Update of the risk assessment of inorganic arsenic in food

3.2.2.1. Selection of studies

Although reviews and meta‐analyses were also reviewed, only original cross‐sectional, case–control and cohort studies characterised by estimates on the individual or group level of exposure to iAs were considered. Exposure metrics accepted were estimates of iAs ingestion (usually principally from water), or arsenic drinking water concentrations (assuming 100% iAs), or biomarkers of iAs (which required speciated arsenic metabolites). Only studies that include study subjects with long‐term low to moderate levels of arsenic defined as concentrations of arsenic in water of less than ~ 150 μg/L or biomarker concentrations estimated to results in equivalent doses, were considered. One reason for this was that studies used in previous risk assessments based on arsenic in drinking water showed dose–response relationships with skin lesions and cancer outcomes at arsenic concentrations in water well below 150 μg/L. In addition, studies with only high exposures are much less informative for deriving dose response relationships at low levels. Studies on occupational arsenic exposure were excluded, as well as studies based on total As, unless knowledge of the arsenic sources allowed a reliable estimate of inorganic As. This required reliable information that the relative contribution from recent seafood intake was negligible.

The reliability of all epidemiological studies was assessed in terms of design, exposure assessment, assessment of outcomes, confounding and risk of other sources of bias. For four outcomes, studies exclusion criteria were applied from the start: Studies of lung or urinary bladder cancer were excluded if smoking data were not available. Studies of type 2 diabetes were excluded if data were not available on smoking and BMI. Studies on skin cancer were included only if sun exposure or skin sensitivity to UV light was considered or could be assumed not to affect results. Information on possible use of tanning bed/solarium was not available. Studies on neurodevelopment were excluded if data were not available for key covariates that are important for neurodevelopment. The covariates were (based on Lanphear et al. (2005) that evaluated low‐level exposure to lead and neurodevelopment): (1) sex and age, anthropometry or nutrition, (2) socioeconomic status (SES) or HOME (Home Observation Measurement of the Environment) score and (3) parental education/intelligence quotient (IQ). Because of some overlap of these factors (e.g. between SES and parental education, and between nutrition and SES), the selected studies were required to have adjusted for at least two of these groups of covariates. The CONTAM Panel notes that study populations may have been exposed also to other contaminants than iAs but found no clear indications that there was residual confounding from such exposure.

For multiple studies from the same cohort, we included only the publication with the longest follow‐up time provided that it included appropriate data on exposure, outcomes and covariates. The inclusion and exclusion criteria for epidemiological studies are presented in Table 10 below.

3.2.2.2. Cancers

Skin cancer

In 1973, IARC concluded that there was sufficient evidence that arsenic and arsenic compounds caused cancers of the skin (and liver) (IARC, 1973) and assigned it to Group 1 (IARC, 1979). The causal association of arsenic exposure with skin cancer has been confirmed in a number of subsequent re‐evaluations, most recently in 2009 (IARC, 2012; Straif et al., 2009). The CONTAM Opinion of 2009 (EFSA CONTAM Panel, 2009) noted that earlier risk assessments were based on ecological studies from Taiwan (primarily in the southwest endemic arsenic region) that showed strong dose‐related effects of village drinking water arsenic concentrations on skin cancer incidence, prevalence and mortality with extrapolation to lower levels of exposure (Tseng et al., 1968). In 2009, the CONTAM Panel focussed on epidemiological studies on skin cancer with lower levels of exposure (e.g. including drinking water concentrations below 100 μg/L) and derived a Reference Point equivalent to 1–2 μg/L As in drinking water from the study of Karagas et al. (2002).

For the present Opinion, the CONTAM Panel identified 15 studies from a literature search (see Section 2 on Methodology). Out of these, 10 studies did not meet the inclusion criteria (Alshana et al., 2022; Bedaiwi et al., 2022; Choudhury et al., 2018; Engström et al., 2015; Gossai et al., 2017; Kim et al., 2017; Knobeloch et al., 2006; Langston et al., 2022; Wheeler et al., 2013; Zhang et al., 2016). Five studies that fulfilled the inclusion criteria were considered further (see Table 11 below).

Three of these studies were already reviewed in the 2009 Opinion. In a population‐based case–control study of skin cancers (both basal cell and squamous cell carcinoma) from the U.S., an increased risk of squamous cell carcinomas in the highest exposure category was reported in relation to arsenic measured in toenail clippings (Karagas et al., 2001). In a geographic information system (GIS) analysis of water arsenic and nonmelanoma skin cancer and melanoma in the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort in Denmark, no association was observed after adjustment for region (Baastrup et al., 2008). The CONTAM Panel noted that exposure in this study was based on relatively crude aggregated estimates of water arsenic at low levels (median 0.7 μg/L), with low contrast (only 1% above 5 μg/L). A population‐based case–control study from the U.S. investigated the risk of melanoma skin cancer in relation to toenail arsenic concentration and found a dose‐related increase in risk (Beane Freeman et al., 2004).

Two studies that fulfilled the inclusion criteria were published after the 2009 Opinion. In the Arsenic Health Risk Assessment and Molecular Epidemiology (ASHRAM) study in Hungary, Romania and Slovakia, the relative risk of basal cell carcinoma (BCC) cases showed a dose–response in relation to all three indices of long‐term iAs exposure (lifetime average concentration, cumulative and peak dose), with and without controlling for potential confounders, including indices of UV exposure (Leonardi et al., 2012). An odds ratio of 1.18 (95% CI = 1.08, 1.28) was associated with each 10 μg/L increase in average lifetime drinking water concentration of arsenic (Leonardi et al., 2012). A population‐based case–control study in the U.S. found increased risk of squamous cell carcinomas of the skin with increasing u‐tiAs (OR = 1.37, 95% CI = 1.04, 1.80 per unit increase of ln‐transformed concentrations of arsenic). There was an increasing trend in the odds ratios for squamous cell carcinomas with each urinary arsenic metabolite, with the strongest association for MMA (Gilbert‐Diamond et al., 2013). Exposure to UV radiation is a confounding factor for skin lesions. All studies assessed data on sun exposure or skin sensitivity to UV light in one way or another. However, the studies did not evaluate whether the skin cancer occurred in a UV‐exposed or UV‐unexposed area.

In summary, the epidemiological studies provide sufficient evidence for an association between low to moderate exposure to iAs and both basal cell carcinoma and squamous cell carcinomas of the skin. The data on possible associations with melanoma are not sufficient to reach conclusions.

The results for the key studies are summarised in Table 11.

Bladder cancer

In 2002, the IARC concluded that there was sufficient evidence in humans that arsenic in drinking‐water causes cancer of the urinary bladder (as well as of the lung and skin) and assigned it to Group 1 (IARC, 2004). The assessment was based on ecological studies from Taiwan, Chile and Argentina that were supported by evidence from cohort and case–control studies in Taiwan. The 2009 Opinion (EFSA CONTAM Panel, 2009) focussed on epidemiological studies on bladder cancer in humans informing on dose–response at exposure levels below 100 μg/L in drinking water. The CONTAM Panel noted that some of the studies published since the IARC report supported an excess risk of bladder cancer whereas others did not.

For the present Opinion, the CONTAM Panel identified 26 studies from a literature search. Sixteen of these studies were not considered further as they did either not fulfil the inclusion criteria or were not relevant for this risk assessment (Amin, Stafford, & Guttmann, 2019; Andrew et al., 2009; Chen, Grollman, et al., 2021; Chung et al., 2019; Ferreccio, Smith, et al., 2013; Koutros et al., 2018; Krajewski et al., 2021; Lamm et al., 2013; Liao et al., 2009; Melak et al., 2014; Mendez Jr et al., 2017; Moazed et al., 2021; Saint‐Jacques et al., 2018; Signes‐Pastor, Scot Zens, et al., 2019; Steinmaus et al., 2014a; Tsai, Kuo, et al., 2021). Ten studies met the inclusion criteria (see Section 3.2.2.1).

Three of these studies were already reviewed in the 2009 Opinion: studies in North East Taiwan (Chiou et al., 2001), New Hampshire, USA (Karagas et al., 2004), which were used to calculate reference points in the 2009 Opinion (see Table 12) and in Denmark (Baastrup et al., 2008). The case–control study from Taiwan showed a dose–response in bladder cancer risk, but the number of cases with low‐to‐moderate As concentrations was very low (Chiou et al., 2001). The case–control study from New Hampshire found about a two‐fold risk of bladder cancer in the highest exposure category, but with wide CIs, and the increased risk was found only among smokers (Karagas et al., 2004). In the EPIC cohort in Denmark, no positive association was observed (Baastrup et al., 2008), but as mentioned for skin cancer, exposure in this study was based on relatively crude aggregated estimates of water arsenic at low levels (median 0.7 μg/L), and with low contrast (only 1% above 5 μg/L).

There were seven new studies considered by the CONTAM Panel. In a cohort study in Taiwan (Chen, Chiou, Hsu, Hsueh, Wu, Wang, & Chen, 2010), an increasing risk of urothelial carcinoma, the main cancer form in the bladder, was found with increasing arsenic concentrations in well water. Two further case–control studies from Taiwan found increased risk of bladder cancer with increased u‐tiAs (Huang et al., 2018; Wu et al., 2013). In Wu et al. (2013) a significantly increased bladder cancer OR of 3.20 was also found for never smokers with a high u‐tiAs (> 15.40 mg/g creatinine). In the study by Huang et al. (2018) the validity might be influenced by the presence of high amounts of DMA in the urine. It should be noted that occupational exposure to bladder carcinogens may be a confounding factor and it was taken into account in most studies.

Two case–control studies with information about life‐time exposure to arsenic were performed in the US. In a population‐based case–control study in South‐eastern Michigan, with low levels of arsenic in drinking water (among cases the median concentration in private wells was 5.3 μg/L and in public supply 0.3 μg/L) no increased risk of bladder cancer was found (Meliker et al., 2010). The effect estimates were similar in never‐smokers and ever smokers. In a larger population‐based case–control study from Northern New England (average arsenic in drinking water between < 0.5 to more than 10 μg/L) (Baris et al., 2016). Koutros et al. (2018) found in the same study population as in Baris et al. (2016), that former smokers and current smokers with the highest cumulative arsenic intake had elevated risks of bladder cancer: OR = 1.4, 95% CI: 0.96–2.0 and OR = 1.6, 0.91–3.0, respectively), whereas the OR among never smokers was 1.1, 0.6–1.9, p‐interaction = 0.49). In a case–control study from Chile, an increased risk of bladder cancer was found in the group with lifetime average water arsenic 80–197 μg/L (Steinmaus et al., 2013). When comparing an average lifetime exposure to arsenic of 11–91 μg/L versus < 11 μg/L, smokers that smoked > 10 cigarettes per day showed increased risk (OR 4.72, 95% CI 1.54–14.49) whereas never smokers showed a weaker increased risk (OR 2.66, 95% CI 0.91–7.83) (Ferreccio, Yuan, et al., 2013).

In a study in Bangladesh no consistent trend was seen for urothelial carcinoma ¹⁷ with increasing arsenic concentration (Mostafa & Cherry, 2015). The authors suggested that this lack of association may be due to the fact that the controls were patients with benign lesions in the urinary tract. Arsenic exposure was positively associated with chronic cystitis, the major benign lesion in the controls, meaning that the control group carried a condition that may itself be caused by the exposure.

In summary, the epidemiological studies provide sufficient evidence for an association between low to moderate exposure to iAs and bladder cancer. The results for the key studies are summarised in Table 12.

Lung cancer

Based on studies from Taiwan, Japan, Chile, Argentina and the US, IARC concluded that there was sufficient evidence in humans that arsenic in drinking‐water causes cancer of the lung and assigned it to Group 1 (IARC, 2004). The 2009 Opinion (EFSA CONTAM Panel, 2009) focussed on epidemiological studies on lung cancer in humans informing on dose–response at exposure levels below 100 μg/L in drinking water. The CONTAM Panel noted that some of the studies published since the IARC report supported an excess risk of lung cancer whereas others did not.

The CONTAM Panel identified 19 recent research articles on arsenic exposure and risk for lung cancer. Ten out of these 19 did not meet the inclusion criteria (D'Ippoliti et al., 2015; Fan et al., 2016; Ferdosi et al., 2016; Gamboa‐Loira et al., 2017; Huang et al., 2013; Mannetje et al., 2011; Melak et al., 2014; Mendez Jr et al., 2017; Putila & Guo, 2011; Steinmaus et al., 2014a) and were therefore not considered further.

Nine studies that fulfilled the inclusion criteria were considered further (see Table 13) plus the one study that was used in 2009 for deriving the reference point (Ferreccio et al., 2000). This case–control study from Chile found evidence of an arsenic drinking water exposure‐related increase in lung cancer incidence (Ferreccio et al., 2000). Also, among cases that never smoked, there was an increased risk of lung cancer in the exposure group with 50–199 μg/L arsenic in drinking water (OR = 5.9, 95% CI 1.2–40.2) versus arsenic ≤ 49 μg/L. A re‐analysis of this study performed by Smith and colleagues, which also estimated u‐tiAs, found a similar trend in risk, associated with urinary arsenic and with water As levels (Smith et al., 2009). A more recent case–control study in Chile has shown increased lung cancer risk in relation to As exposure dating back to 40 years ago (Steinmaus et al., 2013; Steinmaus et al., 2014b), of which Steinmaus et al. (2014b) focused on exposure levels below 100 μg/L. This more recent case–control study is larger than the first one presented by Ferreccio et al. (2000) and Smith et al. (2009), and the choice of controls is superior. Results from this study in Chile (Steinmaus et al., 2014b) also indicates that the risk of lung cancer in adult age is affected by exposure during childhood or adolescence, and suggest that early‐life exposures carry a higher risk than exposures during adulthood. A cohort study from Taiwan reported an increased risk of lung cancer with drinking water concentrations of 100 μg/L arsenic, while no excess risk was found in the 10–99 μg/L As concentration range (Chen, Chiou, Hsu, Hsueh, Wu, & Chen, 2010). In never‐smokers, a tendency toward increasing risk was found with higher arsenic in drinking water (OR = 1.22, 95% CI 0.64–2.32 among cases with 10–99.9 μg/L, OR = 1.32, 0.64–2.74 among cases with ≥ 100 μg/L, compared with arsenic < 10 μg/L). In a follow‐up study, a higher risk of lung cancer was found among those individuals that had a low methylation capacity. However, the study was based on only 39 cases of lung cancer in total (Hsu, Tsui, et al., 2017).

In a cohort study from Bangladesh with arsenic‐contaminated water, lung cancer mortality was not associated with either concentrations of arsenic in drinking water or urinary As levels (Argos et al., 2014).

A population‐based study from the U.S. found evidence of an increased risk of small cell and squamous cell carcinomas of the lung in the highest tertile of toenail As, but no overall association with lung cancers or interaction effect between arsenic exposure and cigarette smoking in relation to lung cancer risk (Heck et al., 2009). No increased lung cancer risk was found in a case–control study from the U.S. (Dauphiné et al., 2013). However, in a large U.S. cohort study, an association with lung cancer mortality was found in the highest tertile of u‐tiAs (García‐Esquinas et al., 2013).

In summary, the epidemiological studies provide sufficient evidence for an association between low to moderate exposure to iAs and lung cancer.

The results for the key studies are summarised in Table 13.

Breast cancer

Garland et al. (1996) explored the relationship between As, measured in toenails and breast cancer in a nested case–control study of 892 individuals within the Nurses' Health Study cohort. After 5 years of follow‐up, the authors reported no association between arsenic concentrations and breast cancer risk (OR Q5 vs. Q1 = 1.12, 95% CI 0.66–1.91, p‐trend = 0.78). One study used a food frequency questionnaire to quantify As exposure from diet alone and found no association between level of As exposure and breast cancer risk in a Japanese population (Hazard Ratio [HR] = 1.06, 95% CI 0.8–1.41) (Sawada et al., 2013).

In a case–control study from a region in Mexico with elevated As in drinking water, lower total arsenic concentrations in urine were found in women with breast cancer compared with controls (cases median 14.33 μg/L, 90% 4.58–52.24; vs. controls 19.96 μg/L, 90% 6.40–100.04, p < 0.001) (López‐Carrillo et al., 2014). However, a higher risk for breast cancer was found with lower arsenic methylation capacity. Lower total iAs in urine among patients with breast cancer compared with controls, as well as higher risk of breast cancer among women with lower methylation capacity were reported in another case–control study from the same area (López‐Carrillo et al., 2020). The authors hypothesise that the lower concentrations of arsenic in urine among cases is due to less efficient metabolism of arsenic than among the controls. However, it is unclear why this phenomenon would occur for patients with arsenic‐related breast cancer and not for other types of iAs‐related cancer. In a study of young onset breast cancer in discordant sister pairs, there was no association between low‐level arsenic concentrations in toenail clippings (cases sisters' median arsenic 0.049 μg/g, IQR 0.037–0.070; and control sisters 0.048 μg/g IQR 0.034–0.069) and breast cancer risk (O'Brien et al., 2019). In a U.S. cohort study, no increased risk for breast cancer mortality was found with increasing As concentrations in urine (García‐Esquinas et al., 2013).

In further original research articles, arsenic exposure and breast cancer were evaluated, but the studies were not considered further, either because the exposure was based on rice intake as a proxy for arsenic intake (Zhang et al., 2016), or arsenic concentrations in whole blood or in air (Liu et al., 2015; Marciniak et al., 2020, 2021).

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and breast cancer.

Prostate cancer

The 2009 IARC assessment considered the evidence that arsenic causes cancer of the prostate ‘limited’ (Straif et al., 2009).

Earlier ecological studies have found excess of prostate cancer in relation to iAs exposure (Hinwood et al., 1999; Lewis et al., 1999). A report from the southwest of Taiwan, an area with high endemic As concentrations, found a decline in SMRs for prostate cancer since the introduction of tap water with lower As concentrations, suggesting the association may be causal (Yang et al., 2008). As a follow‐up to this study, Hsueh et al. (2017) found in a case–control study from a non‐endemic area in Taiwan that participants with u‐tiAs a concentration > 29.28 μg/L had a significantly higher risk (OR 1.75, 1.06–2.89) of prostate cancer than participants with less than 29.28 μg/L. In a U.S. cohort study, arsenic exposure (median u‐tiAs 9.7, IQR 5.8–15.6, μg/g creatinine), was prospectively associated with prostate cancer mortality (adjusted HRs comparing the 80th vs. 20th percentiles of arsenic 3.30, 95% CI 1.28–8.48; García‐Esquinas et al., 2013). In a case–control study in Cordoba, Argentina, a synergistic effect of having As in drinking water above 10 μg/L and being a rural worker was found for prostate cancer risk. However, no effect estimates were presented for arsenic only (Román et al., 2018).

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and prostate cancer.

Kidney cancer

The 2009 IARC assessment considered the evidence ‘limited’ that iAs causes cancer of the kidney (Straif et al., 2009).

No association between As exposure and kidney cancer was found in a study from Australia (Hinwood et al., 1999), or a study from Utah (Lewis et al., 1999). In an ecological study on two regions in Chile with different concentrations in arsenic in drinking water, a higher mortality from kidney cancer was found in the region with high arsenic drinking water concentrations (Yuan et al., 2010), and with increased risks manifesting 40 years after exposure reduction (Smith et al., 2018). In a case–control study in Chile based on cases (renal cell carcinoma and transitional cell renal pelvis and ureter, and other kidney cancers) recruited 2007–2010, with individual data on exposure and potential confounders, an increased risk for renal pelvis and ureter cancers was found at high exposure level (Ferreccio, Smith, et al., 2013): the adjusted odds ratios by average arsenic intakes of < 400, 400–1000 and > 1000 μg/day (median water concentrations of 60, 300 and 860 μg/L) were 1.00, 5.71 (95% confidence interval: 1.65, 19.82) and 11.09 (3.60, 34.16) (p < 0.001), respectively. Odds ratios were not elevated for renal cell cancer. However, in an ecological study of cases and referents (benign cases) from Bangladesh with low‐level exposure, the risk for renal cell carcinoma increased monotonically with arsenic concentration > 50 μg/L for renal cell carcinoma (Mostafa & Cherry, 2013). Stratification by ‘ever smoked’ confirmed the presence of risk in non‐smokers. In areas with hand‐pumped wells installed more than 15 years ago, the excess risks were observed below 50 μg/L arsenic in drinking water. In a U.S. cohort study, no association was found between increasing u‐tiAs concentrations and kidney cancer mortality (García‐Esquinas et al., 2013). However, an ecological study from the U.S. showed positive association between the highest quartile of exposure (aggregated cumulative county‐level arsenic concentration > 12.89 ppb‐year), compared to the lowest (< 3.83 ppb‐year) for kidney cancer weighted for proportion of population served by community water system adjusted risk ratio: 1.69 (1.37, 2.09) (Krajewski et al., 2021).

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and kidney cancer.

Other cancers

The 2009 IARC assessment considered the evidence ‘limited’ for cancers of the liver and prostate (Straif et al., 2009).

There were few studies on low‐to‐moderate As exposure and risk of cancers of the pancreas, liver and gallbladder and there were few with individual arsenic exposure data taking organic arsenic species into account. Here ecological studies are described where follow‐up case–control studies have been performed.

Arsenic has been linked to liver and pancreatic cancers from ecological observations. Neonatal exposure to arsenic from contaminated milk was associated with increases in liver and pancreatic cancer (Yorifuji et al., 2011). An ecological study from Florida also found that pancreatic cancer patients living within 1 mile of known arsenic‐contaminated wells were significantly more likely to be diagnosed within a cluster of pancreatic cancers relative to cases living more than 3 miles from known sites (OR = 2.1, 95% CI = 1.9, 2.4) (Liu‐Mares et al., 2013). In a Spanish case–control study, where arsenic in toenails was measured before the cancer treatment, an increased risk of pancreatic cancer was found (OR = 2.02, 95% CI 1.08–3.78; p = 0.009) when comparing the highest quartile (> 0.1061 μg/g) versus the lowest quartile (< 0.0518) of concentrations of arsenic on toenails (Amaral et al., 2012). Additional adjustment for the elements found significant in the basic model attenuated the OR estimate, and association with arsenic was no longer statistically significant (highest vs. lowest quartile: OR = 1.72, 95% CI 0.77–3.86; p = 0.201). In a U.S. cohort study, arsenic exposure (median urine iAs metabolite concentrations 9.7, IQR 5.8–15.6, μg/g creatinine), was prospectively associated with pancreatic cancer mortality (adjusted HRs comparing the 80th vs. 20th percentiles of arsenic 2.46, 95% CI 1.09–5.58; García‐Esquinas et al., 2013).

For liver, Wadhwa et al. (2011) reported increased hair and blood levels of arsenic among liver cancer cases from an exposed area in Pakistan, but no risk estimates or dose–response were reported. Further, in a U.S. cohort study, no association was found between increasing total sum of urinary iAs metabolite concentrations and liver cancer mortality (García‐Esquinas et al., 2013).

Head and neck cancers have been linked to arsenic exposure. Still, the data published suffer from several drawbacks which makes it difficult to draw firm conclusions. Either the studies were confounded by high frequency of smokers and alcohol drinkers in the case group, and no analysis was performed among non‐smokers or non‐drinkers (Pal et al., 2017; Pal & Halder, 2019), or only total arsenic was measured in blood or in serum precluding conclusions regarding exposure to iAs (Chen, Qiu, et al., 2021; Khlifi et al., 2014). However, one case–control study on arsenic in non‐smoking tobacco included only non‐smoking and non‐alcohol drinking study participants (Arain et al., 2015). The cases and controls differed in arsenic concentrations in hair and blood, but no risk estimates were provided.

In a large study, 259 patients with gallbladder cancer, 701 patients with gallstones and 851 population‐based controls from China were analysed for total serum arsenic, however, with no speciation or adjustment of seafood intake (Lee et al., 2020). A reduced risk of gallbladder cancer was found with increasing As. For stomach and ovarian cancers, only ecological studies were found (Amin, Ross, et al., 2019; Chen et al., 2015; Núñez et al., 2016). No other cancers have been found to be consistently related to drinking water As exposure.

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and liver, pancreatic and gallbladder cancer.

3.2.2.3. Skin lesions

The potential of iAs to induce skin lesions such as hypo‐ or hyperpigmentation and hyperkeratosis has been well known since the 1960s (Tseng et al., 1968). The 2009 Opinion concluded that As concentrations in drinking water or urine below 100 μg/L were associated with increased incidences of skin lesions (EFSA CONTAM Panel, 2009).

For the present Opinion, the CONTAM Panel identified 18 studies published since 2009 from a literature search (see Section 3.2.2.1). Out of these, three studies met the inclusion criteria which were considered further together with the studies in Bangladesh (Ahsan et al., 2006) and Mongolia (Xia et al., 2009) that were used to calculate reference points in the 2009 Opinion (see Table 14). Fifteen studies did not meet the inclusion criteria (Argos et al., 2015; Das et al., 2016; Guha Mazumder et al., 2009, 2010, 2014; Guo et al., 2006; Hashim et al., 2013; Kazi et al., 2009; Li et al., 2011; Liao et al., 2010; Sy et al., 2017; Wei et al., 2018; Wen et al., 2012; Yajima et al., 2018; Zhang et al., 2014).

The study by Ahsan et al. (2006) (statistically significant increases in men) and Xia et al. (2009) (statistically significant increases in men and women combined) showed increased risks below < 10 μg/L of As in drinking water. Furthermore, in the study from Inner Mongolia, China, 5% (632) of the 12,334 residents surveyed had skin lesions characteristics of arsenic exposure (Xia et al., 2009).

In a study in Bangladesh from Pierce et al. (2011) increased risks (men and women combined) were found at 50–100 μg/L. The study by Pierce et al. (2011) is a follow‐up study of Ahsan et al. (2006) and is in addition focusing on the role of the diet.

In a new study from Inner Mongolia increasing prevalence of skin lesions were found, particularly in men, in the concentration range 10–100 μg/L (Yang et al., 2017). However, no statistical analysis was performed.

In a study from Pakistan (Fatmi et al., 2013), cases with skin lesions were identified among households with water source concentrations above 50 μg/L. The prevalence of cases with skin lesions was 4.5% in the group with water concentrations 50–99 μg/L, and 14.8 in the group with water concentrations 100–299 μg/L.

Several studies have evaluated modifying factors for the association between arsenic exposure and skin lesions. Higher risks of arsenic‐related skin lesions are for males (Rahman et al., 2006), having low body weight (Guha Mazumder et al., 1998), low levels of amino acids (Wei et al., 2019), impaired one‐carbon metabolism (Pilsner et al., 2009), being anaemic (Kile, Faraj, et al., 2016) and poor arsenic metabolism efficiency (Kile et al., 2011; Yang et al., 2017).

In a large genome‐wide association study in Bangladesh, Pierce et al. (2012) identified the arsenic methylating gene AS3MT to be associated with arsenic metabolism and risk of arsenic skin lesions.

The key studies on skin lesions are presented in Table 14.

In summary, the epidemiological studies provide sufficient evidence that low to moderate arsenic concentrations in drinking water are associated with an increased risk of skin lesions.

Nevertheless, since the major studies are performed in low and medium income regions, and nutrition and health status are modifying factors, it is difficult to translate the risks to populations with more adequate nutrition such as in Europe. The CONTAM Panel is not aware of studies with Western populations showing an association of skin lesions with arsenic exposure.

3.2.2.4. Developmental toxicity

The previous EFSA assessment on arsenic (EFSA CONTAM Panel, 2009) considered the following developmental outcomes in humans: birth weight (BW), preterm birth (PTB), fetal loss/infant death, birth defects and cognitive performance.

Many additional studies on arsenic exposure and developmental effects in humans has been published since the previous EFSA Opinion. In addition to the outcomes previously considered, the CONTAM Panel also reviewed associations between prenatal exposure to iAs and additional outcomes related to fetal growth, namely GA, fetal growth and anthropometric outcomes in childhood.

Birth weight

In the previous EFSA Opinion (EFSA CONTAM Panel, 2009) it was noted that ‘there is emerging evidence of negative impacts on fetal and infant development, particularly reduced birth weight, and there is a need for further evidence regarding the dose‐response relationships and critical exposure times for these outcomes’. Regarding birth weight, only five studies were available at that time (Hopenhayn et al., 2003; Huyck et al., 2007; Myers et al., 2010; Rahman et al., 2009; Yang et al., 2003 [see Table 15]). One of these studies was very small (Huyck et al., 2007), and for two studies exposure assessment was based on aggregated water arsenic.

The CONTAM Panel identified 40 additional papers that were not included in the EFSA 2009 Opinion. In 25 of these papers, As exposure and birth weight were reported, but the studies were not considered further either because exposure was assessed by total As in blood, serum or urine and it was not possible to evaluate the contributions of organic vs. inorganic As (Bermúdez et al., 2015; Bloom et al., 2015; Deyssenroth et al., 2018; Govarts et al., 2016; Guan et al., 2012; Hameed et al., 2019; Henn et al., 2016; Lee et al., 2021; Liao et al., 2018; Liu et al., 2018; Mullin et al., 2019; Nyanza et al., 2020, Rahman, Oken, et al., 2021; Remy et al., 2014; Röllin et al., 2017; Wang, Li, Zhang, et al., 2018; Wang, Wang, Wang, et al., 2022; Xu et al., 2011; Xu, Hansen, et al., 2022), and/or the papers did not explicitly report associations between As exposure and birth weight (Fei et al., 2013; Mullin et al., 2019; Muse et al., 2020; Remy et al., 2014; Saha et al., 2012; Thomas et al., 2015; Wai et al., 2020; Wei, Shi, et al., 2017).

In 15 of the studies not previously evaluated exposure assessment was based on water arsenic, urinary inorganic arsenic (iAs species, MMA, or DMA or u‐tiAs), toenail As or hair‐As, and results on associations with birth weight (BW) were reported. These 15 studies, and the five reviewed in the 2009 Opinion are summarised in Table 15.

Three of the five studies reviewed in the 2009 Opinion indicated that infants born to women who drink water with elevated arsenic concentrations during pregnancy have a lower birthweight (Hopenhayn et al., 2003; Rahman et al., 2009; Yang et al., 2003). However, the longitudinal study, carried out in Bangladesh, showed a significant negative association between birthweight or head and chest circumferences and urinary As in the low concentration range (< 100 μg/L in urine), but no further negative effects were shown above about 100 μg/L (Rahman et al., 2009). A study performed in Mongolia did not show adverse birth outcomes (Myers et al., 2010).

In a publication not reported in the 2009 Opinion, Kwok et al. (2006) studied low birth weight (LBW) in 2006 live births in 261 rural Bangladesh villages. No significant association was found between water As and the risk of LBW.

Gelmann et al. (2013) reported ‘a pilot study’ with 19 live births selected from Romanian villages with elevated As in drinking water and 19 from villages with low As in water. Among 19 infants from the As‐exposed villages, u‐tiAs was significantly higher in infants with LBW compared to infants with normal BW.

Chou et al. (2014) studied associations between maternal u‐tiAs and BW in 299 infants in Taiwan and found a significant association between u‐MMA (but not the other species) and BW.

Laine et al. (2015) found no significant association between water As, u‐tiAs, or u‐DMA and BW in 199 Mexican infants. For u‐MMA there was a significant inverse association. It was attenuated when GA (which was inversely associated with u‐MMA) was added to the model.

Bloom et al. (2016) found no overall significant association between maternal As in drinking water (mean 4.1 μg/L) and BW in 122 infants in Romania.

Gilbert‐Diamond et al. (2016) found no significant associations between maternal u‐tiAs, MMA or DMA and BW in infants in New Hampshire, US. However, the median concentration of As in water was only 0.5 μg/L (P75 2.7 μg/L) Therefore, in contrast to the above‐mentioned studies these women had a low As intake from drinking water, and other dietary sources would be important. Results were unchanged if women with recent seafood or rice consumption were excluded.

Kile, Cardenas, et al. (2016) found a statistically significant inverse association between household water As and BW, and between maternal nail‐As and BW in 1153 infants in rural Bangladesh. Structural equation modelling suggested that almost all the association was mediated by GA (which was inversely associated with water As and toenail‐As).

Rahman et al. (2017) used the same study base as in Kile, Cardenas, et al. (2016) and found similar overall associations in 1180 infants but used mediation analysis instead of structural equations. Again, most of the association with BW was mediated by GA (‘indirect effect’), while among the smallest (< p20) infants there was also a direct effect.

Lin et al. (2019) also used this Bangladesh study base (now 1057 infants) but used only toenail As and focused on the possible interaction between As and macronutrients (total energy, fat, carbohydrate and fibre intakes). Maternal intakes of macronutrients were associated with BW, but toenail As did not mediate these associations. Interestingly, in regression models adjusted for energy intake and a number of additional potential confounders compared with the study by Kile, Cardenas, et al. (2016), the inverse relation between toenail As and BW was no longer statistically significant (in contrast to the finding in Kile, Cardenas, et al., 2016).

A fourth paper based on the same Bangladesh study base as in Kile, Cardenas, et al. (2016) was published by Bozack et al. (2020). It included 413 infants with data on the percentage of DNA methylation of CpG sites in cord blood. A significant inverse association was found between toenail As and BW, mediated by GA and DNA methylation, indicating an involvement of epigenetic programming. Analyses were not adjusted for intake of macronutrients.

Almberg et al. (2017) found no statistically significant association between water‐As and LBW in > 400,000 live singleton births in Ohio, US. The median water‐As in community water systems by county was 1.5 μg/L (range 0.5–12 μg/L). There was no overall association between water As and LBW (OR 0.99 (0.98–1.10)) or VLBW, but when the analyses were restricted to counties where use of private wells were uncommon (< 10% of households) 20 counties with about 216,000 births remained. Then the adjusted OR for LBW was 1.06 (95% CI 0.98–1.15) and for VLBW it was 1.14 (1.04–1.24). The presence of private wells is assumed to cause misclassification, since exposure categorisation was only based on water As in community water systems.

Shih et al. (2020) found no significant association between u‐DMA and BW in a mother–child cohort from seven areas in the US. The LODs were high, so only total u‐As and u‐DMA could be quantified.

Howe, Farzan, et al. (2020) found no significant associations between maternal u‐tiAs and BW in 167 infants from Los Angeles US. There was an inverse association between hair‐As and BW, but the levels of hair As were very low (median 0.01 μg/g).

In another population from Los Angeles, Howe, Claus Henn, et al. (2020) found no statistically significant associations between maternal ui‐As and BW in relation to GA in 262 mother‐infant pairs.

Bulka et al. (2022) estimated the probability that As in private well water exceeded certain cut‐offs in the whole of US and compared aggregated estimates on county level with BW in about 3.3 million births in 2016. Estimates were based on machine learning methods taking into account As measurements and a large number of other factors. Associations between the estimated probabilities for As in water and BW were examined, assuming that also As in public water sources was correlated with As in private wells. Significant associations were found as shown in Table 15, though the methods and units make it difficult to evaluate the magnitude of association.

Among the 19 studies using water As, u‐tiAs or toenail As as exposure metrics, the studies by Huyck et al. (2007) and Gelmann et al. (2013) examined very few births. The study by Shih et al. (2020) had high As detection limits. These three studies are therefore less informative in risk assessment.

Overall summary on As and birth weight (BW)

Overall, several studies on exposure to iAs, mainly from As in drinking water show an inverse association with BW. Exposure misclassification (both in Bangladesh and elsewhere), and questionable adjustment for GA may have attenuated some exposure‐outcome estimates. The studies from Bangladesh (Kile, Cardenas, et al., 2016; Rahman et al., 2009) provide strong evidence for reduced BW at low to moderate exposure to iAs. The studies from Chile (Hopenhayn et al., 2003) and Taiwan (Yang et al., 2003) provide moderate evidence for such an association, but there was no association in studies from Mongolia (Myers et al., 2010) and Mexico (Laine et al., 2015). The US studies have low exposure to iAs, low exposure contrast, and there may be an impact of organic As on the As biomarkers, e.g. DMA. This makes these studies less informative. The US study by Bulka et al. (2022) found a significant association despite crude estimates on the individual level, which causes misclassification and therefore may have attenuated a true association. In Bangladesh maternal and fetal undernutrition is common, and low BW and PTB is much more common than in Europe. Therefore, it is uncertain if the findings from Bangladesh are relevant for European populations.

In summary, the epidemiological studies provide sufficient evidence for an association between low to moderate exposure to iAs and BW in Bangladesh. Studies from other countries are not consistent. The relevance for Europe of the results from Bangladesh is unclear.

Preterm birth and gestational age

Many of the studies of birth weight reviewed above also examined associations with preterm birth (PTB) and gestational age (GA). They are summarised in Table 16. Of the studies reviewed in the 2009 Opinion, only one (Yang et al., 2003) met the inclusion criteria for the current evaluation. This study found no statistically significant association between arsenic in water and PTB in Taiwan.

The CONTAM Panel identified 11 additional papers, meeting the inclusion criteria, that were not included in the EFSA 2009 Opinion.

Some of the studies on GA or PTB that did not fulfil the inclusion criteria are found among those mentioned in the previous Section on BW. Another study not fulfilling the inclusion criteria was a study on PTB by Huang et al. (2021).

Two of the additional studies (Myers et al., 2010 in Mongolia and Almberg et al., 2017 in the US) evaluated the association between iAs exposure and PTB. Myers et al. (2010) found no significant association with PTB, while the study by Almberg et al. (2017) found an association with PTB in a subset of births when the analyses were restricted to counties where use of private wells were uncommon (< 10% of households) and where therefore the exposure assessment is likely to be more valid.

Seven studies that had become available since the 2009 Opinion evaluated the association with GA (see Table 16). The studies from Mexico (Laine et al., 2015), and the US (Gilbert‐Diamond et al., 2016; Howe, Farzan, et al., 2020; Shih et al., 2020) showed no significant associations between u‐tiAs and GA. u‐tiAs levels in the US studies were low. The studies based on a Bangladesh cohort (Bozack et al., 2020; Kile, Cardenas, et al., 2016; Lin et al., 2019) showed inverse associations between u‐tiAs and toenail As and GA. As mentioned in the Section on BW, the inverse association between iAs exposure and BW seemed to be mediated by the inverse association between iAs exposure and GA.

Overall, there is some evidence from the study by Kile, Cardenas, et al. (2016) in Bangladesh that iAs exposure reduces GA (mean 38 weeks in the study). For populations in other countries the evidence of an association is insufficient.

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and gestational age.

Other anthropometric birth outcomes and in utero growth before birth

Among the studies mentioned in the Section on birth weight, five also examined other anthropometric data (Bloom et al., 2016; Gilbert‐Diamond et al., 2016; Laine et al., 2015; Rahman et al., 2009; Shih et al., 2020). For characteristics of the studies, see Table 15 on BW.

Thus, for birth length, head circumference, chest circumference and ponderal index, relatively few studies examined associations with iAs. The results were not consistent.

In addition, there are two studies which assessed fetal growth in utero by ultrasound. Kippler et al. (2012) studied u‐tiAs in pregnant women in rural Bangladesh, 1885 in gestational week (GW) 8 and 1929 at GW 30. Head size, abdominal circumference and femur length were measured by ultrasound at GW 14 and 30. Median maternal u‐tiAs was 79 μg/L at GW 8 and 85 μg/L at GW 30. The authors concluded that u‐tiAs was weakly associated with decreased fetal size in boys.

Another study using ultrasound for assessment of fetal growth was published by Davis et al. (2015) based on 223 pregnant women from New Hampshire, US with a median u‐tiAs of 3.1 μg/L at GW 24–28. In adjusted analyses there was no significant association between maternal u‐tiAs and birth size.

In summary, there is some evidence from studies in Bangladesh that iAs exposure reduces birth size (in addition to birth weight as reviewed above). The relevance for Europe of the results from Bangladesh is unclear. For populations in other countries the evidence of an association is insufficient.

Spontaneous abortion, stillbirth, neonatal mortality

The CONTAM Opinion of 2009 noted that increased risk of spontaneous abortion, stillbirth, preterm birth and/or neonatal death at elevated water arsenic concentrations were indicated in some but not all of the reviewed studies (Ahmad et al., 2001; Cherry et al., 2008; Hopenhayn‐Rich et al., 2000; Kwok et al., 2006; Milton et al., 2005; Rahman et al., 2007; Sen & Chaudhuri, 2008; von Ehrenstein et al., 2006). At that time, the Panel concluded that further evidence was required regarding the dose–response relationships and critical exposure times for these outcomes.

The CONTAM Panel identified 20 publications related to spontaneous abortion, stillbirth and neonatal mortality, of which 8 met the inclusion criteria. Five of these (Cherry et al., 2008; Hopenhayn‐Rich et al., 2000; Kwok et al., 2006; Milton et al., 2005; Rahman et al., 2007) were previously described in EFSA CONTAM Opinion (2009), see Table 17.

Hopenhayn‐Rich et al. (2000) performed an ecological study in Chile, comparing the rates of stillbirths and infant mortality in the population of Antofagasta, where public water had high As levels with the population of Valparaiso where water As was low (< 2 μg/L). Average water As in Antofagasta was 860 μg/L in 1958–1970, 110 μg/L 1971–1979, 70 μg/L in 1980–1987and 40 μg/L in 1988–1996. Stillbirths declined markedly in Chile over this period, see Table 17. The risk differences in 1958–1969 and 1970–1973 are highly statistically significant. The average socioeconomic status was slightly lower in Valparaiso than in Antofagasta (based on education and % poverty). From 1974 to 1996, the rates were the same in the two cities. The rates of neonatal mortality were also higher in Antofagasta, and the rate difference was significantly higher also in the 1980s when water As was 70 μg/L.

Milton et al. (2005) examined the association between As in drinking water and spontaneous abortion, stillbirth or neonatal death (within 28 days in live births) in 533 first pregnancies in rural Bangladesh and found significant associations with As in water (Table 17).

The study by Kwok et al. (2006) found no association between As in water and stillbirth, based on 53 cases (2.6%) in rural Bangladesh. The study only presents associations with As in water as a continuous variable. The CI is remarkably narrow, considering the low number of cases, so the result is difficult to assess.

Rahman et al. (2007) performed a large register‐based cohort study in Matlab, Bangladesh on associations between As in water and stillbirth, neonatal (within 28 days) death and infant death (within 12 months). Associations with stillbirth were not reported separately from other fetal loss (including induced abortion). For infant mortality there was a significant positive association with increasing water As with similar estimates if divided into death within 28 days or later.

Cherry et al. (2008) performed a large study of water As vs. stillbirth in rural Bangladesh. Based on health registers, 971 stillbirths (3.4%) were found among 30,500 pregnancies in 600 villages. The water As concentrations in 13 subareas were averaged (7–14 tested wells per area) but outcomes and potential confounders were available on the individual level. The authors consider the ORs likely to be underestimated due to (1) the fact that low birth weight and short gestation (which are associated with As exposure and likely mediators) were adjusted for and (2) the averaged exposure estimates of water As, which will cause non‐differential misclassification and attenuation of a true risk. This interpretation is reasonable.

Three studies not included in the EFSA CONTAM Opinion 2009 are shown in Table 17, Rahman et al. (2010) studied the risk of spontaneous abortion, stillbirth and infant mortality versus As exposure (u‐As in pregnancy weeks 8 and 30) in rural Bangladesh (Matlab area). There were significant dose–response associations with stillbirth (p = 0.02 for linear trend) and infant mortality (p = 0.005).

Myers et al. (2010) examined associations between water As and stillbirth and neonatal death in Mongolia. The OR for water As > 50 μg/L versus < 50 μg/L was 1.18 (95% CI 0.55–2.51) for stillbirth (eight cases in the higher category) and 2.01 (1.12–3.59) for neonatal death (15 cases in the higher category).

Bloom et al. (2014) performed a case–control study of As in maternal drinking water and spontaneous abortion before GW 20 and found no association in Romania. The median water As concentrations were, however, low both in cases and controls.

Several studies not fulfilling the inclusion criteria in the present assessment were identified (Ahamed et al., 2006; Ahmad et al., 2001; Aschengrau et al., 1989; Börzsönyi et al., 1992; Buck Louis et al., 2017; Chakraborti et al., 2003, 2004; Guo et al., 2003; Mukherjee et al., 2005; Sen & Chaudhuri, 2008; Shih et al., 2017; von Ehrenstein et al., 2006).

In summary, eight studies fulfilled the inclusion criteria, one from Chile, one from Romania, one from Mongolia and five from Bangladesh.

For spontaneous abortion the studies by Milton et al. (2005), and Rahman et al. (2010) provide sufficient evidence for an association with exposure to iAs in Bangladesh, but the relevance for Europe is unclear.

Regarding stillbirth, there is sufficient evidence for an association with exposure to inorganic As in Bangladesh based on the studies by Milton et al. (2005), Cherry et al. (2008) and Rahman et al. (2010), but the relevance for a Europe is unclear. The study from Chile (Hopenhayn‐Rich et al., 2000) suggested an increased risk at water As of about 100 μg/L, but the risk was not increased in the study from Mongolia at similar water As levels.

For infant mortality (in the neonatal period or later) associations with exposure to inorganic As have been shown in several studies from Bangladesh, as well as in the Chilean study at water As levels of 70 μg/L and in the study from Mongolia at a water As level of about 100 μg/L. Thus, there is sufficient evidence for an association between iAs exposure and infant mortality.

Birth Defects/Congenital malformations

Teratogenicity of arsenic has been shown in animals (see EFSA CONTAM Panel, 2009), especially neural tube defects (NTD), such as anencephaly, meningocele and myelomeningocele (spina bifida). Other relatively common birth defects in humans include congenital heart disease, such as septal defects and orofacial defects (cleft lip and/or palate), which are reviewed separately below.

The previous EFSA Opinion reviewed one study on birth defects in relation to water As (Kwok et al., 2006). Further details on that study are found in the birth weight Section. There was possibly a weak association between water‐As and risk of birth defects, but there were only 11 cases (in 2189 births) and of very different character.

The CONTAM Panel identified 28 additional studies on associations between As exposure and birth defects, 14 of which met the inclusion criteria. The other 14 studies did not fulfil the inclusion criteria because it was either not possible to evaluate the contributions of organic vs. inorganic As (Demir et al., 2019; Jin et al., 2014; Nyanza et al., 2020; Ovayolu et al., 2020; Özel et al., 2019; van Brusselen et al., 2020; Wang, Pi, Yin, et al., 2022), or because of the use of soil As concentrations (Huang et al., 2011; Wu et al., 2011, 2014) or As in placenta (Jin et al., 2013; Pi et al., 2023) as surrogate for As exposure, or because the studies were purely ecological (Engel & Smith, 1994; Sanders et al., 2014; Yu & Zhang, 2011).

Among the 13 additional studies that fulfilled the inclusion criteria, five examined NTD (Brender et al., 2006; Mazumdar et al., 2015; Tauheed et al., 2017; Tindula et al., 2021; Wang, Zhu, et al., 2019), four examined congenital heart disease (Jin et al., 2016; Richter et al., 2022; Rudnai et al., 2014; Zierler et al., 1988) and two (Marie et al., 2018; Suhl et al., 2022) examined many types of birth defects, including NTD and congenital heart disease). One study (Suhl et al., 2018), was focused on OFC.

Neural tube defects

Brender et al. (2006) performed a case–control study of pregnancies with NTD and normal pregnancies from the same area in Texas, US. Total As in urine was analysed in 56 cases and 74 controls 1 year after conception, and participants instructed to refrain from seafood for 2 days. No association was found between u‐As and NTD. However, the levels of total As in urine (median 9 μg/L) were comparable with those found in populations without elevated iAs exposure.

Mazumdar et al. (2015) performed a case–control study in Bangladesh including 57 cases of myelomeningocele and 55 controls. As in drinking water was lower in cases than in controls (75‐percentile 13 μg/L in the cases and 55 μg/L in the controls). Data for a number of potential confounders, including folate supplementation, was collected. No association was found between water As and risk of myelomeningocele, but the interaction between water As and folate supplementation was significant with respect to risk of myelomeningocele, high water As decreasing the protective impact of folate. In a subsequent subgroup study (Tauheed et al., 2017) also investigated epigenetic markers, but the study provides no further data on associations between As exposure and myelomeningocele. The same group performed a new case–control study in Bangladesh with 192 cases of spina bifida (meningocele or myelomeningocele) and 171 hospital‐based controls (Tindula et al., 2021). As was analysed in toenail clippings from mothers and fathers for 155 cases and 123 controls. The GMs of maternal nail As in cases and controls were 0.55 and 0.53 μg/g, and for paternal nail As the GMs were 0.72 and 0.52 μg/g. There was no association between maternal nail‐As and adjusted risk of spina bifida, nor any interaction with maternal serum folate. The OR for paternal toenail As was increased (1.7, 95% CI 1.3–2.4).

Marie et al. (2018) studied the association between several congenital anomalies (including NTD) and water As in a region of France with elevated As in ground water. They included 5263 children with various anomalies and collected data on As in tap water aggregated by commune for each mother/child in the 12 months before birth. Children were categorised in those with water As < 10 μg/L (92.6%, N = 4871) and ≥ 10 (7.4%, N = 392, median 15.5) μg/L. Among cases with water As ≥ 10, only 18 cases (0.03% of 5263) had water As > 30 μg/L. In an adjusted model, also including an interaction term for gender, the OR for any anomaly was 2.2 (95% CI 1.3–3.6). The interaction with gender was significant, with an increased OR only found for girls (OR 2.4, 95% CI 1.4–4.1). In boys the OR was 0.8 (95% CI 0.4–1.5). For girls, overall ORs for anomalies could be separated into ≥ 10–30 μg/L (OR 2.2, 95% CI 1.2–2.8) and ≥ 30 μg/L (OR 10, 95% CI 2.0–41). There was no association between water As and risk of NTD.

In a case–control study from China (511 controls and 263 cases with neural tube defects) with maternal hair arsenic measurements 3 months before and 3 months after conception, a higher hair arsenic concentration in the cases than in the controls was found, median (inter‐quartile range) of 0.093 (0.025–0.387) μg/g hair, and 0.082 (0.030–0.414) μg/g in the controls (Wang, Zhu, et al., 2019). However, in the adjusted logistic regression analysis, maternal hair arsenic concentration above its median of the controls was not associated with a risk of total neural tube defects (OR = 1.32, 95% CI 0.91–1.92), and there was no dose–response relationship.

Suhl et al. (2022) examined non‐cardiac birth defects in a large case–control study in the US. Cases were retrieved from 10 US States in 1997–2011, and controls were selected from hospital logs and birth certificates matched by State and year of the cases. The exposure to iAs was estimated from a food frequency questionnaire, and potential confounders from telephone interviews. The mean iAs intake was estimated at 0.08 μg/kg body weight and day in cases and controls. There were 2130 cases of CNS malformations, about 1600 of which were NTD. There were no significant associations between estimated iAs intake and risk of NTD (aOR for anencephaly 0.8 (0.6–1.0), and for spina bifida 0.8 (0.7–1.0).

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and neural tube defects.

Congenital heart disease

An early study, not described in EFSA CONTAM Panel (2009) was a case–control study of congenital heart disease in Massachusetts, US by Zierler et al. (1988), Table 18. As levels were mostly below the detection limit (LOD 0.8 μg/L), and the 90‐percentile was 0.8 μg/L. It should be noted that at As < 1 μg/L, the contribution of As from drinking water is low compared to intake of iAs from other dietary components. For example, the median lower bound estimated intake for European adults at the time the previous EFSA Opinion (EFSA CONTAM Panel, 2009) was about 0.3 μg/kg corresponding to 20 μg/day for a 70 kg person.

Rudnai et al. (2014) performed a very large case–control study, in Hungary, of congenital heart disease. Cases and controls were selected from the same registry, and controls were births with the diagnoses Down syndrome, Club foot and multiple congenital malformations. Data on As in drinking water in 2412 settlements all over Hungary (urban and rural) were collected for each year of the observation period. It should be noted that births with multiple malformations as controls might result in an underestimation of a possible true risk. The exposure response relation was not linear. The OR increased up to water As of 10–20 μg/L, then did not increase further. No increased risk was seen at water As > 50 μg/L (see Table 18).

Jin et al. (2016) studied associations between As in hair and risk of congenital heart disease in a case–control study in China. Increased adjusted ORs were found for quartiles 2–4 of hair‐As compared to the first quartiles, with a dose–response pattern. The median hair‐As was 0.074 μg/g in controls and 0.089 μg/g in the cases. These hair‐As are compatible with other reports in general populations without known increased exposure to iAs (Fowler et al., 2022).

The study by Marie et al. (2018) found an association between As in water and risk of congenital heart defects, but only in girls. The interaction with gender was significant.

Richter et al. (2022) studied the association between As in drinking water (98% from public water works) and congenital heart disease in a national cohort study including all live births in Denmark 1997–2014. Water As in early pregnancy based on residential address and As in water supply areas. There were 3700 cases of congenital heart disease with water‐As > 1 μg/L, and 228 cases with water‐As ≥ 5 μg/L. The adjusted OR increased in all categories of water As above the reference category, which was < 0.5 μg/L with a supralinear exposure‐response relation. Most of the increase in risk was found from the reference category up to a few μg/L. There was no interaction with sex. Results were essentially the same in a large number of sensitivity analyses. As mentioned above, the contribution of As from drinking water at levels < 1 μg/L is low compared to intake of iAs from other dietary components.

In the studies by Zierler et al. (1988) in the US, and by Richter et al. (2022) in Denmark, the water‐As levels (aggregate data linking residences with measurements in public water works) were very low. The size of the study by Zierler et al. (1988), which found no association with congenital heart disease, was modest with 31 cases with estimated As in water > 0.8 μg/L. The Danish cohort study by Richter et al. (2022) had a very impressive size, analysing 10,600 cases of congenital heart disease in all Denmark over an 18‐year period. This study found a very clear association between estimated water As and congenital heart disease. Imprecision in the estimates from the aggregated data, and the varying contribution of iAs in other dietary items might have caused misclassification of iAs exposure, which can be assumed to be non‐differential, and therefore the true association between iAs exposure and the outcome may be even stronger. A caveat is that the exposure‐response found is strange, since the increase in relative risk per increase of water‐As is highest at the very low water‐As levels. This may be less biologically plausible.

In the studies by Rudnai et al. (2014) in Hungary, and Marie et al. (2018) in France there were larger proportions of the populations with high water As as compared to the studies by Zierler et al. (1988) and Richter et al. (2022). The results in these two studies support an association with congenital heart disease at water‐As > 10 μg/L, although Marie et al. (2018) found an increased risk only in girls.

All studies reviewed above find associations between iAs exposure and congenital heart disease. In summary, the epidemiological studies provide sufficient evidence for an association between low to moderate exposure to iAs and congenital heart disease. However, the exposure‐response relations in key studies were atypical.

Table 18 depicts the studies on As and congenital heart defects.

Water As and orofacial cleft

Suhl et al. (2018) performed a case–control study of OFC and As exposure in Iowa, US. 370 cases 1067 controls. There was no association between drinking water As and risk of OFC, but only about 4% had water As > 5 μg/L. Therefore, the impact of water As on total exposure to iAs should be modest. This single study with modest water‐As levels and no increased risk does not permit any conclusion on evidence regarding this birth defect.

In summary there is insufficient evidence of an association between exposure to iAs and OFC.

Water As and any birth defect

The study by Kwok et al. (2006) is too small to permit any conclusions. The study by Marie et al. (2018) found an association between water‐As and ‘any birth defect’ but only in girls.

In summary there is insufficient evidence of an association between exposure to iAs and ‘any’ birth defect.

Growth after birth

Four cohort studies including mother–child pairs investigated the associations between iAs and children's growth, two of which were from Bangladesh and one from the US (Gardner et al., 2013; Igra et al., 2021; Muse et al., 2020; Saha et al., 2012). One of the studies measured concentrations of iAs in urine only in mothers during pregnancy, whereas the other two measured it also in the children. A study from Bangladesh who followed the children until the age of 2 years, found no associations when the quintiles with the highest concentrations of iAs in urine were compared with the quintiles with the lowest concentrations (Saha et al., 2012). Another study from Bangladesh found an association between concentration of iAs in urine and the children's weight at 5 years of age, most apparent among girls (Gardner et al., 2013). However, no association was observed between prenatal exposure (maternal urine at early gestation) and children's weight at 5 years of age. A third study from Bangladesh which followed the children until the age of 10 years did not observe associations between iAs and growth (Igra et al., 2021). A study from the US observed an association between concentrations of iAs in urine and increased length during the first year of life (Muse et al., 2020). However, although the association was statistically significant the magnitude of the effect was very small, each doubling of tAs increased the length among males with 0.12 cm and among females with 0.13 cm. In addition, no associations were observed in the study between concentrations of iAs in urine and weight gain, change in weight‐for‐length or head circumference.

Table 19 depicts the studies on children's growth.

In summary, there is insufficient evidence that iAs is associated with children's growth.

Neurodevelopmental effects

The previous EFSA CONTAM Opinion (EFSA CONTAM Panel, 2009), described eight studies investigating associations between iAs exposure and developmental effects (cognitive performance) (Calderon et al., 2001; Rosado et al., 2007; Tofail et al., 2009; Tseng et al., 2006; von Ehrenstein et al., 2007; Wang, Wang, et al., 2007; Wasserman et al., 2004, 2007). Based on those studies, the CONTAM Panel concluded ‘Taken together, these studies provide some evidence for neurobehavioural effects of arsenic during childhood, also at exposure levels occurring in areas with elevated arsenic exposure via drinking water. More longitudinal studies are warranted to evaluate the most critical windows of exposure, the type of effects and dose‐response relationships’.

Since then, many studies have been published but most are cross‐sectional and relatively small. Different tests for cognitive, behavioural or motor/sensory function in children have been applied and at different ages, but the majority of studies have evaluated arsenic exposure vs. cognitive function and in the age span 5–11 years. Further studies have evaluated arsenic exposure and the neurodevelopmental outcomes autism, attention deficit hyperactivity disorder (ADHD).

For the current evaluation, the CONTAM Panel identified 59 studies, of which 20 met the criteria, including five (Rosado et al., 2007; Tofail et al., 2009; von Ehrenstein et al., 2007; Wasserman et al., 2004, 2007) that were already reviewed in the 2009 Opinion.

Of the selected studies, 12 were cross‐sectional, one was case–control and seven were cohort studies (i.e. prospective mother–child cohorts). Thirty‐nine further studies were not selected as they did not meet the inclusion criteria (Calderon et al., 2001; Cusick et al., 2018; de Assis Araujo et al., 2022; de Water et al., 2022; do Nascimento et al., 2015; Forns et al., 2014; Freire et al., 2018; Hsueh et al., 2020; Huang et al., 2011; Jiang et al., 2018, 2022; Khan et al., 2012, 2019; Langley et al., 2015; Levin‐Schwartz et al., 2019; Lin et al., 2013; Lucchini et al., 2019; McDermott et al., 2012; Merced‐Nieves et al., 2022; Nahar, Inaoka, & Fujimura, 2014; Nahar, Inaoka, Fujimura, Watanabe, et al., 2014; Nozadi et al., 2021; Nyanza et al., 2021; Parajuli et al., 2013, 2014, 2015; Rahman et al., 2015; Renzetti et al., 2021; Rocha‐Amador et al., 2007; Rodríguez‐Barranco et al., 2016; Saxena et al., 2022; Valeri et al., 2017; Vibol et al., 2015; Wang, Wang, et al., 2007; Wang, Liu, et al., 2018; Wang, Wang, & Yan, 2022; Wasserman et al., 2011; Wright et al., 2006). Table 20 summarises the selected studies from both the 2009 Opinion and the current evaluation.

Prenatal arsenic exposure and cognition

In a study from India on iAs and intellectual function in 5–15 year‐old children assessed both arsenic exposure during pregnancy (mean water arsenic concentrations 110 μg/L) and childhood (mean 147 μg/L) (von Ehrenstein et al., 2007). There were no significant associations with either exposure time window when using arsenic in water as the exposure matrix. Associations were found with child urinary As, but it was not speciated for arsenic metabolites.

Rodrigues et al. (2016) evaluated water As concentrations from pregnancy up to 20–40 months in two regions (Sirajdikhan and Pabna) in Bangladesh and neurodevelopment. No associations were found between concurrent water arsenic (median Sirajdikhan 1.5 μg/L and Pabna 25.7 μg/L) and children's scores for receptive and expressive language, or gross motor domains (not in table). Increasing concurrent water arsenic concentrations were associated with decreasing cognitive scores in Pabna but not in Sirajdikhan. Similar and slightly stronger associations were observed between first trimester water arsenic concentrations and cognitive scores in Pabna.

In a large and well‐characterised mother–child cohort in Matlab in rural Bangladesh, arsenic concentrations were evaluated vs. neurodevelopment at different ages: at 7 months (Tofail et al., 2009), 18 months (Hamadani et al., 2010), 5 years of age (Hamadani et al., 2011), and 10 years of age (Vahter et al., 2020, in table). In Vahter et al. (2020) cognitive abilities of children were assessed in relation to arsenic in maternal urine in early pregnancy (gestational week 8), and in child urine at 5 and 10 years, and in hair at 10 years. Both early maternal urine (82 μg/L) and child arsenic exposure (57 μg/L) at 10 years, were at low levels (around 50 μg/L) inversely associated with cognitive abilities measured as full developmental score at 10 years. Associations were found in the 2nd and 3rd quintiles. Models with children's hair arsenic concentrations showed similar results (not in Table 20).

Soler‐Blasco et al. (2022) evaluated maternal urinary arsenic (geometric mean 7.78, 95% CI 7.41, 8.17 μg/g creatinine), in the first trimester and neurodevelopment (McCarthy Scales of Children's Abilities) in Spanish 4–5 years‐old children. They found inverse associations between MMA concentrations and the scores for the general, verbal, quantitative, memory and working memory scales. However, no associations were found between total iAs and developmental scales in multivariate models.

Postnatal exposure and cognition

In studies from Araihazar in Bangladesh, exposure to arsenic from drinking water (mean appr. 120 μg/L across the studies) was evaluated in relation to intellectual function at different ages. In 9–10‐year‐old children, increasing water arsenic was associated with reduced intellectual function (full scale score), in a dose‐dependent manner (Figure 1 in article, though the effect estimate for the second quartile was not provided; Wasserman et al., 2004). In 6‐year‐old children, increasing water arsenic was negatively associated with full scale scores, and the associations were less strong than in the previously studied 9–10‐year‐olds (Wasserman et al., 2007). In 14–16‐year‐old children, urinary arsenic showed negative associations with all Wechsler test scores, but it is unclear whether the urinary arsenic was corrected for organic arsenic and there are no effect estimates for the dose response analysis (Wasserman et al., 2018).

Rosado et al. (2007) found in 6–7‐year‐old Mexican children inverse associations between urinary arsenic (mean 58.1 μg/L) and cognition measured as visual–spatial abilities with Figure Design, the Peabody Picture Vocabulary Test, the WISC‐RM Digit Span subscale, visual search and letter sequencing tests.

In a study of Taiwanese children with and without developmental delay increased risk (with very wide confidence interval) for developmental delay was found in the highest tertile (> 24.71 μg/creatinine) of child urinary arsenic (Hsieh et al., 2014).

In a study on 3–5 grade school children (Wasserman et al., 2014) in the USA, associations between drinking water As (mean 9.88 μg/L) and IQ were examined. Water arsenic was negatively associated with Full Scale IQ. This association was strongest in the exposure category 5–10 μg/L arsenic in water, but no dose–response was found.

Three studies from Uruguay (Desai et al., 2018; Desai, Barg, Vahter, Queirolo, Peregalli, Mañay, Millen, Yu, Browne, & Kordas, 2020; Desai, Barg, Vahter, Queirolo, Peregalli, Mañay, Millen, Yu, & Kordas, 2020) were based on the same study population of 5–8‐year‐old children (median appr. 10 μg/L). Two of the studies did not find associations between arsenic and cognitive functions (Desai et al., 2018; Desai, Barg, Vahter, Queirolo, Peregalli, Mañay, Millen, Yu, Browne, & Kordas, 2020), whereas one found adverse associations with different measures of executive function (Desai, Barg, Vahter, Queirolo, Peregalli, Mañay, Millen, Yu, & Kordas, 2020).

Postnatal exposure motor/ sensory functions and behaviour

Parvez et al. (2011) investigated the relation between arsenic in water (mean 43.3 μg/L) and motor function in 8–11‐year‐old children and found lower total motor composite scores with increasing arsenic concentrations.

Signes‐Pastor, Vioque, et al. (2019) evaluated arsenic exposure measured in urine and neuropsychological development among Spanish children of 4–5 years of age. Urinary As concentrations (median < 10 μg/L) were negatively associated with the scores of global, gross and fine motor function. They found no associations between iAs and cognitive function.

Emotional and behavioural symptoms at the subclinical level raise the risk of subsequent development of mental disorders and can be assessed during pre‐adolescence through internalising and externalising problems indicators. Roy et al. (2011) found in a study of Mexican 6–7‐year‐old children an association between urinary arsenic (median 55.2 μg/L) in the 2nd quartile (range 36–55.2 μg/L) and teachers' ratings of oppositional behaviour, but there was no dose–response.

Khan et al. (2011) evaluated water arsenic concentrations and classroom behaviour via teacher's report of internalising and externalising problems in 8–11‐year‐old children in Bangladesh, and found no association with As and these outcomes.

In summary, epidemiological studies published since the EFSA Opinion 2009 provide sufficient evidence for a causal association between low to moderate exposure to iAs and impaired cognition. The study by Vahter et al. (2020) is the largest and longitudinal study which reports results both for prenatal exposure and postnatal exposure and there may be different BMD depending on the timing of exposure.

There is insufficient evidence for associations between low to moderate exposure to iAs and motor function and behaviour.

Arsenic and autism spectrum disorder

For the present Opinion, the CONTAM Panel identified 27 studies on associations between As exposure and autism spectrum disorder (ASD) from a literature search (see Section 2 on methodology). Out of these, 16 studies did not meet the inclusion criteria and were thus not considered further (Adams et al., 2013; Amadi et al., 2022; Baj et al., 2021; Elsheshtawy et al., 2011; Fiore et al., 2020; Geier et al., 2012; Li, Li, et al., 2018; Majewska et al., 2010; Rahbar et al., 2012, 2021; Rezaei et al., 2022; Skogheim et al., 2021; Vergani et al., 2011; Wang, Hossain, et al., 2019; Yorbik et al., 2009; Zhao et al., 2022). Eleven studies that fulfilled the inclusion criteria are described in detail in Table 21.

In a case control study with Arab children, children with ASD had higher concentrations of As in hair compared with controls (Blaurock‐Busch et al., 2011). Also, in an expanded population of the previous study, hair As concentrations were significantly higher in children with ASD compared with controls (Blaurock‐Busch et al., 2012). In two further case control studies of Arab children, one study showed higher hair As among children with ASD compared with controls (Al‐Ayadhi, 2005), but no differences were found in the second study (Fido & Al‐Saad, 2016).

In case–control studies performed in Poland, USA and China, children with ASD had higher hair As concentrations compared with controls (Fiłon et al., 2020; Obrenovich et al., 2011; Zhai et al., 2019).

In a case–control study in Russia, hair As was not significantly different in children with ASD compared with controls (Skalny et al., 2017). In another and larger case control study from the same authors (possibly related to the previous study although there is no statement in the manuscript), hair As concentrations were not significantly different in children with ASD compared with controls (Skalny et al., 2016). Also, in a study in Italy, hair As concentrations did not differ between children with ASD and controls (De Palma et al., 2012), whereas in a US study, hair As concentrations were significantly lower children with ASD than in matched controls (Kern et al., 2007).

Overall, six studies found higher hair As levels in ASD cases, while four did not find any significant differences in hair As levels and one found lower levels of hair As in ASD cases compared to controls. All studies are case–control studies with relatively few cases/controls. This is important as the case–control study is not a relevant study design for a disease likely affecting life and family/child habits. All studies used hair As as exposure metric. It should be noted that hair As in some of the studies was surprisingly high, whereas hair As levels in most of the studies where no association was seen with autism as well as other investigations presented in this Opinion are below 0.1 μg/g.

In summary, there is insufficient evidence for an association between low to moderate exposure to iAs and ASD.

3.2.2.5. Effects on male fertility

Associations between iAs and male fertility were not discussed in EFSA CONTAM Opinion (2009). For the current Opinion, the CONTAM Panel identified eight studies from a literature search, of which one study from China, summarised in Table 22, met the inclusion criteria (Wang et al., 2016). The study did observe a significant association between u‐tiAs and unexplained male infertility. However, very wide confidence intervals clearly indicate the uncertainty in the risk estimates.

Six additional studies were also evaluated but did not fulfil the inclusion criteria and were accordingly not considered further (Calogero et al., 2021; Oguri et al., 2016; Sengupta et al., 2013; Sukhn et al., 2018; Tian et al., 2021; Xu et al., 2012).

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and effects on male fertility.

3.2.2.6. Neurotoxicity

EFSA CONTAM Panel (2009) concluded that the available epidemiological studies indicated a relationship between high level oral exposures to iAs and sensitive endpoints for peripheral and central neurotoxicity.

Peripheral neuropathy is observed both at acute and chronic high exposure to arsenic. The clinical features of the arsenic‐linked neuropathy are paresthesias, numbness and pain, particularly in the hands and the soles of the feet. In many cases a symmetrical peripheral neuropathy is one of the earliest symptoms of arsenic poisoning.

The CONTAM Panel identified 21 studies, of which 10 studies met the criteria. The relevant studies are summarised in Table 23. In 11 papers, arsenic and neurotoxicity were reported but the studies were not considered further, as they did not meet the inclusion criteria (Guha Mazumder et al., 2010; Ishii et al., 2017, 2018; Khan et al., 2019; Mathew et al., 2010; Guha Mazumder et al., 2010; OʼBryant et al., 2011; Peters et al., 2021; Shiue, 2015; Szabo et al., 2016; Valappil & Mammen, 2019; Yorifuji et al., 2016).

Mukherjee et al. (2014) found in pre‐menopausal women of eastern India a higher risk of symptoms from the peripheral nervous system (for example burning sensation in extremities, tingling or numbness, reduced sense of taste and reduced sense of smell) in women who had been drinking water contaminated with As concentrations of 11–50 μg/L compared with women drinking water with As < 10 μg/L. Mochizuki et al. (2019) found in a study on adults and children in Myanmar that more subjective symptoms, possibly due to peripheral neuropathy, were reported at arsenic concentrations in drinking water > 10 μg/L compared with at concentrations < 10 μg/L. Objective peripheral nerve disturbances of both small and large fibres occurred at > 50 μg/L. However, no adjustments were performed in this study and the exposure assessment was crude.

Low chronic exposure to arsenic and associations with impairment of the central nervous system have been evaluated. Mukherjee et al. (2014) found a higher prevalence of depression and other neurobehavioral symptoms in the same pre‐menopausal women mentioned above. Rahman, Niemann, and Yusuf (2022) evaluated in the NHANES survey for years 2015–16, associations between arsenic concentrations in urine and sleep disturbance. Compared with individuals with urinary arsenous acid below the lower level of detection, those with urinary arsenous acid at or above the detection limit had lower odds of sleep disturbance. There was no association for any other metabolites (total arsenic metabolites was not evaluated in the statistical analysis). Liu et al. (2017) reported in a study of Chinese men and women above 40 years, higher risk of cognitive impairment with higher arsenic in drinking water and the impairment was observed already at water arsenic concentrations of 11–50 μg/L compared to concentrations < 10 μg/L. Cognitive impairment with higher arsenic concentrations in drinking water was also found in adults from Bangladesh (Karim et al., 2019) exposed to drinking water with arsenic ranging 5.32–167.94 μg/L. Wang, Huang, et al. (2021) showed in a study from China higher risk of cognitive impairment with higher arsenic concentrations in hair, with a dose–response. The exposure assessment is though difficult in this study as the study participants lived in an area with historically high arsenic concentrations in drinking water polluted from a nearby plant that closed in 2011.

In three studies from Texas, arsenic exposure was estimated by a method based on geographic information system using information on arsenic in ground water (Edwards, Hall, et al., 2014; Edwards, Johnson, et al., 2014; Gong et al., 2011). The authors also reported on a comparison between estimated and measured arsenic values from seven wells and found a reasonable agreement (Edwards, Johnson, et al., 2014). They associated the estimated arsenic levels with different outcomes related to neurophysiological functions. However, the exposure assessment was rather crude. In Edwards, Johnson, et al. (2014), participants with Alzheimer's disease and mild cognitive impairment, were compared with participants with normal cognition. In the whole group, arsenic concentrations were positively associated with language abilities, but also associated with poorer verbal memory, immediate and delayed, as well as poorer visual memory, immediate and delayed. No association was found with depression. In Edwards, Hall, et al. (2014), where rural dwelling adults and elders over 40 years were examined with a battery of neurophysiological tests, estimated higher arsenic concentrations in water were negatively associated with language and executive functioning, and weakly with cognition. Gong et al. (2011) also evaluated estimated arsenic concentrations vs. global cognition in the same study area and same age group (it is not clear if these two latter studies represent independent samples) and found that those with estimated arsenic exposure > 10.0 μg/L had lower MMSE scores than those with arsenic exposure ≤ 10.0 μg/L). However, there were no adjustments in these analyses and no effect estimates presented (not in table).

Several studies have analysed the associations between arsenic and levels of different neurotransmitters or cholinesterase. Ali et al. (2010) investigated the association between arsenic exposure (measured in water) and the activity of plasma cholinesterase in Bangladeshi adults. When dividing the study participants in three exposure groups (low < 129 μg/L arsenic in drinking water; medium 130–264 μg/L and high > 265 μg/L), there was a dose–response with the highest plasma cholinesterase activity in the low exposure group followed by the medium and the high exposure groups (all group comparisons were statistically significant, but no exact values were provided for these comparisons). When dividing study participants according to ≤ 50 and > 50 μg/L As in drinking water, also a lower plasma cholinesterase activity was found in the higher arsenic exposure group. Plasma norepinephrine is used as a measure of sympathetic activity. In the already mentioned Mukherjee et al. (2014) study, levels of plasma neurotransmitters epinephrine, norepinephrine and serotonin were higher among premenopausal women exposed to water arsenic concentrations ranging 11–50 μg/L compared with women exposed to concentrations < 10 μg/L. The authors interpreted these findings as a stress response of the nervous system. In Karim et al. (2019), serum levels of the brain‐derived neurotrophic factor (BDNF), a growth and survival factor for nervous cells, was lower with higher arsenic in water. BDNF and the Mini‐Mental State Examination correlated positively, suggesting that arsenic may mediate its neurotoxicity via lower BDNF levels.

Overall, the available epidemiological studies (particularly Mukherjee et al., 2014; and Liu et al., 2017) indicate a relationship between low level of exposure to iAs and sensitive endpoints for both peripheral and central neurotoxicity. Moreover, studies including biomarkers of effect or mediation support the associations between low‐level arsenic and neurotoxicity. Nevertheless, further epidemiological studies are warranted. It should be noted that in many of the studies, study participants may have been exposed since early in life, and the effects observed may be the result of exposure during developmental periods of the central and the peripheral nervous systems. In Table 23 the studies on neurotoxicity in adults are described.

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and neurotoxicity.

3.2.2.7. Effects on the cardiovascular system

In the previous EFSA Opinion, 13 studies on cardiovascular disease were reviewed. All of them were retrieved from a systematic review by Navas‐Acien et al. (2005). Eight of the 13 studies were performed in Taiwan with very high exposure to As in drinking water. Among the other five studies four (Engel & Smith, 1994; Lewis et al., 1999; Ruiz‐Navarro et al., 1998; Varsanyi et al., 1991) did not fulfil our present inclusion criteria. One of them (Zierold et al., 2004) fulfilled the inclusion criteria, but this study has the risk of information bias since heart disease and hypertension were self‐reported, and many of the 1185 respondents may have known if their water‐As was high.

For the present Opinion, the CONTAM Panel identified 87 additional studies from a literature search and from nine systematic reviews (Abhyankar et al., 2012; Bao et al., 2022; Chowdhury et al., 2018; Karachaliou et al., 2021; Moon et al., 2012, 2017; Tsuji et al., 2014; Xu, Mondal, & Polya, 2020; Zhao et al., 2021). Out of these, 53 studies (Afridi et al., 2011; Ameer et al., 2015; Castiello et al., 2020; Cheng et al., 2021; Dastgiri et al., 2010; Desai et al., 2021; Dong et al., 2022; Gong & O'Bryant, 2012; Guha Mazumder et al., 2012; Gunduz et al., 2015; Guo et al., 2007; Hall et al., 2017; Jovanović et al., 2012; Liao et al., 2012; Lin, Hsu, et al., 2018; Lisabeth et al., 2010; Liu et al., 2022; McLeod et al., 2018; Meliker et al., 2007; Merrill et al., 2017; Nong et al., 2016; Nunes et al., 2022; Pichler et al., 2019; Qu et al., 2022; Rahman et al., 1999, 2014; Rahman, Niemann, & Munson‐McGee, 2022c; Shiue, 2014; Sohel et al., 2009; Soria et al., 2021; Stea et al., 2016; Tang et al., 2022; Wang et al., 2002; Wang et al., 2011; Wang, Hao, et al., 2020; Wang, Karvonen‐Gutierrez, et al., 2021; Wang, Wu, et al., 2007; Wei, Yu, et al., 2017; Wen et al., 2019; Wu et al., 2010; Xu et al., 2021; Xu, Liu, et al., 2022; Xu, Polya, et al., 2020; Yao et al., 2021; Yen et al., 2022; Yildiz et al., 2008; Yu et al., 2017; Yuan et al., 2017; Zhang et al., 2013, 2022; Zhong et al., 2019) did not fulfil the inclusion criteria.

The 34 additional studies that fulfilled the inclusion criteria examined ischemic heart disease (including the diagnoses of myocardial infarction and coronary heart disease), stroke, hypertension/blood pressure, atherosclerosis and overall cardiovascular disease.

Ischemic heart disease (IHD)

The 15 studies are summarised in Table 24. Among those based on As in water the studies performed in Inner Mongolia (Wade et al., 2009, 2015), Bangladesh (Chen, Graziano, et al., 2011; Chen, Wu, Liu, et al., 2013; Wu et al., 2015), Colorado (USA), (James et al., 2015) provide support for an association between water As and risk of IHD. The large study in Denmark (Monrad et al., 2017) provides only limited support for an association (in the Aarhus subcohort). Among the studies based on As in urine, the Strong Heart Study in the USA (Kuo et al., 2022; Moon et al., 2013), and the study of US NHANES (Nigra et al., 2021) also support an association with IHD. Analyses based on As content in toenails (Farzan et al., 2015; Wade et al., 2015) show conflicting results. The ecological (Medrano et al., 2010) and semi‐ecological (D'Ippoliti et al., 2015) studies are considered less reliable, as are the studies with a small/unknown number of self‐reported disease cases (Butts et al., 2015; Zierold et al., 2004).

Two studies were performed in the Ba Men region of Inner Mongolia, China, an agricultural region with elevated levels of As in well water. The first study was a retrospective cohort study in a village with about 12,000 inhabitants (Wade et al., 2009). Cause‐specific mortality was examined during 1997 and 2004. As in the primary water source of each household was analysed. Adjusted rate ratios for mortality in heart disease was analysed by water‐As categories. The adjusted incidence rate was significantly increased when water‐As was treated as a continuous variable. For those exposed from the same water source since 1995, the IRR was 1.19 (95% CI 1.05–1.33) per increase of water As by 50 μg/L, based on 161 deaths. IRRs by categories of water As are shown in Table 24.

The next study from Ba Men was a case–control study (Wade et al., 2015). Cases were hospital patients with sign of ischemic heart disease (myocardial infarction, decreased ejection fraction without valvular disease or ECG changes indicating angina pectoris), and controls were patients from the same hospital with conditions unrelated to As. Cases and controls were recruited 2006–2011. The adjusted OR per 10 μg/L increase of water As was 1.19 (1.03–1.38). The adjusted OR per 1 μg/g increase of toenail As was 1.16 (0.98–1.38). ORs by categories of water As and nail‐As are shown in Table 24.

Three studies were performed in Bangladesh as part of the HEALS project (Chen, Graziano, et al., 2011; Chen, Wu, Liu, et al., 2013; Wu et al., 2015). The first one (Chen, Graziano, et al., 2011) was a prospective cohort study of about 12,000 inhabitants in a defined area, recruited in 2000–2002. Mortality was followed up until 2009 (validated verbal autopsy). As in water was measured in the tube wells (about 6000 wells) used by cohort members, and 96% of participants also delivered a spot urine sample of analysis of total As. The adjusted trend for the HR regarding IHD mortality (71 cases), with water As as a continuous variable, was statistically significant (p = 0.029), and for u‐tiAs as a continuous variable the corresponding p‐value was 0.059. HRs by categories of water As and u‐tiAs are shown in Table 24.

In the second paper (Chen, Wu, Liu, et al., 2013), the incidence of CVD (211 cases, mainly IHD) was studied using a case cohort design. The adjusted HR with water As as a continuous variable was statistically significant. HRs by categories of water As are shown in Table 24. There was also a significant association between MMA% in urine and HR for IHD.

In the third study of the HEALS cohort (Wu et al., 2015), the follow‐up time was longer (2012) and the analysis was based on 237 cases of IHD. There was a significant association between water As as a continuous variable and the adjusted HR for CHD. HRs by categories of water As are shown in Table 24.

A case cohort study by James et al. (2015) in Colorado, US included 96 cases of CHD and showed a strong association between the time‐weighted average lifetime As in water and risk of IHD. The HR was significantly increased already at 30–45 μg/L.

Monrad et al. (2017) performed a very large cohort study in two Danish cities (Copenhagen and Aarhus). There were 2707 incident cases of myocardial infarction. The As levels in drinking water were, however, low. In Copenhagen, about 60% of the population had As in water 0.5 μg/L, and none had levels > 2 μg/L. In Aarhus about 80% had As in water 2 μg/L, about 5% higher than that, and the remaining had As in water < 2 μg/L. Therefore, the analysis of the total cohort will to a large extent be a comparison between two cities. Even if the Aarhus cohort includes a larger range of water As levels, the exposure contrast is still limited. The medians in the third and the fourth quartile of water As is the same.

Two studies using u‐iAs as exposure metric are based on the Strong Heart Study among American Indians in the US. The study by Moon et al. (2013) shows significantly increased risk of CHD incidence and mortality in the upper quartile (Table 24). The study by Kuo et al. (2022) found a significant association between water As and the broader outcome of CVD. The u‐iAs were not very high; the median in upper quartile was 22 μg/g creatinine (Moon et al., 2013). According to the authors, As in water was the main source of iAs exposure in two of the three Strong Heart Study areas (Arizona and the Dakotas). This is consistent with the median inorganic fraction of u‐iAs (8%), and the median MMA of 15% and DMA 78%.

The study by Nigra et al. (2021) based on u‐tiAs after exclusion of those with u‐AsB > 1.2 μg/L provides limited support for an association with heart disease mortality, but the study is relatively small.

Most of the longitudinal studies described above are large, and all of them adjusted their risk estimates for the most important potential confounders. There are no obvious sources of selection bias or information bias.

Several of the studies mentioned above (James et al., 2015; Wade et al., 2015; Wu et al., 2015 ) present results for relatively large data sets with strata of exposure to inorganic As, as concentrations in water. These studies were performed in several different countries (China, US, Bangladesh). The only European study was the Danish one, which also had the lowest exposure levels, but had a narrow range of exposure. Other studies, using As in water, were either based on the same study areas but with fewer cases (Chen, Graziano, et al., 2011; Chen, Wu, Liu, et al., 2013; Wade et al., 2009) or considered having a higher risk of bias (Butts et al., 2015; D'Ippoliti et al., 2015; Medrano et al., 2010; Zierold et al., 2004). Next to Wade et al. (2015), James et al. (2015) and Wu et al. (2015), two studies using u‐tiAs (Moon et al., 2013; Nigra et al., 2021) have also been considered having an appropriate design.

The studies on ischemic heart disease meeting the inclusion criteria are described in Table 24.

In summary, the epidemiological studies provide sufficient evidence for an association between low to moderate exposure to iAs and IHD.

Stroke

The eight studies are summarised in Table 25. Among those based on As in water, the HEALS study in Bangladesh (Wu et al., 2015) showed an association with stroke incidence in the highest stratum (water As 86–864, mean 191 μg/L), while in the stratum of 17–85 μg/L the point estimate was positive (HR 1.14) but with a broad confidence interval (0.65–1.98). The study performed in Inner Mongolia (Wade et al., 2009) showed no association with stroke mortality up to 100 μg/L. The large Danish study (Ersbøll et al., 2018) showed an association in the highest quartile in Aarhus, but the dose–response is difficult to interpret since most cases in the upper quartile had the same water As concentration (2.1 μg/L). The ecological study by Medrano et al. (2010) and the study with an unknown number of self‐reported disease cases (Zierold et al., 2004) are considered less informative. The US studies based on u‐As (Tsinovoi et al., 2018) or As in toenails (Farzan et al., 2015) had very few cases. The studies on stroke are described in Table 25.

In summary, the epidemiological studies provide insufficient evidence for an association between low to moderate exposure to iAs and stroke.

Atherosclerosis

The eight studies are summarised in Table 26. Among those based on As in water, the Taiwanese studies (the largest one by Hsieh et al., 2011) showed clear associations with carotid artery atherosclerosis (increased intima‐media thickness (IMT and/or plaque). The study in Bangladesh (Chen, Wu, Graziano, et al., 2013) also provides some support, but the association with intima‐media thickness (IMT) was not quite significant on the 5% level.

Among the studies based on As in urine the study by Chen, Wu, Graziano, et al. (2013) found a significant association with IMT, and the study in Spain (Grau‐Perez et al., 2022) found a significant association with IMT and plaque, while in the US SHS study (Mateen et al., 2017) and the US MESA study with very low u‐iAs (Sobel et al., 2020) there was no association with IMT or plaque. The study by Grau‐Perez had very low levels of u‐tiAs but high u‐AsB. Although the authors ‘corrected’ for the potential contribution of DMA from seafood to u‐tiAs, the impact of iAs exposure on the carotid artery findings remain uncertain.

The data on atherosclerosis in the lower limbs (ABI) and the coronary arteries (CAC score) are limited.

In summary, the epidemiological studies based on As in water provide sufficient evidence for a causal association between moderate exposure to iAs and carotid artery atherosclerosis. For other arteries, the evidence is insufficient.

Blood pressure and hypertension

The 11 studies are summarised in Table 27. Eight of them (Chen et al., 2007; Hossain et al., 2017; Islam, Khan, Attia, et al., 2012; Li, Li, Xi, Zheng, Lv, & Sun, 2013; Li, Li, Xi, Zheng, Wang, & Sun, 2013; Mendez et al., 2016; Mordukhovich et al., 2012; Zierold et al., 2004) were cross‐sectional. Two studies (Jiang et al., 2015; Spratlen et al., 2018) had a longitudinal design, and one (Kaufman et al., 2021) presented both cross‐sectional and longitudinal analyses. In most of the studies, the outcome was hypertension, but the longitudinal study by Jiang et al. (2015) and the cross‐sectional studies by Mordukhovich et al. (2012) and Li, Li, Xi, Zheng, Wang, and Sun (2013), examined associations between iAs exposure and blood pressure.

The study by Zierold et al. (2004) which had high risk of information bias, and the study by Hossain et al. (2017) which had few cases were considered less informative.

The longitudinal study of blood pressure by Jiang et al. (2015) provides relatively strong support for the hypothesis that iAs exposure increases blood pressure. Blood pressure was measured with appropriate methods at baseline and on several follow‐up examinations in > 10,000 individuals. Systolic and diastolic blood pressure increased more in individuals with baseline water As in quartiles 2–4 than in the referent quartile (As < 12 μg/L), but there was no monotonic exposure‐response relationship (Table 27). The finding is supported by the smaller cross‐sectional study by Li, Li, Xi, Zheng, Wang, & Sun (2013), which found significant partial correlation coefficients between u‐tiAs (and iAs, MMA, DMA) and SBP, and the study of nail As (Mordukhovich et al., 2012), but not by the study by Kaufman et al. (2021) which found no association between baseline u‐tiAs and change of systolic blood pressure (SBP) or diastolic blood pressure (DBP).

An association between iAs exposure and increased blood pressure is biologically plausible. Vascular toxicity at As exposure is well‐known (causing the so‐called Blackfoot disease). In vitro work has shown that arsenic promotes inflammatory activity, oxidative stress and endothelial dysfunction through several mechanisms, including the activation of stress‐response transcription factors such as activator protein‐1 and nuclear factor‐κB (Abhyankar et al., 2012).

Although there is considerable support for an association, it is based on relatively few studies, one study (Kaufman et al., 2021) showed no association, and the study with the highest quality (Jiang et al., 2015) showed no monotonic relationship. Therefore, the epidemiological studies provide insufficient evidence of a causal association between iAs exposure and blood pressure.

Hypertension

Among the studies of hypertension, two studies from Bangladesh, a very large one by Chen et al. (2007) and a small study by Islam, Khan, Attia, et al. (2012) found no associations with As in water. Two studies from China in areas with known As contamination of drinking water found clear associations between prevalence of hypertension and As in urine (Li, Li, Xi, Zheng, Wang, & Sun, 2013) and water (2013b). A study in Mexico (Mendez et al., 2016) found a somewhat elevated odds ratio for hypertension in the upper quartile of water As, but it was not statistically significant. Spratlen et al. (2018) found no increased incidence of hypertension versus u‐tiAs in a follow‐up of the US Strong Heart Study. Kaufman et al. (2021) analysed an extension of the same cohort (Strong Heart Family Study) and found a significant association between u‐tiAs and prevalent hypertension at baseline, but no increased risk of incident hypertension in the longitudinal analysis. This study had, however, very low u‐iAs concentrations.

The studies of hypertension adjusted for important confounders. The study by Chen et al. (2007) excluded hypertension cases on medication, but such medication was rare (Jiang et al., 2015). The study by Li, Li, Xi, Zheng, Lv, and Sun (2013) adjusted (for unknown reasons) the analysis of hypertension risk vs. As in water by cumulative (years of exposure x As concentration) which likely attenuated the association. The findings in the cross‐sectional studies of hypertension were not consistent. Longitudinal studies are stronger when assessing causality. The longitudinal analyses of incident hypertension in the US SHS and SFHS (Kaufman et al., 2021; Spratlen et al., 2018) showed no association with u‐tiAs.

In summary, the epidemiological studies provide insufficient evidence of an association between iAs exposure and hypertension.

The studies on blood pressure and hypertension are described in Table 27.

3.2.2.8. Respiratory disease

In the previous EFSA Opinion (EFSA CONTAM Panel, 2009) studies on respiratory disease and arsenic exposure were not identified. For the present opinion, the CONTAM Panel identified 26 studies from a literature search. Out of these, eight studies did not meet the inclusion criteria (Bhattacharyya et al., 2014; Liao et al., 2011; Majumdar & Guha Mazumder, 2012; Sanchez et al., 2018; Scannell Bryan et al., 2021; Shih et al., 2019; Smith et al., 2011; Zhou, Wang, et al., 2022). The 19 studies that fulfilled the inclusion criteria (see Section 3.2.2.1) criteria were further considered (Table 28).