Arsenic Risk Assessment and Cancer Risk

Chemical-Specific Risk Assessment Methodology — Arsenic

Arsenic Risk Assessment Methodology

RoseyFields conducts Arsenic risk assessments using the following methodology. This document summarizes the chemical-specific hazard identification, dose-response, and toxicity value derivation approach applied by RoseyFields when arsenic is a chemical of concern at a site or in an exposure reconstruction, litigation support, or regulatory compliance matter. It is intended for use by RoseyFields risk assessors, testifying experts, and reviewing clients.

1. Chemical Identity, Regulatory Background, and Exposure Sources

Arsenic (As, atomic number 33) is a naturally occurring metalloid distributed throughout the Earth’s crust in more than 200 mineral species. In environmental and toxicological contexts, the health-relevant form is inorganic arsenic (iAs), occurring predominantly as trivalent arsenite (AsIII) and pentavalent arsenate (AsV) in soil and water; organic arsenicals such as arsenobetaine and arsenosugars, common in seafood, are of substantially lower toxicological concern and are treated separately by RoseyFields in dietary exposure reconstructions.

Regulatory background. Inorganic arsenic has been evaluated under EPA’s Integrated Risk Information System (IRIS) program since 1991 (noncancer reference dose) and 1995 (cancer classification and slope factor). RoseyFields’ current methodology incorporates EPA’s most recent IRIS reassessment, which represents a substantial departure from the legacy approach. The original 1991/1995 toxicity values were both anchored to a single historical dataset: skin lesions (hyperpigmentation, hyperkeratosis) and skin cancer observed in a chronically arsenic-exposed population in the Blackfoot-disease-endemic region of Taiwan (Tseng et al., 1968; Tseng, 1977). The current assessment instead derives organ/system-specific toxicity values from independent bodies of epidemiological literature on cardiovascular disease, diabetes, and reproductive/developmental outcomes, and updates the cancer basis to bladder and lung cancer dose-response meta-analyses. RoseyFields tracks both the legacy and current values because many active sites, historical exposure windows, and remediation goals (e.g., maximum contaminant levels, preliminary remediation goals) were established under the earlier toxicity values, and the change in basis and magnitude is often material to causation and damages analyses.

Exposure sources and pathways. Oral exposure is the dominant pathway of concern in most RoseyFields matters. Drinking water is a principal source, particularly for populations served by private or domestic wells drawing on arsenic-rich geologic formations, which are not subject to the same regulatory oversight as public water systems. Dietary intake is a second major pathway; rice and rice-based products are a disproportionately important source of inorganic arsenic for infants and young children during the transition to solid foods (Signes-Pastor et al., 2016). Incidental soil ingestion and dust inhalation are relevant at contaminated sites (e.g., former smelting, wood-treatment, or pesticide-manufacturing facilities), while occupational inhalation exposure remains relevant in legacy smelter and glass-manufacturing cohorts. RoseyFields’ exposure assessments identify the dominant pathway(s) at a given site and reconstruct dose using site-specific environmental concentrations, land-use and activity patterns, and, where available, biomonitoring data.

2. Point-of-Departure (POD) Selection: Critical Studies and Endpoints

RoseyFields follows the multi-endpoint approach adopted in the current IRIS assessment, under which organ/system-specific points of departure are derived independently for each health outcome with a sufficiently robust epidemiological evidence base, rather than relying on a single critical study. Four noncancer endpoints and two cancer endpoints anchor the current toxicity values.

Ischemic heart disease (IHD). The POD is derived from a Bayesian hierarchical dose-response meta-analysis of five epidemiological studies: a Bangladeshi prospective cohort (Chen et al., 2013), a U.S. Strong Heart Study prospective cohort (Moon et al., 2013), a Colorado case-cohort study (James et al., 2015), an Inner Mongolian case-control study (Wade et al., 2015), and a Taiwanese cohort examining arsenic-related atherosclerosis risk (Wu et al., 2010). These studies were selected because they met high- or medium-confidence study evaluation criteria and, in combination, span diverse populations, exposure ranges, and case ascertainment methods (incident-only versus incident-plus-fatal IHD).

Diabetes (type 2). The POD derives from a four-study Bayesian meta-analysis: the Strong Heart Family Study (Grau-Perez et al., 2017), a Colorado case-cohort study (James et al., 2013), a Mexican cross-sectional study (Coronado-González et al., 2007), and a Bangladeshi cohort (Pan et al., 2013). As with IHD, these studies were judged to demonstrate a causal, dose-responsive relationship between iAs exposure and disease incidence at exposure levels relevant to background U.S. populations.

Birth weight. RoseyFields relies on a three-study meta-analysis (Kile et al., 2016; Liu et al., 2018; Rahman et al., 2009) supplemented by consideration of an additional Mexican pregnancy cohort (Laine et al., 2015), using a hybrid benchmark-response approach applied to the pooled association between maternal urinary total arsenic (tAs) and decreased fetal weight. Rahman et al. (2009), a prospective cohort of 1,578 mother-infant pairs in Matlab, Bangladesh, exemplifies the underlying data: maternal urinary arsenic collected at approximately gestational weeks 8 and 30 was inversely associated with birth weight. Because the single study POD derived from Kile et al. (2016) was more health-protective (lower) than the three-study meta analysis POD, RoseyFields — consistent with the current IRIS approach — carries the Kile et al. (2016) value forward as the organ-specific POD.

Developmental neurocognitive effects (IQ). The POD is based on a three-study meta-analysis of the association between prenatal iAs exposure (maternal urinary tAs) and childhood IQ score: a Spanish birth cohort (Soler Blasco et al., 2022), a U.S. HOME Study cohort (Signes-Pastor et al., 2022), and a Chinese cohort in a low arsenic-background area (Chen et al., 2023).

Cancer endpoints. Bladder and lung cancer remain the accepted cancer hazards of oral inorganic arsenic exposure. RoseyFields uses dose-response meta-analyses of the epidemiological literature underlying these outcomes (see Section 6) rather than the historical Taiwanese skin cancer dataset that anchored the 1995 slope factor.

The resulting PODs, expressed as benchmark dose lower confidence limits (BMDLs) converted to an equivalent total chronic oral dose, are summarized below.

3. Pharmacokinetic Modeling and Dose Conversion

Because the governing PODs for birth weight and IQ are derived from maternal urinary total arsenic biomarker concentrations rather than administered dose, RoseyFields converts these internal-dose PODs to an equivalent oral daily dose using a validated physiologically based pharmacokinetic (PBPK) model for inorganic arsenic, the El-Masri-Kenyon model (El-Masri and Kenyon, 2008; El-Masri et al., 2018). This model was selected over several alternative arsenic PBPK models because it incorporates more complex, independently derived metabolic parameters governing the sequential oxidative methylation of inorganic arsenic to monomethylated (MMA) and dimethylated (DMA) species (Kenyon, 2021).

The El-Masri-Kenyon model has been calibrated and evaluated against two large, independent human datasets — approximately 11,000 subjects in Bangladesh and 500 subjects in Nevada — with matched drinking-water exposure and urinary arsenic excretion measurements. Model evaluation across these populations established an approximately 1:1 empirical relationship between daily oral inorganic arsenic dose and daily total urinary arsenic excretion at exposures relevant to background and moderately elevated populations (El-Masri et al., 2018). RoseyFields uses this relationship bidirectionally: to convert epidemiological urinary-arsenic PODs into an equivalent daily oral dose for RfD derivation, and, in individual exposure reconstructions, to estimate a claimant’s historical daily intake from measured or reported urinary arsenic biomarker data.

RoseyFields does not rely on interspecies (animal-to-human) pharmacokinetic scaling for arsenic, because all governing noncancer and cancer PODs are derived directly from human epidemiological data. This is a deliberate and consequential methodological choice: arsenic metabolism (the balance of methylation versus retention of more toxic trivalent intermediates) varies substantially and unpredictably across species, such that rodent and primate metabolic patterns are poor predictors of human metabolism (Drobná et al., 2010). Reliance on human data for both hazard identification and dose-response quantification avoids this source of uncertainty and is the basis for setting the interspecies uncertainty factor (UFA) to 1 (see Section 4).

4. Uncertainty Factor Derivation

RoseyFields applies the standard five-factor uncertainty analysis framework (interspecies, intraspecies, subchronic-to-chronic, LOAEL-to-NOAEL, and database completeness) to each organ/system-specific candidate toxicity value. Because the governing PODs are drawn from human dose-response meta-analyses rather than single studies or animal bioassays, several of the traditional uncertainty factors are reduced relative to a conventional chemical-specific assessment. The factor values and RoseyFields’ basis for each are summarized below; identical values and rationale are applied across the diabetes, IHD, birth weight, and IQ score candidate values.

The resulting composite uncertainty factor (UFC) applied to each POD is 3. RoseyFields views the reduction of UFH from a conventional default of 10 to 3 as the most consequential and most carefully documented step in the uncertainty analysis, because it rests on a quantitative leave-one-out sensitivity analysis of the underlying meta-analyses rather than a categorical judgment, and because it is expressly limited to endpoints where the contributing studies already reflect substantial population heterogeneity and adjustment for known effect modifiers (smoking status, sex, nutritional status, genetic methylation capacity).

5. Derivation of the Oral Reference Dose (RfD)

Organ/system-specific reference doses (osRfDs) are calculated by dividing each POD by its composite uncertainty factor of 3: diabetes, 0.058 μg/kg-day (high confidence); IHD, 0.057 μg/kg-day (medium-high confidence); birth weight, 0.079 μg/kg-day, based on Kile et al. (2016) (medium-low confidence); and IQ score, 0.105 μg/kg-day (medium confidence). RoseyFields selected the Kile et al. (2016) single-study value over the higher (less protective) three-study birth-weight meta-analysis value (0.47 μg/kg-day) on health protectiveness grounds, notwithstanding a general preference for meta-analytic PODs, because the single study value was approximately six-fold lower.

RoseyFields’ overall RfD for inorganic arsenic is 0.06 μg/kg-day, based jointly on the diabetes and IHD organ specific values, which round to the same value at one significant figure (0.058 and 0.057 μg/kg-day, respectively) and carry the highest confidence ratings among the four candidate endpoints. This RfD represents an estimate of a chronic daily oral exposure, inclusive of sensitive subpopulations and life stages, that is not expected to be associated with an appreciable increased lifetime risk of diabetes or IHD above background. Overall confidence in the final RfD is medium-high.

RoseyFields flags for every matter that this value represents an approximately five-fold reduction from EPA’s legacy 1991 RfD of 0.3 μg/kg-day, which was based on a no-observed-adverse-effect level (NOAEL) for hyperpigmentation, keratosis, and possible vascular effects in the Taiwanese Blackfoot-disease population (Tseng et al., 1968; Tseng, 1977) with a composite uncertainty factor of only 3. Where a matter involves historical exposure periods or prior regulatory decisions premised on the legacy value, RoseyFields analyzes both values and clearly identifies which value is being applied and why.

6. Carcinogenicity Classification and Cancer Slope Factor Derivation

Inorganic arsenic is classified by EPA as a Group A human carcinogen based on sufficient evidence from human data, a classification first assigned in the 1995 IRIS assessment and unchanged since. The original 1995 cancer slope factor (CSF) of 1.5 (mg/kg-day)⁻¹ was derived from skin cancer prevalence in the same Taiwanese Blackfoot-disease population underlying the legacy RfD (Tseng et al., 1968; Tseng, 1977), using a multistage model with time as a covariate.

RoseyFields’ current methodology instead derives cancer slope factors from bladder and lung cancer, using Bayesian dose-response meta-analyses of the epidemiological literature for each tumor type. Slope factors are calculated as the slope of the linear trendline fit to the 95% upper confidence limit (UCL) on extra cancer risk at doses below approximately 0.2 μg/kg-day: bladder cancer, 1.76 × 10⁻² per μg/kg-day; lung cancer, 2.13 × 10⁻² per μg/kg-day. The 0.2 μg/kg-day upper bound was selected because it approximates the 95th percentile U.S. drinking-water arsenic exposure (Mendez et al., 2017, reporting a 95th-percentile drinking water concentration of 15.4 μg/L, translating to roughly 0.19 μg/kg-day at a standard water-consumption rate, plus an assumed 0.02 μg/kg-day dietary contribution), and therefore bounds the exposure range for which a linear approximation is reasonably reliable.

Because a claimant or site population may be at risk of either or both tumor types, RoseyFields calculates a combined cancer slope factor treating the two endpoint-specific slope factors as independent, normally distributed estimates. The combined CSF is calculated as the sum of the maximum-likelihood-estimate slopes plus 1.645 times the composite standard deviation of the two endpoint-specific estimates, yielding a combined CSF of 3.17 × 10⁻² per μg/kg-day (equivalently, approximately 31.7 per mg/kg-day).

Above 0.2 μg/kg-day, RoseyFields does not apply the linear CSF, because the underlying dose-response relationship for both tumor types becomes measurably nonlinear at higher doses and a linear extrapolation increasingly overstates or understates upper-bound risk. Instead, RoseyFields applies quadratic polynomial trendlines fit to the combined (bladder-and/or-lung) upper-bound lifetime extra risk, valid to approximately 1.0 μg/kg-day: extra risk ≈ 0.0243d² + 0.035d − 0.0001, where d is the daily oral dose in μg/kg-day. This dual approach — linear CSF below 0.2 μg/kg-day, polynomial risk equation above it — is used consistently across all RoseyFields cancer risk calculations for arsenic.

7. Practical Application: Hazard Quotients, Excess Lifetime Cancer Risk, and Litigation Support

Hazard Quotient / Hazard Index

For noncancer risk characterization, RoseyFields calculates a hazard quotient (HQ) as the estimated chronic daily oral dose (μg/kg-day) divided by the applicable reference dose. The overall RfD of 0.06 μg/kg-day is used for screening-level and general noncancer risk characterization. Where a cumulative risk assessment involves multiple co-occurring chemicals acting on a shared target organ or system (for example, a site with both arsenic and other cardiometabolic or developmental toxicants), RoseyFields uses the organ/system-specific osRfDs (diabetes, IHD, birth weight, or IQ, as appropriate) to construct a hazard index (HI) that sums HQs across chemicals sharing that endpoint, consistent with EPA cumulative risk assessment guidance.

Excess Lifetime Cancer Risk (ELCR)

For doses below 0.2 μg/kg-day, RoseyFields calculates excess lifetime cancer risk as the estimated chronic daily oral dose multiplied by the combined cancer slope factor (3.17 × 10⁻² per μg/kg-day), or by the tumor specific slope factor where only one tumor type is at issue. For doses above 0.2 μg/kg-day, RoseyFields substitutes the quadratic polynomial risk equation described in Section 6, since the linear approximation is not considered reliable at higher exposures.

Exposure and Dose Reconstruction

In litigation and historical exposure matters, RoseyFields reconstructs individual or population-level dose from available environmental sampling data (soil, groundwater, air), site-use and residency timelines, and, where available, biomonitoring data (urinary total arsenic), applying the El-Masri-Kenyon dose-urinary-excretion relationship (Section 3) to convert between biomarker concentrations and estimated daily intake. Reconstructed doses are then compared to the relevant osRfD or CSF to characterize noncancer hazard or excess cancer risk attributable to the exposure period at issue.

Regulatory and Litigation Support

Because the choice and derivation of toxicity values is frequently central to admissibility challenges (e.g., Daubert or Frye motions) and to disputes over cleanup levels, drinking-water standards, and causation, RoseyFields documents the full evidentiary chain supporting each toxicity value used in a matter: the principal studies, their individual and evidence-base confidence ratings, the POD derivation method, and the specific uncertainty-factor rationale. RoseyFields separately tracks how the shift from the legacy 1991/1995 toxicity values to the current values affects prior regulatory determinations (maximum contaminant levels, preliminary remediation goals, and risk-based cleanup criteria) so that experts can explain, where relevant, why a conclusion reached under an earlier toxicity value may or may not remain valid today.

8. Key Uncertainties and Sensitive Populations

RoseyFields identifies several uncertainties that risk assessors must account for when applying these values to a specific case.

• Benchmark response (BMR) selection for IHD required a judgment call to weight incidence-only studies (Wade et al., 2015; Wu et al., 2010) against studies including both incident and fatal cases

(Chen et al., 2013; Moon et al., 2013; James et al., 2015); RoseyFields applies a 5% BMR to incident cases and a 1% BMR to fatal cases, weighted by study sample size, and treats this as a source of moderate residual uncertainty in the IHD-based osRfD.

• The birth-weight osRfD rests on a single study (Kile et al., 2016) conducted in a Bangladeshi population with a substantially higher background prevalence of low birth weight (approximately 21%) than the U.S. population (approximately 8.3%), raising uncertainty about direct extrapolation of the observed effect magnitude to U.S.-relevant populations, even though the population is considered a reasonable, health-protective sensitive subpopulation.
• Genetic variability in arsenic methylation capacity (AS3MT-mediated) is a recognized modifier of individual susceptibility; individuals who retain a higher proportion of more toxic trivalent methylated intermediates (MMAIII) are believed to be at elevated risk relative to efficient methylators (Drobná et al., 2010).
• Nutritional status (notably folate status) and smoking are established effect modifiers in the cardiovascular and diabetes epidemiological literature and are only partially, non-uniformly controlled for across the contributing studies.
• Sensitive populations warranting particular attention in case-specific work include: pregnant women and fetuses (birth weight and neurodevelopmental endpoints); infants and young children, for whom rice-based dietary sources are disproportionately important (Signes-Pastor et al., 2016); populations served by unregulated private wells; and individuals with pre-existing cardiovascular or metabolic disease, for whom the IHD- and diabetes-based osRfDs are most directly relevant.

Finally, RoseyFields notes that the current toxicity values are themselves subject to revision as EPA’s evidence base and dose-response methodology continue to evolve; risk assessors should confirm the current status of these values before relying on them in a new matter.

References