Appendix F
Review of Past Risk Assessments of Electric Arc Furnace Slag

This appendix presents detailed observations resulting from the committee’s review of five health risk assessments of electric arc furnace (EAF) slag used in residential areas: Proctor et al. (2002), Exponent (2007), ToxStrategies (2011), Streiffer and Thiboldeaux (2015), and Mittal et al. (submitted).¹

EXPOSURE ASSESSMENT

The committee’s review considerations regarding exposure assessment align with the relevant definitions and issues laid out in the U.S. Environmental Protection Agency’s (EPA’s) Guidelines for Human Exposure Assessment (EPA, 2019). The primary purposes were to identify and evaluate exposure parameters used in the five available health risk assessments of EAF slag used in residential areas. The slag constituents included in the assessments, their concentrations, intrinsic and extrinsic exposure parameters, whether parameters could be effectively quantified, potential confounding factors, and opportunities for bias and uncertainty were considered. To add to the foundation for recommendations for improving the representativeness of exposure parameters in EAF slag risk assessments, the committee evaluated the extent to which

• data used in the risk assessments accurately characterized the variability of exposures experienced by residents;
• questions about the precision, reliability, and value of the slag constituent measurements were addressed; and
• key data gaps were recognized.

Data Quality

Evaluating the quality of the data for an exposure assessment is a crucial step. It is a critical element in determining the uncertainty associated with analytic methods and assessment results. Discussion of data quality according to standard criteria covers soundness, applicability and utility, and clarity and completeness. Using these criteria, the committee found the following:

• Soundness: The assessments’ technical methods and models used then-current approaches, sometimes in an appropriately tiered manner. The environmental exposure parameters could not be adequately evaluated because information was lacking about the sampling design and locations, and the relevance of the sampling results to exposed populations.
• Applicability and utility: Streiffer and Thiboldeaux (2015) was the assessment most limited in geographic scope and was the one most applicable for addressing its stated purpose. The other four assessments were conducted to address risk-related issues on a much broader

___________________

¹ While the committee was conducting its study, Deborah Proctor (ToxStrategies) and colleagues were carrying out a probabilistic risk assessment of residential EAF slag applications for covering driveways and unpaved roads. Preliminary materials provided to the committee regarding that assessment comprise the following:

• Proctor and Antonijevic (2022);
• Presentation to the committee on June 29, 2022;
• Crystal Ball input data submitted by Deborah Proctor on February 21, 2023; and
• Manuscript submitted to Risk Analysis for peer review entitled “Probabilistic Risk Assessment of Residential Exposure to Electric Arc Furnace (EAF) Steel Slag Using Bayesian Model of Relative Bioavailability and PBPK Modeling of Manganese.”

• geographic scale, relying on data that may or may not have adequately characterized the exposure parameters on such a scale.
• Clarity and completeness: The data, assumptions, methods, and analysis used in these assessments were documented to varying extents. Streiffer and Thiboldeaux (2015) provided considerable detail.

Constituents and Concentrations in EAF Slag

The two earliest EAF slag risk assessments (Proctor et al., 2002; Exponent, 2007) relied on the same 45 samples from within a set of slag samples collected from 58 operating steel mills in the United States and Canada (Proctor et al., 2000). For ToxStrategies (2011), 22 EAF slag samples were collected, and for Mittal et al. (submitted), 38 carbon steel slag samples were collected from steel mills and slag processing operations in the United States. Those four assessments did not provide specific details about the plant locations, how slag was disposed and stored on the sites, where the samples were taken on the plant sites or elsewhere, and the number of EAF slag samples taken per site. Streiffer and Thiboldeaux (2015) evaluated the health risks associated with a specific EAF site, operated by Charter Steel in Saukville, Wisconsin; the slag sampling strategy and number of samples for this site were not presented.

The methods used to determine the presence and concentrations of EAF slag constituents were presented in all of the risk assessments, based on standardized methods (e.g., EPA and ASTM² protocols). Up to 26 constituents were found in processed slag and in samples divided by particle size (ToxStrategies, 2011). All five assessments reported concentrations for manganese and iron; several assessments reported levels for arsenic, cadmium, hexavalent chromium, copper, and nickel.

The most frequently quantified exposure parameters used in the risk assessments were the concentrations measured in slag samples. However, the statistics for the concentrations were not consistent across the risk assessments; some assessments reported maximums or 95 percent upper confidence levels (or both) for the arithmetic mean, and some showed means or unspecified “averages.” Parameter values for other exposure variables used in exposure equations were intended to represent the central tendency exposure and reasonable maximum exposure (RME) (e.g., 50th and 90th percentiles).

Slag composition values were reported for different kinds of EAF samples; some were from processed slag, slag pots, or open slag piles (ready for sale), and some were from samples separated by particle size. Streiffer and Thiboldeaux (2015) reported EAF results with and without data collected as early as 1994. The heterogeneity of the results reduces comparability across risk assessments. Assessments often relied on published slag risk assessments or EPA screening levels to determine which slag constituents merited further investigation within their scopes.

Receptor Scenario and Populations

All of the risk assessments stated their purposes and scopes. Essential questions were translated into relevant scenarios; some assessments indicated that their decisions were based on peer-reviewed literature (e.g., Exponent, 2007). Problem formulations noted the general adult and child populations in the scope but did not include conceptual models.

In a conventional point estimate (deterministic) risk characterization, estimates of exposure point concentrations are combined with a set of central tendency exposure and RME parameter values for other exposure variables in dose equations (see Chapter 4) to represent the typical and high-end exposures among the general population. Additionally, relevant time and spatial scales appropriate for the outcomes of interest need to be incorporated into risk assessment scenarios. When an assessment is not site specific, exposure parameters are based on regulatory default values or professional judgment. This was done in all but the Streiffer and Thiboldeaux (2015) assessment, which focused on EAF slag sold from one steel mill and distributed locally.

___________________

² ASTM, formerly known as American Society for Testing and Materials, is now known as ASTM International.

The variability and uncertainty of the slag constituent measurements add importance to characterizing central tendency exposure and RME, as well as related risks among the most vulnerable populations. The populations studied varied across the assessments. Some evaluated both adults and children, while others focused on specific age groups expected to be the most sensitive to EAF slag exposures. Four of the assessments examined the potential health risks to children, but “children” was defined using a variety of age ranges. For example, Exponent (2007) used data for persons “under the age of 17,” while Streiffer and Thiboldeaux (2015) modeled risks for “worst case” scenarios, such as children 1–6 years of age for soil ingestion, and newborns (birth to 1 month) in their nasopharyngeal swallow and PM_2.5 inhalation scenarios. Rationales for age ranges were not always clear, but assessments often based their choices on EPA guidance and practices.

Recognition of vulnerable subpopulations is an important feature of risk assessment; it is based on the synthesis of a constellation of exposure factors. Attention is paid to differential distribution of exposure factors (e.g., routes of exposure, duration of exposure, proximity to sources of exposure, pre-existing health conditions, socioeconomic status, and behaviors) that may relate to exposure and/or outcomes. Key exposure characteristics of sensitive subgroups are noted here and discussed in more detail in Chapters 5 and 6.

Intrinsic exposure parameters refer to “intrinsic biological factors such as lifestage, genetic polymorphisms, prior immune reactions, disease state or prior damage to cells or systems” (EPA, 2019). Other than age and body weight, most of these factors were not considered in the five risk assessments the committee considered. ToxStrategies (2011) noted that children with soil-pica behaviors may be at elevated risk. Demographic factors, life stage, health status, behaviors, and practices are discussed in Chapter 6.

Furthermore, extrinsic exposure factors, such as socioeconomic status and inequalities, access to health care, setting (such as neighborhood), housing quality, and proximity to the source of EAF slag (EPA, 2005), are discussed in Chapter 6. Proximity of residences to roads or other sites (“distance”) where unencapsulated EAF slag was used was quantified in all but the earliest risk assessment (Proctor et al., 2002). Estimates of road lengths and widths, assumed volume and weights of vehicles on roads and other surfaces per day, and modeled PM emission heights were used in several of the exposure assessments. Mittal et al. used vehicle traffic data available from three states (Montana, New York, and Virginia) and particle emission factors based on EPA data.

Exposure Routes

The routes of exposure examined in the EAF slag risk assessments included ingestion and dermal contact with and inhalation of particulate matter (PM). Although all the assessments considered both inhalation and ingestion routes, some focused on one route as the most relevant source of exposure for the population of concern.

All five risk assessments considered soil ingestion in the form of a quantifiable rate of ingestion; two assessments assumed that 50 percent of soil is slag (Exponent, 2007; Streiffer and Thiboldeaux, 2015), while the other assessments assumed that all soil ingested was slag.

• Proctor et al. (2002) and ToxStrategies (2011) used a constituent-specific dermal absorption factor, while Exponent (2007) used a soil-to-skin adherence factor in their assessment.
• Quantitative inhalation/air intake rates were used in all of the risk assessments; Exponent (2007) and ToxStrategies (2011) incorporated quantitative inhalation exposure rates. All assessments considered risks associated with particle size, but not all assessments had measurement data to use in their equations.
• ToxStrategies (2011) and Streiffer and Thiboldeaux (2015) addressed the potential for risks related to nasopharyngeal swallow of large particles (PM₁₀) by using airborne PM₁₀ concentration data and assuming that 90 percent of the impacted particles were swallowed.

Exposure Parameters

The bases and rationale for each parameter value have to be clearly documented in a risk assessment. When reliable site-specific data do not exist for a parameter, assessments often use assumptions, modifying factors, models, defaults, or proxies to create a quantitative value. Typically, these choices are based on EPA risk assessment guidance. Otherwise, professional judgment is used. The choice of data and summary statistics used to define a parameter value may introduce a high or low degree of bias in estimates of exposure and risk. Standard risk assessment practice, consistent with EPA’s definition of an RME, is to make choices that ensure health risks are characterized for the most sensitive members of the population.

This section discusses selected exposure factors that were modified in the risk assessments to account for potential differences between EAF slag and conventional assessments of soil and dust: particle size distribution, exposure time factors, and bioaccessibility and bioavailability.

Particle Size Distribution

Particle size is an important factor to consider when assessing health outcomes. Size relates to how far slag will disperse from its source and where it will make contact with human tissues. Larger particles do not travel as far as fine particles, and the smaller sizes deposit more deeply in the respiratory tract. Over time, assessments have incorporated particle size in more detail. Proctor et al. (2002) adjusted total suspended PM for PM₅ to address this issue. Exponent (2007) and Streiffer and Thiboldeaux (2015) used meteorological data to model PM dispersion with the goal of developing more refined estimates of slag PM exposures. ToxStrategies (2011) and Mittal et al. (submitted) conducted analyses of three sizes of PM.

Exposure Time Factors

The relevance of time factors associated with the outcomes of interest is another important consideration. For example, cancer outcomes are typically associated with longer periods of exposure than for noncancer outcomes.

• All the assessments applied a “frequency” time factor, expressed as the number of days or percentage of time exposed to slag per year. Mittal et al. (submitted) calculated a percent time exposed value by applying adjustment factors to data from EPA’s then-current Exposure Factor Handbook.
• Four of the assessments included exposure durations defined by age groups (e.g., 6 years for children and 30 years for adults). Streiffer and Thiboldeaux (2015) used 1 year duration in their exposure scenarios.
• Three of the assessments assumed the number of hours per day (“exposure time”) that people spent outdoors, potentially in contact with slag. This metric was based on EPA guidance about the number of hours children and adults spent outdoors in residential areas or near roadsides. However, these same assessments dismissed the potential for continual indoor dust exposure, which would be expected to occur daily.

Bioaccessibility and Bioavailability

Bioaccessibility and bioavailability were included as exposure parameters in the five assessments. Proctor et al. (2002) used a range of values up to 45 percent for dermal bioaccessibility, but the other four assessments relied on EPA’s published values for oral and inhalation bioaccessibility. Although Streiffer and Thiboldeaux (2015) assumed 25 percent bioavailability in two child incidental soil ingestion scenarios, 100 percent bioaccessibility or bioavailability was otherwise used as the most conservative value to account for susceptible groups. Based on animal data, Mittal et al. (submitted) used a Bayesian linear regression

model and Monte Carlo simulation to define the characteristics of the relative bioavailability (RBA) of Mn; the results were used in their probabilistic risk assessment (PRA).

Potential for Bias

Bias in the assessment design or analysis may obscure or distort true relationships between exposures and outcomes. Incomplete identification of sources of bias, limited data quality, or inconsistent use of evidence in analysis can reduce the value of exposure characterization. Because systematic errors challenge the internal validity of an assessment, factors contributing to bias require identification and accountability. Bias may explain a nonsignificant risk finding. When bias is high, confidence in the assessment results is seriously diminished.

Some of the most important sources of bias involve confounding factors: variables that are associated with both the exposure and outcome while not acting in the causal chain between them. A list of all potential confounders may be developed through literature reviews in the planning stage of an assessment. If they can be quantified, they need to be controlled during analysis (e.g., in a sensitivity analysis). When confounders cannot be quantified, they may be represented by reasonable proxies that can be measured. The extent to which and direction in which confounders are known or thought to alter the estimate of the true exposure call for discussion in a risk assessment. For example, Mittal et al. (submitted) adjusted oral ingestion for the dietary intake of manganese in order to estimate risk more closely to that posed only by slag-related sources of that metal. Mention of other potential confounding factors (such as obesity, or body mass index, for manganese-related risks) was not found in the EAF slag risk assessments.

When exposure parameters could not be quantified from available measurements, assessments used existing risk assessment guidance and data sources as well as professional judgment to create parameters that could be used in exposure equations. Five elements of judgments are important: assumptions, the models used, defaults and modifying factors, professional judgment, and independent evaluation and review.

Assumptions

Throughout all the assessments, assumptions had to be made to quantify exposure factors. As early as Proctor et al. (2002) such assumptions were described to differing degrees. They assumed that plant-site samples were representative of EAF slag used in residential settings and nearby roads. In the absence of slag particle size data, many assessments assumed that soil data could be used to represent slag. Exponent (2007), ToxStrategies (2011), and Mittal et al. (submitted) assumed the number of vehicles per day on roads near residences to estimate particle emission factors. ToxStrategies (2011) also assumed that indoor air concentrations of slag were the same as outdoor levels. The most detailed discussion of assumptions was found in Streiffer and Thiboldeaux (2015).

Models Used

Data combined from more than one source were used to create new variables for equations. For example, PM emission factors were estimated by incorporating meteorological data (e.g., Exponent, 2007; Streiffer and Thiboldeaux, 2015). Proctor et al. (2002) calculated exposure-point concentrations for PM by dividing the concentration in slag by the particulate emission factor (PEF). In other cases (e.g., Exponent, 2007), conservative, upper bound values following EPA guidance were used to develop exposure parameters. Typically, time outdoors and time in contact with soil, drawn from EPA’s then-current Exposure Factor Handbook, were used to estimate time exposed to slag on residential properties. The Centers for Disease Control and Prevention’s national demographic, dietary, and other exposure-related data were frequently used to develop exposure parameters. By dividing age-group-specific air intake rates by body weight, Streiffer and Thiboldeaux (2015) found that neonates (birth to 1 month of age) were the most susceptible subpopulation for airborne slag exposures.

Defaults and Modifying Factors

All of the assessments relied on default values published by one or more agencies—for example, EPA’s risk assessment and exposure factor guidance as well as the Agency for Toxic Substances and Disease Registry’s (ATSDR’s) public health assessment guidance. ToxStrategies (2011) used defaults published in EPA’s risk assessment guidance; they also used age-dependent adjustment factors to estimate the risk of cancer from Cr(VI) exposures. In their PRAs, Mittal et al. (submitted) applied EPA’s three-fold modifying factor to account for nondietary sources of Mn, following ToxStrategies’ use of the same factor.

Professional Judgment

Streiffer and Thiboldeaux (2015) estimated variations in PM parameters based on state university and agency recommendations and aerial photographs, as well as on professional judgment. Among assessments, Proctor et al. (2002) and Mittal et al. (submitted) noted that they used professional judgment to refine parameters for use in their exposure calculations.

Independent Evaluation and Review

Proctor et al. (2002) was published in a peer-reviewed journal. Mittal et al. (submitted) had been sent for peer review at the time this report was written. Review of the other three assessments was not documented.

While data limitations were acknowledged by assessments and resulted in their use of scenarios, assumptions, defaults, and adjustments, the rationales for these choices were not always clearly presented. One adjustment for database limitations, a ten-fold modifying factor, was used by ToxStrategies (2011). In each of the assessments, two to three scenarios were built on several assumptions to study residential exposures to EAF slag. Scenario parameters were developed to represent slag exposures from residential driveways, nearby roadsides, parking lots, and/or agricultural uses. Choices were usually intended to characterize the highest potential exposure, thereby permitting estimation of the risks to the most sensitive subpopulations. However, the omission of some intrinsic, extrinsic, and confounding exposure factors may have biased the quantitative EAF slag health risk assessment results.

Uncertainty Analysis

The potential for sources of uncertainty and variability, which can distort exposure-response estimates in either direction from their true value, is high in these assessments. Uncertainty can arise, for example, from sampling errors, misclassification of exposures, mischaracterization of exposure pathways or bioavailability, inappropriate assumptions, bias, and/or analytic misinterpretations. Each of the assessments noted at least one uncertainty; some conducted qualitative evaluations (e.g., Proctor et al., 2002), while others conducted quantitative examinations (e.g., ToxStrategies, 2011; Mittal et al., submitted).

The lack of knowledge about precise exposure parameters is a serious concern; Proctor et al. (2002) noted that there were multiple sources of uncertainties in their assessment. With more accurate information, uncertainty can be reduced but variability cannot (EPA, 2001). Variability is inherent in factors representing real-world conditions, exposures, and populations, so it affects the value of exposure parameters. Use of high-end estimates was more likely to capture the exposures of the most sensitive subpopulations. Statistical analyses to evaluate the extent and characterize the nature of uncertainty and variability in the data and constructed parameters were described in some of the assessments. Deterministic models, which rely on point values, were less powerful in identifying uncertainty and variability than were probabilistic models that depend on distributions of parameter values. An example of the value of PRA compared to deterministic risk assessment (DRA) can be found in ToxStrategies (2011). PRA was conducted by ToxStrategies (2011) and Mittal et al. (submitted) for a sensitivity analysis of the uncertainty and variability

of exposure parameters. Mittal et al. (submitted) noted that their results indicated that reduction of uncertainties should be pursued in future assessments. Mittal et al. (submitted) used physiologically based pharmacokinetic (PBPK) modeling to address uncertainties in their assessment of Mn.

Toxicity Values

After chemicals of potential concern (COPCs) are identified, critical effect health outcomes and corresponding toxicity values are selected using a hierarchy of sources. The risk assessments evaluated both cancer and noncancer health effects and applied standard toxicity values (e.g., oral cancer slope factors and chronic oral reference doses [RfDs]) in risk calculations. Most assessments used toxicity values available from EPA’s Integrated Risk Information System (IRIS). Noncancer toxicity values were also derived from a lowest observed adverse effective level reported in the literature (ToxStrategies, 2011) and from ATSDR’s minimum risk level (MRL) (Mittal et al., submitted).

The two COPCs used as examples in this report are hexavalent chromium and manganese. Prior risk assessments evaluated both cancer and noncancer outcomes for Cr⁶⁺ and noncarcinogenic effects for Mn; the latter is not considered a carcinogen. Relevant toxicity values for inhalation and oral exposures of these two COPCs were extracted from each risk assessment; the values are shown in Table F-1 for Cr⁶⁺ and Table F-2 for Mn. These values primarily relied on EPA’s IRIS database and ATSDR public health risk assessment guidance and toxicological profiles.

TABLE F-1 Toxicity Values for Hexavalent Chromium Used in Prior Risk Assessments of EAF Slag

Issues in the four assessments that included Cr⁶⁺ related to the extent to which this COPC is bioaccessible and bioavailable. Proctor et al. (2002), Streiffer and Thiboldeaux (2015), and Mittal et al. (submitted) used 100 percent as the maximum (most conservative) value in their equations. Streiffer and Thiboldeaux also used 25 percent bioavailability in two of their six scenarios to consider how much a low-end value for this factor would affect their calculations and risk characterization. Proctor et al. (2002) stated that Cr⁶⁺ is more soluble and bioaccessible in fairly neutral conditions, such as in the intestines. ToxStrategies (2011) pointed out that Cr⁶⁺ concentrations are higher in larger PM, which is more likely to deposit in the upper respiratory tract. They cautioned that only a draft oral cancer slope factor was available at the time of their assessment and that significant data gaps limited knowledge for determining Cr⁶⁺’s carcinogenic mode of action. As a result, a variety of uncertainties may have affected their dose estimates. Mittal et al. (submitted) stated that the acidic condition of the stomach rapidly converts Cr⁶⁺ to Cr³⁺, but they assumed 100 percent bioaccessibility of Cr⁶⁺ to ensure conservatism.

TABLE F-2 Toxicity Values for Manganese Used in Prior Risk Assessments of EAF Slag

All five of the prior risk assessments examined noncarcinogenic risks related to Mn exposures; each mentioned at least one issue in estimating dose levels. As for Cr⁶⁺, the authors included toxicity values drawn from the most recent EPA and ATSDR resources. ToxStrategies (2011) noted that only Mn, not its various chemical forms, was considered; none of the prior assessments considered chemical forms of Mn.

The bioaccessibility and bioavailability of Mn were handled variously in the five risk assessments. Two of the assessments incorporated values for bioaccessibility and three included bioavailability in their calculations. Proctor et al. (2002) and ToxStrategies (2011) based their determinations of Mn-specific bioaccessibility on minimum, average or mean, 95 percent upper confidence level, and/or maximum values from animal studies. Proctor et al. (2002, Table 6) derived bioaccessibility values for oral and dermal pathways of exposure, while ToxStrategies (2011, Table 6.3) developed distributions of bioaccessibility for oral exposures to Mn. Exponent (2007) stated that they determined the value for Mn oral bioavailability; for example, in their Appendix E, using RME parameters, the authors showed a unitless oral bioavailability factor of 0.16 for Mn in the adult resident on-site incidental soil ingestion pathway. Streiffer and Thiboldeaux (2015) used 25 percent and 100 percent bioavailability in their scenarios of children’s exposures to Mn. As noted above, Mittal et al. (submitted) used Monte Carlo simulations to obtain the RBA of Mn.

The oral RfD varied across the studies based on then-current knowledge and agency risk assessment guidance. Proctor et al. (2002) said that they converted the oral RfD to a systemic dose by applying an absorption factor. Exponent (2007) stated that incidental ingestion was a major pathway of Mn exposure; they also noted that dietary intake of Mn affects its absorption. However, they did not indicate how they handled this additional Mn source in their dose and risk calculations. ToxStrategies (2011) and Mittal et al. (submitted) applied the three-fold modifying factor to account for nondietary sources of Mn.

The inhalation route of exposure required several issues to be addressed. Some assessments (e.g., Proctor et al., 2002; Exponent, 2007; ToxStrategies, 2011; Mittal et al., submitted) considered particle size, the concentration of Mn by particle size (higher levels on larger particles), and where in the respiratory tract particles by size would deposit Mn (more deeply for smaller particles). Proctor et al. (2002) relied on EPA’s approach to inhalation toxicity. Exponent (2007) converted EPA’s PM₁₀ data for Mn to estimate its PM₅ levels. Using EPA’s standard assumptions for body weight and inhalation rates, they then converted the Mn reference concentration (RfC) to a Mn RfD. ToxStrategies (2011) used much the same approach but noted that there is a large degree of uncertainty in the Mn RfC. Mittal et al. (submitted) used PBPK modeling to address uncertainty in Mn risk assessment.

How Were Relevant Toxicity-Related Issues Addressed?

Toxicity-related issues require examination of the tools and methods used to assess the relationships between exposure factors, estimated doses, and adverse health outcomes. Furthermore, key assumptions, uncertainties, and parameters used to develop dose estimates are addressed in this section. Later authors inevitably built their work upon earlier assessments, so the tools, methods, and related issues were often

similar across the toxicity and exposure assessment portions of the five risk assessments. Both cancer and noncarcinogenic effects were evaluated for EAF slag exposures in residential settings.

Tools

Cancer: Streiffer and Thiboldeaux (2015) did not assess risks for cancer. The other risk assessments used EPA’s then-current cancer slope factor. ToxStrategies (2011) applied an inhalation unit risk factor as well.

Noncarcinogenic health effects: The earliest EAF slag risk assessment (Proctor et al., 2002) estimated noncancer health risks by calculating hazard indices (HIs) for each slag constituent identified as “of concern.” The no-effects level for each constituent was adjusted by applying an uncertainty factor. Assessments relied on RfDs and RfCs published by EPA. Exponent (2007) and Mittal et al. (submitted) also used ATSDR’s tolerable dose intake values and/or MRLs in their assessments. Streiffer and Thiboldeaux (2015) used guidance and data published by ATSDR. Some assessments produced HIs (Proctor et al., 2002; Exponent, 2007; ToxStrategies, 2011; Mittal et al., submitted), while Streiffer and Thiboldeaux (2015) calculated hazard quotients (HQs).

Methods

Hazard identification: Most of the assessments documented a two-step process to identify COPCs. The two earliest studies (Proctor et al., 2002; Exponent, 2007) conducted (1) comparisons of slag sampling results to background levels in U.S. soil; and (2) for the COPCs exceeding background levels, comparisons to published federal and state health-based screening levels.

In their second step, ToxStrategies (2011) was the first EAF slag risk assessment to identify COPCs by comparing maximum concentrations of metals in slag samples to EPA regional screening levels. Streiffer and Thiboldeaux (2015) also used this hazard identification approach.

Dose estimation: Developing an appropriate dose metric requires combining data for key aspects of the COPCs and exposure scenarios; therefore, the metric is dependent on the data available for these aspects and on the assessments’ abilities to utilize the data effectively. Judgment was needed to adjust data to the specific needs of EAF slag risk assessment. Key elements of a dose metric include the COPC and its concentration in the environmental medium of concern (EAF slag applied in residential areas), potential dose (the amount ingested, inhaled, or in contact with skin), delivered dose (available to interact at the cellular or organ level), and dose levels (such as average and peak doses).

The dermal route of exposure was studied in the two earliest risk assessments (Proctor et al., 2002; Exponent, 2007). Inhalation and ingestion pathways were the focuses of the more recent assessments. Exponent studied all three pathways.

The agents of interest in all assessments were metals found in EAF slag samples, judged to be relevant to residential uses; most often only the elemental forms of the metals were assessed. The COPCs were not examined as combined doses. Concentrations of metals found in the samples that exceeded background or screening levels were used in dose calculations, including for potential and delivered doses.

Average, mean, or most likely exposure (MLE) values were used in all but the earliest and latest risk assessments (Proctor et al., 2002; Mittal et al., submitted). Exponent (2007) included the fraction of metals released from the EAF slag matrices to calculate exposure-point concentrations.

A variety of factors were used to estimate the dose at the outer barrier of human contact—for example, skin, oral, or respiratory tract tissue. Proctor et al. (2002) used the simplest approach—that is, exposure-point concentrations. Later assessments developed equations to combine several factors for estimating dose; these included contact rates, intake rates, time parameters (such as frequency and duration of exposure, fraction of time spent outdoors in contact with slag), and exposure factors. Mittal et al. (submitted) used Monte Carlo simulations to obtain stable values for several of these factors.

Assessments used different methods to determine the amount of slag ingested, inhaled, and/or adhered to skin. Exposure-pathway specific contact rates calculated in Proctor et al. (2002) included soil

ingestion rate, inhalation rate, and skin adherence. (To err on the side of conservatism, they chose to use the adult slag-to-skin adherence factor for children as well.) Exponent (2007) estimated chronic dermal absorption by multiplying the COPC-specific oral RfD and the calculated gastrointestinal absorption factor: This approach resulted in low absorption levels for all COPCs. ToxStrategies (2011) examined intake rates by particle size, noting that only a small percentage of slag particles appeared to be ingested, while fine particles are more likely to be inhaled. Streiffer and Thiboldeaux (2015) developed a “dose exposure” equation that was calculated for each exposure pathway, COPC, and scenario. Their equations required relevant quantitative values for contaminant concentration, exposure factor, intake rate, body weight, and bioavailability. Mittal et al. (submitted) used distributions for each parameter in their probability risk assessment model; these included body weight (adult and child), exposure time, exposure duration, soil ingestion rate (adult and child), inhalation rate (adult and child), and a scenario-specific PEF. Clearly, over time the models for estimating doses have become more complex in terms of the number of parameters and the statistical methods to combine their values into relevant dose estimates.

Bioaccessibility was included in most assessments (Proctor et al., 2002; Exponent, 2007; ToxStrategies, 2011; Mittal et al., submitted) but in different forms. Oral and ingested exposures were assumed to be 100 percent bioaccessible in assessments by Proctor et al. (2002), but they were calculated in Mittal et al. (submitted). Smaller percentages of COPCs released from slag were used in all but the earliest assessment by Proctor et al. (e.g., Streiffer and Thiboldeaux [2015] used 25 percent bioavailability as a lower bound in two of their scenarios). For most metals in their PRA model, Mittal et al. (submitted) calculated RBA from in vitro bioaccessibility.

Formulas used to determine doses for cancer risk estimates incorporated environmental, exposure, and biological factors, drawing on then-current EPA guidance. For each COPC, Proctor et al. (2002) and Exponent (2007) calculated lifetime average daily doses (LADDs) by dividing a numerator of exposure-related factors (e.g., concentration of a COPC in air, PEF, exposure frequency and duration, time spent in the slag-contaminated area, bioaccessibility of the COPC) by body weight and averaging time (70 years for children and adults) in the denominator.

Calculations of average daily doses (ADDs) were used by Proctor et al. (2002) and Exponent (2007) to estimate chronic COPC exposures related to the risk of noncarcinogenic effects. The formulas for ADDs were much the same as for LADDs; however, the averaging time in the denominator was shorter (30 years for adults and 6 years for children). Proctor et al. (2002) used oral absorption for most of their COPCs and developed COPC-specific chronic dermal RfDs. Using EPA guidelines, they calculated systemic doses by combining the COPC-specific oral RfD with the fraction that would be absorbed through the oral pathway. To be conservative, ToxStrategies (2011) applied the three-fold modifying factor in their oral RfD for manganese. Streiffer and Thiboldeaux (2015) used formulas to estimate doses for incidental ingestion (with nasopharyngeal swallow) and dust inhalation scenarios. To build their equations, they relied on regional soil screening levels (EPA Region 3) and ATSDR’s national body weight and inhalation rate data. Their numerator included multiple values for COPC concentration, intake rate (ingestion and inhalation), and exposure and conversion factors, while their denominator combined body weight and a bioavailability factor.

Toxicity assessment: Evaluation of dose–response relationships requires data for specific health outcomes, dose levels, time factors, routes of exposure, and relevant statistical means to link the health outcomes with the estimated doses.

The relationship of estimated doses to cancer outcomes was modeled using a COPC-specific cancer slope factor in low-dose extrapolation models, based on data from animal and/or human studies. Using children’s estimated low exposure levels to EAF slag, Exponent (2007) calculated their lifetime cancer risks in incidental soil ingestion and dermal contact with soil scenarios. In addition to using oral and inhalation cancer slope factors, ToxStrategies (2011) included an inhalation unit risk factor in their COPC-specific probability risks. Mittal et al. (submitted) created PRA distributions to identify average (50%) and most likely (90%) Mn exposure levels for their cancer risk estimates. They relied on guidance and data from both EPA and ATSDR resources.

In all prior EAF slag risk assessments, calculations of noncarcinogenic responses to COPC doses depended on RfDs and RfCs; most assessments noted that to protect the most sensitive subpopulations, these values were derived by applying an uncertainty, modifying, and/or safety factor to an observed no- or low-adverse effect level. Extrapolation of risks at low-level exposures was conducted using distributions of exposure factors (such as average hours/day) in PRA (e.g., ToxStrategies, 2011) and in HQ equations (e.g., Mittal et al., submitted). To examine risks at a range of conditions, Streiffer and Thiboldeaux (2015) used COPC-specific minimum and maximum bioavailability, and average and maximum concentrations in year-round and “winter” seasons. Mittal et al. (submitted) ran Monte Carlo simulations to generate both 50th and 90th percentile risk values.

Key Assumptions

Slag characteristics: The most common assumptions about EAF slag were that its characteristics and dispersion patterns are similar to soil, resulting in the use of soil data as a proxy for slag data. ToxStrategies (2011) assumed that slag transported from mills was neither diluted nor mixed with other materials before it was used at or near residential properties. The test results of mill samples, taken to identify constituents of concern in slag, were assumed to be representative of the slag used at residential sites and nearby unpaved roads.

Up to 50 percent of PM or soil was assumed to be slag. ToxStrategies (2011) assumed that 67 percent of PM₁₀ was PM₅ and that a portion of PM was slag. Streiffer and Thiboldeaux (2015) relied on EPA’s AERMOD models to generate reasonable distributions of PM, and therefore slag, for their assessment.

Exposure characteristics: Assessments typically assumed that estimated slag exposures reflected the upper bounds of actual exposures. A key assumption in most of the risk assessments was that exposure rates to slag were the same as for soil or for sand and gravel—for instance, in terms of daily direct contact time, skin adherence, and inhalation and ingestion rates.

Proctor et al. (2002) used conservative assumptions and approaches. They noted that “health-based screening levels were protective of long-term chronic average exposures” and that their formulas were sufficient to overestimate potential health risks. Stating that slag particles are large and not likely to adhere to skin or be ingested, they assumed they had overestimated exposures “for the majority of the population.” Importantly, however, this assessment would not have estimated risks for sensitive subpopulations.

Exponent (2007) chose the 95th percentile of children’s time outdoors in contact with soil for their EAF slag exposure model, assuming that this level would overestimate actual contact time. These authors assumed that the volume of vehicular traffic and related factors (such as residents’ distance from unpaved roads) in their models would reflect residents’ actual exposure experiences.

ToxStrategies (2011) assumed that children’s frequency of exposure to slag was the same as or greater than the frequency for adults. In the absence of residential data, the authors assumed that indoor exposures to EAF slag-related dust would be the same as outdoor exposures. Also, they assumed that using state data for daily vehicular volume would provide RMEs and MLEs, and that 90 percent of respirable slag particles were swallowed. They concluded that incidental EAF slag ingestion related to living near unpaved roads was not important when estimating slag-related health risks.

ToxStrategies (2011) stated that the three-fold modifying factor for Mn would overestimate exposure; Mittal et al. (submitted) used an Mn exposure level corrected for dietary exposures but not nondietary exposures.

ToxStrategies (2011) stated that incidental ingestion of slag related to unpaved roads was not important; later Mittal et al. (submitted) asserted that dermal contact was not soluble or readily available for absorption, so this exposure pathway was deemed not important for estimating Mn exposure.

Toxicokinetics: Oral bioaccessibility was assumed to be 100 percent in the Exponent (2007) risk assessment. Except for vanadium and Mn, Exponent assumed that oral bioavailability was 100 percent.³

The authors of this assessment also assumed that inhalation bioavailability was 100 percent, as did Streiffer and Thiboldeaux (2015) in four of their six scenarios (e.g., the scenarios involving PM₁₀ nasopharyngeal swallow and the PM_2.5 respirable fraction). Furthermore, Streiffer and Thiboldeaux (2015) assumed that incidental ingestion bioavailability was 100 percent.

Mittal et al. (submitted) noted that iron affects manganese absorption and limits Mn uptake, so they used a Bayesian model to determine appropriate RBA values.

Populations at risk: Published population data for body weight and intake rates (e.g., EPA’s Exposure Factor Handbook and ATSDR’s public health assessment guidelines) were used in all of the assessments, based on the assumption that federal guidelines for their use would adequately protect sensitive subpopulations. Proctor et al. (2002) and Streiffer and Thiboldeaux (2015) differed from the other three in their use of these data. The former assumed that average population data were sufficient to estimate risks. Assuming that national data applied to their population of interest, the latter examined the range of children’s body weights and intake rates to identify the subpopulation (infants) most at risk for health effects.

Parameters

Parameters used in equations to estimate dose and exposure varied somewhat across the five risk assessments. All authors included body weight, exposure frequency (days/year), and a conversion factor in their equations. The four industry-sponsored assessments utilized similar parameters: all included bioaccessibility and exposure duration (years). Except for Proctor et al. (2002), the other three of these four risk assessments included PEFs. In the two later assessments (ToxStrategies, 2011; Mittal et al., submitted), exposure-point concentrations were replaced by exposure concentrations, using more advanced statistical methods.

Streiffer and Thiboldeaux (2015) used some parameters that were different from the other four risk assessments. Instead of bioaccessibility they included a bioavailability factor. They also developed a PM emission rate for their analyses rather than using PEFs.

Except for Proctor et al. (2002), the other four risk assessments included parameters related to soil ingestion or intake, inhalation of PM, and exposure time (hours/day). Only Exponent (2007) examined soil-to-skin adherence. All of the industry-sponsored assessments included exposure duration (years) and averaging time (days) for cancer and noncarcinogenic health effects.

Uncertainties

The authors of the five prior EAF slag risk assessments addressed the uncertainties in their results to varying degrees and in different ways. Proctor et al. (2023) declared that in previous Mn risk assessments (noting only Proctor et al., 2002) “uncertainty in the use of BA [bioaccessibility] measures for RBA [relative bioavailability] exist in these studies [presumably meaning all prior EAF slag risk assessments which included Mn].” Proctor et al. (2002) stated that their qualitative analysis could not be precise because generic values were used in their exposure estimates.

Although Exponent (2007) used assumptions and adjustments to account for inherent uncertainties, the authors noted that “each assumption and parameter introduces uncertainty into quantitative risk assessment” and that several of their overestimated parameters likely overestimated risk. Uncertainties discussed in the Exponent risk assessment included whether soil and slag characteristics, dispersion, contact times, and exposure rates are similar; what percentage of soil is made up of slag; to what extent slag is

___________________

³ See Appendix E of Exponent (2007) for the oral and dermal bioavailability factors used in incidental soil ingestion and dermal contact with soil scenarios for adults and children.

bioavailable; whether PM emissions related to vehicles are representative of residents’ inhalation exposures to slag; and whether safety factors and conservative assumptions appropriately account for uncertainties.

ToxStrategies (2011) noted that in their DRA, (1) conservative assumptions are compounded, and (2) “each parameter and assumption … introduces uncertainty into the quantitative risk and hazard estimates and resulting RBCs [risk-based concentrations].” They conducted a PRA, using probability distributions, to evaluate parameter and model uncertainties and variabilities. In this analysis, they used conservative assumptions and modifying and safety factors to account for additional uncertainties, not only improving on their DRA results but also being more protective of sensitive subpopulations.

Streiffer and Thiboldeaux (2015) and Mittal et al. (submitted), in part, examined their calculations using sensitivity analyses. The former risk assessment commented on dietary and other forms of Mn intake and said that these sources of uncertainty could not be considered.

Mittal et al. (submitted) used a PBPK model to evaluate uncertainties in the Mn RfD. They conducted Monte Carlo simulations to reduce uncertainty in their estimates of Mn exposure levels.

Forms of Toxics Considered

Chromium was the toxic metal for which different forms were most often considered. Trivalent chromium was included in the earliest two risk assessments (Proctor et al., 2002; Exponent, 2007). Hexavalent chromium received analytic attention in all of the assessments except Exponent (2007).

Exponent (2007) noted under their Table 2-5 that nickel in slag was likely nickel oxides, but their tables and text do not indicate that oxide forms of nickel were assessed.

Streiffer and Thiboldeaux (2015) mentioned thallium salts, but they did not calculate risk estimates for these potential exposures.

HOW HAVE PREVIOUS RISK ASSESSMENTS ADDRESSED SUSCEPTIBILITY AND SENSITIVITY FACTORS?

The factors most often incorporated in the five EAF slag risk assessments (Proctor et al., 2002; Exponent, 2007; ToxStrategies, 2011; Streiffer and Thiboldeaux, 2015; Mittal et al., submitted) were childhood and time in direct contact with EAF slag. Body weight was included in all of the risk assessments. Other exposure parameters were considered but infrequently entered into equations as quantitative variables.

Susceptible and Sensitive Subpopulations

Susceptible subgroups were noted in the risk assessments but, except for “children,” these groups were not assessed separately. ToxStrategies (2011) mentioned “sensitive subpopulations” and stated that conservative human health risk assessment is protective of susceptible people. To account for related health risks, ToxStrategies considered sensitive subpopulations in the development of their oral RfDs and inhalation RfCs.

Demographic Characteristics

The demographic factor most often incorporated in the risk assessments was age. All of the assessments included “adults” (usually relying on data for 18- to 64-year-olds). Mittal et al. (submitted) included body weight data for adults over age 17 and inhalation rates for persons ages 31 up to 40 years.

Neonates (birth up to 1 month of age) were noted in four of the assessments (Exponent, 2007; ToxStrategies, 2011; Streiffer and Thiboldeaux, 2015; Mittal et al., submitted). Most authors omitted this group based on the assumption that these individuals had no substantial contact with EAF slag. For two of their worst-case scenarios, Streiffer and Thiboldeaux used neonatal intake rate and body weight data (EPA,

2011) to represent the most sensitive subpopulation. Infants (1 month to 1 year of age) were mentioned by ToxStrategies but not included in their risk calculations.

Four assessments considered risks to children separately from adults (Proctor et al., 2002; Exponent, 2007; ToxStrategies, 2011; Mittal et al., submitted); one assessed children only, considering them a sensitive subgroup more susceptible to adverse health outcomes (Streiffer and Thiboldeaux, 2015). However, authors used different sources of data, thereby creating different age groups. Exponent used the broadest childhood category (ages 1 to 17), based on EPA’s Exposure Factors Handbook (2002) data for duration of contact with sand, gravel, and soil. ToxStrategies used EPA’s interim child-specific exposure data for 6- to 11-year-olds. The age range for “young children” in Streiffer and Thiboldeaux appeared to be 1–6 years, based on their use of data from EPA (2011) and ATSDR (2005). Mittal et al. (submitted) used body weight data for children ages 1–6, based on the National Health and Nutrition Examination Survey (NHANES) (shown as dated 2011–2012); the age range for children used in their equations to estimate biological levels of manganese is not clear.

The other assessments described geographic locations of potentially exposed populations in general terms (such as North America or 58 mills in the United States) or gave no indication. Streiffer and Thiboldeaux (2015) conducted their assessments for areas affected by slag from one steel production site in Wisconsin.

Demographic factors not included in the assessments were race, ethnicity, income, and occupation.

Genetic Variability

The potential impacts of this group of factors were not included in the available risk assessments.

Life Stage

Childhood was the life stage most frequently considered. Point estimates were used for children’s body weight and time outdoors (i.e., potentially in contact with EAF slag). Data for children were drawn, for example, from EPA (2002, 2011) and ATSDR (2005) resources. These data were used to calculate HIs, as well as toxicity, dose, and exposure assessments. ToxStrategies (2011) applied age-dependent adjustment factors to account for early-life sensitivities in their DRA.

Stages of life not included in any of the assessments were in utero, neonatality, infancy, puberty, women of child-bearing age (i.e., not distinguished from adults ages 18-64), and pregnancy. Although Mittal et al. (submitted) used NHANES (National Center for Health Statistics, 2012) body weight data for persons ages 18 and over, they did not analyze the old age (65 and older) life stage separately in their risk assessment.

Health Status

The IRIS Toxicological Review of Hexavalent Chromium(VI) (EPA, 2022) noted a variety of health conditions; these could also affect risks related to EAF slag exposures. The document indicates that psychological stress, elevated body mass index, frailty, nutritional status, and chronic diseases may need to be considered in Cr⁶⁺ risk assessments.

Moreover, EPA (2022) listed 12 specific conditions that could alter toxicity (e.g., diabetes, asthma, low stomach acid, anemia, non-alcoholic fatty liver disease, iron deficiency, and infections). Most of these health conditions and others were not addressed in the available EAF slag risk assessments.

Proctor et al. (2002) noted that pre-existing health conditions may affect EAF slag-related risks; in their discussion of oral bioaccessibility, they considered gastric juices in the stomach. Exponent (2007), ToxStrategies (2011), and Mittal et al. (submitted) commented that stomach acidity could affect manganese uptake, but in their risk calculations these assessments did not account for the proportion of the exposed population that had this condition.

Body weight values were used for groups of children and adults, defined in different ways in the prior risk assessments. Proctor et al. (2002) used body weight as one factor in their dose calculations. Exponent (2007) used body weight as a variable to determine dermal, ingestion, and inhalation doses and to calculate risks for cancer and noncancer outcomes.⁴ ToxStrategies (2011) used normal distributions for child and adult body weights to calculate risk-based concentrations in their DRAs and target HQs in their PRAs. Streiffer and Thiboldeaux (2015) used body weight as a variable in their exposure dose equations for incidental soil ingestion, inhalation with nasopharyngeal swallow, and respirable fraction (PM_2.5) inhalation. Mittal et al. (submitted) stated that they used custom continuous age ranges for age groups 1–6 and 18+⁵ in their PRA equations. Body weight is used in EPA’s standard ADD formulas, which therefore argues for continuing its use in risk assessment. However, the increasing rates of obesity in the U.S. population and the links between obesity and adverse health outcomes related to COPCs (e.g., Eick and Steinmaus, 2020), which are found in EAF slag, suggest that alternative body composition metrics need to be monitored and considered for potential improvements, compared to the use of body weight data, as a means for more accurately identifying subgroups susceptible to increased risks related to EAF slag exposures.

Behaviors or Practices

Dietary and non-food sources of exposure, hand-to-mouth behaviors, and time spent outdoors were the behavioral factors most often mentioned in the EAF slag risk assessments. However, the lack of data to support the development of quantitative behavioral variables limited the risk assessment equations to relying on assumptions, proxy data, and/or modifying factors.

Exponent (2007) stated that “exposure parameters and assumptions were selected to overestimate, rather than underestimate, potential human … exposures.” They and ToxStrategies (2011) noted that among the important uncertainties is the assumption that slag presents characteristics similar to those of soil. Exponent and ToxStrategies assumed exposure rates to soil and slag were similar. ToxStrategies assumed that 100 percent of soil ingestion was contributed by slag. Exponent and Streiffer and Thiboldeaux (2015) assumed that 50 percent of soil particles derived from slag. Exponent also assumed that 50 percent or more of inhaled PM was slag-related.

To ensure that risk assessments accounted for various uncertainties and were adequately conservative, some assessments⁶ incorporated EPA’s recommended three-fold modifying factor for nonfood exposures to constituents of interest.⁷ ToxStrategies (2011) and Mittal et al. (submitted) used the threefold modifying factor to account for non-dietary exposures to manganese.

Hand-to-mouth behaviors were most extensively addressed by Streiffer and Thiboldeaux (2015). They commented on children’s potential contact with slag particles present on toys, domestic animals, household members’ clothing and shoes, indoor surfaces, and car interiors. In the absence of slag-specific data, Proctor et al. (2002) included variables for soil ingestion rate and soil loading rate to estimate dermal exposure to slag.

Differing lengths of time spent outdoors were used in most of the assessments; these variables estimated the extent of direct exposure to EAF slag. In their dose calculations, Proctor et al. (2002) used 1 out of 16 hours outdoors per day to represent childhood exposures. In their oral and dermal risk assessments, Exponent used a value of 0.125 (equivalent to 2 hours in a 16-waking-hour day) to represent the 95th percentile of time a child (ages 17 or younger) would spend playing outdoors in sand, gravel, dirt, or grass. ToxStrategies combined hours/day for children ages 6–11 and hours/day spent handling sand and gravel,

___________________

⁴ Respectively, LADD and ADD.

⁵ Data from NHANES 2011–2012 (National Center for Health Statistics, 2012).

⁶ For example, ToxStrategies (2011) and Streiffer and Thiboldeaux (2015).

⁷ For example, see EPA Region 3 Screening Table, User Guide, May 2014, http://www.epa.gov/reg3hwmd/risk/human/rb-concentration_table/usersguide.htm, as cited in Streiffer and Thiboldeaux (2015).

resulting in an adjustment factor of 0.56.⁸ Mittal et al. (submitted) used the same factor, explaining that in a 24-hour period a typical child spent 60 minutes playing in sand and gravel during 107 minutes outdoors at their residence. Streiffer and Thiboldeaux (2015) estimated the extent of outdoor exposure expected by their study population based on the seasonal lengths of day and the number of days/year potentially outdoors in Wisconsin.

Factors in this category that were not included in the EAF slag assessments were behaviors and practices including pica, alcohol consumption, smoking, and subsistence and recreational hunting and fishing.

Social Determinants

Most of the social determinants listed in the IRIS Cr(VI) draft were not mentioned in any of the risk assessments of EAF slag. This category includes income, socioeconomic status, neighborhood factors, and health care access, as well as social, economic, and political inequality.

The proximity of residents to EAF slag-covered roads and/or parking lots was the neighborhood factor considered in most of the assessments (Exponent, 2007; ToxStrategies, 2011; Streiffer and Thiboldeaux, 2015; Mittal et al., submitted).

REFERENCES

___________________

⁸ EPA’s EFH (2009) data.

Proctor, D. M., and T. Antonijevic. 2022. “Refined health risk assessment for residential exposures to manganese in EAF steel slag.” Poster presented at Society of Toxicology Annual Meeting, San Diego, CA, March 2022.

Proctor, D. M., K. A. Fehling, E. C. Shay, J. L. Wittenborn, J. J. Green, C. Avent, R. D. Bigham, M. Connolly, B. Lee, T. O. Shepker, and M. A. Zak. 2000. “Physical and chemical characteristics of blast furnace, basic oxygen furnace, and electric arc furnace steel industry slags.” Environmental Science & Technology 34(8):1576–1582. https://doi.org/10.1021/es9906002.

Proctor, D. M., E. C. Shay, K. A. Fehling, and B. L. Finley. 2002. “Assessment of human health and ecological risks posed by the uses of steel-industry slags in the environment.” Human and Ecological Risk Assessment: An International Journal 8(4):681–711. https://doi.org/10.1080/20028091057150.

Proctor, D. M., S. N. Vivanco, and A. D. Blanchette. 2023. “Manganese relative oral bioavailability in electric arc furnace steel slag is influenced by high iron content and low bioaccessibility.” Toxicological Sciences 1–10. https://doi.org/10.1093/toxsci/kfad037.

Streiffer, A., and R. Thiboldeaux. 2015. Charter Steel Slag Health Assessment. P-01283, Wisconsin Department of Health Services.

ToxStrategies. 2011. Human Health Risk Assessment for Iron and Steel Slag. Rancho Santa Margarita, CA: ToxStrategies, Inc.