are consumed by a given household during one year.

The probabilistic approach of the contamination of each product unit is here mainly related to

the uncertainty in both the physical quantities that control the migration and in the initial

concentration within the considered packaging materials.

Contribution

The current work focus on the assessment at household scale of the exposure to styrene

related to the consumption of yogurt pots. The exposure to styrene offers a high

methodological concern since the link between the migrating substance (styrene) and its use in

styrene based materials is clearly established. Besides, styrene oxide has been believed to

promote lung tumors in mice and could promote cancer of the forestomach in rats at high

doses (Cohen 2002). The International Agency for Research on Cancer (1994) evaluates the

styrene-7, 8-oxide as “possibly carcinogenic to humans” (group 2B). A provisional maximum

7/42

150

155

160

165

170

Page 9: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

tolerable daily intake of 40 µg⋅kg-1 bodyweight⋅day-1 has been set by the joint FAO/WHO

Expert Committee on Food Additives (World Health Organization 1984). An accurate

appraisal of exposure to styrene at household scale and focused on both vulnerable and high

consumers of dairy products such as children is therefore also highly desirable.

The contamination of all yogurt pots consumed by 5,473 households during one year is

derived from their yearly volume and frequency of purchases. Individual exposure at

household scale is calculated by weighting results according to household sizes. The assessed

exposure at the scale of tested population is finally compared by gathering the information

available at household scale (proposed strategy) and by crudely combining the sampled

distributions of contamination and consumption at the scale of the tested population

(conventional strategy). With the developed example on styrene, we additionally identify

some important gaps in knowledge and we are able to suggest some risk mitigation strategies

for a specific packed food product of concern.

2. Quantitative methodology for exposure assessment

The granular approach for exposure assessment of households, which is used in this work, is

illustrated in figure 1. This detailed risk scheme assumes that members of a given household

consume only yogurt pots purchased and stored by themselves. It is assumed i) that there is no

interaction between households (no products exchange and no crossed consumption) and ii)

that all purchased yogurts are consumed. Yogurts consumed in institutional catering or fast

food chain are not considered here. As a result of the migration process, the contamination of

yogurts by styrene increases during storage. The contribution of the storage period on

exposure is studied here.

8/42

175

180

185

190

195

200

Page 10: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

The concentration in styrene in each yogurt at the date of its consumption is calculated for a

given contact time and a probable distribution of the initial concentration in wall pots. The

exposure at household scale is assumed to be equal to the cumulative amount of styrene that

has migrated into all consumed yogurts.

The time list of purchases of each household is extracted from long marketing national

surveys and is used as input in the risk assessment procedure. Since the date of consumption

of each yogurt is unknown, contact times are reconstructed as point estimates from likely

scenarios of household storage. Due to a significant uncertainty in predicted styrene

concentrations in yogurts, the contamination of each consumed yogurt is calculated via a

probabilistic approach that takes into several sources of uncertainty. The resulting

concentration in each yogurt is known via a distribution of probabilities. Finally, all these

distributions are combined to derive the exposure at household scale. The so-calculated

exposure is also a distribution, which represents both the different sources of uncertainty and

the variability due to different contact times between consumed yogurts.

2.1 Exposure quantification at household scale: kE

Exposure at household scale is expressed in mg⋅day-1⋅pers-1 and inferred from the cumulative

amount of migrant ingested during one year within by the considered household and averaged

over the household size. For a householdk , of size kP , the exposure of the “typical”

consumer, kE , that consumes kN yogurts contained in individual pots of weight 0 iM , is

efficiently calculated by factorizing kE as a sum of independent variables:

( ), , ,*0

1

1365

ki

NFo Bi K Leq

k i F i ik iE M C v

P == × × ×

× å (1)

where the notation iX stands for the quantity X taken for the thi pot (i.e. food product unit).

*v is the migration level, it is ranged between 0 (no migration) and 1 (thermodynamical

equilibrium) and is related to 4 dimensionless numbers: the dimensionless time, Fo , the ratio

9/42

205

210

215

220

225

Page 11: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

between internal and external resistances to mass transfer (also known as mass Biot number),

Bi , the partition coefficient between food and packaging materials, K , the dilution factor, L ,

defined as the ratio between the mass of packaging and food materials P

F

VV

P

F

Lrr

= . The

quantity eqFC is related to the concentration in yogurt at thermodynamical equilibrium (when

* 1v = ) and is derived from a mass balance in styrene assuming an initial concentration 0c

(weight/weight) in the packaging material:

01 1

eqF

cC

K L

=+ (2)

Equations (1) and (2) are valid for any food, any contaminant and any packaging material. The

only assumptions are that: i) the contamination is controlled by the diffusion of the

contaminant from the packaging material into the food product, ii) the packaging material is

the only source in the considered contaminant, iii) no contaminant losses occur during the

storage. Besides, as previously discussed in Vitrac (2003), equation (1) assumes that K and L

have a much higher effect on eqFC than on *v for similar products with similar packaging. The

effect of possible variations in K and L are thus neglected in *v but are taken into account in

eqF iC via equation (2). Since no information is available of the value of Bi for each food

product unit, a typical value is assumed for all units.

For yogurts sold in France, additional simplifications are possible since most yogurts

packaged in thermoformed polystyrene pots have a same weight of 125 g (volume 120 ml).

This description is still currently valid and was valid at the year of considered purchases

(1998). In addition, according to the French legal definition of yogurts, such labeled products

are only made from coagulated milk resulting from the lactic fermentation by only two strains:

Lactobacillus bulgaricus and Streptococcus thermophilus. From these considerations,

equation (1) is realistically simplified as:

10/42

230

235

240

245

10

Page 12: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

( ), , ,0 *

1365

ki

NeqFo Bi K LF

k ik i

M CE v

P =

×= ×

× å (3)

2.2 Migration model: ( ), , ,* iFo Bi K L

iv

The migration level, *v , also defined by eqF FC C , where FC which is the concentration in food,

is calculated from a general physical model of the migration similar to the one used in Vitrac

et al. (2005). The main assumptions are detailed in Vitrac (2003) and in Vitrac and Hayert

(2005). For plastic materials non-subjected to swelling, the transport of additives in polymer

matrices is assumed to be controlled by a molecular diffusion process. At the interface

between food and packaging materials, the pervious contact between food and packaging is

assumed to be controlled by a generalized Robin boundary condition, which accounts for both

a possible partition effect on both sides of the interface and an external mass transfer

resistance controlled by the texture of the food product. All properties are assumed to be

constant during storage.

As depicted in figure 2, the expected value of *v of a yogurt is expected to vary significantly

with the dimensionless contact time (or Fourier number), 2D tFol×= , and the Biot number,

h lBi

D×= where D is the diffusion coefficient (with SI units in m2·s-1), t is the contact time, l

is the packaging thickness and h is the mass transfer coefficient on the food side of the

packaging-food interface (with SI units in m·s-1). Bi is the mass Biot number expressing the

ratio of mass transfer resistances on both sides of the food-packaging interface. For solid food

products, Bi is expected to be close to 1 whereas it is greater than 1 for liquid food products.

The limit condition Bi ® ¥ corresponds to a situation where only an internal resistance to

mass transfer within the packaging material occurs. For a same Fo value, v * is increasing

with Bi up to a limit curve that depends on K and L values. In particular, for Fo values

11/42

250

255

260

265

270

Page 13: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

lower than 1 (case of yogurts), the overall migration kinetics change from a variation of v *

proportional to the square root of time to a first order kinetic when the Bi ® ¥ condition is

changing towards 1Bi>® . Consequently, the shape of the distribution of exposure values is

expected to vary in a significant extent with the considered boundary condition. In solid like

food materials (related to low Bi values), the effect of contact time is expected to be higher

than in liquid like food materials (related to high Bi values) and would consequently lead to

more spread values of exposure at household scale.

In the current study, a likely value of D , noted D , is assumed to be available and applied to

all consumed packed products (i.e. yogurts packed in PS pots). By contrast, since few

estimates of h are available for real food products and in realistic conditions of storage, only

magnitude orders of Bi are considered. The true value of D of the considered packaging (i.e.

yogurt pot) remains however unknown since it can vary significantly with the type of

polystyrene (formulation and processing) and the temperature of storage (typically between

2°C and 10°C). The sources of uncertainty in the D value are taking into account via a

stochastic resolution of the migration model discussed in sub-section 2.4. Since the migration

level does not vary regularly with Bi , the uncertainty in the true Bi value due to possible

different food textures (e.g. gelified or stirred yogurts) is tested by comparing results obtained

for a set of typical Bi values.

2.3 Assumptions for the reconstruction of contact times between food and packaging

material: 1..i NkFo

=

Likely dimensionless contact times defined by 1..

1.. 2i Nk

i Nk

D tFo

l=

=

×= , are derived from contact

times, 1..i Nkt= , reconstructed from the frequency and the amount of household purchases as

12/42

275

280

285

290

295

Page 14: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

recorded in marketing databases. The methodology used to reconstruct 1..i Nkt= related to each

food product unit (i.e. yogurt pot) is similar to the approach described in Vitrac et al. (2005)

and applied to 10 different real food products. The contact time due to household storage is

defined as the time between the date of purchase and consumption. Since only the date of

purchase is known, the date of consumption is estimated assuming a regular consumption

between two consecutive purchases. Two scenarios are considered:

Scenario H1: all yogurts bought at the previous purchase date are consumed by the considered

household before the date of the next purchase of yogurts;

Scenario H2: the number of consumed yogurts between two consecutive purchases cannot

exceed the number of purchased units at the date of the last purchase. According to this

scenario, an accumulation of yogurts is possible between consecutive purchases. They are

consumed in the order “first bought, first consumed” so as to minimize the overall duration of

storage.

Since scenarios H1 and H2 do not take into account the contribution of distribution and

retailing on contact times, an upper bound, noted scenario H3, is provided by assuming that all

products are consumed at the sale-by-date, that is:

1.. 2i Nk

D SDFol=

×= (4)

where SD is the contact time at the sale-by-date.

2.4 Probabilistic description of exposure

The probability density function (pdf) of kE is derived from the pdfs of independent quantities

eqFC and 1 ki Nv = K . The distribution of eq

FC values is only related to the distribution of 0c between

all packaging materials. The main source of residual styrene in the wall of yogurt pots is the

thermal degradation of polystyrene due to partial depolymerization, its residual concentration

13/42

300

305

310

315

320

Page 15: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

is assumed to be concentrated around a likely value controlled by the processing technology of

polystyrene. Thus, the dimensionless quantity 0 0 0/c c c* = is assumed normally distributed

with an average of 1 and a standard deviation, noted 0cs , where 0c is the likely concentration

in the packaging material. Besides, the distribution of each 1 ki Nv = K is related to the uncertainty

in the corresponding dimensionless contact times 1i NkFo

= K written as:

1 1i Nk i Nk

Fo Fo Fo= =

*= ×K K

(5)

where *Fo D* = is a normalized random variation of contact times mainly related to the

uncertainty inD value. As discussed in sub-section 2.2, this uncertainty includes both the

unknown variability in the PS matrix and possible change in temperature during storage. The

random contribution is assumed to be log-normally distributed as ( )log Norm(1, )DD s* : . This

choice is discussed in Vitrac and Hayert (2005).

Mass balance considerations between food and packaging materials demonstrated that 1 ki Nv = K

are Beta distributed with parameters ( )Beta ,i ia bb b (Vitrac and Hayert 2005). This approach is

illustrated in figure 2 for 3 distributions of Fo values, corresponding to a same prescribed

distribution of *Fo values and 3 different Fo values. It is underlined that the mode of Fo pdfs

coincides with Fo . The Since iab and ibb values vary continuously with Fo for a same Ds

value, they are efficiently non-linearly interpolated from pre-calculated tabulated data.

Since { } 1 ki i Nv * = K are not identically distributed, the sequence 1

kN

iiv *

=å is besides not expected to be

normally distributed even for high consumer patterns with large kN values. The pdf of 1

kN

iiv *

=å is

efficiently calculated as a kN convolution product of independent 1 ki Nv = K densities. As

convolutions are equivalent in Fourier space to simple multiplications with associativity and

14/42

325

330

335

340

Page 16: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

commutativity properties, the final result is efficiently calculated iteratively by partitioning

previously calculated densities in same order as their variance. At each step, this algorithm

aims at calculating two-by-two the convolution of densities with similar variation coefficient.

Each two-by-two convolution product is numerically calculated starting from FFT (Fast

Fourier Transforms) transforms of previously scaled densities. Discrete FFT transforms were

performed with a normalized cut-off frequency of 1/216. Since all calculations can be

effectively calculated in parallel for all households and all packaging units, no Monte Carlo

sampling is required. This approach is therefore particularly conservative of all pertinent

information available at household scale.

Finally, the distribution of product of random normalized quantities 01

kN

ii

c v* *

=

æ ö÷ç ÷× ç ÷ç ÷÷çè øå is inferred

from the distribution of its log-transform written as ( )10 0 101

log logkN

ii

c v* *

=

æ ö÷ç ÷+ ç ÷ç ÷÷çè øå , which is

straightforwardly calculated as the convolution product of ( )10 0log c* and 101

logkN

iiv *

=

æ ö÷ç ÷ç ÷ç ÷÷çè øå

densities.

2.5 Initial assumptions and purchase data

Both contact time and consumption at household scale are based on yearly volume and

frequency of purchases of a panel of 5,473 households (including 14,649 persons) (SECODIP

databases, 1998, France; Combris et al. 2000). Eighty thirteen percents of the French market

of yogurts are estimated to be covered by the present survey (SECODIP sources). All

15/42

345

350

355

360

15

Page 17: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

households belonging to the tested panel performed more than 8 purchases per year. Only

purchases of yogurt in plastic containers with net weight about 0.125 kg are considered. Such

restrictions ensure that almost packaging materials were made in PS. According to former

conditions, 221,190 purchases including 1,930,257 yogurt pots are analyzed.

The values of parameters involved in the exposure assessment procedure are summarized in

table 1. Thermodynamical and transport properties as well as an initial concentration in

styrene monomer ( 0c ) of 500 ppm are chosen according to our own expertise and accordingly

to values given by EC-DG SANCO D3 (2003). It is emphasized that the chosen value of 0c is

3 times lower than values assessed 29 years ago in polystyrene (Withey, 1976) but is of a

same magnitude order as values found by Murphy et al. (1992). Geometry and mass related

quantities are measured starting from typical yogurt products available on the French market.

Values of distribution shape parameters, 0cs and Ds , are assumed to introduce realistic sources

of variability in the analysis. This choice is discussed in Vitrac and Hayert (2005) and in

Vitrac et al. (2005).

2.6 Distribution of individual exposure at the scale of tested population (5,473 households)

Since equation (3) does not separate children and adults, males and females, the calculated

individual exposure was averaged over the household size. When the size of a given

household varied during the considered year, the maximum size of the household was used.

The likely distribution of individual exposure at the scale of all the 5,473 examined

households is derived by accumulating (gathering) all the corresponding individual exposure

distributions calculated at household scale according to a suitable set of weights. In absence of

pertinent information, weights proportional to household sizes are applied. This approach

16/42

365

370

375

380

385

390

Page 18: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

provides likely estimates of the French population exposure to styrene resulting from the

consumption of yogurts.

It is emphasized that the probabilistic estimation of exposure combine two different scales –

food product unit and household scale – and different kind of distributions - delta-

distributions, discrete and continuous ones. At the scale of each food product unit, contact

times are thus delta distributed, whereas concentrations in food are continuously distributed.

At household scale, distributions are either discretely (e.g. contact times) or continuously (e.g.

v * or individual exposure) approximated.

In order to identify trends and deviations between results different strategies relying on the

properties of each explicative variable are used. Contact times, used as explicative variable at

the scale of tested population, are arithmetically averaged over all product units consumed by

each household. Consumption rates and delays between purchases, used as explicative

variable, are averaged over the observed time frame (1 year) utilizing a root mean square

average operator. By contrast, the individual exposure to styrene is mainly discussed

according to either its 50 or 95th percentile values. The 50th percentile defines indeed the likely

individual exposure value and the 95th exposure value defines an upper bound that takes into

account the effects of both “unknown” sources variation and uncertainty (e.g. initial

packaging formulation, material structure, storage conditions).

3. Results and discussion

3.1 Comparison of contact times reconstructed according to H1 and H2 scenarios

Results present first how differ household averaged contact times reconstructed according to

scenarios H1 and H2. Figure 3 compares averaged reconstructed contact times derived from

17/42

395

400

405

410

415

Page 19: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

scenarios H1 and H2. Scenario H1 leads to values asymmetrically distributed between 2 and 16

days with a likely value of 5 days. On the contrary, scenario H2 yields more widely and

symmetrically distributed values between 3 and 22 days with a likely value of 12 days. Due to

possible stock creation, H2 generates strictly higher values than H1 ones. The discrepancy

between both values varies significantly according to household practices. It rises up to 10

days when the expected contact time increases. Low consumption rates give more particularly

both higher contact times and higher discrepancy between averaged reconstructed values.

3.2 Distributions of individual exposure within a household and between households

Individual exposure distributions are presented for some households belonging to typical

consumption patterns and compared to the weighted distribution that is representative for all

the tested population. The main objective is to identify the possible range of variation of the

individual exposure value due to tested scenarios, differences in consumption, unexplained

variations between households, uncertainty in the amount of styrene that may migrate.

Figure 4 plots the cumulative distributions of individual exposure for households belonging to

a similar consumption pattern for all combinations of tested conditions

( ) ( )1 2 3H ,H ,H 1, 100,Bi Bi Bi´ = = ® ¥ . Three typical patterns are considered, based on the 5th,

50th and 95th percentile values of gathered individual consumption rates. For each pattern, the

cumulative distributions of individual exposure of 20 households – with averaged

consumption rates distributed closest to the targeted value – are plotted. Due to the skewness

the distribution of consumption rates (Vitrac et al. 2005) and the finite size of the household

population, only the two last patterns consist in homogeneous samples. The cumulative

distribution of individual exposure accumulated over all the tested population (called

“gathered” distribution) is also depicted. Typical values are collected in table 2.

18/42

420

425

430

435

440

Page 20: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

In particular, the conservative combinations ( )3H 1, 100,Bi Bi Bi´ = = ® ¥ illustrate the effect

of the variation in consumption pattern alone on the exposure dispersion. For a given Bi

value, scenario H3 is indeed responsible for an uniform contamination of all consumed

yogurts.

3.2.1 Gathered distributions of exposure: towards an estimation of the exposure of the general

population

The spreading of the exposure “gathered” over all examined households is assessed by the

ratio between the 95th and 5th percentile values. For similar tested conditions, the spreading is

higher when realistic contact times are considered (figures 4a, 4b, 4d, 4e, 4g and 4h) than

when a rough estimation of contact times (i.e. sale by date) is assumed (figures, 4c, 4f and 4i),

with ratios ranged between 1.5 and 2 decades and ranged between 1 and 1.5 decades

respectively. Upper percentile values calculated with scenario H3 are however up to 2 to 5

times higher than those calculated with scenario H1 and H2 (table 2). It is hence concluded that

accounting for contact times between yogurts and their packaging materials at household scale

modifies significantly the estimation of the individual exposure to styrene of the tested

population.

Since all reconstruction scenarios account for similar uncertainties in 0c and D values, the

bias between the “household” approach (i.e. scenarios H1 and H2) and the “global” approach

(i.e. scenario H3) is only related to the variability in storage practices between households. It is

also demonstrated that the introduction of realistic assumptions combined with an additional

source of uncertainty on storage practices (i.e. discrepancy between H1 and H2 scenarios) does

not increase the whole uncertainty but contributes to a more realistic estimate of an upper

bound of the exposure for the whole tested population.

19/42

445

450

455

460

465

Page 21: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

All 9 “gathered” distributions of exposure have similar shape (asymmetric distributions with a

right decreasing tail) but whose scale values change significantly with the tested condition. At

the scale of the whole tested population, the lowest median exposure, 0.3 µg⋅day-1⋅pers-1, is

obtained with the condition ( )1H 1Bi´ = . The condition ( )3H Bi´ ® ¥ gives the highest

median value, about 35 µg⋅day-1⋅pers-1. As a result, the tested conditions introduce

discrepancies in exposure estimates as high as expected variations within the tested population

(up to 2 decades). The differences between tested conditions will be discussed further but it is

noticeable that assuming unlikely contact times, as scenario H3 does, leads to a positive shift

of exposure values that is higher when Bi values are higher. Such effects illustrate the time

dependence of exposure estimates on the migration process and consequently on household

practices.

3.2.2 Typical distributions of exposure at household scale

Individual exposure distributions at household scale, corresponding to the 5th, 50th and 95th

percentiles of consumption rates, are distributed along the “gathered” exposure scale with

median values that coincide approximatively with the 5th, 50th and 95th percentiles of the

“gathered” exposure values. As a result, individual exposure and consumption rate values

appear highly correlated together. As expected, the dispersion of median individual exposure

is minimal under scenario H3 (figures 4c, 4f and 4i). On the contrary, the dispersion is the

highest for low consumption rates and scenario H2. Results summarized in table 2

demonstrate also that variability between households is higher (up to a factor five) than the

uncertainty (lower than a factor 2) defined by the ratio between the 5th and 95th percentiles

values of the individual exposure. It is consequently demonstrated that other factors, such as

equivalent contact times at a reference temperature, must be taken into account to achieve

realistic exposure estimations to styrene originating from yogurt pots.

20/42

470

475

480

485

490

20

Page 22: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

In this study, the effect of time is not concealed by the considered sources of uncertainty:

either initial concentration in the food packaging material or in the diffusion coefficient. In a

similar way, substituting a reconstruction scenario (either H1 or H2) does not modify

dramatically the conclusions. Only the parameter Bi modifies drastically the expected

dynamics of the migration by changing the limiting transport and the corresponding dynamic

of the contamination of packed food materials.

3.3 Dispersion of household-scale estimated exposure values among the tested population

In order to emphasize on variations between household exposures, figure 5 plots the

distributions of the 50th (likely exposure) and 95th (exposure including uncertainty) percentile

values of styrene exposure estimated at household scale among the tested population (5473

households). The effects of all combinations of tested conditions

( ) ( )1 2 3H ,H ,H 1, 100,Bi Bi Bi´ = = ® ¥ are depicted. The corresponding exposure values at the

maximum of probability are gathered in table 3.

Variable contact times (figures 5a, 5b, 5d, 5e, 5g and 5h) yield distributions of 50th and 95th

percentile exposure values with very similar shapes. The distributions are asymmetrical and

exhibit a decreasing tail. They are conveniently approximated by log-normal distributions. By

contrast, scenario H3 (figures 5c, 5f and 5i) generates distributions presenting a left increasing

tail, which cannot be fitted by log-normal distributions. The alike continuous distribution is a

particular form of generalized Beta distributions. This discrepancy in the shape reveals the

additional effect of contact times on exposure estimates. For a same Bi value, assuming a

similar contact times for all yogurts (scenario H3) shifts drastically the distributions towards

higher exposure values. As a result, the mode of the distribution predicted by scenario H3 is

higher than 90 % of values calculated under H1 or H2.

21/42

495

500

505

510

515

Page 23: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

In details, distributions of either 50th or 95th percentile values are derived from a homothetic

transformation with a factor ranged between 1.2 and 1.6 (see table 3). This ratio is a measure

of the uncertainty in the “likely” exposure value inherent to possible variations in initial

concentration and to the uncertainty in the true diffusion coefficient. By choosing the 95th

percentile exposure value as reference, the likely exposure at household scale varies between

0.4 and 64 µg⋅kg-1⋅pers-1 depending on the tested conditions. Scenarios H2 and H3

overestimates the likely exposure calculated from scenario H1 by a factor ranged between 1.5

and 2.1, and between 2.6 and 4.9, respectively. The overestimation is higher when the effect

of time on yogurt contamination, and therefore on exposure, is almost linear (i.e. when 1Bi>® )

.

The parameter Bi combined with contact times influences significantly the likely exposure.

The exposure is increased by a factor ranged between 35 and 66 when it is increased from 1 to

an infinite value. The case 1Bi = is unlikely as it would assume that the mass transfer

resistance is as high within the yogurt than within the packaging material (from the Bi

definition, it would be stated that the diffusion of styrene in yogurt is only 100 times higher

than in polystyrene). On the opposite, the case Bi ® ¥ is also poorly realistic as it assumes

that the yogurt is a perfectly stirred liquid. It is however realistic to envision different Bi

values between stirred (expected high Bi values) and gelified yogurts (expected lowBi

values), probably about one decade. Since mass transfer coefficients between food and

packaging materials are poorly described in the literature, the choice 100Bi = seems an

acceptable compromise for all cases.

The choice of the partition coefficient, K , is not discussed in this paper as it is an apparent

physicochemical property, which is poorly documented. In addition, it is known to be highly

variable according to the composition of yogurt (mainly due to the fraction of lipid fraction,

22/42

520

525

530

535

540

Page 24: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

maximum 4% w/w on a wet basis) and the possible separation of gel and liquid phases during

storage of yogurt (exudation, gel dissolution). A value of 1 is assumed as an upper bound of

physically realistic values, but results can be straightforwardly extrapolated to other K values

(i.e. K <1) from algebraic equation (2). As an indication, values of K between crystal

polystyrene and different foods are given by Till et al. (1982). It must be noted that a similar

transformation or correction is not possible with Bi because it acts non-trivially and non-

linearly on the contamination of food products.

3.4 Comparison of calculated values with previously published exposure estimates

Few estimations of lifetime exposure to styrene from dietary intake have been published in the

literature comparatively to the large amounts of available data on the migration of styrene into

food and food simulants. According to the investigations of Tang et al. (2000) conducted in

Germany, if all consumed milk products (average consumption rate about 0.338 kg⋅day-1⋅pers-

1) and fat products (average consumption rate 0.072 kg⋅day-1⋅pers-1) were packed in styrene

materials, causing a contamination in styrene ranged between 5 and 30 µg⋅kg-1, the

corresponding daily intake would reach 2 – 12 µg⋅day-1⋅kg-1. In the particular case of yogurts,

the British Ministry of Agriculture Fisheries and Food (1983 and 1995) reports a broader

range of styrene concentrations in yogurt pots that were sampled on the market, with values

between 1 and 200 µg⋅kg-1. By comparison, simulated contamination of yogurt pots assigned

in the current exposure assessment procedure are ranged between 10 and 350 µg⋅kg-1

(respectively 5th and 95th percentile value) for the likelihood condition H2× 100Bi = (see

Vitrac et al. 2005 for details). Svensson (2002) considers that a styrene intake of 15-30 µg⋅kg⋅

pers-1 due to the contact of food with styrenic materials would be a realistic estimate for highly

exposed consumers. Average exposure would range from 0.2 to 9 µg⋅day-1⋅pers-1.

23/42

545

550

555

560

565

Page 25: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

As a result, the 95th percentile exposure value predicted by the current full probabilistic

approach, that is 20 µg⋅day-1⋅pers-1, under the likely condition H2× 100Bi = , is realistic and

possibly consistent with previous studies and in particular for high consumers. According to

this last estimation, exposure to styrene caused by the consumption of yogurts in polystyrene

materials would contribute significantly on the total exposure to styrene estimated to be

between 18.2 and 55.2 µg⋅kg⋅pers-1 reported by Tang et al. (2000) From this observation and

for particular consumers, the ingested amount of styrene could have the similar magnitude as

the inhaled amount. In particular, it is worth to notice that the overestimated exposure (95th

percentile of the household-scale estimated exposure) varies significantly between

households. Under the likely condition H2× 100Bi = , 5% of the most exposed households are

exposed to values above 60 µg⋅kg⋅pers-1 (figure 5).

By comparison, the predictive methodology proposed by Lickly et al. (1995), relying on

migration modeling and consumption factors and applied to a wide range of products, leads to

an average styrene exposure values of 9 µg⋅day-1⋅pers-1 for the US population. According to

this study, styrene exposure related to yogurt pots represented 31 % of the total exposure from

polystyrene food packaging. Concentration in yogurts were calculated assuming a hot fill

(leading to a contact of 30 min at 65°C) of containers and followed by long term storage (60

days) at 4°C.

The raw comparison of simulated results with previously estimates of styrene exposure is

difficult in practice. First, because previous studies were not focused on really consumed

yogurts (i.e. sampled in households) and because significant differences in the consumption of

yogurts and therefore in storage practices are expected between Countries, even within the

EU. Differences in the legal definition of yogurt must be also underlined, as the French

definition of yogurts is very restrictive regarding the fermenting strains (see paragraph 2.1).

24/42

570

575

580

585

590

Page 26: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Secondly, surveys (Withey 1976, Withey and Collins 1978, Gilbert and Startin 1983, Ministry

of Agriculture, Fisheries and Food, 1983, 1995 and 1999), which collect contamination data,

do not report generally the conditions that make it possible to compare results with similar

mass balances (via equation (2)) and kinetic considerations (see section 2.2). The type of

yogurt (composition, texture, and volume) and packaging (thickness, residual concentration in

containers in styrene and in plasticizers) and the duration of contact are rarely available. In

addition, contrary to the study conducted by Lickly et al. (1995), it is often unclear whether

samples were subjected to hot fill, conservative conditions of storage or not (e.g. storage at

8°C or 10°C during 40 days rather than few days or weeks at temperatures ranged between 2

and 4°C).

The use previously published migration data into food simulants (Davies 1974, Varner S. L.

and Breder 1981, Miltz and Rosen-Doody 1984, Snyder and Breder 1985, Linssen et al. 1991,

Murphy et al. 1992, Lehr et al. 1993, Lickly et al. 1995) as alternative to contamination data

are also questionable. Indeed, migration into oil would be a too severe migrant condition as it

is suspected to plasticize packaging materials in polystyrene and consequently to increase the

apparent diffusion coefficient of styrene (Milz and Rosen-Doody 1984, Linssen et al. 1991).

On the opposite, aqueous or acidic simulants, as low viscous liquids, are not satisfying as they

may generate underestimated K values (too much optimistic scenario) and may drastically

increase the yielded Bi value (too much pessimistic scenario). This effect was experimentally

assessed by Till et al. (1982) using radio labeled styrene monomer. Finally, from equation (2),

one must also notice that migration levels assessed in food simulants for a particular L value

(e.g. 6 dm2 of packaging material in contact 1 kg of food) cannot be introduced “as is” into

exposure assessment procedures and must be corrected to match realistic dilution factors.

Previous comments suggest that it would be preferable to use calculated diffusion and

partition coefficients from the literature instead of raw contamination data that may not

directly related to real cases. The probabilistic approach of the migration developed in Vitrac

25/42

595

600

605

610

615

25

Page 27: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

and Hayert (2005) and used in this work can be used to extrapolate the contamination and

therefore the consumer exposure to real conditions of storage ,while taking into account the

inherent uncertainty/variability in published results. As an example of uncertainty, Till et al.

studies (1982 and 1987) demonstrate that the diffusion coefficient of styrene in polystyrene

was independent on the concentration in styrene whereas Miltz and Rosen-Doody (1984)

showed a dependence of the diffusion coefficient with the residual monomer concentration.

Similarly, the effect of mineral oil as possible plasticizer of styrene container (Jickells et al.

1994, International Life Sciences Institute 2002), which may modify diffusion coefficients by

a factor ranged between 5 and 25 (Licky et al. 1995) can be analyzed and introduced in the

exposure assessment as initial uncertainty in D values.

3.5 Correlations between contact times, purchase practices and individual exposure

Possible correlations at household scale between model quantities are analyzed this sub-

section. General recommendations for the development of either robust or simplified

strategies are finally discussed by comparing results derived from the full approach at

household scale and from rough approximations.

Previous results suggested that highly exposed consumers would not only be related to high

consumers but also to particular purchase and storage practices. Since it is expected that

highly consumers are more likely to minimize their storage time and their stock volume, this

part analyzes conditions that modify significantly the 95th percentile exposure value calculated

at household scale. Once again, it is emphasized that this study does not seek absolute

reference values of consumer exposure to styrene but assesses the effects of combined sources

of uncertainty or variation in household behaviors on exposure estimates derived from a full

probabilistic approach and a set on initial inputs (see table 1).

26/42

620

625

630

635

640

645

Page 28: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

3.5.1 Correlations between consumption rates and contact times

Since contact times are expected to be related to consumption rates, figure 6 plots on a log

scale the apparent correlation between contact times and averaged individual consumption

rates for both scenarios H1 (figures 6a and 6c) and H2 (figures 6b and 6d). The individual

consumption is expressed in number of packaging units (pots of 125 g). All scattered data are

averaged over all yogurt pots consumed during one year by each household. The surface of

each symbol is proportional to the 95th percentile of the cumulated individual styrene

exposure. In addition, since the concentration in styrene and subsequently exposure are

expected to vary significantly with the assumed resistance to mass transfer at the interface,

both results inferred from the extreme boundary condition 1Bi = (figures 6a and 6b) and the

likely boundary condition 100Bi = (figures 6c and 6d) are also depicted.

Similar conclusions are derived from both contact time scenarios. As expected, contact times,

ranged between 8 and 22 days, are maximum when consumption rates are minimum. Below

0.1 pot⋅day-1⋅pers-1, contact times are not correlated with consumption rates. For average

consumption rates ranged between 0.1 and 2 pot⋅day-1⋅pers-1, an increase in consumption rate

is related to a significant reduction in contact times. For a same consumption rate, the

variability between households is however very significant; it is ranged between 1 and 12 days

under scenario H1 (figures 6a and 6c) for a typical individual consumption rate of 1 pot (1

yogurt) per day. For a same consumption pattern, these heterogeneities in contact times

suggest significant related variations in styrene exposure.

Both possible combined effects of consumption and storage practices on the individual

exposure to styrene originating from yogurt pots are illustrated by the orientation of symbols

with similar size ((i.e. Related to similar exposure values). High exposure values are in

27/42

650

655

660

665

670

Page 29: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

particular related either to high consumption rates or to long contact times. It is noticeable that

the spreading of exposure values is similar when consumption rate is the only variable

parameter (i.e. when the contact time is fixed) and when contact time is the only free

parameter (i.e. when the consumption rate is fixed). In other words, individual exposure

cannot be realistically estimated by accumulating an amount of styrene that would correspond

to a particular contact time and consequently that would not vary between households. This

conclusion is particularly illustrated in figure 5, where scenario H3 generates distributions of

95th exposure values that are highly dissimilar in shape. A same bias will occur if exposure to

styrene is only based on the measured concentrations in styrene of yogurts that are sampled on

the market or that are subjected to non realistic storage conditions.

In details, the 4 tested conditions lead to very different scales for the estimated 95th percentile

exposure values. The lowest range of exposure values, 10-2 and 2 µg⋅day-1⋅pers-1, is derived for

scenario H1 associated with boundary condition 1Bi = (figure 6a). The highest range, 0.7 and

75 µg⋅day-1⋅pers-1, is obtained for scenario H2 associated with 100Bi = (figure 6d). Based on

scenario H1, the likely situation corresponds to an average consumption rate of 0.25 pot⋅day-1⋅

pers-1 and a contact time of 5 days. This situation corresponds to exposure ranges of 0.4-1 µg⋅

day-1⋅pers-1 and 10-40 µg⋅day-1⋅pers-1 respectively to boundary conditions 1Bi = (figure 4a) and

100Bi = (figure 4c). Scenario H2 related to a likely contact time of 12 days leads to intervals

of 1.5-2.5 µg⋅day-1⋅pers-1 and 20-40 µg⋅day-1⋅pers-1 respectively to boundary conditions 1Bi =

(figure 4a) and 100Bi = (figure 4c).

3.5.2 Identification of household practices that possibly increases the exposure to styrene for a

same consumption rate

28/42

675

680

685

690

695

Page 30: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

As previously detailed, exposure to styrene increases globally with consumption rate. For a

same consumption pattern, household practices, which may affect contact times, can also

worsen exposure. Two parameters influencing contact times are investigated independently:

the averaged volume of purchases (figure 7) and the delay between purchases (figure 8). For

each figure, the scatter of 95th exposure values is plotted versus each explicative value for both

scenarios H1 and H2 and both boundary conditions 1Bi = and 100Bi = . The surface are of

each symbol is proportional to the average consumption rate. A significant correlation with

the explicative variable is identified by an orientation of the scatter.

The exposure increases significantly with the average volume of purchases (figure 7). Since

symbols with a similar size (i.e. related to a similar consumption pattern) are also oriented

along the scatter, it is hinted that this effect is partially independent on consumption rate. High

exposures are thus a consequence of both high consumption rates and high purchase volumes.

Figure 8 generates more complex results. The delay between purchases is globally poorly

correlated to exposure. According to the global orientation of the scatter, the exposure seems

decreasing when the delay between purchases increases. This global effect is accompanied

with a reduction in symbol size, which corresponds a simultaneous reduction of the

consumption rate when the delay between purchases increases. In details, an apposite trend is

discernible for high consumption rates showing by contrast an increase in exposure when the

delay between purchases is increasing. For a same high consumption rate ranged between 2

and 10 pots⋅day-1⋅pers-1 , a variation in the delay between consecutive purchases from 2 day to

20 days increases the individual exposure by a factor ranged between 3 and 7. High

consumption rates and doing infrequent purchases are both worsening factors. The effect of

consumption rates appears at median consumption rates but is indiscernible for very low

consumption rates.

29/42

700

705

710

715

720

Page 31: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

For a same consumption pattern, figures 7 and 8 demonstrate that the average delay between

purchases and the volume of purchases may modify household exposure in all considered

scenarios and conditions. At low consumption rates, the effect of the volume of purchases

dominates while the average delay between purchases is one of the main dominating factor at

high consumption rates. The last effect is expected to be higher if storage is prolonged beyond

28 days.

3.6 Comparison of realistic exposure estimates with simplified quantifications

The simplification of the current approach may be of high concern for domains that require

clear conclusions without reflecting accurately the reality, for instance: risk assessment

regarding a specific application or usage based on worst case scenarios, guidance for further

detailed sanitary survey based on a classification of packaged products. These domains require

conservative conclusions that ensure enough safety margins for the consumer, that is to say:

that introduce overestimations above the previously identified aggregated range of uncertainty

and variability. The addressed questions are therefore what could be the amplitude of the

required safety range/margin ? what could be the appropriated simplifications ?

Within the general framework of the proposed methodology, figures 9 and 10 analyze the

effects of two exposure approximations on exposure estimates: time aspect of migration are

first neglected and all yogurts are consumed at the sale-by-date (figure 9); secondly, the

external resistance to mass transfer is additionally neglected (figure 10). These effects are

analyzed as correlation plots, where realistic estimates calculated for all combinations of

( ) ( )1 2H ,H 1, 100,Bi Bi Bi´ = = ® ¥ are plotted on a log-log scale against corresponding

approximated ones, respectively ( ) ( )1 2H ,H 1, 100Bi Bi´ = = (figure 9) and ( ) ( )1 2H ,H Bi´ ® ¥

(figure 10). Since all tested approximations overestimate the expected exposure, all

30/42

725

730

735

740

745

30

Page 32: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

approximated results are below the straight line. The distance to the line y x= depicts the

corresponding bias or the equivalent safety margin when “worst case” scenarios or conditions

are applied. To illustrate how the bias may fluctuate between consumption patterns, the

surface area of each symbol is proportional to the average consumption rate. Two percentile

values of exposure are plotted : 50 and 95th. It is underlined that the 95th percentile value takes

into account the uncertainty in the real contamination value of yogurts.

At first sight of figure 9, the scatter of 50 and 95th percentile values are spread in a direction

almost parallel to the straight line y x= . The scattering along y rises when the expected

exposure increases, but is noticeable that the ratio between the overestimate and the maximum

corresponding expected value (predicted with realistic and/or detailed assumptions) remains

almost constant when the exposure increases. In other words, assuming that all yogurts are

consumed at the sale-by-date provide a safety ratio regarding the most exposed person, noted

Sr . Sr is independent of the assumed true exposure and varies between 1.4 and 2 depending on

the considered boundary condition and the considered percentile value. Since the additional

storage time due to handling and retailing is disregarded in both scenarios H1 and H2, Sr

ranges may be comparable to the uncertainty in contact times before household storage.

Deviations with similar magnitudes are observed between exposure values estimated either

according to scenarios H1 or H2, or according to 50th and 95th percentiles.

In details, for a same exposure overestimate, Sr is significantly lower than the ratio between

the lowest and the highest expected values. The latter ratio is noted Vr and it is related to the

effect of the variability of behavior between households. From this interpretation, scenario H3

overestimates slightly the expected exposure of the least exposed consumer by a factor up to

one decade and defined by S Vr r× . The overestimation is particularly higher when the exposure

or the consumption rate is high. This trend observed for very different boundary conditions,

1Bi = and 100Bi = , confirms that scenario H3 is only realistic for low consumers while it

31/42

750

755

760

765

770

Page 33: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

deviates significantly from the likely averaged exposure for regularly consumed products. In

particular, it is shown that an analysis not performed at household scale, as scenario H3, does

not take into account that households with high consumption rate are less exposed due to

quicker turnovers of their stock. This effect is identified in figure 9 by vertical gradients in

symbol sizes.

The sensitivity to contact time scenarios is characterized by S Vr r× values lower than 10

verifying the following inequality S Vr r< . By contrast, the sensitivity to boundary conditions,

as recapitulated in figure 10, is characterized by higher total deviations S Vr r× (up to 30 for

1Bi = , figures 10a and 10b) and a different inequality: S Vr r³ . As a result, the Bi value must

be chosen with care if realistic exposure estimates are required. When only exposure

overestimates are required, the appropriated simplifications depend mainly on the food

texture.

For solid food (figures 10a and 10b), the value Bi ® ¥ provides a sufficient margin ( S Vr r>>

) so that the variability between households (contact times) and other sources of uncertainty

(initial concentration, diffusion coefficient) may be disregarded.

For liquid or semi-liquid food (figures 10c and 10d), the condition Bi ® ¥ generates safety

margins, Sr , about 2, which are of similar magnitude as the inter-household variability Vr and

the uncertainty in contamination values (defined by the ratio between the 50th and 95th

individual exposure values). Since the condition Bi ® ¥ does not overestimate dramatically

the migration level (figure 2), the choice of Bi ® ¥ seems an appropriate alternative for food

materials with low viscosities when no pertinent information is available on realistic Bi

values. Neglecting the sources of uncertainty (initial concentration and diffusion coefficients)

would underestimate the 95th expected exposure by a factor 2 which cannot be envisioned

without a significant source of overestimation in addition to the safety margin provided by the

assumption Bi ® ¥ . For example, neglecting the variability between households by assuming

32/42

775

780

785

790

795

Page 34: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

a similar contact times for all yogurts would additionally overestimate the expected exposure

by a factor 3 to 5 on average (figure 9).

In solid food materials that include a liquid phase, as a supernatant or an exudate, the choice

of high Bi values may also be of interest when the risk assessment includes the consumption

of the liquid surrounding the solid food.

As a rule of thumb, the choice of additional overestimations combined with the approximation

Bi ® ¥ is not encouraged, since all sources of uncertainty (in both 0c and D values, Bi

guess, reconstructed contact-times) and the expected variability in storage practices have

similar effects on exposure for a same averaged consumption rate. The particular combination

of approximations total contact time equal to the sale-by-date and no external resistance to

mass transfer would reflect no reality. Indeed, this extreme exposure would correspond to

unlikely associations of practices by consumers that aim at promoting mass transfer between

food and its packaging material such as storage at high temperature (or long term storage)

associated with a mechanical modification of food texture (e.g. mixing). Such estimates do

not fulfill with consistency requirements for risk assessment but can be of high concern for

risk managers to demonstrate that, for a given consumption rate (e.g. 1 kg), a great security

margin exists between the maximum physically realistic accumulated exposure and the dose

that is potentially considered dangerous by toxicologists. Thus, from the cumulative densities

functions plotted in figures 4e and 4i, assuming scenario H3 and Bi ® ¥ , when the reality

would match with scenario H2 and 100Bi = , leads to overestimate the gathered exposure by a

factor up to 8.

4. Conclusions

33/42

800

805

810

815

820

Page 35: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

This work describes a multiscale approach to assess the exposure of consumers to substances

originating from the plastic layer in contact with food. Since at the most granular scale, it is

based on the prediction at the scale of all packaged food product units that are consumed by a

given household or individual during a long period (e.g. 1 year), the approach can be used to

investigate the cumulative intake of particular substances originating from packaging

materials according a given consumption pattern, household composition, household practices

(e.g. frequency of purchases, storage conditions, reuse conditions…). Mechanistic models of

the migration based on consistent physicochemical properties are particularly efficient in

refined exposure assessment procedures as they can predict the possible contamination of food

in realistic conditions of migration by accounting for the true geometry of packaging and food

materials, the conditions of storage (duration, temperature…), the texture of food, the initial

concentration in packaging material... The uncertainty in model parameters is propagated

through a stochastic resolution of transport equations as described in Vitrac and Hayert (2005)

or by performing a sensitivity analysis including interactions that relies on likely intervals of

each parameter. In both situations, efficient algorithms (based on Monte Carlo or pseudo

Monte Carlo sampling, interval algebra) are available in the literature.

This paper examines the application of a “full probabilistic” approach of both contamination

and consumption to assess the exposure to styrene originating from yogurt pots at household

scale. Since a large panel of households (5,473) is considered, the exposure of a given

population is also derived. From the methodological point of view, styrene is particularly

interesting as its main source in both the diet and food contact application is particularly well

identified and makes it possible to a perform a detailed sensitivity analysis of the whole

methodology. Main input data are volume and frequency of purchases extracted from

marketing databases. The main unknown at the scale of each packaging unit is reconstructed

from two likely storage handling scenarios, H1 and H2, already applied in Vitrac et al. (2005).

34/42

825

830

835

840

845

850

Page 36: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

The individual cumulative exposure to styrene is estimated at household scale by

accumulating the amount of styrene that is expected to migrate in all yogurts purchased by a

given household. The concentration in each yogurt is calculated by accounting for the

expected variability in the initial concentration in packaging materials and for the uncertainty

in the diffusion coefficient of styrene monomer in polystyrene. The main source of uncertainty

is related to boundary conditions between food and packaging materials, the non linear effect

of Bi on exposure estimates is analyzed via typical values: { }1,100,Bi = ¥ .

Both scenarios of reconstruction of contact times H1 and H2 show that individual exposure

does not vary independently with the averaged consumption rate and contact time. High

consumers consume products that are up to 5 times less contaminated than those consumed by

low consumers. In a same manner, for a same consumption rate, the individual exposure may

vary by a factor up to 3 according to household practices. This effect is significant since it

higher than the initial uncertainty in input parameters (initial concentration in the packaging

material, diffusion coefficient at conditions of storage). The volume of purchase and the

average delay between purchases are the main explicative parameters of the discrepancy

between similar consumers.

The median individual exposure value (respectively the 95th percentile value) derived from the

gathered distribution varies between 0.3 (0.72) µg⋅day-1⋅pers-1 and 35 (120) µg⋅day-1⋅pers-1

between extreme tested conditions ( )1H 1Bi´ = and ( )3H Bi´ ® ¥ . The likely condition

( )2H 100Bi´ = leads to 12 (35) µg⋅day-1⋅pers-1. Since yogurt pots represent one the main

source of styrene through the diet (Palmer, 1996), it is found that the calculated likely values

have magnitudes consistent with previously published value of diet exposure to styrene. For

consumers combining both high consumption pattern and high storage time, this work shows

however that exposure to styrene originating from consumption of yogurts could be as high as

35/42

855

860

865

870

875

35

Page 37: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

exposure through inhalation. For a regularly consumed products as yogurts, changing the

scenario H2 to H1 yields similar conclusions 8 (21) µg⋅day-1⋅pers-1. On the contrary, the

scenario H3, which assumes that all yogurts are consumed at the sale-by-date give significantly

overestimated values 25 (80) µg⋅day-1⋅pers-1 for a same likely 100Bi = value. For a same

consumption rate, the scenario H3 overestimates coarsely the variability between households.

Thus, for a median consumption rate (respectively 95th percentile of highest consumers), the

95th percentile of individual exposure is expected to be ranged respectively between 9 and 25

(32 and 65) µg⋅day-1⋅pers-1 and under the likely condition ( )2H 100Bi´ = .

As a result, conventional probabilistic approaches of exposure, based on accumulated and

assumed uncorrelated distributions of food contamination and consumption, provide only

overestimates of individual exposure. Indeed, when contamination values are controlled by

time dependent processes, the bias related to the overestimation of food contamination may

arises from both an overestimation of the “true” contact time but also from the increasing

uncertainty on the migration kinetic when time is running. Coarse risk assessments, which do

not include household practices, may therefore conclude that the risk is only related to

consumption factors without identifying potential risky practices or misuses of packaging.

By contrast, risk managers may promote rough overestimates. This work provides first ranges

of uncertainty associated to conventional approximations used to simulate the migration from

contact materials into food starting from migration testing in simulants or numerical

simulations. In descending order, neglecting the external resistance to mass transfer between

food and its packaging overestimate the “true” exposure by a factor ranged between 2 and 30.

Assuming a pessimistic contact time based on the sale-by-date and not on the “true” or likely

distribution of contact time increase the individual exposure by a factor ranged between 1.5

and 3. Uncertainty related to the initial concentration in food packaging material and

conditions of storage adds a factor between 1.2 and 2.

36/42

880

885

890

895

900

Page 38: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

5. Prospects

The proposed probabilistic approach based on modular approaches may provide a suitable

basis for the reevaluation of reduction factors, or for the risk assessment and management

associated to the use of complex materials (e.g. multi-layers materials, active/intelligent

materials) or particular conditions of use (e.g. reuse, cooking purposes) or recycling. Indeed,

the strategy that would assume very conservative assumptions, such as “package in the same

material for lifetime”, “same additives”, “100 % migration”, “100 % market share”, is not

appropriate as it dramatically increase the adverse to risk related to the rejection of a safe

application and could lead to erroneous conclusions for miss-used approximations. Complex

packaging applications need sophistications in the predictive migration models and additional

physicochemical properties such as activation energy and partition coefficients.

The extension of the current approach to screening strategies for a wide range of migrants

requires a significant effort in collection of non-aggregated information on packaging usage

(type, geometry, food application, typical formulation of the packaging material) at EU scale.

A cost effective strategy relies on the use of purchase data alone as those available through

marketing companies. The available information is however disparate in quality (tested

population, product type, time frame…) and incomplete regarding the dependent migration

problem. Algorithms of reconstruction of missing information (contact time, identified

consumers within the tested household…) based on likelihood maximums require to be

validated or completed on purpose with dedicated household surveys.

References

37/42

905

910

915

920

925

Page 39: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Begley, T., Castle, L., Feigenbaum, A., Franz, R., Hinrichs, K., Lickly, T., Mercea, P., Milana,

M., O'Brien, A., Rebre, S., Rijk, R., Piringer, O., 2005, Evaluation of migration models that

might be used in support of regulations for food-contact plastics, Food Additives and

Contaminants, 22(1), 73-90.

Chatwin, P. C. and Katan, L. L., 1989, The role of mathematics and physics in migration

predictions, Packaging Technology and Science, 2, 75-84.

Cranck, J., Mathematics of diffusion, 2nd edition. Oxford University Press, Oxford, 1975.

Cohen, J., 2002, Evaluating the risk to workers and the public from styrene exposure, Harvard

Center for Risk Analysis, 10(3), 1-6.

Combris, P, Delobel and D, Boizot, C., 2000, La consommation alimentaire en 1998:

distribution des quantités consommées à domicile. Rapport du laboratoire de

recherches sur la consommation (France, INRA).

Davies, J. T., 1974, Migration of styrene monomer from packaging material into food:

Experimental verification of a theoretical model, Journal of Food Technology, 9(3),

275-283.

EC-DG SANCO-D3, 2003, Conclusions of the Project FAIR CT98-4318 “Recyclability:

Programme on the recyclability of food packaging materials with respect to food safety

considerations – polyethylene terephthalate (PET), paper and board and plastics covered

by functional barriers”.

European Commission, 2003, Food contact materials: a practical guide for users of European

Directives. http://europa.eu.int/comm/food/fs/sfp/food_contact/practical_guide_en.pdf

Gauchi, J.-P. and Leblanc, J.-P., 2002, Quantitative assessment of exposure to the mycotoxin

Ochratoxin A in food. Risk Analysis, 22(2), 219-234.

Gibney, M. J. and van der Voet, H., 2003, Introduction to the Monte Carlo project and the

approach to the validation of probabilistic models of dietary exposure to selected food

chemicals, Food Additives and Contaminants, 20,1-7.

38/42

930

935

940

945

950

Page 40: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Gilbert, J., and Startin, J. R., 1983, A survey of styrene monomer levels in foods and plastic

packaging by coupled mass spectrometry-automatic headspace gas chromatography.

Journal of the Science of Food and Agriculture, 34, 647-652.

Hamey, P. and Harris, C.A., 1999, The variation of pesticide residues in fruits and vegetables

and the associated assessment of risk. Regulatory Toxicology and Pharmacology, 30,

S34–S41.

Huggett, A., Petersen, B. J., Walker, R, Fisher, C. E., Notermans, S.H.W., Rombouts, R. M.,

Abbott, P., Debackere, M. , Hathaway, S. C., Hecker, E. F. F., Knaap, A. G. A.,

Kuznesof , K. M., Meyland, I., Moy, G.,Narbonne, J.-F., Paakkanen, J., Smith, M. R.,

Tennant , D., Wagstaffe, P., Wargo, J. and Würtzen, G., 1998, Towards internationally

acceptable standards for food additives and contaminants based on the use of risk

analysis, Environmental Toxicology and Pharmacology , 5, 227–236.

International Life Sciences Institute, 2002, Packaging materials: 2. polystyrene for food

packaging applications, Report, ILSI Europe, Brussels.

International Agency for Research on Cancer, 1994. Some industrial chemicals. IARC

Monographs on the evaluation of carcinogenic risks to humans, 60, 338-339.

Jickells, S. M., Nichol, J. and Castle, L., 1994, Migration of mineral hydrocarbons into foods.

5. Miscellaneous applications of mineral hydrocarbons in food contact materials, Food

Additives and Contaminants, 11(3), 333-341.

Lambe J., 2002, The use of food consumption data in assessments of exposure to food

chemicals including the application of probabilistic modelling, Proceedings of the

Nutrition Society, 61(1), 11-18.

Lau, O.-W. and Wong, S.-K., 2000, Contamination in food from packaging material, Journal

of Chromatography A, 882, 255–270.

Leber, A. P. , 2001, Human exposures to monomers resulting from consumer contact with

polymers, Chemico-Biological Interactions, 135–136, 215–220.

39/42

955

960

965

970

975

Page 41: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Lehr, K. M., Welsh, G. C., Bell, C. D., and Lickly, T. D., 1993, The “vapour-phase” migration

of styrene from general-purpose and high-impact polystyrene into cooking oil. Food and

Chemical Toxicology, 31(11), 793-798.

Lickly, T. D., Breder C. V. and Rainey, M. L., 1995, Model for estimating the daily dietary

intake of a substancefrom food-contact articles: styrene from polystyrene food-contact

polymers, Regulatory Toxicology and Pharmacology, 21, 406-417.

Linssen, J. P. H., Reitsma, J. C. E. and Roozen, J. P., 1991, Effect of sampling method on the

level of styrene monomer migrated from polystyrene packaging material, Packaging

Technology and Science, 4, 171-175.

Miltz, J. and Rosen-Doody, V., 1984, Migration of styrene monomer from polystyrene

packaging materials into simulants, Journal of Food Processing and Preservation, 8,

151-161.

Ministry of Agriculture Fisheries and Food, 1983, The Eleventh Report of the Steering Group

on Food Surveillance, Survey of styrene levels in food contact materials and in foods,

HMSO, London.

Ministry of Agriculture Fisheries and Food, 1995, Food Safety Directorate Food Surveillance

Information Sheet No.38.

Ministry of Agriculture Fisheries and Food, 1999, Food Safety Directorate Food Surveillance

Information Sheet No.189.

Murphy, P. G., Macdonald, D. A., Lickly, T. D., 1992, Styrene migration from general-

purpose and high-impact polystyrene into food-simulating solvents, Food Chemistry

Toxicology, 30(3), 225-232.

Palmer, M., 1996, Assessment of food exposure to packaging in EU countries (Belgium, MPA

International).

Snyder R. C. and Breder C. V., 1985, New FDA migration cell used to study migration of

styrene from polystyrene into various solvents. Analytical Chemistry, 68, 770-775.

40/42

980

985

990

995

1000

40

Page 42: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Svensson, K., 2002, Exposure from food contact materials. Summary report, International

Life Sciences Institute Europe, Brussels.

Tang W., Hemm I. and Eisenbrand G., 2000, Estimation of human exposure to styrene and

ethylbenzene, Toxicology, 144, 39-50.

Till, D. E., Ehnthoff, D. J., Reld, R. C., Scharwz, P. S., Schwope, A. D., Sidman, K. R. and

Whelan, R. H, 1982 Migration of styrene monomer from crystal polystyrene to foods

and food simulating liquids, Industrial Engineering and Chemistry Fundamentals, 21,

161-168.

Till D. E., Schwope A. D., Ehntholt D. J., Sidman K. R., Whelan R. H., Schwartz P. S. and

Reid, R. C., 1987, Indirect food additive migration from polymeric food packaging

materials. CRC Critical Reviews in Toxicology, 18, 215-243.

Treybal, R.E., 1980, Mass transfer operations. 3rd edition, McGraw-Hill, Boston.

Varner S. L. and Breder C. V., 1981, Headspace sampling and gas chromatographic

determination of styrene migration from food-contact polystyrene cups into beverages

and food simulants. Journal of the Association of Official Analytical Chemists, 64,

1122-1130.

Vergnaud, J.M., 1991, Liquid transport process in polymeric materials: modeling and

industrial applications (Polymer Science and Engineering Series). Prentice Hall,

Englewood Cliffs.

Vermeire, T. G., van der Poel, P., van de Laar, R. T. H. and Roelfzema, H., 1993, Estimation

of consumer exposure to chemicals: application of simple models, The Science of the

Total Environment, 136, 155-176.

Vitrac O., Challe B., Leblanc J.-C. and Feigenbaum A.F., 2005, Risk of contamination of

packed foods by substances from the plastic contact layer: a generic quantitative

methodology at the scale of private households. submitted to Food Additives and

Contaminants.

41/42

1005

1010

1015

1020

1025

Page 43: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Vitrac O. and Hayert M., 2005, Risk assessment of migration from packaging materials into

foodstuffs. AIChE J., 51(4), 1080-1095.

Vitrac, O., 2003. Generally recognized migration models : conditions of use of softwares,

limitations and perspectives. Proceedings of the use of mathematical modelling software

applications for the prediction of migration from food contact materials, EUR 20666EN

(Ed. Piccini P., European Communities), 31-107.

Withey, J. R., 1976. Quantitative Analysis of styrene monomer in polystyrene and foods

including some preliminary studies on the uptake and pharmacodynamics of the

monomer in rats, Environmental Health Perspective, 17, 125-153.

Withey, J. R. and Collins, P. G., 1978, Styrene monomer in foods. A limited Canadian survey.

Bulletin of Environmental Contamination and Toxicology, 19, 86-94.

World Health Organization, 1984, Toxicological evaluation of certain food additives and food

contaminants. 28th Report of the joint FAO/WHO Expert Committee on Food Additives.

WHO Food Additives, 19.

42/42

1030

1035

1040

Page 44: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

List of figures

Figure 1. Risk scheme detailed at household scale. A and B are two arbitrary households that

do not exchange food products. Yogurt pots are darkened respectively their expected

contamination value in styrene.

Figure 2. Dimensionless migration kinetics ( *v ) for a) 1Bi = , b) 100Bi = , c) Bi ®¥

assuming 1K = and 210L -= . The developed stochastic approach is illustrated for the

condition { }0.1, 0.5,1Fo = and 0.15Ds = .

Figure 3. Comparison of contact times estimated according to scenarios H1 and H2. Each

symbol represents averaged results for all yogurt pots consumed by a given household during

one year. The averaged household consumption is proportional to the surface of each symbol.

The straight line y=x is plotted in continuous line.

Figure 4. Distributions of individual exposure to styrene for 3 typical consumption patterns

(5th, 50th and 95th percentiles) according to scenarios a,d,g) H1, b,e,h) H2, c,f,i) H3 and boundary

conditions: a,b,c) 1Bi = , d,e,f) 100Bi = , g,h,i) Bi ®¥ . Each consumption pattern is based on

20 households with similar individual consumption rate.

The corresponding distribution weighted over all households is depicted in dashed line.

Figure 5. Distributions of 50th and 95th percentiles of individual exposure to styrene based on

scenarios a,d,g) H1, b,e,h) H2, c,f,i) H3 and boundary conditions: a,b,c) 1Bi = , d,e,f) 100Bi = ,

g,h,i) Bi ®¥ . The distributions are based on 5473 households.

List of Tables & Figures – Paper 04-JG-FAC_090 version 2 1/4

5

10

15

20

25

Page 45: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 6. Effect of individual consumption rates on contact times reconstructed based on

scenarios a,c) H1 and b,d) H2. Each symbol represents averaged results for all yogurt pots

consumed by a given household during one year. The surface of each symbol is proportional

to the 95th percentile of the individual exposure to styrene. Two boundary conditions are

considered: a,b) 1Bi = and c,d) 100Bi = .

Figure 7. Effect of individual consumption rate and volume of purchase on exposure to

styrene based on scenarios a,c) H1 and b,d) H2 and boundary conditions: a,c) 1Bi = and b,d)

100Bi = . Each symbol represents averaged results for all yogurt pots consumed by a given

household during one year. The surface of each symbol is proportional to the averaged volume

of purchase of each household.

Figure 8. Effect of individual consumption rate and delay between purchases on exposure to

styrene based on scenarios a,c) H1 and b,d) H2 and boundary conditions: a,c) 1Bi = and b,d)

100Bi = . Each symbol represents averaged results for all yogurt pots consumed by a given

household during one year. The surface of each symbol is proportional to the averaged delay

between purchases of each household.

Figure 9. Comparison of individual exposure to styrene (50 and 95th percentile) based on

scenarios a,c) H1 and b,d) H2 and boundary conditions: a,c) 1Bi = and b,d) 100Bi = with

values derived from the worst case scenario H3. The straight line y=x is plotted in continuous

line. The 50th percentile values are plotted with black filled symbols and 95th percentile values

with open symbols.

Figure 10. Comparison of individual exposure to styrene (50 and 95th percentile) based on

scenarios a,c) H1 and b,d) H2 and boundary conditions: a,c) 1Bi = and b,d) 100Bi = with

List of Tables & Figures – Paper 04-JG-FAC_090 version 2 2/4

30

35

40

45

50

Page 46: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

values derived from the worst boundary condition Bi ®¥ . The straight line y=x is plotted in

continuous line. The 50th percentile values are plotted with black filled symbols and 95th

percentile values with open symbols.

List of Tables & Figures – Paper 04-JG-FAC_090 version 2 3/4

Page 47: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

List of tables

Table 1. Geometrical, physical and technological parameters used to assess the consumer

exposure to styrene from yogurt pots.

Table 2. 50th and 95th percentile values of individual exposure derived from distributions

calculated for “gathered” households and typical consumption patterns as depicted in figure 4.

Consumption patterns 1, 2 and 3 consist respectively in low (5th percentile), medium (50th

percentile) and high (95th percentile) consumers including 20 households each. Minimum and

maximum values are given for all considered consumption patterns.

Table 3. Likely values of the distributions of 50th and 95th percentiles of exposure calculated at

household scale and depicted in figure 5.

List of Tables & Figures – Paper 04-JG-FAC_090 version 2 4/4

55

60

65

Page 48: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Table 1

quantity meaning value unitsl thickness 10-4 mL dilution factor 1/110 –K partition coefficient 1 –D likely diffusion coefficient 10-16 m2⋅s-1

Ds distribution shape related factor 0.2 –Bi mass Biot number { }1,100,¥ –

0clikely initial concentration in the

packaging material 500 mg⋅kg-1

0cs distribution shape related factor 0.15 –0M mass of yoghurt 0.125 kg

Page 49: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Table 2

conditionH1 H2 H3

50th

percentile95th

percentile50th

percentile95th

percentile50th

percentile95th

percentile

1Bi =

gatheredpattern 1pattern 2pattern 3

0.300.01–0.070.15–0.40.50–1.3

0.720.02–0.090.19–0.510.63–1.6

0.500.03–0.090.32–0.820.88–2.1

1.40.04–0.110.41–1.01.1–2.7

1.10.07–0.170.93–1.23.8–4.0

3.60.10–0.231.2–1.54.8–5.1

100Bi =

gatheredpattern 1pattern 2pattern 3

80.37–1.84.8–10 16–35

210.53–2.4

6–1320–44

120.81–2.38.6–1926–51

351.1–2.911–2432–63

251.8–3.922–2582–85

802.5–5.128–31

102–106

Bi ®¥

gatheredpattern 1pattern 2pattern 3

160.87–3.611–2038–67

431.2–4.614–2547–84

231.7–4.217–3154–88

652.3–5.321–39

67–110

353.3–6.436–39

129–134

1204.4–8.346–48

162–167results are expressed in µg⋅day-1⋅pers-1

Page 50: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Table 3

conditionH1 H2 H3

50th

percentile95th

percentile50th

percentile95th

percentile50th

percentile95th

percentile1Bi = 0.3 0.4 0.7 0.85 1.5 1.9100Bi = 10 12 16 20 28 40

Bi ®¥ 19 25 28 38 41 64results are expressed in µg⋅day-1⋅pers-1

Page 51: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 1

Page 52: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 2

Page 53: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 3

Page 54: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 4

Page 55: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 5

Page 56: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 6

Page 57: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 7

Page 58: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 8

Page 59: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 9

Page 60: Food Additives & Contaminants - AgroParisTechmodmol.agroparistech.fr/pub/PUB_RISK3_Vitrac_Leblanc.pdf · separately the effects of variability and uncertainty on quantitative risk

Figure 10