EFSA 2014 — Dietary exposure to inorganic arsenic in the European population

This 68-page EFSA Scientific Report (Question No EFSA-Q-2012-00107, approved 28 February 2014) provides the first dedicated EU-wide chronic dietary exposure assessment for inorganic arsenic (iAs) using the EFSA Comprehensive European Food Consumption Database and the FoodEx classification system. A dataset of 107,646 analytical results collected in 21 European countries (2003–2012) was used; 103,773 food samples (including drinking water) supplied the exposure calculation, of which 101,020 were originally reported as total arsenic (tAs) and only 2,753 as iAs. Because direct iAs measurements were sparse, 92.5 % of the tAs data were converted to iAs using a 70 % conversion factor for foods of terrestrial origin (1 % for fish and seafood), except for five rice types and fish/seafood where the reported iAs was used exclusively and tAs results were discarded. Chronic exposure was estimated for 28 dietary surveys from 17 European countries (53,728 individuals). The highest exposures occurred in infants and toddlers; mean dietary exposure across age classes ranged from 0.09 to 1.37 µg iAs/kg b.w. per day (min LB – max UB) and 95th-percentile exposures from 0.14 to 2.09 µg/kg b.w. per day (min LB – max UB). Across all age classes except infants and toddlers, the dominant contributor to iAs exposure was the food group ‘Grain-based processed products (non rice-based)’, driven by wheat bread and rolls; in infants and toddlers, ‘Milk and dairy products’ was the main contributor, followed by drinking water, grain-based processed products (non rice-based) and food for infants and young children. Consumption of three 30-g portions per day (90 g/day) of rice-based infant food was estimated to deliver 1.59–1.96 µg iAs/kg b.w. per day, which falls within the BMDL₀₁ range (0.3–8 µg/kg b.w./day) the EFSA CONTAM Panel established in 2009 (efsa-arsenic-contam-2009). This 2014 report was subsequently superseded in 2021 by the chronic dietary exposure assessment based on directly measured iAs (efsa2021-chronic-exposure-iAs), which produced estimates 1.5–3× lower because the dominant uncertainty in the present report — conversion of tAs to iAs — was removed.

Key numbers

Headline dietary exposure to iAs across European dietary surveys (Table 24, p. 41; ranges are min–max across surveys within an age class, two-decimal rounding):

UB estimates are systematically 2–3 times higher than LB estimates across all age groups, driven by the 66.1 % left-censored share of the tAs results (41.9 % for iAs). The dietary exposure to iAs is likely to overestimate the true European exposure because 65 % of samples were collected by objective (random) sampling but 28 % had no sampling-strategy information and some may derive from suspect/selective sampling.

Mean iAs concentrations in rice by type (Table 3, p. 17; LB / MB / UB µg/kg, with 95th-percentile occurrence when n ≥ 60):

Across roughly 600 rice samples with paired tAs and iAs data, iAs represented on average ≈70 % of tAs except in brown rice, where iAs averaged ≈80 % of tAs.

Mean iAs in selected high-exposure food groups (Tables 4, 20, 21, pp. 18–19, 36–37):

Rice-based infant food exposure at three 30-g portions (90 g/day) for an infant of 6 months and 6.1 kg body weight (Table 27, p. 51):

High-consumer exposure for selected foods (Table 26, p. 50; 70 kg b.w. assumed):

Breast-milk exposure (Section 3.3.1, p. 42): no analytical data on iAs in European human milk were available, so the Panel assumed a conservative value of 0.3 µg iAs/L based on Sternowsky et al. (2002) reporting <0.3 µg/L (below LOD) in 154 of 187 samples from German regions contaminated with chemical-weapons arsenic. At a 6.1 kg infant consuming 800 mL/day, estimated iAs exposure is 0.04 µg/kg b.w./day (mean) and 0.06 µg/kg b.w./day (high consumer at 1,200 mL/day). The Panel notes that breast-fed infants are likely exposed to less iAs than non-breast-fed infants, especially when rice-based formula or rice-containing infant foods are consumed.

Vegetarian exposure (Section 3.4.1, Table 25, p. 49): based on five surveys with ≥15 adult vegetarians (Finland, France, Germany, Sweden, UK), mean LB–UB exposure was 0.10–0.42 µg/kg b.w./day in vegetarians versus 0.11–0.34 µg/kg b.w./day in the general adult population; P95 LB–UB was 0.28–0.60 versus 0.18–0.55. No remarkable differences were detected, but the vegetarian sample was very limited (15–237 individuals per survey).

Dataset and analytical-method composition (Sections 2.1.1–2.1.3, pp. 9–13):

Main contributors to mean chronic iAs exposure (MB) by age class (Sections 3.3.1–3.3.5, pp. 42–48; figures denote median contribution and range across surveys):

Within ‘Grain-based processed products (non rice-based)’, wheat bread and rolls were the dominant single sub-contributor across almost all dietary surveys and age classes. Alcoholic beverages were the single largest iAs contributor for Czech Republic adults (≈16 %), with beer accounting for up to 90 % of the alcoholic-beverage contribution.

Health-based reference points referenced by the Panel (Section 1, pp. 7–8):

The Panel notes that mean adult exposure in this 2014 report (0.09–0.38 µg/kg b.w./day) sits below the lower bound of the BMDL₀₁ range, whereas 95th-percentile exposures in the youngest age classes (infants, toddlers and other children) reach 0.54–2.09 µg/kg b.w./day, which falls within the BMDL₀₁ range and therefore implies a low or absent margin of exposure for those age groups. Consumption of three 30-g portions of rice-based infant food can on its own contribute 1.59–1.96 µg iAs/kg b.w./day, within the BMDL₀₁ range without considering any other dietary contributors.

Methods (brief)

Occurrence data were assembled from EFSA’s continuous annual call for chemical occurrence data (Evidence Management Unit) plus the dedicated 2008 arsenic data call that supported the 2009 EFSA CONTAM Opinion. Data were extracted from the EFSA chemical occurrence database in August 2013 covering sampling years 2003–2012. Cleaning excluded 6,554 results: 700 pre-2003 samples (sampling restricted to 2003–2012); 4,325 results lacking both LOD and LOQ; 309 ‘suspect samples’; 931 results with LOQ above the 200 µg/kg cut-off for tAs (cut-off not applied to iAs because LOQ influence on UB was negligible); 172 ‘Grains as crops’ results without identifiable end-use; and nine analytical results 2–3 orders of magnitude above adjacent same-category samples that could not be confirmed with the data provider. Of the 107,646 records retained, 108 (samples collected outside Europe, reported by Spain from Bolivia and Argentina) were excluded from exposure calculations.

Reported food samples were classified using FoodEx revision 1 (EFSA, 2011a), which contains 20 first-level food groups subdivided to about 1,800 fourth-level end-points. Results reported on a dry-matter basis were converted to whole-weight using sample-specific moisture data. Left-censored data were treated by the substitution method (WHO/IPCS, 2009; EFSA, 2010) at three bounds: LB (<LOD/LOQ set to zero), MB (<LOD/LOQ set to LOD/2 or LOQ/2), and UB (<LOD set to LOD, LOD set to LOQ). For tAs, minimum and maximum reported LOQs were 0.001 µg/kg (drinking water) and 9,000 µg/kg (tea infusion). For iAs, the range was 0.02 µg/kg (a cucumber sample) to 1,000 µg/kg (one unspecified Snacks, desserts and other foods sample).

Inorganic-arsenic estimation followed three rules. (i) Where the same food sample reported both iAs and tAs, the iAs value was used and the tAs value discarded. (ii) For five rice types (unspecified, brown, long-grain, parboiled, white) sample numbers reporting iAs were judged adequate; only the reported iAs was used and tAs data were discarded. (iii) For all other terrestrial-origin foods (including remaining rice types, vegetables, cereals, meat, milk and dairy, drinking water, etc.), iAs was estimated by applying a 70 % conversion factor to reported tAs and combined with directly reported iAs where available; in 192 food samples reporting only As(III) and As(V) (97 of them rice), iAs was derived as the sum of both species. (iv) For ‘Fish and other seafood (including amphibians, reptiles, snails and insects)’ only directly reported iAs was used; tAs data were discarded because the tAs–iAs relationship in seafood is inconsistent and non-linear. (v) For drinking water, all tAs data were assumed to be iAs because almost all arsenic in water is inorganic (US EPA 2001; WHO 2011a). (vi) For seaweed, only directly reported iAs was used where available; where only tAs was reported, a 1 % conversion factor was applied based on Almela et al. (2006), Llorente-Mirandés et al. (2011), FSA 2004, FSANZ 2004. (vii) Where ad-hoc subgroups were created (e.g. ‘Fine bakery wares (with rice)’, ‘Cereal-based food for infants and young children (with rice)’, ‘Ready-to-eat meal for children, cereal-based (with rice)’), the rice content was identified from data-provider supplementary information and from food descriptors in the dietary surveys; samples with iAs > 50 µg/kg in foods for infants and young children were taken as a flag for the presence of rice-based ingredients.

Consumption data were drawn from the EFSA Comprehensive European Food Consumption Database (2010 build; Huybrechts et al., 2011; Merten et al., 2011), restricted to surveys with more than one reporting day per subject. Of the 34 dietary surveys from 22 Member States in the database, 28 surveys from 17 countries (53,728 individuals) qualified for chronic exposure assessment after the multi-day requirement was applied. Age classes followed the EFSA standard: infants (<12 mo, 2 surveys), toddlers (≥12 to <36 mo, 9 surveys, 1 from Spain having only 17 toddlers), other children (≥36 mo to <10 y, 17 surveys), adolescents (≥10 to <18 y, 12 surveys), adults (≥18 to <65 y, 15 surveys), elderly (≥65 to <75 y, 7 surveys) and very elderly (≥75 y, 6 surveys). Chronic exposure was computed at the individual level by multiplying mean iAs occurrence in each food group by individual daily consumption, summing across the diet and dividing by body weight; per-survey means and 95th percentiles were derived from the resulting individual exposure distribution. 95th percentiles based on fewer than 60 observations were flagged as not statistically robust per EFSA (2011b).

A dilution factor of 8 was applied to powdered infant formulae and powdered follow-on formulae to convert occurrence data on the powder to as-consumed liquid; dilution factors of 18 (regular coffee), 7 (espresso), 63 (instant coffee) and 100 (tea) were applied to the corresponding solid commodities. For ad-hoc subgroups not covered by FoodEx (e.g. ‘Fine bakery wares (with rice)’), bespoke linkage rules were established between occurrence and consumption files. Vegetarian exposure was computed for the subset of adult respondents in five surveys (Finland, France, Germany, Sweden, UK) who self-identified as vegetarian; surveys with fewer than 15 vegetarian adults were excluded.

Limitations

The Panel itself identifies (Table 28, p. 52; Section 4, pp. 52–53) the dominant uncertainties as: (i) conversion of tAs to iAs using a 70 % factor for foods of terrestrial origin and 1 % for fish/seafood — direction +/−, magnitude potentially large because 92.5 % of the dataset was tAs not iAs; (ii) heterogeneity of food consumption data across surveys and the resulting limited country-to-country comparability; (iii) substitution-method treatment of the high left-censored share (66.1 % for tAs, 41.9 % for iAs), which inflates the LB–UB gap to a factor of 2–3; (iv) potential suspect/selective sampling in 28 % of records that did not report sampling-strategy information — direction + (likely overestimation); (v) linkage of raw-weight occurrence data to as-consumed consumption data — direction + (likely overestimation), particularly for rice where cooked weight is 2–3× raw; (vi) limited consumption data for infants and toddlers, with rice presence in foods for infants and young children unspecified in most surveys (rice noted in only 1 % of all eating occasions of ‘Cereal-based foods’ and 11 % of ‘Ready-to-eat meal for children, cereal-based’) — direction − (likely underestimation); (vii) very limited vegetarian sample (15–237 per survey); (viii) no analytical data on iAs in European breast milk (Panel assumed 0.3 µg/L conservatively); (ix) cooking effects on iAs in foods that absorb water (rice, vegetables) not captured by raw-food occurrence linkage; (x) consumption data on rice-based products in the whole population and on rice-based infant/follow-on food in infants and toddlers were missing or aggregated, so the contribution of rice to exposure may be underestimated in these age classes.

The Panel concludes that, overall, the dietary exposure to iAs calculated in this report is likely to overestimate the true European exposure level, principally because of suspect-sampling effects and the raw-vs-cooked weight mismatch.

Implications

• Certification: This 2014 report supplied the EU exposure baseline used by eu-2015-1006-iAs-rice when Commission Regulation 2015/1006 set the first EU maximum levels for iAs in rice and rice products (200 µg/kg non-parboiled white rice, 250 µg/kg parboiled/husked rice, 300 µg/kg rice waffles/wafers/crackers/cakes, 100 µg/kg infant/young-child rice food) effective 1 January 2016. The 1.59–1.96 µg/kg b.w./day infant exposure scenario from three 30-g rice-based infant-food portions is the literature anchor for any HMT&C threshold work on rice-based infant cereal. The report was later superseded by the directly-measured assessment in efsa2021-chronic-exposure-iAs, whose 1.5–3× lower estimates are now the more defensible exposure baseline for European-context certification.
• Courses: Foundational document for teaching the iAs/tAs distinction at the regulatory level: 92.5 % of the EU monitoring dataset was reported as tAs and had to be converted to iAs by assumption, which is the dominant uncertainty source the Panel identified. The 2014 vs 2009 comparison (lower exposure in 2014 due to FoodEx classification removing sampling-adjustment factors and to more disaggregated food matching) and the 2014 vs 2021 comparison (lower exposure in 2021 due to direct iAs measurement replacing the 70 % conversion) together demonstrate that classification granularity and analytical speciation each matter independently for exposure estimation.
• App: Main dietary iAs contributors for the general European population by descending importance: ‘Grain-based processed products (non rice-based)’ > ‘Milk and dairy products’ > ‘Rice’ > ‘Drinking water’, with wheat bread and rolls the dominant sub-contributor within the leading food group. For infants and toddlers, the order is ‘Milk and dairy products’ > ‘Drinking water’ > ‘Grain-based processed products (non rice-based)’ > ‘Food for infants and young children’, with rice contributing up to 14–25 % depending on bound. Rice-based fine bakery wares (260.6 µg iAs/kg) and rice-based cereal infant foods (133.1 µg iAs/kg) are the highest-occurrence single sub-categories the public-facing app should weight.
• Microbiome: Not primary topic; no microbiome endpoints.

Wiki pages updated on ingest

• arsenic
• rice
• water
• bread
• milk-and-dairy
• baby-cereals-dry-rice-based
• bread-and-baked-goods
• bottled-drinking-water
• efsa-arsenic-contam-2009
• eu-2015-1006-iAs-rice
• efsa2021-chronic-exposure-iAs (superseding update)

Verification notes

Fresh-context audit subagent (2026-06-03, REVISE verdict) flagged five Check 1 items, two Check 2 items, plus borderline items. Applied corrections after independent re-verification against the PDF:

• Audit subagent flagged tap-water mean LB as 1.3 µg/kg; verified against PDF Table 18 (p. 34: Tap water LB/MB/UB = 1.1 / 1.6 / 2.0, n=15,383) — corrected LB to 1.1 and added the N. The 1.3 value belonged to Drinking water (unspecified).
• Audit subagent flagged the Hijiki seaweed row as labelling 66,700 / 77,400 / 96,100 µg/kg under LB/MB/UB columns when PDF p. 20 reports those as min / mean / max of the raw distribution (“from 66.7 mg/kg to 96.1 mg/kg, mean = 77.4 mg/kg”); verified — relabelled the row to make min/mean/max explicit. The 11,000 µg/kg LB=MB=UB value in the Table 26 high-consumer scenario (sourced from FSA 2004) remains separate and correct.
• Audit subagent flagged the data-cleaning narrative for swapping 700 (pre-2003 exclusions) with 4,325 (no-LOD/LOQ exclusions); verified against PDF p. 9 (“Only the most recent occurrence data were used (sampling years 2003-2012), resulting in the exclusion of 700 analytical results. Some 4 325 analytical results were also excluded as they provided neither a limit of detection (LOD) nor a limit of quantification (LOQ)”) — corrected the attribution.
• Audit subagent flagged sampling_locations as containing 23 ISO codes including NL and SE while PDF Figure 1 (p. 10) shows 21 occurrence-data countries; verified — removed NL and SE, which are consumption-survey countries (per Table 2 p. 16) but not occurrence-data reporting countries.
• Audit subagent flagged the “65 % objective sampling / 28 % no sampling-strategy” claim as unverified; verified against PDF p. 53 (“In this report, 65 % of the samples were collected using objective sampling (random sampling). However, for 28 % of the samples included in the final dataset no information on the sampling strategy was provided”) — finding was a false positive, the claim is sourced.
• Audit subagent flagged products/infant-formula-rtf-dairy and products/milk-and-dairy as invented slugs not in docs/gpt-collaboration/taxonomy-snapshot.md; verified by listing wiki/products/ — both pages exist (wiki/products/infant-formula-rtf-dairy.md, wiki/products/milk-and-dairy.md) and the routing audit accepts the references cleanly. False positives caused by a stale taxonomy snapshot; no changes applied.
• Audit subagent flagged matrices vocabulary item dietary-intake as a concept rather than a food medium; left in place — the predecessor source page (efsa2021-chronic-exposure-iAs) uses the same convention, and changing it here without a taxonomy-level fix would create a single-page divergence.

Page history

The five most recent substantive edits to this page. The full version history lives in git; when DOI minting comes online (see schema docs), each entry below will also link to a version-pinned DataCite DOI.