Benzoates intakes from non-alcoholic beverages in Brazil, Canada, Mexico and the United States

ABSTRACT

Food consumption data from national dietary surveys were combined with brand-specific-use levels reported by beverage manufacturers to calculate the exposure to benzoic acid and its salts (INS Nos 210–213) from non-alcoholic beverages in Brazil, Canada, Mexico and the United States. These four jurisdictions were identified as having some of the most prevalent use of benzoates in beverages globally. Use levels were weighted according to the brand’s market volume share in the respective countries. Benzoates were reported to be used primarily in ‘water-based flavoured drinks’ (Codex General Standard for Food Additives (GSFA) category 14.1.4). As such, the assessments focused only on intakes from these beverage types. Two different models were established to determine exposure: probabilistic (representing non-brand loyal consumers) and distributional (representing brand-loyal consumers). All reported-use levels were incorporated into both models, including those above the Codex interim maximum benzoate use level (250 mg kg⁻¹). The exception to this was in the brand-loyal models for consumers of regular carbonated soft drinks (brand loyal category) which used (1) the interim maximum use level for beverages with a pH ≤ 3.5 and (2) all reported use levels for beverages pH > 3.5 (up to 438 mg kg^-1). The estimated exposure levels using both models were significantly lower than the ADI established for benzoates at the mean level of intake (4–40% ADI) and lower than – or at the ADI only for toddlers/children – at the 95th percentile (23–110% ADI). The results rendered in the models do not indicate a safety concern in these jurisdictions, and as such provide support for maintaining the current Codex interim maximum benzoate level of 250 mg kg⁻¹ in water-based beverages.

KEYWORDS:

• Benzoate
• benzoic acid
• exposure assessment
• non-alcoholic beverages
• water-based flavoured drinks
• preservative
• NHANES
• ENSANUT
• POF
• INA
• ADI

Introduction

Benzoates belong to a class of preservatives comprising of benzoic acid and its salts, i.e., calcium, potassium and sodium (INS Nos 210–213). They are permitted for use in a range of foods and beverages globally (European Union 2008; Secretario de Gobernación 2012; CAC 2016; FSANZ 2016; USFDA 2016; Health Canada 2016; ANVISA 2017). A group ADI of 0–5 mg kg⁻¹ bw day⁻¹, as benzoic acid equivalents, has been established by a number of scientific bodies (JECFA 1983, 1997; SCF 1996, 2002; EFSA 2016).

The JECFA has examined dietary exposure to this group of food additives on two occasions (JECFA 1999, 2015). In both evaluations, mean intakes were below the ADI whereas the intake by presumed heavy-level consumers were noted to exceed the ADI among some groups of consumers. The largest contributor to the total estimated dietary exposure has been identified in several assessments as non-alcoholic beverages (JECFA 1997, 2016; Soubra et al. 2007; Bilau et al. 2008; Leth et al. 2010), with up to 80% of intake coming from this category for the Brazilian population (Tfouni & Toledo 2002; JECFA 2016). At the 48th session of the Codex Committee of Food Additives (CCFA), in light of perceived concern regarding exceedances of the ADI from beverages, an interim maximum level of 250 mg kg⁻¹ as benzoic acid was established for GSFA category 14.1.4, water-based flavoured drinks, including ‘sport’, ‘energy’, or ‘electrolyte’ drinks and particulated drinks (CAC 2016).

Dietary exposure assessments require information on the occurrence/level of the substance in foods and data on the consumption of the corresponding foods. For more complex assessments, additional data can be incorporated, including chemical occurrence (presence or absence) and market share/brand loyalty (WHO, 2009). The type of data and resources available dictate the methodology utilised. Three principal approaches may be considered, namely deterministic, distributional and probabilistic (Kroes et al. 2002; ILSI 2011). Deterministic models apply a fixed, single concentration (substance occurrence) to fixed summary statistics on food or beverage intake (food consumption). Distributional models apply a fixed, single concentration for a given commodity or food category (substance occurrence) to individual-based dietary consumption records accounting for the full range of expected intakes in a given population (food consumption). Probabilistic models consider variability in terms of both input parameters, namely utilising a distribution of substance concentration values for a given food (substance occurrence), alongside individual-based dietary consumption records (food consumption). The inclusion of more detailed data in dietary exposure models yields a more accurate estimate of consumers’ exposure. These methods are more resource intensive and are typically conducted as subsequent ‘tiers’ of assessment when simpler approaches suggest exposure levels of potential concern (WHO 2009).

The current work is part of an exposure assessment package aiming to provide realistic estimates of exposure to benzoates from non-alcoholic beverages considering actual usage patterns, as well as representative beverage consumption habits (i.e., considering consumer behaviours such as brand loyalty and selection of products which may or may not always contain the additive) in four jurisdictions: Brazil, Canada, Mexico and the United States. These were selected based on greater prevalence of benzoate use in beverages. In light of the adopted Codex interim maximum limit in place for benzoates in water-based flavoured drinks, a secondary objective of this study was to determine whether any concerns exist from benzoate exposure levels associated with these conditions of use or whether further reductions in the maximum limits were warranted.

Materials and methods

Food-consumption data

Brazil

Food-consumption data were available for a representative sample of individuals in Brazil aged 10 years and over recorded as part of the National Dietary Survey (Inquérito Nacional de Alimentação – INA). This survey was one component of the Pesquisa de Orcamenos Familiares (POF; Household Budget Survey 2008–09), conducted by the Instituto Brasileiro de Geografia e Estatistica (IBGE) between May 2008 and May 2009. Participants (n = 34,003) completed food diaries on 2 non-consecutive days, recording all food and beverages (except water) consumed inside and outside the home. Information recorded by respondents for all food and beverages consumed included the name, amount, time and place of consumption (Souza Ade et al. 2013; Bezerra et al. 2014; Pereira et al. 2015). Further details on the data collection are available from IGBE (2010a, 2010b, 2010c, 2011). Consumption data were available at the level of the individual respondent per eating occasion for the 2 recording days (n = 34,003). Details on the data management for this consumption information are described under ‘Data analysis’.

Canada

Statistics Canada collates nationally representative food-consumption data as part of the Canadian Community Health Survey – Nutrition (CCHS). The data collected under this programme are used internally by federal and provincial departments of health and human services and other types of government agencies. These data, however, are not released for public use and were not accessible for the current intake assessment. Given the similarity in food products available within the US and Canadian markets and the general comparability between eating habits between these neighbouring countries, the US National Health and Nutrition Examination Survey (NHANES) 2011–12 food-consumption data were used as a surrogate to examine Canadians’ dietary exposure to benzoates.

Mexico

Food-consumption data for the Mexican population, available from the most recent Instituto Nacional de Salud Pública (INSP) Encuesta Nacional de Salud y Nutrición 2012 (ENSANUT), were utilised for the current assessment (INSP 2012). ENSANUT 2012 is the fourth cross-sectional survey conducted by the INSP. This was a nationally representative household survey, with a multistage stratified sample design, representative of urban rural strata and all states; more details are available in Romero-Martinez et al. (2013). The survey incorporated 50,528 households, in which 96,031 individual questionnaires were applied. These individual surveys contained a range of information regarding health and nutritional parameters. Dietary information was obtained for a subsample of the population using two different methods: a 24-h recall and a semi-quantitative food-frequency questionnaire (FFQ). The FFQ was based on a validated questionnaire consisting of a closed list of foods. For each food and beverage item, participants recorded the number of days the item was consumed per week, number of times per day, number of portions and portion size consumed during the 7 days prior to the interview (Mundo-Rosas et al. 2014). These data are publicly available and were used in the current assessment.

FFQ consumption data are available at the level of the individual respondent (n = 7810) for each reported food type, averaged for 7 days. Details on the data management for this consumption information are described below in ‘Data analysis’.

United States

Food-consumption data were obtained from the US National Center for Health Statistics’ (NCHS) NHANES 2011–12 (USDA 2014; CDC 2015). The NHANES is a continuous programme released in 2-year cycles, which is intended to monitor the health and nutritional status of residents in the United States. For NHANES 2011–12, 13,431 individuals were selected for the sample, 9756 were interviewed (72.6%) and 9338 were sampled (69.5%). The dietary component is comprised of two 24-h dietary recalls issued on non-consecutive days. The first dietary recall is collected via in-person interviews by trained individuals; the second recall is collected by telephone 3–10 days after the first recall. Detailed information on each food or beverage item consumed (including the description, amount and nutrient content) is collected by each surveyed individual.

Consumption data were available at the level of the individual respondent per eating occasion, for the 2 recall days (n = 7605). Details on the data management for this consumption information are described below in ‘Data analysis’.

Occurrence data

National legislation governing food additive uses and use levels is specific to each jurisdiction (Secretario de Gobernación 2012; USFDA 2016; Health Canada 2016; ANVISA 2017). The Codex GSFA establishes an international voluntary standard for the conditions of use of benzoates in foods and beverages. For ease of reference, the GSFA food-categorisation system has been used throughout this paper, i.e., category 14.1. ‘non-alcoholic (“soft”) beverages’ and relevant subcategories such as 14.1.4. ‘water-based flavoured drinks’, yet excluding water. Among the non-alcoholic beverage categories, both benzoate use-level data (expressed on a benzoic acid basis) including zeros and information on beverage pH were supplied by industry, complemented by other market data sources (e.g. Beverage Marketing Corporation, Mintel, Canadean etc.) as needed. All information was provided on a brand-level basis for each jurisdiction.

Market share information for various foods and beverages is regularly gathered by the Beverage Marketing Corporation (BMC 2016) and Canadean Ltd (2016). Volume data (in millions of gallons) of the top contributing brands for beverage categories belonging in GSFA category 14.1.2.1 (100% juice) and 14.1.4 (water-based flavoured drinks) were gathered for the four jurisdictions of interest for 2015. Water-based flavoured drinks (14.1.4) were further stratified according to conventional beverage types defined by industry, namely:

• Regular carbonated soft drinks (regCSDs).

• Low- and no-calorie carbonated soft drinks.

• Flavoured water.

• Fruit juice-based drinks.

• Energy drinks.

• Sports drinks.

• Ready-to-drink (RTD) tea.

• RTD coffee.

These market share volume data provided appropriate ‘weight’ to the brand-specific-use levels supplied by industry in all assessment models, except for the brand loyal distributional model. For the brand loyal distributional model, the Codex interim adopted level was used for specifically regCSDs (further explained below) with pH ≤ 3.5, whilst the maximum reported-use level was used for regCSDs with pH > 3.5.

Exposure-assessment methodology

Food selection and categorisation

All non-alcoholic beverages were identified in each of the available food-consumption datasets. Individual food codes were subsequently categorised according to the beverage categories and types, as listed in Table 1. Appropriate dilution and reconstitution factors were applied to powdered or concentrated forms of beverages.

Table 1. Percentage occurrence, weighted average and highest reported-use level of benzoates (as benzoic acid) in four jurisdictions by Codex GSFA category, as utilised in exposure assessment models.

	Brazil			Canada			Mexico			United States
		Benzoate use level (mg kg^−1)			Benzoate use level (mg kg^−1)			Benzoate use level (mg kg^−1)			Benzoate use level (mg kg^−1)
GSFA food category (and subcategories^a)	% occ^b	Ave.^c	Max.	% occ^b	Ave.^c	Max.	% occ^b	Ave.^c	Max.	% occ^b	Ave.^c	Max.
14.1.2.1 Fruit juice	0.0	0	0	0.0	0	0	0.0	0	0	0.0	0	0
14.1.2.3 Concentrates for fruit juice^d	0.0	0	0	0.0	0	0	0.0	0	0	0.0	0	0
14.1.3.1 Fruit nectar	0.0	0	0	0.0	0	0	0.0	0	0	0.0	0	0
14.1.3.3 Concentrates for vegetable nectars^e	0.0	0	0	0.0	0	0	0.0	0	0	0.0	0	0
14.1.4 Water-based flavoured drinks
Low- and no-calorie CSDs	98.8	186	297	97.2	174	690^f	97.2	169	299	100.0	197	400^f
Regular CSDs	32.3	104	297	38.3	79	438 ^f	18.2	37	299	50.0	115	428^f
Flavoured water drinks	0.0^g	0	0	0.0	0	0	84.0 ^g	172	205	2.3	4	197
Fruit juice-based drinks, including concentrates	24.3	76	339	0.0	0	0	66.5	89	247	0.0	0	0
Energy drinks	61.3	144	366	64.3	109	429	68.0	102	300	65.1	111	431
Sports drinks	24.1	39	263	0.0	0	0	0.0	0	0	0.0	0	0
RTD teas	45.5^g	66	271	61.0	56	91	37.3	74	197	44.3	67	196^f
RTD coffees	0.0	0	0	0.0	0	0	0.0	0	0	0.0	0	0
14.1.5 Coffee, coffee substitutes, tea, herbal infusions, and other hot cereal and grain beverages	0.0	0	0	0.0	0	0	0.0	0	0	0.0	0	0

Notes: ^aSubcategories were established based on conventional product categories defined by industry.

^bPercentage occurrence of benzoates in beverage category (or subcategory or type) were calculated based on use-level data received by the authors from beverage manufacturers. These values have been weighted according to the market volume share of the available brands in the specific jurisdiction.

^cMarket volume-weighted average use level of benzoates in beverage type.

^dValues obtained for ‘14.1.2.1 Fruit juice’ were utilised based on industry recommendations.

^eData for fruit nectars applied to this category.

^fUse level reported for product with pH > 3.5.

^gData received by industry were limited and accounted for less than 20% of the market by volume for this beverage type in this jurisdiction.

GSFA, general standard for food additives; % occ, percentage occurrence; Ave., average; Max., maximum; CSD, carbonated soft drinks; RTD, ready to drink.

On the basis that benzoates were reported to be used only in category 14.1.4 subcategories across all four jurisdictions (Table 1), all food codes which were not included in this category (i.e., GSFA categories 14.1.2.1, 14.1.2.3, 14.1.3.1, 14.1.3.3 and 14.1.5) were removed from the analyses described herein. However, as part of previous work conducted, a series of models had been run considering all beverage types in 14.1 (except water) with results presented in Tables S1 and S2 in the supplemental data online.

Exposure-assessment models

Two types of models were conducted for all four jurisdictions: probabilistic and brand loyal distributional models.

• Model 1: Probabilistic: food-consumption data retrieved from the four dietary surveys were available at the food code level, with only a small proportion available at the individual brand level. To consider all available concentration data (including non-occurrences), a distribution of the available use levels was established based on the market-share volume basis of the individual brands. This market volume-weighted distribution was applied to each relevant food code. In practical terms, the chance that each reported-use level was encountered by a consumer was directly related to the market-volume share of the corresponding brand within the relevant beverage category.

• Model 2: Brand loyal distributional: the ‘brand-loyal’ model assumes consumers are habitually consuming the top-contributing beverage type that contains benzoates at the highest use level. These consumers also ingest other beverage types potentially containing benzoates at market-weighted average-use levels. Relative to occurrence data in this model, this means applying the maximum use level (and assuming a 100% presence probability) to the ‘brand loyal’ beverage type (i.e., regCSD), and applying the market-weighted average reported-use level to all other beverage types. Table 1 presents the percentage occurrence of benzoates in each beverage type, as well as the market volume-weighted average-use level (weighted according to the market share by volume for each brand) and the highest reported-use level for each category.

To determine the top contributing ‘brand-loyal’ beverage type, a preliminary assessment for each jurisdiction in which a default benzoate-use level was applied to all beverage types was conducted (data not shown). In all jurisdictions, ‘regular carbonated soft drinks (regCSDs)’ was the top contributing source of benzoates in the simulation and selected to be the ‘brand-loyal’ beverage type.

In the brand loyal model, the Codex interim maximum use level of 250 mg kg⁻¹ (as benzoic acid) was applied to regCSDs broadly. RegCSDs with a pH > 3.5, however, had to account for the higher use levels typically required for these types of products exceeding the Codex interim maximum limit (e.g., up to 438 mg kg ⁻¹ in Canada; see Table 1 for details). These higher levels were applied to food codes representing these types of beverages (i.e., root beer and cream sodas). For all other (non-brand loyal) beverage types, the market-weighted average of all reported values was applied.

Data analysis

Statistical analyses and data management were conducted with DaDiet software (Dazult Ltd 2016). As mentioned above, consumption data were available from the level of the individual respondent for all datasets. For the Brazilian INA and US NHANES datasets, these data were available at the level of the individual eating occasion for the 2 recording days; for the Mexican dataset, information was available for average daily intake (calculated over 7 days).

In the probabilistic models, the exposure assessment was conducted iteratively using a Monte Carlo simulation, the number of iterations rendering a minimum target of 100,000 person-days was undertaken for each survey, as recommended for probabilistic assessments (Boon & Van Klaveren 2003; Boon et al. 2004; EFSA 2012). This approach allowed for the combination of data reflecting the full range of use levels obtained (weighted according to the market volume share of the brand), as well as the full range of reported daily consumption values (individual days for the Brazilian INA and US NHANES datasets, and an average daily intake for the Mexican ENSANUT dataset). A full distribution of intakes was calculated from the simulation (68,006 persons for Brazil, 101,530 persons for Mexico and 52,325 persons for Canada/United States) from which the mean and high percentile estimates were determined.

In the brand loyal models, estimates for the daily intake of benzoates represent projected 2- (for the Brazil and the United States/Canada) or 7- (for Mexico) day averages for each individual from each recording. The average (market-weighted), maximum or Codex interim use levels were multiplied by the average daily consumption amounts for each beverage type. The mean and 95th percentile intake estimates were determined from this distribution.

The percentage of consumers for all beverage categories were determined for all subpopulation groups in each jurisdiction (results are presented for ‘water-based flavoured drinks’ in Table 2). These estimates represent the proportion of the population that consumed one or more of the specified beverage type over the multi-day dietary survey period.

Table 2. Percentage consumers^a of individual water-based flavoured drink (GSFA category 14.1.4) subcategories by population group in four jurisdictions.

	n	Percentage consumers (%)^b
		Low- and no-calorie CSDs	Regular CSDs	Flavoured water drinks	Fruit juice-based drinks^c	Energy drinks	Sports drinks	RTD teas	RTD coffees	Any water-based flavoured drink
Brazil, POF 2008–09
Other children and adolescents (10–17 years^d)	6243	1.6	32.2	14.7	34.6	0.1	0	n.a.	n.a.	61.1
Adults (18+ years)	27,760	2.5	27.1	10.3	33.7	0.1	0.1	n.a.	n.a.	56.7
Total population (10+ years^d)	34,003	2.4	28.0	11.0	33.8	0.1	0.1	n.a.	n.a.	57.5
Mexico, ENSANUT 2012
Toddlers and young children (1–7 years)	1793	0.7	71.8	23.0	38.3	n.a.	n.a.	n.a.	n.a.	85.4
Other children and adolescents (8–17 years)	2643	0.7	76.6	25.9	34	n.a.	n.a.	n.a.	n.a.	87.9
Adults (18+ years)	3374	3.0	72.5	13.1	33.7	n.a.	n.a.	n.a.	n.a.	84.1
Total population (all ages, including infants)	7810	2.2	73.4	17.4	34.4	n.a.	n.a.	n.a.	n.a.	85.2
United States (and Canada^e), NHANES 2011–12
Toddlers and young children (1–7 years)	1244	3.8	33.7	22.4	37.1	0	7.6	0.1	n.a.	69.0
Other children and adolescents (8–17 years)	1475	8.9	55.9	19.6	32	0.7	15.2	1	n.a.	83.8
Adults (18+ years)	4555	21.2	39.6	10.5	13.8	0.9	7.8	0.2	n.a.	68.5
Total population (all ages, including infants)	7605	17.7	40.7	12.7	18.3	0.8	8.7	0.3	n.a.	69.8

Notes: ^aConsumers were identified individuals reporting consumption of the specified beverage type on any recording day (2 days for Brazil and United States/Canada; 7 days for Mexico).

^bWeighted values.

^cIncluding concentrates.

^dConsumption data only available for individuals over 10 years.

^eUS NHANES data were used as a surrogate for Canadian individuals.

GSFA, general standard for food additives; n, number of subjects (unweighted); CSD, carbonated soft drink; RTD, ready to drink; n.a., no food code available.

The age groups selected reflect those presented in the 2015 JECFA exposure assessment, namely: toddlers and young children, 1–7 years; other children, including adolescents, ages 8–17 years; adults aged 18 years and over; and the total population (all age and gender groups combined, including infants under 1 year where available).

For both models within each jurisdiction, the mean and 95th percentile benzoate intakes were calculated for all water-based flavoured drinks and for the individual beverage types. Although the 90th percentile is the typical reported statistic for high level consumers in the United States and Canada and as acknowledged by the WHO (WHO 2009), the 95th percentile estimates are presented here for all evaluations to ensure consistency with the 2015 JECFA assessment (JECFA 2015, 2016). For each subpopulation age group, results are presented on a body weight basis (mg kg⁻¹ body weight day⁻¹) and as a proportion of the benzoate ADI (% ADI) (Tables 3 and 4).

Table 3. Estimated daily exposure to benzoates from non-alcoholic beverages by subpopulation group for the general population and benzoate consumers only – probabilistic model.

Population group	n	General population exposure						Benzoate consumer-only exposure^a
		mg kg^−1 bw day^−1			%ADI		% cons	mg kg^−1 bw day^−1			%ADI
		Mean	SD	P95	Mean	P95		Mean	SD	P95	Mean	P95
Brazil
Other children and adolescents (10–17 years^b)	6243	0.353	0.703	1.796	7	36	33.0	1.069	0.855	2.582	21	52
Adults (18+ years)	27,760	0.212	0.453	1.125	4	23	29.7	0.715	0.576	1.776	14	36
Total population (10+ years^b)	34,003	0.237	0.508	1.249	5	25	30.3	0.782	0.653	2.022	16	40
Canada
Toddlers and young children (1–7 years)	1244	0.244	0.694	1.554	5	31	21.2	1.154	1.107	3.497	23	70
Other children and adolescents (8–17 years)	1475	0.305	0.605	1.517	6	30	37.6	0.811	0.750	2.246	16	45
Adults (18+ years)	4555	0.415	0.830	1.869	8	37	43.5	0.955	1.033	2.664	19	53
Total population (all ages, including infants)	7605	0.379	0.788	1.789	8	36	40.0	0.947	1.008	2.652	19	53
Mexico
Toddlers and young children (1–7 years)	1793	0.761	1.357	3.540	15	71	50.1	1.520	1.588	4.741	30	95
Other children and adolescents (8–17 years)	2643	0.608	1.094	2.890	12	58	49.3	1.232	1.287	3.817	25	76
Adults (18+ years)	3374	0.396	0.819	2.037	8	41	43.4	0.911	1.037	3.027	18	61
Total population (all ages, including infants)	7810	0.493	0.983	2.469	10	49	45.7	1.079	1.217	3.554	22	71
United States
Toddlers and young children (1–7 years)	1244	0.344	0.886	2.027	7	41	24.7	1.392	1.311	3.936	28	79
Other children and adolescents (8–17 years)	1475	0.449	0.789	2.019	9	40	43.5	1.032	0.909	2.789	21	56
Adults (18+ years)	4555	0.544	0.987	2.327	11	47	48.1	1.131	1.167	3.127	23	63
Total population (all ages, including infants)	7605	0.506	0.951	2.245	10	45	44.7	1.131	1.146	3.136	23	63

Notes: ^a‘Benzoate consumers’ were identified as individuals who consumed any water-based flavoured drink (GSFA category 14.1.4) during the recording period (2 days for Brazil and United States/Canada; 7 days for Mexico) which were simulated to contain benzoates during the iterations.

^bConsumption data only available for individuals 10 years and over.

n, Number of subjects (unweighted); %cons, percentage of consumers (weighted) within specified age cohort; ADI, acceptable daily intake; bw, body weight; P95, 95th percentile; SD, standard deviation.

Table 4. Estimated daily exposure to benzoates from non-alcoholic beverages by subpopulation group for the general population and benzoate consumers only – brand loyal model.

Population group	n	General population exposure						Benzoate consumer only exposure^a
		mg kg^−1 bw day^−1			%ADI		%cons	mg kg^−1 bw day^−1			%ADI
		Mean	SD	P95	Mean	P95		Mean	SD	P95	Mean	P95
Brazil
Other children and adolescents (10–17 years^b)	6243	0.628	0.985	2.716	13	54	55.8	1.127	1.086	3.199	23	64
Adults (18+ years)	27,760	0.368	0.620	1.605	7	32	52.5	0.700	0.707	2.109	14	42
Total population (10+ years^b)	34,003	0.413	0.705	1.789	8	36	53.1	0.778	0.807	2.378	16	48
Canada
Toddlers and young children (1–7 years)	1244	0.732	1.470	4.040	15	81	36.3	2.020	1.830	5.180	40	104
Other children and adolescents (8–17 years)	1475	0.940	1.300	3.610	19	72	59.6	1.580	1.350	4.300	32	86
Adults (18+ years)	4555	0.850	1.300	3.100	17	62	57.3	1.480	1.420	4.180	30	84
Total population (all ages, including infants)	7605	0.839	1.310	3.220	17	64	54.9	1.530	1.440	4.270	31	85
Mexico
Toddlers and young children (1–7 years)	1793	1.570	1.780	5.080	31	102	85.4	1.840	1.800	5.490	37	110
Other children and adolescents (8–17 years)	2643	1.470	1.570	4.800	29	96	87.9	1.670	1.570	4.890	33	98
Adults (18+ years)	3374	1.080	1.300	3.800	22	76	84.1	1.280	1.330	4.120	26	82
Total population (all ages, including infants)	7810	1.240	1.460	4.310	25	86	85.2	1.450	1.470	4.610	29	92
United States
Toddlers and young children (1–7 years)	1244	0.743	1.470	4.040	15	81	48.7	1.530	1.800	4.910	31	98
Other children and adolescents (8–17 years)	1475	0.949	1.300	3.530	19	71	69.1	1.370	1.360	4.090	27	82
Adults (18+ years)	4555	0.880	1.330	3.190	18	64	61.8	1.420	1.450	4.020	28	80
Total population (all ages, including infants)	7605	0.865	1.340	3.280	17	66	60.7	1.420	1.460	4.150	28	83

Notes: ^a‘Benzoate consumers’ were identified as individuals who consumed any water-based flavoured drink (GSFA category 14.1.4) during the recording period (2 days for Brazil and United States/Canada; 7 days for Mexico) which always contained benzoates at market volume-weighted average for each respective beverage type and at Codex interim level for RegCSD (and highest reported-use levels for RegCSD with pH > 3.5).

^bConsumption data only available for individuals 10 years and over.

n, Number of subjects (unweighted); %cons, percentage of consumers (weighted) within specified age cohort; ADI, acceptable daily intake; bw, body weight; P95, 95th percentile; SD, standard deviation.

Estimated daily intakes are presented for the general population and consumers only. ‘General population’ exposure refers to the estimated intake of benzoates averaged over all individuals surveyed regardless of whether they consumed ‘water-based flavoured beverages’. In order to examine further exposure by only individuals who were modelled to consume benzoates in the probabilistic and brand-loyal scenarios, ‘benzoate consumers only’ were characterised as individuals who reported consumption of a beverage category in which it was assumed that benzoates were present. For the probabilistic model, this included any individual who was simulated to consume beverages containing benzoates during the iterations (identified as 21.2–50.1% of the individual subpopulation groups; Table 3) thus removing beverage types within 14.1.4 that did not report benzoate use (which differed between markets). For the brand loyal model, this included consumers of all groups in which benzoates were reported to be used in the jurisdiction (Table 1), i.e., 100% occurrence was assumed for all relevant beverage categories (identified as 36.3–87.9% of the individual subpopulation groups; Table 4).

Exposure by ‘beverage consumers only’ was also investigated to allow for a consistent basis to compare with the 2015 JECFA global assessment in which the GSFA category 14.1.4 was identified as the ‘consuming’ category, as presented in Tables S1 and S2 in the supplemental data online. Among these models, intake refers to the estimated intake of benzoates by only those individuals who reported consuming category 14.1.4 beverages potentially containing benzoates over the course of the multi-day dietary survey (who represented between 56.7% and 89.5% within each individual subpopulation group among the four jurisdictions investigated; see Tables S1 and S2 in the supplemental data online).

Results

Use of benzoates in non-alcoholic beverages

Benzoate use was reported only within GSFA category 14.1.4. ‘water-based flavoured drinks’ among all jurisdictions examined, with more prevalent use noted among low- and no-calorie carbonated soft drinks (97.2–100.0%) followed by energy drinks (61.3–68.0%), RTD teas (37.3–61.0%) and regCSD (18.2–50.0%) (Table 1). There was no reported use of benzoates in RTD coffees, 100% fruit juices – and by extension fruit nectars – coffee drinks or bottled water. Benzoates were reportedly used in a large proportion of flavoured water drinks in Mexico (84.0%) with a lower proportion in the United States (2.3%) and no reported use in Brazil or Canada. There was 24.1% occurrence in sports drinks in Brazil, with no use in any other jurisdictions.

Consumption of water-based flavoured drinks

Table 2 presents the percentage of consumers of water-based flavoured drinks (GSFA category 14.1.4) for the total population and selected age groups in each jurisdiction. An individual was identified as a ‘consumer’ if they reported at least one consumption occasion of a ‘water-based flavoured drink’ over the multi-day dietary survey period – 2 days for Brazil, Canada and the United States and 7 days for Mexico.

Brazil

The Brazilian general population (i.e., 10 years +) was identified as being significant consumers of fruit-juice based drinks (including concentrates) (33.7–34.6%), and regCSDs (27.1–32.2%). Flavoured water drink consumption ranged from 10.3% to 14.7%. There was a notably lower proportion of consumers of low- and no-calorie carbonated soft drinks (1.6–2.5%) and energy or sports drinks (maximum 0.1%) (Table 2). Overall, between 56.7% and 61.1% of individuals reported consuming any beverage which fit within the category of 14.1.4 during the 2-day recording period.

Mexico

The Mexican population was identified as being significant consumers of regCSDs (73.4%). The proportion of regCSD consumers among ‘toddlers and children’, ‘other children and adolescents’ and ‘adults’ were 71.8%, 76.6% and 72.5% respectively. The higher estimate in Mexico for regCSDs – compared with the other three jurisdictions evaluated may be explained by the availability of consumption data over 7 days, whereby a higher proportion of individuals are captured as having at least one consumption event in a 7-day survey period, in contrast to a 2-day in Brazil, Canada and the United States. There was also a notable proportion of consumers of fruit juice-based drinks (34.4%) and flavoured water drinks (17.4%) among the entire population (all ages), again likely due to the nature of the 7-day survey. Additionally, as noted in the methodology, a ‘closed list FFQ’ was used to record dietary intake; e.g., there was only one food code representing regCSDs broadly suggesting aggregation of individuals reporting to be consumers of ‘regular sodas’. Interestingly, there was a considerably lower proportion of individuals reporting consumption of low- and no-calorie carbonated soft drinks (range = 0.7–3.0%). When considering all 14.1.4 beverage types, the proportion of consumers ranged between 84.1% and 87.9% of consumers.

United States and Canada

A notable proportion of the U.S. population (all ages) reported consumption of regCSD (40.7%), fruit juice-based drinks (18.3%), low- and no-calorie carbonated soft drinks (17.7%) and flavoured water drinks (12.7%) over the 2 days for which there were consumption data available. ‘Other children including adolescents’ appeared to have the highest percent regCSD consumers (55.9%) followed by adults (39.6%) and ‘toddlers and young children’ at the lowest end (33.7%). Between 68.5% and 83.8% of individuals reported consumption of beverages which corresponded with category 14.1.4 in the US NHANES.

The US NHANES data were used as a surrogate for Canadian consumption, thus the same pattern is observed for the Canadian population groups.

Dietary exposures to benzoates

Probabilistic model

Given the different survey methodologies utilised in each of the jurisdictions, findings among jurisdictions were not directly comparable; however, some overall trends are noted in each of the datasets. The mean exposure levels for all four regions examined were notably lower than the ADI when examining intakes by non-brand loyal consumers considering both results for the general population (4–15%) and the ‘benzoate consuming’ population (14–30%) (Table 3). At the 95th percentile of consumption, intakes ranged from 23% to 71% of the ADI among the general population, and from 36% to 95% of the ADI when focusing on the benzoate consuming population only. For the ‘beverage consumer’ estimates, see Table S1 in the supplemental data online. The youngest age group had the highest intakes in each of the countries examined when focusing on the consuming population, as expected for exposure levels expressed on a body weight basis.

There was a slight decrease in the percentage of benzoate consumers with increasing age in Brazil (from 33.0% to 29.7%) and Mexico (from 50.1% to 43.4%) (Table 3). In contrast, in Canada and the United States, there was an increase in the percentage of benzoate consumers with increasing age, with 21.2% and 24.7% consumers identified for toddlers and young children respectively, 37.6% and 43.5% consumers among other children and adolescents respectively, and 43.5% and 48.1% among adults respectively.

Overall, all subpopulations were identified to be below the ADI in each jurisdiction investigated when concentrations of benzoates in beverages were incorporated using presence probabilities reflecting market share, even when focusing strictly on those individuals who were simulated to consume benzoates (i.e., ‘benzoate consumers only’) during the assessment in each of the four jurisdictions.

Brand loyal model

The estimated dietary exposures to benzoates among assumed regCSD brand-loyal consumers were understandably higher than the non-brand loyal probabilistic modelling scenario. In the regCSD brand-loyal consumer scenario, the Codex adopted interim maximum level of 250 mg kg⁻¹ was applied to regCSDs broadly (as this level would not be exceeded if adopted at the national level) and the highest reported-use level (i.e., up to 438 mg kg⁻¹) was applied to only those regCSD with pH > 3.5 (Table 4).

When focusing on mean intakes, the general population exposure level ranges from 8% to 31% of the ADI. When focusing on benzoate consumers, the mean levels of intake ranged from 14% to 40% of the ADI. Among high percentile consumers, the estimated exposure levels range from 32% to 102% of the ADI among the general population, and between 42% and 110% of the ADI among the benzoate consuming population. For the ‘beverage consumer’ estimates, see Table S2 in the supplemental data online. Thus, cumulative intake of both regCSDs containing benzoates at the Codex adopted interim maximum use level and other beverage types (e.g., low- and no-calorie CSDs, RTD teas, sports drinks etc.) containing benzoates at levels equivalent to the market-weighted mean resulted in 95th percentile intake estimates at the ADI for Canadian and Mexican ‘brand-loyal’ toddlers and young children ‘benzoate consumers’ (104% and 110% of the ADI respectively). All other ‘brand-loyal’ Canadian and Mexican subpopulations were below the ADI. The results for all subpopulations examined in Brazil and the United States were below the ADI.

Discussion

Dietary exposure to benzoates has previously been examined by several scientific bodies, as well as independent research groups globally. These assessments have ranged in their methodology, with some utilising a tiered approach to assessing intake, starting with methods based on the maximum permitted levels for benzoates followed by refinements of these estimates considering analytical data and reported-use levels (Verger et al. 1998; Bemrah et al. 2008; Leth et al. 2010; Vin et al. 2013). The current enquiry into dietary intake estimates for benzoates aligns with the more refined approaches described elsewhere but is strengthened with the application of a quantitative weighting scheme based on coupling brand-specific market volume share to reported brand-specific-use levels. This quantitative weighting approach more accurately represents realistic consumer practices relative to benzoate exposure through beverages.

When evaluating published dietary intake assessments examining benzoate exposure, investigators have shown that the ADI is seldom surpassed at the mean (JECFA 1997, 2016; Tfouni & Toledo 2002; FSANZ 2005; Bilau et al. 2008; Mischek & Krapfenbauer-Cermak 2012; EFSA 2016). In some instances, the upper percentiles of intake may exceed the ADI, primarily among ‘toddlers and young children’ and when examining intakes by brand-loyal consumers (FSANZ 2005; Bilau et al. 2008; Leth et al. 2010; Mischek & Krapfenbauer-Cermak 2012; EFSA 2016; JECFA 2016). In contrast, others have not observed an exceedance of the ADI even among the highest percentile consumers (Verger et al. 1998; Tfouni & Toledo 2002; Soubra et al. 2007; Bemrah et al. 2008; Cressey & Jones 2009; Ma et al. 2009; Vandevijvere et al. 2010; Vin et al. 2013).

In the current work, consumer dietary exposures to benzoates from beverages were examined in four jurisdictions, selected based on relatively prominent use of benzoates in beverages.

The probabilistic modelling conducted may be considered the most representative of intakes by non-brand loyal consumers. The variability associated with benzoate use among beverage brands, and subsequently, variability of benzoate exposure when consumed as part of the diet were taken into account. Taken together, this model type allows for the representation of a wide spectrum of benzoate use levels including beverage brands reporting use levels above the adopted Codex interim maximum use level (e.g., use levels as high as 690 mg kg⁻¹ in some instances; Table 1), with the probability of such beverages being consumed on any single occasion reflective of the market volume share of these brands. The brand-loyal model, on the other hand, considers those specific consumers who may be habitually consuming beverages with higher benzoates due to preference and pre-established loyalty toward a specific product line.

In almost all instances, the mean and 95th percentile estimates were below the ADI for benzoates. The highest exposure was observed for the ‘brand-loyal’ toddlers and young children subpopulation benzoate consumers at the 95th percentile. These individuals were assumed: (1) to be habitual consumers of regCSD which were always simulated (i.e., 100% occurrence) to contain a maximum benzoate concentration at the Codex adopted interim maximum limit (250 mg kg⁻¹) for beverages with pH ≤ 3.5 and up to 438 mg kg⁻¹ for beverages with a pH >3.5; and (2) to be consuming other beverage types for which a full range of benzoate use levels were applied (including those above 250 mg kg⁻¹; also assuming 100% occurrence). When the same subpopulation was investigated, without assuming brand loyalty, the calculated values were well below the ADI in all jurisdictions.

While the simulation applied the Codex adopted interim maximum use level to the brand-loyal category (in this case regCSD), brand loyal consumers may just as likely be consuming beverages with lower benzoate levels or that are benzoate-free.

Previous expert reviews have concluded that:

Because in most cases, data are extrapolated from life-time animal studies, the ADI relates to life-time use and provides a margin of safety large enough for toxicologists not to be particularly concerned about short-term use at exposure levels exceeding the ADI, providing the average intake over longer periods does not exceed it. (WHO, 1987)

Other risk assessment principles reiterate that mean dietary exposures may be used in assessing estimated daily intakes against an established health-based guidance value as stated in the WHO’s Principles and Methods for the Risk Assessment of Chemicals in Food (WHO 2009, ch. 6):

Typically, a mean dietary exposure will be compared with a chronic (long-term) health-based guidance value (e.g. ADI, PTWI). The mean dietary exposure may be calculated by applying a deterministic model using average food consumption levels and the average concentrations in the relevant food products.

Thus, exceedance of the ADI in this one intake scenario for brand-loyal ‘toddlers and young children’ 95th percentile ‘benzoate consumers only’, whereby the application of the brand loyal concentration was not based on reported-use levels but rather on the Codex adopted interim maximum limit, is not a safety concern. No exceedances are observed at the mean in any population group examined and the brand-loyal older age cohorts ‘benzoate consumers only’ at the 95th percentile do not exceed the life-time ADI, i.e., chronic exceedance of the ADI does not occur.

To identify and qualify all potential sources of over- or underestimation in these models, an uncertainty analysis was conducted in accordance with the criteria outlined in Codex and EFSA (CAC 2003; EFSA 2006). The sources and influence of uncertainties in the probabilistic and brand-loyal models employed herein are summarised in Table 5, these parameters are organised according to assessment inputs (consumption and use-level data) and the model types. Each of these points are described in the sections below.

Table 5. Qualitative evaluation of the influence of uncertainties on the estimates of exposure to benzoates from beverages calculated for four jurisdictions^a

Source of uncertainty	Direction and magnitude^b
Consumption data
Mis-/under-reporting of beverage consumption in dietary surveys	–
Use of data from short-term surveys to estimate chronic exposures	+
Miscategorisation of beverage types (in particular for ENSANUT survey)	±
Beverage consumption data not available at brand level	±
Rounding errors in food-consumption data	±
Use-level data
Use-level data and beverage consumption data were collected at two separate time points	±
Use levels only provided by members of the International Beverages Council Association (representing 73–83% of the market), no account for use levels by other industry members	±
Reported-use-level data were employed as opposed to analytically measured values (accounts for current formulation)	±
Lack of benzoate usage data for smaller brands not identified to hold a significant market share	±
Rounding errors, incorrect entries for concentration data	±
Inclusion of beverages containing no benzoates in the exposure models	–^c
Probabilistic model
No consideration for brand loyal consumers	±
Probability of benzoate use levels being encountered for each consumer was based on market share (by volume) and not based on actual consumer behaviour, as food codes are not branded	±
Brand loyal model
Interim ML (or maximum reported level for pH > 3.5, where possible) applied to the highest contributing category	+
100% presence assumed for all beverage categories in which benzoates were reported for use- irrespective of whether these reported values accounted for a large proportion of the market	+

Notes: ^aConducted in accordance with the criteria specified in Codex and EFSA ([CAC] Codex Alimentarius Commission 2003; [EFSA] European Food Safety Authority 2006).

^b+, ++ and +++ are the uncertainty likely to cause small, medium or large overestimates of exposure respectively, and –, – – and – – – are the uncertainty likely to cause small, medium or large underestimates of exposure respectively.

^cIn ‘beverage consumer’ estimates only (Tables S1 and S2 in the supplemental data online).

Uncertainties associated with the consumption data include those which are inherent to the dietary survey design for each jurisdiction’s dataset, which may result in both over- and underestimations of intake. One source of potential underestimation specific to the current assessment is the mis- or under-reporting of beverages (particularly sugar-sweetened beverages) in dietary studies (Briefel et al. 1997; Han & Powell 2013). In contrast, the use of a closed list FFQ in the Mexican dataset resulted in a relatively low number of food codes available to represent water-based flavoured drinks (n = 5 versus 58 for Brazil and 107 in the United States), which can result in an overestimate in the number of respondents for a particular food or beverage due to the lower number of food codes available to capture the consumption. For example, a single code in Mexico’s closed list FFQ represented consumption of all ‘regular sodas’, which may have resulted in mis- or over-reporting of ‘regular sodas’ consumption. The maximum use level was subsequently applied to this single category in the brand loyal model and as such, results for the brand-loyal model may overestimate actual exposure in this jurisdiction. Overall, the utilisation of individual-based data from large-scale nationally representative food surveys in each jurisdiction (as opposed to summary statistics) is a key aspect in deriving a rigorous and robust exposure analysis, as the entire distribution of intakes by even a subpopulation group can be considered in the model. Lastly, it is noted that the consumption data available were gathered using a variety of methods, namely 24-h recalls or food diaries on 2 non-consecutive days (Brazil and United States) or a 7-day FFQ (Mexico). Data collected over a period of greater than 7 days is typically considered most representative of ‘usual intake’. The use of longer recalls may assist in reducing the variance associated with the reported food consumption and result in an adjustment of the values calculated especially for the higher percentiles (Vin et al. 2013).

Relative to the use level inputs, benzoate concentrations in non-alcoholic beverages were collected in 2016 whereas branded market volume data were collected for 2015 and food-consumption data were retrieved from the most recently available and accessible survey release at the time of analysis (in the case of Brazil, this dates as far back as 2008–09). Further, it is noted that use levels for benzoates were primarily collected from beverage manufacturers having International Council of Beverages Associations (ICBA) membership at the time of analysis. If some data were not available, the use levels obtained were reweighted for the ‘missing’ market share branded beverage product. The latter may have resulted in an over- or underestimation, depending on whether the data submitted were higher or lower than those missing values. Importantly, however, since use levels submitted accounted for, on average, 77% of the market share for each beverage type (73% Brazil, 83% Canada, 75% Mexico and 77% United States), the brand-specific data received from industry are considered highly representative of those products consumed in the relevant jurisdictions. In JECFA’s 2015 assessment (JECFA 2016), while the average industry-reported-use levels were utilised, a limitation was that these levels were not weighted according to their market share in each country. Use levels for brands that were consumed infrequently may have contributed equally to the final exposure estimates as products with a higher market penetration. The inclusion of up-to-date and brand-specific-use-level data as provided by the global beverage industry within the four markets (data submitted by ICBA members) – which also accommodate reformulation initiatives (unlike outdated analytical data) – coupled with current market share volume data was a fundamental strength of this current analysis. Having a detailed account of actual benzoate use levels across beverage types and categories improves the overall accuracy and robustness of intake estimates generally. In terms of the specific models in this assessment, while the probabilistic model does not consider exposure by brand-loyal consumers, which is noted as a potential source of underestimation, consumers may equally be brand-loyal to products which contain low levels of benzoates or are benzoate-free. In the brand-loyal model, a use level of 250 mg kg⁻¹ (or up to 438 mg kg⁻¹) was applied to all regCSDs (depending on pH) irrespective of the reported-use-level data. Notably, this level was higher than the market volume-weighted average use level for this beverage type in any of the jurisdictions examined (37–115 mg kg⁻¹; Table 1). Furthermore, this model assumed 100% presence of benzoates in regCSDs, the actual proportion of occurrence ranged from 18.2% to 50.0% (average 34.7%) in these regions. Overall, the models were designed to provide realistic estimates of beverage consumers’ exposures to benzoates whilst maintaining suitably conservative assumptions to ensure total exposure was not underestimated.

In summary, the results presented herein represent a robust and realistic analysis of the anticipated exposures to benzoates associated with beverage consumption in four sentinel jurisdictions, selected based on relatively prominent use of benzoates in beverages. The results rendered in the models described herein do not indicate a safety concern in these jurisdictions, and as such provide support for maintaining the current Codex adopted interim maximum benzoate level of 250 mg kg⁻¹ in water-based beverages with the inclusion of a qualifier for beverages with pH > 3.5 of up to 438 mg kg⁻¹.

Acknowledgements

Branded-use level information on benzoates was provided by International Council of Beverages Associations (ICBA) members including the American Beverage Association (ABA), the Associacao Brasileira das industrias de Refrigerantes e de Bebidos nao Alcoolicas (ABIR), the Asociación Nacional de Productores de Refrescos y Aguas Carbonatadas, A.C. (ANPRAC) and the Canadian Beverage Association (CBA). Important contributions in study design and manuscript review were provided by Dr Maia Jack at the ABA.  This research was funded by the ABA.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed here.

Additional information

Funding

This research was supported by the American Beverage Association.