National Primary Drinking Water Regulations: Perchlorate

ENVIRONMENTAL PROTECTION AGENCY 40 CFR Parts 141 and 142 [EPA-HQ-OW-2018-0780; FRL-9994-68-OW] RIN 2040-AF28 National Primary Drinking Water Regulations: Perchlorate AGENCY:

Environmental Protection Agency (EPA).

ACTION:

Proposed rule, request for public comment.

SUMMARY:

The Environmental Protection Agency (EPA) is proposing a drinking water regulation for perchlorate and a health-based Maximum Contaminant Level Goal (MCLG) in accordance with the Safe Drinking Water Act (SDWA). The EPA is proposing to set both the enforceable Maximum Contaminant Level (MCL) for the perchlorate regulation and the perchlorate MCLG at 0.056 mg/L (56 µg/L). The EPA is proposing requirements for water systems to conduct monitoring and reporting for perchlorate and to provide information about perchlorate to their consumers through public notification and consumer confidence reports. This proposal includes requirements for primacy agencies that implement the public water system supervision program under the SDWA. This proposal also includes a list of treatment technologies that would enable water systems to comply with the MCL, including affordable compliance technologies for small systems serving 10,000 persons or less.

DATES:

Comments must be received on or before August 26, 2019. Under the Paperwork Reduction Act (PRA), comments on the information collection provisions are best assured of consideration if the Office of Management and Budget (OMB) receives a copy of your comments on or before July 26, 2019.

ADDRESSES:

Submit your comments, identified by Docket ID No. EPA-HQ-OW-2018-0780, at https://www.regulations.gov. Follow the online instructions for submitting comments. Once submitted, comments cannot be edited or removed from Regulations.gov . The EPA may publish any comment received to its public docket. Do not submit electronically any information you consider to be Confidential Business Information (CBI) or other information whose disclosure is restricted by statute

-0780, at https://www.regulations.gov. Follow the online instructions for submitting comments. Once submitted, comments cannot be edited or removed from Regulations.gov . The EPA may publish any comment received to its public docket. Do not submit electronically any information you consider to be Confidential Business Information (CBI) or other information whose disclosure is restricted by statute. Multimedia submissions (audio, video, etc.) must be accompanied by a written comment. The written comment is considered the official comment and should include discussion of all points you wish to make. The EPA will generally not consider comments or comment contents located outside of the primary submission ( i.e., on the web, cloud, or other file sharing system). For additional submission methods, the full EPA public comment policy, information about CBI or multimedia submissions, and general guidance on making effective comments, please visit http://www2.epa.gov/dockets/commenting-epa-dockets.

FOR FURTHER INFORMATION CONTACT:

Samuel Hernandez, Office of Ground Water and Drinking Water, Standards and Risk Management Division (Mail Code 4607M), Environmental Protection Agency, 1200 Pennsylvania Avenue NW, Washington, DC 20460; telephone number: (202) 564-1735; email address: hernandez.samuel@epa.gov.

SUPPLEMENTARY INFORMATION:

In addition to the proposed regulation, the EPA is requesting comment on three alternatives: (1) Whether the MCL and MCLG for perchlorate should be set at 0.018 mg/L (18 µg/L), (2) whether the MCL and MCLG for perchlorate should be set at 0.090 mg/L (90 µg/L), or (3) whether instead of issuing a national primary drinking water regulation, the EPA should withdraw the Agency's February 11, 2011, determination to regulate perchlorate in drinking water based on new information that indicates that perchlorate does not occur in public water systems with a frequency and at levels of public health concern and there may not be a meaningful opportunity for health risk reduction through a drink

ng a national primary drinking water regulation, the EPA should withdraw the Agency's February 11, 2011, determination to regulate perchlorate in drinking water based on new information that indicates that perchlorate does not occur in public water systems with a frequency and at levels of public health concern and there may not be a meaningful opportunity for health risk reduction through a drinking water regulation. Under this last alternative, the final action would be a withdrawal of the determination to regulate and there would be no MCLG or national primary drinking water regulation for perchlorate. This proposed rule is organized as follows:

I. General Information A. What is the EPA proposing? B. Does this action apply to me? II. Background A. What is perchlorate? B. Statutory Authority C. Statutory Framework and Regulatory History III. Assessment and Modeling of the Health Effects of Perchlorate A. 2008 Preliminary Regulatory Determinations B. 2009 Supplemental Request for Comment and 2011 Final Regulatory Determination C. Science Advisory Board Recommendations D. Perchlorate Model Development and Peer Reviews E. Sensitive Population for Deriving MCLG F. BBDR Model Specification for the Sensitive Population G. Epidemiological Literature H. Identifying a Point of Departure for Developing the MCLG I. Translate PODs to RfDs J. Translate RfD Into an MCLG IV. Maximum Contaminant Level Goal and Alternatives V. Maximum Contaminant Level and Alternatives VI. Occurrence VII. Analytical Methods VIII. Monitoring and Compliance Requirements A. What are the proposed monitoring requirements? B. Can States grant monitoring waivers? C. How are system MCL violations determined? D. When must systems complete initial monitoring? E. Can systems use grandfathered data to satisfy the initial monitoring requirements? IX. Safe Drinking Water Act Right to Know Requirements A. What are the Consumer Confidence Report requirements? B. What are the public notification requirements? X. Treatment Technologies A

tes grant monitoring waivers? C. How are system MCL violations determined? D. When must systems complete initial monitoring? E. Can systems use grandfathered data to satisfy the initial monitoring requirements? IX. Safe Drinking Water Act Right to Know Requirements A. What are the Consumer Confidence Report requirements? B. What are the public notification requirements? X. Treatment Technologies A. What are the best available technologies? B. What are the small system compliance technologies? XI. Rule Implementation and Enforcement A. What are the requirements for primacy? B. What are the State record keeping requirements? C. What are the State reporting requirements? XII. Health Risk Reduction Cost Analysis A. Identifying Affected Entities B. Method for Estimating Costs C. Method for Estimating Benefits D. Comparison of Costs and Benefits XIII. Uncertainty Analysis A. Uncertainty in the MCLG Derivation B. Uncertainty in the Economic Analysis XIV. Request for Comment on Proposed Rule XV. Request for Comment on Potential Regulatory Determination Withdrawal XVI. Statutory and Executive Order Reviews A. Executive Order 12866: Regulatory Planning and Review and Executive Order 13563 Improving Regulation and Regulatory Review B. Executive Order 13771: Reducing Regulations and Controlling Regulatory Costs C. Paperwork Reduction Act D. Regulatory Flexibility Act (RFA) E. Unfunded Mandates Reform Act F. Executive Order 13132: Federalism G. Executive Order 13175: Consultation and Coordination With Indian Tribal Governments H. Executive Order 13045: Protection of Children From Environmental Health and Safety Risks I. Executive Order 13211: Actions That Significantly Affect Energy Supply, Distribution, or Use J. National Technology Transfer and Advancement Act of 1995 K. Executive Order 12898: Federal Actions To Address Environmental Justice in Minority Populations and Low-Income Populations XVII

public water systems (PWSs) with a frequency and at levels of public health concern, and (3) in the sole judgment of the Administrator, regulation of such contaminant presents a meaningful opportunity for health risk reduction for persons served by PWSs.

Second, as explained in more detail below, the EPA is soliciting comment on two alternative MCLG/MCL values of 18 µg/L and 90 µg/L respectively. Third, in light of new considerations that have come to the EPA's attention since it issued its positive regulatory determination in 2011, including information on lower levels of occurrence of perchlorate than the EPA had previously believed to exist and new analysis of the concentration that represents a level of health concern, this action also discusses and requests comment on an alternative action under which the EPA would withdraw its 2011 determination to regulate perchlorate. Under this alternative, there would be no MCLG or NPDWR for perchlorate.

B. Does this action apply to me?

Entities that could potentially be affected include the following:

Category Examples of potentially affected entities Public water systems Community water systems: Non-transient, non-community water systems. State and tribal agencies Agencies responsible for drinking water regulatory development and enforcement. This table is not intended to be exhaustive, but rather provides a guide for readers regarding entities that could be affected by this action. To determine whether your facility or activities could be affected by this action, you should carefully examine this proposed rule. If you have questions regarding the applicability of this action to a particular entity, consult the person listed in the FOR FURTHER INFORMATION CONTACT section.

II. Background

A. What is perchlorate?

Perchlorate is a negatively charged inorganic ion that is comprised of one chlorine atom bound to four oxygen atoms (ClO 4− ), which is highly stable and mobile in the aqueous environment. Perchlorate comes from both natural and manmade sources

this action to a particular entity, consult the person listed in the FOR FURTHER INFORMATION CONTACT section.

II. Background

A. What is perchlorate?

Perchlorate is a negatively charged inorganic ion that is comprised of one chlorine atom bound to four oxygen atoms (ClO 4− ), which is highly stable and mobile in the aqueous environment. Perchlorate comes from both natural and manmade sources. It is formed naturally via atmospheric processes and can be found within mineral deposits in certain geographical areas. It is also produced in the United States, and the most common compounds include ammonium perchlorate and potassium perchlorate used primarily as oxidizers in solid fuels to power rockets, missiles, and fireworks. For the general population, most perchlorate exposure is through the ingestion of contaminated food or drinking water.

B. Statutory Authority

Section 1412(b)(1)(A) of the SDWA requires the EPA to establish NPDWRs for contaminants that may have an adverse effect on the health of persons; that are known to occur or there is a substantial likelihood that the contaminant will occur in public water systems with a frequency and at levels of public health concern; and where in the sole judgment of the Administrator, regulation of such contaminant presents a meaningful opportunity for health risk reduction for persons served by public water systems.

C. Statutory Framework and Regulatory History

Section 1412(b)(1)(B)(i) of the SDWA requires the EPA to publish every five years a Contaminant Candidate List (CCL). The CCL is a list of drinking water contaminants that are known or anticipated to occur in public water systems and are not currently subject to the EPA drinking water regulations. The EPA uses the CCL to identify priority contaminants for regulatory decision-making and information collection. Contaminants listed on the CCL may require future regulation under the SDWA. The EPA included perchlorate on the first, second, and third CCLs published in 1998, 2005, and 2009

cipated to occur in public water systems and are not currently subject to the EPA drinking water regulations. The EPA uses the CCL to identify priority contaminants for regulatory decision-making and information collection. Contaminants listed on the CCL may require future regulation under the SDWA. The EPA included perchlorate on the first, second, and third CCLs published in 1998, 2005, and 2009.

Once listed on the CCL, the Agency continues to collect data on CCL contaminants to better understand their potential health effects and to determine the levels at which they occur in drinking water. Section 1412(b)(1)(B)(ii) requires that, every five years, the EPA, after public comment, issue a determination whether or not to regulate at least five contaminants on the CCL. For any contaminant that the EPA determines meets the criteria for regulation, under Section 1412(b)(1)(E), the EPA must issue a proposed national primary drinking water regulation within two years and issue a final regulation 18 months after the proposal (which may be extended by 9 months).

As part of its responsibilities under the SDWA, the EPA implements section 1445(a)(2), “Monitoring Program for Unregulated Contaminants.” This section requires that once every five years, the EPA issue a list of no more than 30 unregulated contaminants to be monitored by public water system. This monitoring is implemented through the Unregulated Contaminant Monitoring Rule (UCMR), which collects data from community water systems (CWS) and non-transient, non-community water systems (NTNCWS). The UCMR collects data from a census of large water systems (serving more than 10,000 people) and from a statistically representative sample of small water systems. On September 17, 1999, the EPA published its first UCMR (64 FR 50556) which required all large systems and a representative sample of small systems to monitor for perchlorate and 25 other contaminants (USEPA, 1999, 2000b)

ems (NTNCWS). The UCMR collects data from a census of large water systems (serving more than 10,000 people) and from a statistically representative sample of small water systems. On September 17, 1999, the EPA published its first UCMR (64 FR 50556) which required all large systems and a representative sample of small systems to monitor for perchlorate and 25 other contaminants (USEPA, 1999, 2000b).

The EPA and other federal agencies asked the National Research Council 1 into the thyroid by a protein molecule knows as the sodium/iodide symporter (NIS), which may lead to decreases in two hormones, thyroxine (T3) and triiodothyronine (T4) and increases in thyroid-stimulating hormone (TSH) (National Research Council (NRC), 2005b). Additionally, the NRC concluded that the most sensitive population to perchlorate exposure are “the fetuses of pregnant women who might have hypothyroidism or iodide deficiency” (p. 178). The EPA established a reference dose (RfD) consistent with the recommended National Research Council RfD of 0.7 µg/kg/day for perchlorate. The reference dose is an estimate of a daily exposure to humans that is likely to be without an appreciable risk of adverse effects. This RfD was based on a study (Greer, Goodman, Pleus, & Greer, 2002) of perchlorate's inhibition of radioactive iodine uptake in healthy adults and the application of an uncertainty factor of 10 for intraspecies variability (USEPA, 2005b).

1 For the purposes of this FRN, “iodine” will be used to refer to dietary intake before entering the body. Once in the body, “iodide” will be used to refer to the ionic form.

In October 2008, the EPA published a preliminary regulatory determination not to regulate perchlorate in drinking water and requested public comment (73 FR 60262)

factor of 10 for intraspecies variability (USEPA, 2005b).

1 For the purposes of this FRN, “iodine” will be used to refer to dietary intake before entering the body. Once in the body, “iodide” will be used to refer to the ionic form.

In October 2008, the EPA published a preliminary regulatory determination not to regulate perchlorate in drinking water and requested public comment (73 FR 60262). In that preliminary determination, the EPA tentatively concluded that perchlorate did not occur with a frequency and at levels of public health concern and that development of a regulation did not present a meaningful opportunity for health risk reduction for persons served by public water systems. The EPA derived and used a Health Reference Level (HRL) of 15 μg/L based on the RfD of 0.7 µg/kg/day in making this conclusion (USEPA, 2008a). Based primarily on the UCMR 1 occurrence data, the EPA estimated that less than 1% of drinking water systems (serving approximately 1 million people) had perchlorate levels above the HRL of 15 µg/L. Based on this information the Agency determined that perchlorate did not occur frequently at levels of health concern. The EPA also determined that there was not a meaningful opportunity for a NPDWR to reduce health risks.

In January 2009 the EPA published an interim health advisory for perchlorate of 15 µg/L, consistent with the HRL derivation for perchlorate of 15 µg/L described above. Health Advisories are non-enforceable and non-regulatory and provide technical information to state agencies and other public health officials on health effects, analytical methodologies, and treatment technologies associated with drinking water contamination. Health Advisories provide the public, including the most sensitive populations, with a margin of protection from a lifetime of exposure. For perchlorate, the health advisory was developed for subchronic exposure (USEPA 2008d)

tion to state agencies and other public health officials on health effects, analytical methodologies, and treatment technologies associated with drinking water contamination. Health Advisories provide the public, including the most sensitive populations, with a margin of protection from a lifetime of exposure. For perchlorate, the health advisory was developed for subchronic exposure (USEPA 2008d).

In August 2009, the EPA published a supplemental request for comment with a new analysis that derived potential alternative HRLs for 14 life stages, including infants and children. The analysis used the RfD of 0.7 μg/kg/day and life stage-specific bodyweight and exposure information (74 FR 41883; USEPA, 2009a). After careful consideration of public comments on the October 2008 and August 2009 notices, on February 11, 2011, the EPA published its determination to regulate perchlorate (76 FR 7762; USEPA, 2011a). The Agency stated then that when considering the alternative HRL benchmarks described in the 2009 notice, the likelihood of perchlorate to occur at levels of concern had significantly increased in comparison to the levels described on the 2008 preliminary negative determination. The EPA concluded that as many as 16 million people could potentially be exposed to perchlorate at levels of concern, up from 1 million people originally described in the 2008 notice.

In its 2011 determination, the Agency found that perchlorate may have an adverse effect on the health of persons, that it is known to occur in public drinking water systems with a frequency and at levels that present a public health concern, and in the judgment of the Administrator, regulation of perchlorate presented a meaningful opportunity for health risk reduction for persons served by public water systems. As a result of the determination, and as required by Section 1412(b)(1)(E), the EPA initiated the process to develop an MCLG and NPDWR for perchlorate as described in this notice.

In September 2012, the U.S

ic health concern, and in the judgment of the Administrator, regulation of perchlorate presented a meaningful opportunity for health risk reduction for persons served by public water systems. As a result of the determination, and as required by Section 1412(b)(1)(E), the EPA initiated the process to develop an MCLG and NPDWR for perchlorate as described in this notice.

In September 2012, the U.S. Chamber of Commerce (the Chamber) submitted to the EPA a Request for Correction under the Information Quality Act regarding the EPA's regulatory determination. In the request, the Chamber claimed that the UCMR 1 data did not comply with data quality guidelines and were not representative of current conditions. In response to this request, the EPA reassessed the data and removed certain source water samples that could be paired with appropriate follow-up samples located at the entry point to the distribution system. The EPA also updated the UCMR 1 data for systems in California and Massachusetts using state compliance data to reflect current occurrence conditions after state regulatory limits for perchlorate were implemented.

In response to a lawsuit brought to enforce the deadlines in Section 1412(b)(1)(E), the U.S. District Court for the Southern District of New York entered a consent decree, requiring the EPA to propose an NPDWR with a proposed MCLG for perchlorate in drinking water no later than October 31, 2018, and finalize an NPDWR and MCLG for perchlorate in drinking water no later than December 19, 2019. The deadline for the EPA to propose an NPDWR with a proposed MCLG for perchlorate in drinking water was later extended to May 28, 2019. The consent decree is available in the docket for today's proposed rule.

III. Assessment and Modeling of the Health Effects of Perchlorate

Perchlorate inhibits uptake of iodide into the thyroid gland by competitively binding to the NIS (ATSDR, 2008; Greer et al., 2002; NRC, 2005; SAB 2013; Taylor et al., 2013)

proposed MCLG for perchlorate in drinking water was later extended to May 28, 2019. The consent decree is available in the docket for today's proposed rule.

III. Assessment and Modeling of the Health Effects of Perchlorate

Perchlorate inhibits uptake of iodide into the thyroid gland by competitively binding to the NIS (ATSDR, 2008; Greer et al., 2002; NRC, 2005; SAB 2013; Taylor et al., 2013). Iodide is necessary for the synthesis of thyroid hormones and decreased iodide uptake into the thyroid can adversely affect thyroid hormone production (SAB for the U.S. EPA, 2013; Blount et al., 2006; Steinmaus et al., 2007, 2013, 2016, McMullen et al., 2017; Knight et al., 2018). These changes in thyroid hormone levels in a pregnant woman may be linked to changes in the neurodevelopment of her offspring (SAB for the U.S. EPA, 2013; Korevaar et al., 2016; Fan and Wu, 2016; Wang et al., 2016; Alexander et al., 2017; Thompson et al., 2018). In addition, alterations in thyroid homeostasis may impact other body systems including the reproductive (Alexander et al., 2017; Hou et al., 2016; Maraka et al., 2016) and cardiovascular systems (Asvold et al., 2012; Sun et al., 2017).

The mode of action of perchlorate toxicity has been proposed as follows: exposure to perchlorate is known to inhibit the uptake of iodide by the thyroid gland through the NIS (NRC, 2005; SAB for the U.S. EPA, 2013). A sufficient inhibition of iodide uptake results in iodide deficiency within the thyroid. Given that T3 and T4 require iodide for production, a decrease in intra-thyroidal iodide can result in decreased production of these hormones. This could in turn result in increased TSH, the hormone that acts on e.g., fetuses, neonates, and children), disruptions in homeostatic thyroid hormone function can result in adverse neurodevelopmental effects (Alexander et al., 2017; Glinoer & Delange, 2000; Glinoer & Rovet, 2009; SAB for the U.S. EPA, 2013)

decrease in intra-thyroidal iodide can result in decreased production of these hormones. This could in turn result in increased TSH, the hormone that acts on e.g., fetuses, neonates, and children), disruptions in homeostatic thyroid hormone function can result in adverse neurodevelopmental effects (Alexander et al., 2017; Glinoer & Delange, 2000; Glinoer & Rovet, 2009; SAB for the U.S. EPA, 2013). Specifically, decreased maternal thyroid hormone levels during pregnancy, including in the hypothyroxinemic range, 2 have been linked to decrements in neurocognitive function in offspring (Alexander et al., 2017; Thompson et al., 2018; Wang et al., 2016). There is also limited evidence to suggest an association with other adverse neurodevelopmental outcomes including ADHD, expressive language delay, reduced school performance, autism, and delayed cognitive development (Alexander et al., 2017; Ghassabian, Bongers-Schokking, Henrichs, Jaddoe, & Visser, 2011; Gyllenberg et al., 2016; Henrichs et al., 2010; Korevaar et al., 2016, Noten et al., 2015; Pop et al., 2003, 1999; SAB for the U.S. EPA, 2013; van Mil et al., 2012).

2 Maternal hypothyroxinemia is defined as TSH in the reference range and fT4 in the lower percentiles. The SAB notes that hypothyroxinemia has been defined by a “variety of cutoffs . . . ranging from fT4 below the 10th or 5th percentiles to below the 2.5th percentile” (SAB, 2013, p.10) in the population.

The difficulty in estimating the likelihood and magnitude of the potential implications of perchlorate's mode of action on expressed neurodevelopmental health effects in humans exposed to perchlorate during development is the lack of robust epidemiological studies, especially in sensitive populations. Therefore, based on the known mode of action of perchlorate the Agency estimated potential health risks using a novel approach suggested by the EPA's Science Advisory Board (SAB for the U.S. EPA, 2013)

mode of action on expressed neurodevelopmental health effects in humans exposed to perchlorate during development is the lack of robust epidemiological studies, especially in sensitive populations. Therefore, based on the known mode of action of perchlorate the Agency estimated potential health risks using a novel approach suggested by the EPA's Science Advisory Board (SAB for the U.S. EPA, 2013). The EPA's approach to estimating perchlorate risks has evolved over time with improved research and modeling capabilities. The following sections describe information sources the EPA used in its assessment as well as the regulatory process followed by the Agency in its decision making.

A. 2008 Preliminary Regulatory Determinations

In 2005, at the request of the EPA and other federal agencies, the NRC evaluated the health implications of perchlorate ingestion. The NRC concluded that perchlorate exposure could inhibit the transport of iodide into the thyroid, leading to thyroid hormone deficiency (NRC, 2005). A significant inhibition of iodide uptake results in intra-thyroid iodide deficiency, decreased synthesis of T3 and T4, and increased TSH. The NRC also concluded that a prolonged decrease of thyroid hormones is potentially more likely to have adverse effects in sensitive populations ( e.g., the fetuses of pregnant women who might have hypothyroidism or iodide deficiency). Based on these findings, the NRC recommended a reference dose of 0.7 µg/kg/day.

Based on NRC's analysis, the EPA established a perchlorate reference dose (RfD) of 0.7 µg/kg/day in 2005 (USEPA, 2005). This value was based on a no observed effect level (NOEL) of 7 µg/kg/day identified from a study (Greer, Goodman, Pleus, & Greer, 2002) of perchlorate's inhibition of radioactive iodine uptake in healthy adults and the application of an uncertainty factor of 10 for intraspecies variability

RC's analysis, the EPA established a perchlorate reference dose (RfD) of 0.7 µg/kg/day in 2005 (USEPA, 2005). This value was based on a no observed effect level (NOEL) of 7 µg/kg/day identified from a study (Greer, Goodman, Pleus, & Greer, 2002) of perchlorate's inhibition of radioactive iodine uptake in healthy adults and the application of an uncertainty factor of 10 for intraspecies variability.

As discussed above, in 2008, the EPA derived an HRL of 15 μg/L using the RfD of 0.7 μg/kg/day, a default bodyweight of 70 kg, a default drinking water consumption rate of 2 L/day, and a perchlorate-specific relative source contribution (RSC) of 62 percent that was derived for a pregnant woman (USEPA, 2008a) (73 FR 60262). The RSC is the percentage of the RfD remaining for drinking water after other sources of exposure to perchlorate ( i.e., food) have been considered. The EPA's HRL was calculated to offer a margin of protection against adverse health effects to the subpopulation identified by the NAS as likely the most sensitive to the effects of perchlorate exposure, fetuses.

B. 2009 Supplemental Request for Comment and 2011 Final Regulatory Determination

The EPA received over 33,000 comments in response to its 2008 preliminary determination to not regulate perchlorate (USEPA, 2011a). After reviewing the comments, the EPA developed alternative HRLs for other sensitive populations in addition to fetuses of pregnant women. The EPA developed alternative HRLs for 14 life stages including infants and children. The EPA also evaluated the occurrence of perchlorate at levels above these alternative HRLs using the UCMR 1 occurrence data.

The analysis used the RfD of 0.7 μg/kg/day and life stage-specific bodyweight and exposure information ( i.e., drinking water intake, RSC) for each of the 14 life stages evaluated. The resulting HRLs ranged from 1 μg/L to 47 μg/L. In August 2009, the EPA published a supplemental request for comment with the new analysis and HRLs (74 FR 41883; USEPA, 2009a)

HRLs using the UCMR 1 occurrence data.

The analysis used the RfD of 0.7 μg/kg/day and life stage-specific bodyweight and exposure information ( i.e., drinking water intake, RSC) for each of the 14 life stages evaluated. The resulting HRLs ranged from 1 μg/L to 47 μg/L. In August 2009, the EPA published a supplemental request for comment with the new analysis and HRLs (74 FR 41883; USEPA, 2009a). After careful consideration of public comments, on February 11, 2011, the EPA published its final determination to regulate perchlorate (76 FR 7762; USEPA, 2011a).

C. Science Advisory Board Recommendations

As required by Section 1412(d) of the SDWA, as part of the NPDWR development process, the EPA requested comments from the Science Advisory Board (SAB) in 2012, seeking guidance on how best to consider and interpret the life stage information, the epidemiologic and biomonitoring data since the NRC report, physiologically-based pharmacokinetic (PBPK) analyses, and the totality of perchlorate health information to derive an MCLG for perchlorate. The SAB recommended the following:

• Derive a perchlorate MCLG that addresses sensitive life stages through physiologically based pharmacokinetic/pharmacodynamic (PBPK/PD) modeling based upon perchlorate's mode of action rather than the default MCLG approach using the RfD and specific chemical exposure parameters;

• expand the modeling approach to account for thyroid hormone perturbations and potential adverse neurodevelopmental outcomes from perchlorate exposure;

• utilize a mode-of-action framework for developing the MCLG that links the steps in the proposed mechanism leading from perchlorate exposure through iodide uptake inhibition—to thyroid hormone changes—and finally to neurodevelopmental impacts; and

• “Extend the [BBDR] model expeditiously to . .

hormone perturbations and potential adverse neurodevelopmental outcomes from perchlorate exposure;

• utilize a mode-of-action framework for developing the MCLG that links the steps in the proposed mechanism leading from perchlorate exposure through iodide uptake inhibition—to thyroid hormone changes—and finally to neurodevelopmental impacts; and

• “Extend the [BBDR] model expeditiously to . . . provide a key tool for linking early events with subsequent events as reported in the scientific and clinical literature on iodide deficiency, changes in thyroid hormone levels, and their relationship to neurodevelopmental outcomes during sensitive early life stages” (SAB for the U.S. EPA, 2013, p. 19).

This SAB-proposed framework would incorporate the previous endpoint of iodide uptake inhibition that was the basis for the RfD as part of a broader and more comprehensive framework that links perchlorate exposure to adverse neurodevelopmental outcomes. It also focuses on the smaller changes in thyroid hormones (specifically free T4 (fT4)) that are associated with maternal hypothyroxinemia and subsequent

D. Perchlorate Model Development and Peer Reviews

To address the SAB recommendations, the EPA revised an existing PBPK/PD model that describes the dynamics of perchlorate, iodide, and thyroid hormones in a woman during the third trimester of pregnancy (Lumen, Mattie, & Fisher, 2013; USEPA, 2009b). The EPA also created its own Biologically Based Dose Response (BBDR) models that included the additional sensitive life stages identified by the SAB, i.e., breast- and bottle-fed neonates and infants (SAB for the U.S. EPA, 2013, p. 19).

To determine whether the Agency had implemented the SAB recommendations for modeling thyroid hormone changes, the EPA convened an independent peer review panel to evaluate the BBDR models in January 2017 (External Peer Reviewers for USEPA, 2017)

luded the additional sensitive life stages identified by the SAB, i.e., breast- and bottle-fed neonates and infants (SAB for the U.S. EPA, 2013, p. 19).

To determine whether the Agency had implemented the SAB recommendations for modeling thyroid hormone changes, the EPA convened an independent peer review panel to evaluate the BBDR models in January 2017 (External Peer Reviewers for USEPA, 2017). In addition to estimating effects on breast fed infants, several reviewers recommended that the EPA shift the primary focus of its analysis to modeling the exposure implications to the fetus during early pregnancy. This was based on the knowledge that fetuses lack a functioning thyroid gland until approximately 16 gestational weeks and the substantial epidemiological evidence linking early pregnancy low fT4 levels with adverse neurodevelopmental outcomes (Endendijk et al., 2017, Korevaar et al., 2016; Morreale de Escobar, Obregón, & Escobar del Rey, 2004, Pop et al., 1999; Pop et al., 2003). Specifically, the SAB recommended that the EPA use specific sensitive populations to develop the MCLG for perchlorate: “the fetuses of hypothyroxinemic pregnant women, and infants exposed to perchlorate through either water-based formula preparations or the breast milk of lactating women” (SAB for the U.S. EPA, 2013, p. 19).

The EPA considered all recommendations from the 2017 peer review. The previously developed BBDR model describing perchlorate's effects in the third trimester (Lumen, Mattie, & Fisher, 2013; USEPA, 2009b) was calibrated only for that phase of pregnancy, not for the first trimester, and lacked a description of TSH signaling (feedback) that becomes significant as individuals become hypothyroxinemic or hypothyroid. In particular, this signaling was considered necessary to accurately predict responses of women with very low iodine intake, which was also part of the 2017 peer review recommendations

09b) was calibrated only for that phase of pregnancy, not for the first trimester, and lacked a description of TSH signaling (feedback) that becomes significant as individuals become hypothyroxinemic or hypothyroid. In particular, this signaling was considered necessary to accurately predict responses of women with very low iodine intake, which was also part of the 2017 peer review recommendations. Therefore, the Lumen et al., (2009b) model needed to be revised to address these recommendations and the EPA implemented those changes needed to increase the scientific rigor of the model and modeling results. These modifications include:

• Extending the model to early pregnancy;

• Incorporating biological feedback control of hormone production via TSH signaling, such that the model can describe lower levels of iodide nutrition;

• Calibrating the model and evaluating its behavior for upper and lower percentiles of the population, as well as the population median; and

• Conducting an uncertainty analysis for key parameters.

The EPA convened a second independent peer review panel in January 2018 to evaluate these updates to the BBDR model. The EPA also presented several approaches in the draft Proposed Approaches to Inform the Derivation of a Maximum Contaminant Level Goal for Perchlorate in Drinking Water (MCLG Approaches Report) to link the thyroid hormone changes in a pregnant mother predicted by the BBDR model to neurodevelopmental effects using evidence from the epidemiological literature (External Peer Review for U.S. EPA, 2018). The 2018 peer review identified a variety of strengths and limitations of the modeling (to be discussed in more detail later in this notice)

hlorate in Drinking Water (MCLG Approaches Report) to link the thyroid hormone changes in a pregnant mother predicted by the BBDR model to neurodevelopmental effects using evidence from the epidemiological literature (External Peer Review for U.S. EPA, 2018). The 2018 peer review identified a variety of strengths and limitations of the modeling (to be discussed in more detail later in this notice). The peer review panel was largely supportive of the efforts described in the MCLG Approaches Report, as evidenced by the following from the peer review final report:

Overall, the panel agreed that the EPA and its collaborators have prepared a highly innovative state-of-the-science set of quantitative tools to evaluate neurodevelopmental effects that could arise from drinking water exposure to perchlorate. While there is always room for improvement of the models, with limited additional work to address the committee's comments [in the peer-reviewed report], the current models are fit-for-purpose to determine an MCLG (External Peer Reviewers for U.S. EPA, 2018, p. 2).

The EPA also presented an alternative, population-based approach evaluating the shift in the proportion of the population that would fall below a hypothyroxinemic cut point, given exposure to perchlorate (Section 7 of the MCLG Approaches Report). This approach does not directly connect the BBDR output to a neurodevelopmental endpoint. However, for pregnant women in early pregnancy, this shift could be related to avoiding an increase in the population of offspring's risk of adverse neurodevelopmental impacts. The 2018 peer review identified strengths associated with this approach, including

lorate (Section 7 of the MCLG Approaches Report). This approach does not directly connect the BBDR output to a neurodevelopmental endpoint. However, for pregnant women in early pregnancy, this shift could be related to avoiding an increase in the population of offspring's risk of adverse neurodevelopmental impacts. The 2018 peer review identified strengths associated with this approach, including

(1) the central premise, that hypothyroxinemia is associated with adverse neurodevelopmental effects is supported by a large number of studies, including categorical studies; (2) this approach encompasses a variety of adverse neurodevelopmental outcomes, as indicated by these studies, rather than focusing on one or a limited number of adverse outcomes, as with the two-stage approach; and (3) this approach avoids all of the uncertainties associated with determining a quantitative relationship between a specific maternal fT4 level and the magnitude an adverse neurodevelopmental effect. ( External Peer Reviewers for U.S. EPA, 2018, p. 7)

The peer reviewers expressed concern about hypothyroxinemia being a precursor effect, rather than an adverse health outcome, which they argued may create difficulties in explaining the basis for an MCLG based on this approach to some audiences. However, the EPA has used precursor effects as the basis for setting regulatory and non-regulatory limits previously. The peer-review panel also expressed concern that a standard definition of hypothyroxinemia has not yet been established, as clinicians use varying fT4 thresholds to define their own working definition of the condition. This also could lead to difficulties communicating the population at risk for developing this precursor effect as a result of perchlorate exposure.

Ultimately, the EPA chose to develop the MCLG using dose-response functions from the epidemiological literature to estimate neurodevelopmental impacts in the offspring of pregnant women exposed to perchlorate

own working definition of the condition. This also could lead to difficulties communicating the population at risk for developing this precursor effect as a result of perchlorate exposure.

Ultimately, the EPA chose to develop the MCLG using dose-response functions from the epidemiological literature to estimate neurodevelopmental impacts in the offspring of pregnant women exposed to perchlorate. The EPA selected this proposed approach because it is consistent with the SDWA's definition of an MCLG to avoid adverse health effects and because it is most consistent with the SAB recommendations. The EPA is requesting public comment in Section XIV on the adequacies and uncertainties of the methodology to derive the MCLG including the decision not to pursue this population-based approach for setting the MCLG.

Based on the comments of the peer reviewers, the EPA's final analysis informing the derivation of the MCLG and benefits of avoided perchlorate exposure is based upon a 2-step

EP26JN19.008

Note: Process figure does not imply the strength of scientific evidence.

E. Sensitive Population for Deriving MCLG

SDWA 1412(b)(4)(A) requires MCLGs to be set at a concentration in water “at which no known or anticipated adverse effects on the health of persons occur and which allows an adequate margin of safety.” SDWA 1412(b)(3)(C)(V) further requires that the EPA “consider the effects of the contaminant on the general population and on groups within the general population such as infants, children, pregnant women, the elderly, individuals with a history of serious illness, or other subpopulations that are identified as likely to be at greater risk of adverse health effects due to exposure to contaminants in drinking water than the general population.” The EPA has interpreted these requirements to establish MCLGs that avoid adverse effects within the portions of the population that are at greater risk of adverse effects from exposure to the contaminant

rious illness, or other subpopulations that are identified as likely to be at greater risk of adverse health effects due to exposure to contaminants in drinking water than the general population.” The EPA has interpreted these requirements to establish MCLGs that avoid adverse effects within the portions of the population that are at greater risk of adverse effects from exposure to the contaminant. The EPA is proposing an MCLG that is developed to protect the fetuses of a first trimester pregnant mother with low-iodine intake levels ( i.e., 75 µg/kg/day), low fT4 levels ( i.e., 10th percentile of an fT4 distribution for individuals with 75 µg/day iodine intake), and weak TSH feedback strength ( i.e., TSH feedback is reduced to be approximately 60 percent less effective than for the median individual). The choice of this population is consistent with discussion by the NRC (2005), and the SAB (2013). The EPA believes that by protecting this population, the other sensitive populations ( i.e., breast- and bottle-fed infants) will also be protected. This conclusion is based on the EPA's analysis of predictions of the impact of perchlorate on fT4 levels from the original EPA BBDR model (which was peer reviewed in January of 2017) and an analysis of the literature on the connection between altered thyroid hormones in these life stages, and neurodevelopmental outcomes.

The EPA's original BBDR model demonstrated that perchlorate had minimal impact on the thyroid hormone levels for 30-, 60-, and 90-day formula-fed infants, even at doses as high as 20 µg/kg/day. Specifically, the model demonstrated that “the range of iodine levels in formula is sufficient to almost entirely offset the effects of perchlorate exposure at 30, 60 and 90 days” (USEPA, 2017; p. 73). As a result of these findings the EPA concluded that any MCLG based on the fetus of the first trimester hypothyroxinemic pregnant mother would also protect the formula-fed infant

gh as 20 µg/kg/day. Specifically, the model demonstrated that “the range of iodine levels in formula is sufficient to almost entirely offset the effects of perchlorate exposure at 30, 60 and 90 days” (USEPA, 2017; p. 73). As a result of these findings the EPA concluded that any MCLG based on the fetus of the first trimester hypothyroxinemic pregnant mother would also protect the formula-fed infant.

To determine if the same would be true for the breast-fed infant, the EPA compared the predicted percent change in fT4 experienced at given doses of perchlorate for both the breast-fed infant and the first trimester pregnant mother at varying doses of iodine intake 3 (50 to 100 µg/day). Assuming 2 or 4 µg/kg/day of perchlorate, the first trimester hypothyroxinemic pregnant mother has a greater percent change in fT4 compared to the 30 and 60 day breast-fed infant at all maternal iodine intake levels evaluated, except for the 30 day breast-fed infant of a mother consuming only 50 µg/day iodine. However, given that the original BBDR model did not have a TSH feedback loop, T4, fT4, T3 and fT3 predictions for lactating mothers with less than 75 µg/day iodine intake were considered highly uncertain because the thyroid hormone levels had fallen into the hypothyroid range.

3 Given that the current version of the BBDR model contains a TSH feedback loop and the infant models previously developed did not contain this feedback loop, this comparison is done with the feedback loop turned off.

The Agency found that there are reports in the scientific literature suggesting that minor perturbations in thyroid hormone levels in the first trimester mother may adversely impact her offspring's neurodevelopment

f the BBDR model contains a TSH feedback loop and the infant models previously developed did not contain this feedback loop, this comparison is done with the feedback loop turned off.

The Agency found that there are reports in the scientific literature suggesting that minor perturbations in thyroid hormone levels in the first trimester mother may adversely impact her offspring's neurodevelopment. Specifically, some studies show that children exposed gestationally to maternal hypothyroxinemia (without hypothyroidism) have a higher risk of reduced levels of global and specific cognitive abilities, as well as increased rates of behavior problems including greater dysregulation in early infancy and attentional disorders in childhood (Kooistra, Crawford, van Baar, Brouwers, & Pop, 2006; Man, Brown, & Serunian, 1991; Pop et al., 2003; Pop et al., 1999). Notably these effects are correlated with both degree (Henrichs et al., 2010; Pop et al., 1999) and duration (Pop et al., 2003) of maternal hypothyroxinemia (SAB for the U.S. EPA, 2013, p. 10).

The EPA did not find analogous evidence linking minor perturbations in thyroid hormones during infancy to adverse neurodevelopmental outcomes in infants. This finding is consistent with conclusions by the California Environmental Protection Agency (CalEPA) in their assessment of a public health goal for perchlorate (California Environmental Protection Agency, 2011, p. 90).

Specifically, two studies evaluated both the impact of maternal hypothyroxinemia and infant fT4 levels on subsequent neurodevelopmental outcomes. Costeira et al. (2011) found that children born to mothers with low fT4 in the first trimester had increased odds of mild-to-severe delays in

The SAB pointed to two lines of evidence supporting their suggestion of the infant as a potentially sensitive population to perchlorate: Preterm infants that experience transient hypothyroxinemia of prematurity (THOP) and infants that experience congenital hypothyroidism (SAB for the U.S. EPA, 2013)

orn to mothers with low fT4 in the first trimester had increased odds of mild-to-severe delays in

The SAB pointed to two lines of evidence supporting their suggestion of the infant as a potentially sensitive population to perchlorate: Preterm infants that experience transient hypothyroxinemia of prematurity (THOP) and infants that experience congenital hypothyroidism (SAB for the U.S. EPA, 2013). Thus, sufficient thyroid hormone levels in infancy are necessary for the infant brain to develop properly. However, the best evidence linking perturbations in thyroid hormone levels to disrupted neurodevelopment for infants are in individuals with significant thyroid deficiencies manifesting as clinical conditions ( e.g., THOP and congenital hypothyroidism). It is unclear and unknown if minor perturbations in thyroid hormones in infants, such as those that could be caused by environmental levels of perchlorate, would result in adverse neurodevelopmental outcomes similar to those seen in the literature for the offspring of first trimester pregnant mothers with hypothyroxinemia. Given the lack of evidence demonstrating minor perturbations in infant fT4 levels as being associated with neurodevelopmental outcomes, the EPA has concluded that it is appropriate to derive the perchlorate MCLG to protect the first trimester fetus of a pregnant mother with low-iodine intake. The EPA concludes that an MCLG calculated to offer a margin of protection against adverse health effects to these fetuses targets the most sensitive lifestage and will be protective of other potentially sensitive life stages as well.

F

mes, the EPA has concluded that it is appropriate to derive the perchlorate MCLG to protect the first trimester fetus of a pregnant mother with low-iodine intake. The EPA concludes that an MCLG calculated to offer a margin of protection against adverse health effects to these fetuses targets the most sensitive lifestage and will be protective of other potentially sensitive life stages as well.

F. BBDR Model Specification for the Sensitive Population

The BBDR model used to develop the proposed MCLG has two main components:

• A pharmacokinetic model for perchlorate and iodide, which describes chemical absorption, distribution, metabolism, and excretion of perchlorate and iodide; and

• A pharmacodynamic model, which describes the joint effect of varying perchlorate and iodide blood concentrations on thyroidal uptake of iodide and subsequent production of thyroid hormones, including fT4.

The pharmacokinetic model component contains a physiological description of a human mother and fetus during pregnancy ( e.g., organ volumes, blood flows) and chemical-specific information ( e.g., partition coefficients, volume of distribution, rate constants for transport, metabolism, and elimination) that enable a prediction of perchlorate and iodide internal concentration at the critical target ( i.e., thyroidal sodium-iodide symporter of the mother) in association with a particular exposure scenario (route of exposure, age, dose level). This component of the model is similar to many other PBPK models. Because perchlorate does not undergo metabolism in vivo (Clewell et al., 2007), potential uncertainty from this factor of the model is avoided since it does not need to be described.

The pharmacodynamic component of the model uses this internal concentration to simulate how the chemical will act within a known mechanism of action to perturb host systems and lead to a toxic effect

er PBPK models. Because perchlorate does not undergo metabolism in vivo (Clewell et al., 2007), potential uncertainty from this factor of the model is avoided since it does not need to be described.

The pharmacodynamic component of the model uses this internal concentration to simulate how the chemical will act within a known mechanism of action to perturb host systems and lead to a toxic effect.

Thus, the BBDR model estimates serum thyroid hormone levels in the mother at specific gestational weeks, given specific levels of iodine intake, the TSH feedback loop strength, and perchlorate doses. As noted above, to be health protective the EPA chose to model a sensitive individual (an adult woman with low iodine through the first trimester of pregnancy) to derive an MCLG, thereby protecting both this target sensitive population with an adequate margin of safety and those who are less sensitive with an even larger margin of safety.

The BBDR model simulates perchlorate's impact on thyroid hormones at each gestational week from conception to week 16. To derive the MCLG, the EPA selected outputs for gestational week 13 to correspond with the thyroid hormone data reported in Korevaar et al., (2016), which is the basis for the Agency's quantitative relationship between maternal thyroid hormone levels and neurodevelopmental impacts.

Individuals with low iodine intake have increased sensitivity to perchlorate's impact on thyroid hormone levels because the functional iodide reserve of the hypothalamic-pituitary-thyroid (HPT) system is limited (Blount et al., 2006, Steinmaus et al., 2007; Leung, Pearce, & Braverman, 2010). The EPA selected an iodine intake level of 75 µg/day to simulate an individual with low-iodine intake. This value represents an intake between the 15th and 20th percentile of the women of child bearing age population distribution of estimated iodine intake from the National Health and Nutrition Examination Survey (NHANES)

unt et al., 2006, Steinmaus et al., 2007; Leung, Pearce, & Braverman, 2010). The EPA selected an iodine intake level of 75 µg/day to simulate an individual with low-iodine intake. This value represents an intake between the 15th and 20th percentile of the women of child bearing age population distribution of estimated iodine intake from the National Health and Nutrition Examination Survey (NHANES). The EPA considered using a lower iodine intake level of 50 µg/day, which represents approximately the 5th percentile of the NHANES distribution. At 50 µg/day of iodine intake, however, the BBDR model predicts TSH levels that would be elevated to within the clinically hypothyroid range before exposure to any perchlorate 4 (TSH ranges between 4.51 and 5.41 milli-international units per liter (mIU/L) at zero dose of perchlorate when evaluating gestational weeks 12 or 13). In contrast, at 75 µg/day iodine, the BBDR modeled concentrations of serum fT4 and TSH are significantly reduced from the population median but are still within the euthyroid range. Thus, the intake of 75 µg/day is a better approximation of the sensitive population—the offspring of pregnant women who have low fT4.

4 For the purposes of this analysis, the EPA evaluated the American Thyroid Association's (ATA's) 2017 recommendations for defining hypothyroidism (Alexander et al., 2017). Specifically the ATA recommends “in the pregnancy setting, maternal hypothyroidism is defined as a TSH concentration elevated beyond the upper limit of the pregnancy-specific reference range” (Alexander et al., 2017, p. 332). ATA goes on to state, in the absence of population- and trimester-specific reference ranges defined by a provider's institute or laboratory, that the TSH reference ranges should be obtained from similar patient populations. From their recommended studies with trimester-specific data on a U.S. population, Lambert-Meserlian et al

pregnancy-specific reference range” (Alexander et al., 2017, p. 332). ATA goes on to state, in the absence of population- and trimester-specific reference ranges defined by a provider's institute or laboratory, that the TSH reference ranges should be obtained from similar patient populations. From their recommended studies with trimester-specific data on a U.S. population, Lambert-Meserlian et al. (2008) is the largest U.S.-based population with a reference range upper bound of 3.37 mIU/L for the first trimester (and 3.35 mIU/L for the second trimester). Therefore, these values were used to compare to BBDR output TSH values in the first trimester (or second trimester in cases of gestational weeks 15 and 16) to determine the presence of hypothyroidism.

TSH increases in response to decreases in T4 have been captured in numerous studies that document the relationship between these hormones (Blount et al., 2006; Steinmaus et al., 2013, 2016). The EPA designed the BBDR model to depict this feedback regulation by adjusting a set of three parameters: The number of sodium-iodide symporter sites, the T4 synthesis rate, and the T3 synthesis rate. The BBDR model allows for variability in the strength of the TSH feedback by varying these parameters with a variable called “pTSH.” For the MCLG analysis, the EPA used a pTSH value of 0.398, which is the ratio of a median value for TSH

Example output from the BBDR model for gestational week 13 and a low TSH feedback coefficient is presented in Table III-1

thesis rate, and the T3 synthesis rate. The BBDR model allows for variability in the strength of the TSH feedback by varying these parameters with a variable called “pTSH.” For the MCLG analysis, the EPA used a pTSH value of 0.398, which is the ratio of a median value for TSH

Example output from the BBDR model for gestational week 13 and a low TSH feedback coefficient is presented in Table III-1.

Table III-1—Summary of BBDR Model Results for fT4 Levels: Pregnant Women at Gestational Week 13, Assuming Low (75 µg/day ) Iodine Intake and with Muted TSH feedback strength a Perchlorate dose (μg/kg/day) Percentile fT4 (pmol/L) b (% decrease from 0 dose) 2.5th 5th 10th 50th 0 5.57 6.09 6.70 8.84 1 5.50 (−1.26%) 6.02 (−1.15%) 6.63 (−1.04%) 8.77 (−0.79%) 2 5.43 (−2.45%) 5.96 (−2.24%) 6.56 (−2.04%) 8.71 (−1.54%) 3 5.37 (−3.59%) 5.96 (−3.28%) 6.50 (−2.98%) 8.64 (−2.26%) 4 5.31 (−4.68%) 5.83 (−4.28%) 6.44 (−3.89%) 8.58 (−2.95%) 5 5.25 (−5.73%) 5.77 (−5.23%) 6.38 (−4.76%) 8.52 (−3.60%) 6 5.19 (−6.73%) 5.72 (−6.14%) 6.33 (−5.59%) 8.47 (−4.23%) 7 5.14 (−7.69%) 5.66 (−7.02%) 6.27 (−6.39%) 8.41 (−4.84%) a pTSH = 0.398; see USEPA, (2018b) for additional information on pTSH. b The 50th percentile is direct output from the BBDR model, and additional percentiles are estimated by assuming a normal distribution with a SD of 1.67. All of the examined study data demonstrated a positive skew, and overall the lognormal function demonstrated a better fit than a normal distribution. Despite this, the available study data only accounted for variation due to gestation week and did not account for variation in perchlorate and iodine intake in the measured populations. Because perchlorate and iodine can affect fT4 levels, and this relationship produced the estimated median BBDR values, the distribution around values estimated by the model from perchlorate and iodine intake should account for a small reduction in variation due to the effect of perchlorate and iodine intake

did not account for variation in perchlorate and iodine intake in the measured populations. Because perchlorate and iodine can affect fT4 levels, and this relationship produced the estimated median BBDR values, the distribution around values estimated by the model from perchlorate and iodine intake should account for a small reduction in variation due to the effect of perchlorate and iodine intake. Additionally, as iodine has a demonstrated lognormal distribution with strong right skew ( e.g., Blount et al., 2007) and is predicted to have a stronger effect on fT4 than perchlorate (see Section 3). The EPA assumed the error around predicted fT4 would likely be closer to normal than lognormal after accounting for perchlorate and iodine intake. When modeling changes in fT4, the baseline level of fT4 affects the magnitude of changes seen as a result of perchlorate exposure. Therefore, to predict the impact of perchlorate exposure on the population distribution of fT4 for the identified sensitive population, the EPA estimated a distribution for fT4 plasma concentrations around the median modeled values based on fT4 data from studies that were used to calibrate the BBDR model (C. Li et al., 2014; Männistö et al., 2011; Zhang et al., 2016). The EPA assumed the variation around predicted fT4 concentrations for women with low fT4 of childbearing age would likely be close to normal after accounting for perchlorate and iodine intake, and thus estimated a combined standard deviation (SD) using the distributional information from each of the studies (C. Li et al., 2014; Männistö et al., 2011; Zhang et al., 2016). The EPA then used the estimated combined SD to predict a distribution of fT4 around the median fT4 estimated by the BBDR model. To protect the most sensitive population from adverse effects, the EPA chose to use the 10th percentile from this distribution of baseline fT4 to conduct its analyses to account for variability in thyroid hormones in the population

nistö et al., 2011; Zhang et al., 2016). The EPA then used the estimated combined SD to predict a distribution of fT4 around the median fT4 estimated by the BBDR model. To protect the most sensitive population from adverse effects, the EPA chose to use the 10th percentile from this distribution of baseline fT4 to conduct its analyses to account for variability in thyroid hormones in the population. 5

5 For a discussion on the details of the BBDR model, including uncertainties associated with the model the reader is directed to section 3.5 of the MCLG Approaches Report.

G. Epidemiological Literature

The SAB recommended that the EPA integrate BBDR model results with data on neurodevelopmental outcomes from epidemiological studies. There is substantial epidemiological evidence that early pregnancy hypothyroxinemia is a risk factor for a variety of adverse neurodevelopmental outcomes, including those related to both cognition and behavior (Costeira et al., 2011; Finken, van Eijsden, Loomans, Vrijkotte, & Rotteveel, 2013; Ghassabian et al., 2014; Gyllenberg et al., 2016; Henrichs et al., 2010; Júlvez et al., 2013; Kooistra, Crawford, van Baar, Brouwers, & Pop, 2006; Korevaar et al., 2016; Y. Li et al., 2010; Oostenbroek et al., 2017; Päkkilä et al., 2015; Pop et al., 2003, 1999; Roman et al., 2013; van Mil et al., 2012). These individual studies showing that maternal hypothyroxinemia is associated with offspring neurodevelopment are also supported by three meta-analyses (including one full systematic review), all of which conclude maternal hypothyroxinemia is associated with increased risk of cognitive delay, intellectual impairment, or lower scores on performance tests when considering the entire body of evidence on this topic (Fan & Wu, 2016; Thompson et al., 2018; Wang et al., 2016). Additionally, the American Thyroid Association concludes that “overall, available evidence appears to show an association between hypothyroxinemia and cognitive development of the offspring” (Alexander et al., 2017, p. 337)

llectual impairment, or lower scores on performance tests when considering the entire body of evidence on this topic (Fan & Wu, 2016; Thompson et al., 2018; Wang et al., 2016). Additionally, the American Thyroid Association concludes that “overall, available evidence appears to show an association between hypothyroxinemia and cognitive development of the offspring” (Alexander et al., 2017, p. 337).

The EPA did not conduct a full systematic review and weight of evidence evaluation between maternal thyroid hormones and neurodevelopmental outcomes given: (1) The body of scientific literature regarding this association, and (2) the SAB recommendation that the EPA “consider available data on potential adverse health effects (neurodevelopmental outcomes) due to thyroid hormone level perturbations regardless of the cause of those perturbations” (p. 25). Instead, the EPA conducted a “methodologic approach to reviewing the literature” to evaluate the body of literature on this topic. This approach assisted in extrapolating the relationship modeled by the BBDR model to neurodevelopmental outcomes by concentrating on studies that allowed for evaluation of incremental changes in fT4 as they relate to incremental changes in neurodevelopmental outcomes. More specifically, the EPA only used studies that had sufficient data to show a quantitative relationship between maternal fT4 and a neurodevelopmental outcome. The EPA acknowledges that by not giving any weight to the studies that did not show

Ultimately, the EPA developed a dose-response function that estimates incremental changes in a neurodevelopmental endpoint based on a given change in thyroid hormone concentration (fT4), which could be linked to a given dose of perchlorate using the BBDR model.

The specifics of this “methodologic approach to reviewing the literature” follow

ot giving any weight to the studies that did not show

Ultimately, the EPA developed a dose-response function that estimates incremental changes in a neurodevelopmental endpoint based on a given change in thyroid hormone concentration (fT4), which could be linked to a given dose of perchlorate using the BBDR model.

The specifics of this “methodologic approach to reviewing the literature” follow. First, the EPA identified and screened the available 71 epidemiological studies, which potentially pertained to altered maternal thyroid hormone levels and offspring neurodevelopment to identify candidates based on the following criteria:

• Compatible with the sensitive life stages identified by the NRC and SAB;

• Continuous measure of thyroid hormone values (versus categorical values);

• Low risk of bias based on analysis using the National Toxicology Program's Office of Health Assessment and Translation (OHAT) Risk of Bias (ROB) tool score; and

• Access to underlying data.

Second, using these screening steps, the EPA categorized all 71 studies into three groups. One group consisted of studies that were not compatible 6 with extending the BBDR model (40 studies). Another group consisted of papers that were relevant to the pertinent life stages but did not have data from which a dose-response analysis could be conducted (15 studies). This includes studies that compared differences between groups, for example studies of offspring of mothers with hypothyroxinemia versus offspring of mothers without hypothyroxinemia. Consequently, these studies may have provided insight into the maternal thyroid hormone and offspring neurodevelopment relationship but did not have enough information to develop a continuous dose-response function. The last group of papers had data that may inform a dose-response function (16 studies). This last group of papers included publications that may have had categorical analyses but also presented data that assessed fT4 as a continuous variable and the outcome of interest

e and offspring neurodevelopment relationship but did not have enough information to develop a continuous dose-response function. The last group of papers had data that may inform a dose-response function (16 studies). This last group of papers included publications that may have had categorical analyses but also presented data that assessed fT4 as a continuous variable and the outcome of interest. In most instances, the continuous fT4 variable encompassed the full range for fT4 and not just the hypothyroxinemic range. After excluding one paper due to a high risk of bias (Kastakina et al., 2006) 15 papers remained that potentially had dose-response data between a continuous measure of fT4 and various neurodevelopmental outcomes describing cognition, behavior and other outcomes. The EPA notes that by selecting the papers that potentially had dose response data the Agency is deviating from the systematic weight of evidence review approach to identify those studies that the SAB recommended we examine to derive the MCLG.

6 For example, if the study evaluated the impact of only neonatal thyroid hormones ( i.e., at a potentially sensitive life stage), it cannot be used because the BBDR model is specific to early pregnancy. Further, if the study evaluates a population with an existing disease ( i.e., hypothyroidism) that may have a different response to perchlorate compared to the euthyroid population, it was not considered compatible with BBDR model results. Additionally, if the study does not include information on T4 or fT4, it does not assist in understanding the implications of the BBDR modeling results. Another reason for exclusion at this stage include that the study does not have a population with an exposure window ( i.e., when the thyroid hormone measurements are taken) that overlaps with the outputs for the BBDR model. Specifically, the study should evaluate thyroid hormone levels in pregnant mothers between conception and gestational week 16

ng the implications of the BBDR modeling results. Another reason for exclusion at this stage include that the study does not have a population with an exposure window ( i.e., when the thyroid hormone measurements are taken) that overlaps with the outputs for the BBDR model. Specifically, the study should evaluate thyroid hormone levels in pregnant mothers between conception and gestational week 16. The neurodevelopmental outcomes could be measured at any life stage.

Third, from these 15 papers five were selected for dose response assessment—four related to cognition (Finken et al., 2013; Korevaar et al., 2016; Pop et al., 2003, 1999) and one related to behavior (Endendijk, Wijnen, Pop, & van Baar, 2017). The other ten papers were excluded for a variety of reasons including updated analyses being presented in a different paper for which dose-response analysis was being conducted, lack of all the data needed to complete a dose-response assessment ( e.g., dose-response results were presented as “per standard deviation of fT4” but the standard deviation needed to fully interpret the results for a continuous function was not presented in the paper, statistical methods presented in the paper were insufficient to allow for the derivation of a concentration response function), or a lack of a relationship between maternal fT4 as a continuous variable and the outcome of interest evaluated in the paper. For example, Noten et al., (2015) found a relationship between maternal hypothyroxinemia and offspring arithmetic test performance. However, maternal fT4 as a continuous variable across the entire fT4 range was not associated with arithmetic test performance. Given this null finding, as well as the lack of published literature evaluating maternal fT4 as a continuous variable and arithmetic test performance, it would be difficult for the Agency to justify setting an MCLG based on changes in this endpoint

metic test performance. However, maternal fT4 as a continuous variable across the entire fT4 range was not associated with arithmetic test performance. Given this null finding, as well as the lack of published literature evaluating maternal fT4 as a continuous variable and arithmetic test performance, it would be difficult for the Agency to justify setting an MCLG based on changes in this endpoint.

As laid out for the peer reviewers, for each study that met the criteria identified above for dose-response modeling, a relationship between maternal thyroid hormone levels (specifically fT4) and offspring neurodevelopment was derived (see USEPA, 2018b). These relationships were either presented in the original published paper or derived by the EPA through either the digitization of figures or through re-analysis of data provided by the study authors. The EPA used the upper effect estimate (the upper bound of the 95th percent confidence interval) from each study to assure consideration of the populations likely to be at greater risk from the dose of perchlorate associated with a given change in fT4.

Table III-2 provides a summary of the changes in fT4 predicted to produce a 1, 2, and 3 percent decrease in any given neurodevelopmental effect and corresponding perchlorate doses. The choice of 1, 2, and 3% is based on the analyses for IQ, Mental Development Index (MDI), and Psychomotor Development Index (PDI). Specifically, a 1%, 2%, or 3% change from the standardized mean for each test ( i.e., 100 points) equates to a 1, 2, or 3 point change, respectively. The analyses for anxiety/depression score and SD of reaction time are based on a 1%, 2%, or 3% change from the study mean of each measure, which for anxiety/depression is 0.01, 0.02, or 0.03 points, respectively, and for reaction time is 2.7, 5.4, and 8.1 milliseconds (study mean SD of reaction time = 270 ms), respectively (Endendijk et al., 2017; Finken et al., 2013)

3 point change, respectively. The analyses for anxiety/depression score and SD of reaction time are based on a 1%, 2%, or 3% change from the study mean of each measure, which for anxiety/depression is 0.01, 0.02, or 0.03 points, respectively, and for reaction time is 2.7, 5.4, and 8.1 milliseconds (study mean SD of reaction time = 270 ms), respectively (Endendijk et al., 2017; Finken et al., 2013).

These results provide the potential impacts of perchlorate on maternal fT4 (as predicted by the BBDR model) and subsequent neurodevelopmental impacts (derived from the epidemiologic literature 7 ).

7 For a more complete description of all the studies evaluated the reader is directed to Sections 5 and 6 of the MCLG Approaches Report. For a discussion on the uncertainties related to the approach the reader is directed specifically to section 6.5.

EP26JN19.009

EP26JN19.010

H. Identifying a Point of Departure for Developing the MCLG

From the seven analyses presented in Table III-2 above, the EPA chose to use its independent analysis of the Korevaar et al., (2016) data (comprising of 3,600 useable mother/child data pairs) as the basis for calculating the point of departure (POD) for the MCLG. There are three reasons for this selection: (1) There is sufficient quantitative data to derive a health impact function for the sensitive population of interest; (2) the analysis adjusts for an appropriate set of confounders, and (3) the neurodevelopmental endpoint—intelligence quotient (IQ)—is more straightforward to interpret because there is more national and cross-national data available (more on the selection of this endpoint below). The other studies presented in Table III-2 do not provide one or more of these features (USEPA, 2018b)

ion of interest; (2) the analysis adjusts for an appropriate set of confounders, and (3) the neurodevelopmental endpoint—intelligence quotient (IQ)—is more straightforward to interpret because there is more national and cross-national data available (more on the selection of this endpoint below). The other studies presented in Table III-2 do not provide one or more of these features (USEPA, 2018b).

The five identified papers evaluated a variety of endpoints with Korevaar et al., (2016) evaluating IQ, Pop, Kuijpens, et al., (1999) and Pop, Brouwers, et al., (2003) using the Bayley Scale to evaluate PDI and MDI, Finken, van Eijsden, Loomans, Vrijkotte, and Rotteveel (2013) evaluating the SD of reaction time, and Endendijk, Wijnen, Pop, and van Baar (2017) evaluating anxiety/depression scores using the Child Behavioral Check List (CBCL). The SD of reaction time from Finken et al., (2013) was not well-received by the peer reviewers (External Peer Review for U.S. EPA, 2018) because it is difficult to ascertain the true implications of a change in the SD of reaction time. The Endendijk et al., (2017) study was identified after the peer review so no feedback was given on the appropriateness of the endpoint; however, the anxiety/depression raw score is not an intuitively interpretable endpoint. Further, neither the Endendijk et al., (2017) nor the Finken et al., (2013) analyses had functions for the sensitive life stage ( i.e., their analyses were based on the full range of fT4 levels and did not concentrate on the impacts of low-end fT4 levels). For these reasons, the Endendijk et al., (2017) and Finken et al., (2013) papers were not selected for further evaluation.

The Korevaar et al., (2016) original and independent analyses are preferable compared to the Pop, Kuijpens, et al., (1999) and Pop, Brouwers, et al., (2003) studies because neither function derived from the Pop et al., studies was adjusted for confounders

ow-end fT4 levels). For these reasons, the Endendijk et al., (2017) and Finken et al., (2013) papers were not selected for further evaluation.

The Korevaar et al., (2016) original and independent analyses are preferable compared to the Pop, Kuijpens, et al., (1999) and Pop, Brouwers, et al., (2003) studies because neither function derived from the Pop et al., studies was adjusted for confounders. Additionally, both Pop et al., papers have an N <50 compared to the Korevaar et al., analyses, which have an N of greater than 3,600. 8

8 The original Korevaar et al. (2016) analysis included 3,839 mother/child pairs. The EPA reanalysis of the Korevaar et al. (2016) data had a slightly lower N of 3,609 due to the exclusion of subjects with imputed values for maternal fT4.

Although the original Korevaar et al., (2016) analysis was the most rigorous analysis available in the literature to date, the Korevaar et al., (2016) EPA reanalysis was chosen over the original analysis because it included modifications to the analysis at the suggestion of the peer review panel. The e.g., previously included variables such as infant gender, maternal parity, birthweight, mother's body mass index (BMI), and gestational age at blood draw that are not related to both the exposure and the outcome were excluded), thus decreasing the chances of overfitting the estimation of the association between maternal fT4 and child IQ. The EPA was prompted to revisit the original Korevaar et al., (2016) model because of the feedback received during the peer review of the MCLG Approaches Report. Specifically, a member of the peer-review panel expressed the following suggestion:

Korevaar et al., [2016] controlled for instrumental variables (e.g. gestational week at fT4 measurement) as well as variables that are consequences of altered fT4 (e.g. maternal BMI), which may have biased estimates

2016) model because of the feedback received during the peer review of the MCLG Approaches Report. Specifically, a member of the peer-review panel expressed the following suggestion:

Korevaar et al., [2016] controlled for instrumental variables (e.g. gestational week at fT4 measurement) as well as variables that are consequences of altered fT4 (e.g. maternal BMI), which may have biased estimates. This study also assumed a log-linear relation between fT4 and the outcome but it is unclear whether the data fit this functional form better than a linear form. Reanalysis of the data performed by EPA should not include the variables noted above, which may have driven measures of association towards the null, and should investigate the most appropriate functional form to inform decisions about transformation of fT4 values (External Peer Reviewers for U.S. EPA, 2018, pp. 61-62).

The EPA responded to this suggestion by developing a causal model for the effect of maternal fT4 on child IQ to identify the minimum set of confounding variables, testing the proper functional form of the relationship between maternal fT4 and child IQ in the Korevaar et al., (2016) data, and making decisions about data quality and influential data points in the analysis. That is, the EPA determined that there were values of the independent variable of interest, fT4, in the original analysis that were imputed using multiple imputations. This could have impacted the effect estimate of the independent variable of interest with data that were not directly measured. The EPA reanalysis excludes these non-measured values. Subsequently, the EPA selected the Korevaar et al., (2016) reanalysis as the most appropriate function from which to assess the relationship between fT4 and IQ. 9

9 A more complete description of the EPA independent analysis of the Korevaar et al. (2016) data can be found in Section 6.3.2 of the MCLG Approaches Report

not directly measured. The EPA reanalysis excludes these non-measured values. Subsequently, the EPA selected the Korevaar et al., (2016) reanalysis as the most appropriate function from which to assess the relationship between fT4 and IQ. 9

9 A more complete description of the EPA independent analysis of the Korevaar et al. (2016) data can be found in Section 6.3.2 of the MCLG Approaches Report.

As indicated above, the EPA has utilized a health protective approach to this analysis consistent with the SDWA definition of the MCLG. The peer reviewers commented that this approach was fit-for-purpose. In particular, the Agency assumed it could estimate risk reductions based on evidence of a quantifiable relationship between thyroid hormone changes and neurodevelopmental outcomes. The existence of a quantifiable relationship between thyroid hormone changes and neurodevelopmental outcomes has strong support from the literature on the subject; however, not every study identified an association between maternal fT4 and the specified outcome of interest, and the state of the science on this relationship is constantly evolving. As explained earlier, the results of the EPA's dose-response literature review identified 31 studies that evaluated the association between maternal thyroid hormone levels and offspring neurodevelopment, with neurodevelopment defined using a variety of endpoints related to cognition, behavior, and other outcomes such as autism. Among these studies, only 16 were deemed to potentially possess information that could inform a dose-response relationship. The other 15 only presented data on categorical analyses assessing the impact of maternal hypothyroxinemia on the neurodevelopmental outcomes of interest. Therefore, because the data presented was only a comparison of two groups, there was not information that could be used to inform a dose-response function

deemed to potentially possess information that could inform a dose-response relationship. The other 15 only presented data on categorical analyses assessing the impact of maternal hypothyroxinemia on the neurodevelopmental outcomes of interest. Therefore, because the data presented was only a comparison of two groups, there was not information that could be used to inform a dose-response function.

Of the 16 studies that potentially had data to inform a dose-response function, 10 evaluated cognition using a variety of tests including various IQ tests (three papers; Ghassabian et al., 2014; Korevaar et al., 2016; Moleti et al., 2016), Bayley Scales of Infant Development (two papers; Pop et al., 1999; Pop et al., 2003), and other validated tests associated with child cognition such as expressive language delay or test performance (five papers; Finken et al., 2013; Henrichs et al., 2010; Kastakina et al., 2006; Noten et al., 2015; Oken et al., 2009). Six of these papers found a statistically significant relationship between maternal fT4, as a continuous variable, and offspring cognitive outcome (Korevaar et al., 2016; Pop et al., 1999; Pop et al., 2003; Finken et al., 2013; Henrichs et al., 2010, Kastakina et al., 2006). However, there were studies where maternal fT4 as a continuous variable was not significantly associated with the outcome of interest. For example, in Ghassabian et al., (2014) the authors found maternal hypothyroxinemia to be associated with an average of a 4.3-point reduction in IQ in their offspring compared to offspring of non-hypothyroxinemic mothers. Nevertheless, when assessing the relationship between the continuous measure of maternal fT4 as a continuous variable (across the entire range of fT4 levels) and child IQ, the authors did not find a significant relationship. Additionally, Moleti et al., (2016) found the relationship between maternal fT4 and child IQ to be consistently inversely associated with IQ scores, but their assessment failed to reach statistical significance

lationship between the continuous measure of maternal fT4 as a continuous variable (across the entire range of fT4 levels) and child IQ, the authors did not find a significant relationship. Additionally, Moleti et al., (2016) found the relationship between maternal fT4 and child IQ to be consistently inversely associated with IQ scores, but their assessment failed to reach statistical significance. This study included fewer than 60 study participants and was considered by the authors to be a pilot assessment.

In addition to the cognitive effects assessed and modeled, the EPA identified four papers that assessed maternal fT4 status and behavioral outcomes (Endendijk et al., 2017; Ghassabian et al., 2011; Modesto et al., 2015; Oostenbroek et al., 2017), one paper that assessed maternal fT4 status and autism (Roman et al., 2013) and one paper that evaluated odds of a schizophrenia diagnosis as associated with maternal thyroid hormone status (Gyllenberg et al., 2016). From this group of papers, the majority of papers found an association either between maternal hypothyroxinemia or maternal fT4 as a continuous variable and the outcome of interest (Endendijk et al., 2017; Modesto et al., 2015; Oostenbroek et al., 2017; Roman et al., 2013; Gyllenberg et al., 2016). However, this was not always the case as exemplified by Ghassabian et al., (2011) and Gyllenberg et al., (2016). Although Endendijk et al., (2017) found maternal fT4 to have a significant adverse impact on anxiety/depression using the Child Behavioral Check List (CBCL), Ghassabian et al., (2011) did not find any association between maternal thyroid hormone status and offspring score on various components of the CBCL. Additionally, Gyllenberg et al., (2016) found maternal hypothyroxinemia during early to mid-gestation was associated with 70% increased odds of schizophrenia diagnosis in offspring of hypothyroxinemic mothers compared to the offspring of non-hypothyroxinemic mothers

., (2011) did not find any association between maternal thyroid hormone status and offspring score on various components of the CBCL. Additionally, Gyllenberg et al., (2016) found maternal hypothyroxinemia during early to mid-gestation was associated with 70% increased odds of schizophrenia diagnosis in offspring of hypothyroxinemic mothers compared to the offspring of non-hypothyroxinemic mothers. Gyllenberg et al., (2016) also found an association with odds of schizophrenia diagnosis using conditional logistic regression when assessing fT4 as a continuous variable across the entire fT4 range ( i.e., not just the hypothyroxinemic range); however, this relationship was attenuated after controlling for smoking.

Not every paper the EPA located in its literature review found a statistically i.e., the initially identified 16 studies identified as potentially useful to inform a dose-response function) and the neurodevelopmental outcome of interest. However, many studies located in the EPA literature review, several meta-analyses (Fan & Wu, 2016; Thompson et al., 2018 and Wang et al., 2016), the American Thyroid Association (Alexander et al., 2017) and the U.S. EPA's SAB (2013) have concluded there is a relationship between maternal hypothyroxinemia and various neurodevelopmental outcomes. The relationship between maternal fT4 levels and neurodevelopmental outcomes appears strongest in the hypothyroxinemic range, and when looking at the entire range of fT4 as a continuous variable (as opposed to a categorical cut off), the significant relationship between the two variables may dissipate. Therefore, the EPA has concentrated on the neurodevelopmental impacts of changes in fT4 in the lower range of fT4 from the Korevaar et al., (2016) data. In an attempt to minimize uncertainty, the EPA reanalyzed the data collected by Korevaar et al., (2016) using a spline function that estimates a coefficient specifically for the low range of the fT4 data

ip between the two variables may dissipate. Therefore, the EPA has concentrated on the neurodevelopmental impacts of changes in fT4 in the lower range of fT4 from the Korevaar et al., (2016) data. In an attempt to minimize uncertainty, the EPA reanalyzed the data collected by Korevaar et al., (2016) using a spline function that estimates a coefficient specifically for the low range of the fT4 data.

There are a variety of neurodevelopmental endpoints used to examine behavior and cognition in children ( e.g., intelligence quotient (IQ), motor skills, vocabulary and language development, stimulus responsiveness, etc.). The EPA selected IQ decrements because this was the endpoint evaluated in the Korevaar et al., (2016) study. The EPA determined that the Korevaar study was the most rigorous analysis that examined the relationship between decreased thyroid hormones and neurodevelopmental effects. As such, in the derivation of the MCLG, IQ is a surrogate for a suite of potential neurodevelopmental effects that might occur to the offspring of hypothyroxinemic and iodine deficient mothers.

There are several different tests that are widely used to measure IQ in children, including the Stanford-Binet and the Wechsler Intelligence Scale for Children (WISC) (Sternberg et al., 2001). Each of these tests is intended to assess a child's global functioning and uses a numerical IQ point scale (Beres et al., 2000). IQ scores are standardized by age and sex group with a mean score of 100 points and a standard deviation of 15 (Beres et al., 2000). Although the specific tasks differ by test, all IQ tests contain a number of tasks to assess diverse skills (Sternberg et al., 2001). For example, the WISC test evaluates full-scale IQ using a combination of verbal and performance scales (verbal IQ and performance IQ may also be assessed separately) (Beres et al., 2000)

mean score of 100 points and a standard deviation of 15 (Beres et al., 2000). Although the specific tasks differ by test, all IQ tests contain a number of tasks to assess diverse skills (Sternberg et al., 2001). For example, the WISC test evaluates full-scale IQ using a combination of verbal and performance scales (verbal IQ and performance IQ may also be assessed separately) (Beres et al., 2000). The verbal scale includes tasks such as arithmetic, vocabulary, and comprehension, while the performance scale includes tasks such as picture completion, block design, and object assembly (Beres et al., 2000). The WISC was standardized using a sample of 2200 U.S. children aged 6 to 16 years old (Seashore et al., 1950). It has been well validated and has demonstrated high reliability, with a reliability coefficient of 0.96 observed across age groups (Beres et al., 2000).

Associations have been found between IQ scores and both educational achievement and attainment, though observed correlations vary widely. In a review of the literature, Sternberg et al., (2001) suggest that IQ scores explain approximately 25% of the variance in academic achievement. Evidence also suggests that IQ is linked to career outcomes and job performance, with observed correlations ranging from approximately 0.2 to 0.6 (Sternberg et al., 2001). Research suggests that children's rearing environment, including parental education, while growing up may increase IQ scores in adolescence by several points ( e.g., Kendler et al., 2015).

IQ scores have been used to help diagnose disorders such as intellectual disability and to identify children for placement into specialized learning programs (Beres et al., 2000). For example, in the Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-V) IQ scores are used in an individual's comprehensive assessment to determine intellectual disability, which pairs standardized testing of intelligence with a clinical assessment of adaptive functioning

bility and to identify children for placement into specialized learning programs (Beres et al., 2000). For example, in the Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-V) IQ scores are used in an individual's comprehensive assessment to determine intellectual disability, which pairs standardized testing of intelligence with a clinical assessment of adaptive functioning. Intellectual disability is considered for individuals with an IQ score of about 70 or below (American Psychiatric Association, 2013).

The EPA uses a variety of science policy approaches to select points of departure for developing regulatory values. For instance, in noncancer risk assessment the EPA often uses a percentage change in value. When assessing toxicological data, a 10 percent extra risk (for discrete data), or a 1 standard deviation ( i.e., 15 IQ points) change from the mean (for continuous data) is often used (USEPA, 2012). A smaller response to inform a POD has been applied when using epidemiological literature because there is an inherently more direct relationship between the study results and the exposure context and health endpoint. Given the difficulty in identifying a response below which no adverse impact occurs when considering a continuous outcome in the human population, the EPA looked to its Benchmark Dose Guidance (2012) for insight regarding a starting point. Specifically, “[a] BMR of 1% has typically been used for quantal human data from epidemiology studies” (p. 21, USEPA, 2012).

For the specific context of setting an MCLG for perchlorate, the EPA made a policy decision to evaluate the level of perchlorate in water associated with a 1 percent decrease, a 2 percent decrease, and a 3 percent decrease in the mean population IQ ( i.e., 1, 2 and 3 IQ points). The EPA selected IQ as a surrogate for neurodevelopmental effects based upon its evaluation of the epidemiologic literature describe above

ontext of setting an MCLG for perchlorate, the EPA made a policy decision to evaluate the level of perchlorate in water associated with a 1 percent decrease, a 2 percent decrease, and a 3 percent decrease in the mean population IQ ( i.e., 1, 2 and 3 IQ points). The EPA selected IQ as a surrogate for neurodevelopmental effects based upon its evaluation of the epidemiologic literature describe above. The need to utilize the best available peer reviewed data to inform scientific assumptions and policy choices to meet the statutory requirements associated with developing an MCLG under the SDWA highlights the challenges associated with regulating chemicals for which potential effects are indirect, and scientific data do not address all uncertainties. The Agency must make a policy decision informed by science, consistent with statutory requirements even in situations where the data do not provide clear choices. To develop the proposed MCLG for perchlorate, the EPA made a policy decision to use a 2 IQ point decrement in the population-distribution of IQ for the sensitive population. By selecting this approach, the EPA is not establishing a precedent for future Agency actions on other contaminants for which there is concern about potential thyroid effects, either under the SDWA or other statutory frameworks.

Applying these response rates to the results from the reanalysis of Korevaar et al., (2016), results in a POD dose of 3.1 µg/kg/day for a 1 point decrease in the sensitive population's IQ, a POD dose of 6.7 µg/kg/day for a 2 point decrease in the sensitive population's IQ, and a POD dose of 10.8 µg/kg/day for a 3 point decrease in the sensitive population's IQ. These PODs associated with a 1, 2, or 3 point decrease from the standardized mean IQ are calculated for the most sensitive population

lts in a POD dose of 3.1 µg/kg/day for a 1 point decrease in the sensitive population's IQ, a POD dose of 6.7 µg/kg/day for a 2 point decrease in the sensitive population's IQ, and a POD dose of 10.8 µg/kg/day for a 3 point decrease in the sensitive population's IQ. These PODs associated with a 1, 2, or 3 point decrease from the standardized mean IQ are calculated for the most sensitive population. Specifically, the POD is designed to provide an adequate margin of safety for the fetuses of mothers with fT4 at the 10th percentile of a population with iodine intake of 75 µg/day and a TSH feedback loop that is less than 60% as effective as individuals with median

I. Translate PODs to RfDs

When deriving an RfD the EPA evaluates whether to apply uncertainty/variability factors to account for heterogeneity of effect in the target population and data gaps (USEPA, 2002). As presented in A Review of the RfD & RfC Processes (USEPA, 2002) the EPA considers the following uncertainty factors: Inter-individual variability, interspecies uncertainty, extrapolating from subchronic to chronic exposure, extrapolating from a lowest-observed adverse effect level (LOAEL) rather than from a no-observed-adverse-effect-level (NOAEL), and an incomplete database. The factors are intended to account for: (1) Variation in susceptibility among the members of the human population ( i.e., inter-individual or intraspecies variability); (2) uncertainty in extrapolating animal data to humans ( i.e., interspecies uncertainty); (3) uncertainty in extrapolating from data obtained in a study with less-than-lifetime exposure ( i.e., extrapolating from subchronic to chronic exposure); (4) uncertainty in extrapolating from a LOAEL rather than from a NOAEL; and (5) uncertainty associated with extrapolation when the database is incomplete. (U.S. EPA, 2011b) The EPA has considered each of these factors in deriving an RfD to inform an MCLG for perchlorate

lating from data obtained in a study with less-than-lifetime exposure ( i.e., extrapolating from subchronic to chronic exposure); (4) uncertainty in extrapolating from a LOAEL rather than from a NOAEL; and (5) uncertainty associated with extrapolation when the database is incomplete. (U.S. EPA, 2011b) The EPA has considered each of these factors in deriving an RfD to inform an MCLG for perchlorate.

The EPA considered variation and uncertainty in the relationship between exposure and response among the members of the human population ( i.e., uncertainty factor (UF) for within-human variability/inter-individual variability, UF H ). For this analysis a UF of 3 is used. The approach taken to derive the RfD attempts to address variability between the general population and the sensitive population. Specifically, the EPA was able to modify the strength of the TSH feedback loop and iodine intake levels in the BBDR model and concentrate on the dose-response relationship between lower level (as opposed to median level) fT4 and neurodevelopmental outcomes. However, there is still uncertainty in the relationship between perchlorate exposure and subsequent neurodevelopmental outcomes. 10 There are very few toxicokinetic calibration data available for the perchlorate to thyroid hormone relationship described in the BBDR model. On the toxicodynamic side of the BBDR model, aspects such as competitive inhibition at the NIS, depletion of iodide stores under different iodine intake levels and physiological states, and the ability of the TSH feedback loop to compensate for perturbations in thyroid function each have their own uncertain features. There are also uncertainties linking maternal fT4 levels to offspring IQ. These uncertainties include the population for which dose-response information is available ( i.e., no study is U.S

iodide stores under different iodine intake levels and physiological states, and the ability of the TSH feedback loop to compensate for perturbations in thyroid function each have their own uncertain features. There are also uncertainties linking maternal fT4 levels to offspring IQ. These uncertainties include the population for which dose-response information is available ( i.e., no study is U.S. based), a lack of study information on the iodine intake status for the population for which the dose-response information is available, uncertainties around the methods used to assess maternal fT4 measurement during pregnancy, and uncertainties related to the true distribution of fT4 for a given iodine intake.

10 For a more complete discussion on the uncertainties in the analysis the reader is directed to Sections 3.5 and 6.5 of the MCLG Approaches Report.

Further, as discussed in section III.C. of this preamble the EPA believes that protecting the fetus of a hypothyroxinemic woman will protect other identified sensitive life stages. However, there is some uncertainty due to the lack of information linking incremental changes in infant thyroid hormone levels to adverse neuorodevelopmental outcomes. In addition, this analysis is assuming that protecting a first trimester fetus from alterations in maternal fT4 will protect the fetus throughout pregnancy. This is based on epidemiologic evidence that shows the relationship between first trimester maternal fT4 and neurodevelopmental outcomes. This is potentially because before mid-gestation, the mother is the only source of thyroid hormone for the fetus (Morreale de Escobar et al., 2004). Therefore, when evaluating maternal fT4 as associated with neurodevelopmental outcomes it is critical to understand the first-trimester levels. Later in gestation, when the fetal thyroid begins secreting thyroid hormones, maternal fT4 may no longer be a good surrogate for the thyroid hormone levels available to the fetus

y source of thyroid hormone for the fetus (Morreale de Escobar et al., 2004). Therefore, when evaluating maternal fT4 as associated with neurodevelopmental outcomes it is critical to understand the first-trimester levels. Later in gestation, when the fetal thyroid begins secreting thyroid hormones, maternal fT4 may no longer be a good surrogate for the thyroid hormone levels available to the fetus. Given that the fetal thyroid has had little time to develop, its iodine storage is much less than that of an adult, hence there may be more sensitivity to short-term fluctuations in iodine availability and uptake that may have little impact on maternal levels. Therefore, there is some uncertainty about the impact perchlorate may have on the fetal thyroid gland, and subsequent neurodevelopmental impacts, in later trimesters of pregnancy. The immature fetal HPT axis has very limited capacity to increase output of thyroid hormones (Savin, Cvejić, Nedić, & Radosavljević, 2003; van Den Hove, Beckers, Devlieger, De Zegher, & De Nayer, 1999), so the fetal HPT may not be able to adjust output in the face of reduced maternal fT4 supply and perchlorate exposure. Therefore, as described above, the EPA selected an intraspecies UF of 3 to account for the uncertainties in modeling the impacts of perchlorate ingestion on the thyroid hormone levels for pregnant mothers with low iodide intake, and the uncertainties in predicting the neurodevelopmental effects of these thyroid hormone changes on their children.

The EPA considered but did not derive a Data-Dependent Extrapolation Factor (DDEF) for this analysis. As described above, the UFs are applied based on the uncertainties in the perchlorate to thyroid hormone and thyroid hormone to neurodevelopment relationship. 11 As noted above, the Agency has opted to apply a UF of 3 to the POD, which adds an adequate margin of safety to the MCLG derivation. Section 4.4.5.3 (p

ered but did not derive a Data-Dependent Extrapolation Factor (DDEF) for this analysis. As described above, the UFs are applied based on the uncertainties in the perchlorate to thyroid hormone and thyroid hormone to neurodevelopment relationship. 11 As noted above, the Agency has opted to apply a UF of 3 to the POD, which adds an adequate margin of safety to the MCLG derivation. Section 4.4.5.3 (p. 4-42) of A Review of the RfD & RfC Processes recommends reducing the intraspecies UF from a default of 10 “only if data are sufficiently representative of the exposure/dose-response data for the most susceptible subpopulation(s)” (p. xviii, USEPA, 2002). The EPA selected a UF of 3 instead of the full 10 because the modeled groups within the population that are identified as likely to be at greater risk to perchlorate in drinking water ( i.e., the fetus of the iodide deficient pregnant mother) and has selected model parameters to account for the most sensitive individuals in that group ( i.e., muted TSH feedback, low fT4 values, low-iodine intake).

11 As explained in U.S. EPA, 2014 “UFs incorporate both extrapolation components that address variability (heterogeneity between species or within a population) and components that address uncertainty ( i.e., lack of knowledge) . . . whereas DDEFs focus on variability” (p. 7, US EPA, 2014).

Below we list the other uncertainty factors added and the justification.

• Uncertainty in extrapolating animal data to humans ( i.e., interspecies uncertainty) (uncertainty factor, animal-to-human, UF A ). For this analysis an UF of 1 is used because this factor is not applicable since animal studies were

• Uncertainty in extrapolating data obtained in a study with less-than-lifetime exposure to lifetime exposure ( i.e., extrapolating from subchronic to chronic exposure, UF S ). An uncertainty factor of 1 is used

umans ( i.e., interspecies uncertainty) (uncertainty factor, animal-to-human, UF A ). For this analysis an UF of 1 is used because this factor is not applicable since animal studies were

• Uncertainty in extrapolating data obtained in a study with less-than-lifetime exposure to lifetime exposure ( i.e., extrapolating from subchronic to chronic exposure, UF S ). An uncertainty factor of 1 is used. Extrapolating from subchronic to chronic exposures did not occur as the BBDR model was designed to assess long-term steady-state conditions in the non-pregnant woman and week-to-week variation in pregnancy, rather than short-term (hour-to-hour or day-to-day) fluctuations.

• Uncertainty in extrapolating from a LOAEL rather than from a NOAEL (uncertainty factor, LOAEL-to-NOAEL, UF L ). A more sophisticated BBDR modeling approach, coupled with extrapolation to changes in IQ using linear regression, was used to determine a POD that would not be expected to represent an adverse effect. Subsequently an uncertainty factor of 1 is used. LOAELs and NOAELs were not identified or used in this approach.

• Uncertainty factor for database deficiency to address the potential for deriving an inadequately protective RfD in the instance where the available database provides an incomplete characterization of the chemical's toxicity (database deficiency, UF D ; USEPA, 2002). An uncertainty factor of 1 is used as “[t]he mode of action of perchlorate toxicity is well understood” (SAB for the U.S. EPA, 2013, p. 2).

• The product of all the uncertainty factors (UF H ) is 3 (3 × 1 × 1 × 1 × 1).

Below we generate RfD's for each of the points of departure

atabase provides an incomplete characterization of the chemical's toxicity (database deficiency, UF D ; USEPA, 2002). An uncertainty factor of 1 is used as “[t]he mode of action of perchlorate toxicity is well understood” (SAB for the U.S. EPA, 2013, p. 2).

• The product of all the uncertainty factors (UF H ) is 3 (3 × 1 × 1 × 1 × 1).

Below we generate RfD's for each of the points of departure.

Using the POD of 6.7 μg/kg/day based on a 2 percent decrease in the population standardized mean IQ from the EPA's independent analysis of the Korevaar et al., (2016) data, the EPA can derive a RfD by incorporating the UF H , which results in the following:

EP26JN19.011

Using an alternative POD of 3.1 μg/kg/day based on a 1 percent decrease in the population standardized mean IQ from the EPA's independent analysis of the Korevaar et al., (2016) data, the EPA can derive an RfD by incorporating the UF H . This results in the following:

EP26JN19.012

Using an alternative POD of 10.8 μg/kg/day based on a 3 percent decrease in the population standardized mean IQ from the EPA's independent analysis of the Korevaar et al., (2016) data, the EPA can derive an RfD by incorporating the UF H . This results in the following:

EP26JN19.013

J. Translate RfD Into an MCLG

To translate the RfD (μg/kg/day) to a concentration in drinking water (μg/L), the EPA used the following equation:

EP26JN19.014

Where: W = drinking water concentration of perchlorate in micrograms per liter (μg/L); RfD = reference dose (1.03 μg/kg/day for a 1 percent decrease in IQ, 2.23 μg/kg/day for a 2 percent decrease in IQ, or 3.6 μg/kg/day for a 3 percent decrease in IQ); DWI = bodyweight-adjusted drinking water ingestion rate (L/kg/day); and RSC w = relative source contribution of drinking water to overall perchlorate exposure

drinking water concentration of perchlorate in micrograms per liter (μg/L); RfD = reference dose (1.03 μg/kg/day for a 1 percent decrease in IQ, 2.23 μg/kg/day for a 2 percent decrease in IQ, or 3.6 μg/kg/day for a 3 percent decrease in IQ); DWI = bodyweight-adjusted drinking water ingestion rate (L/kg/day); and RSC w = relative source contribution of drinking water to overall perchlorate exposure. To calculate the MCLGs, the EPA selected the 90th percentile body-weight adjusted drinking water ingestion rate specific to women of childbearing age ( i.e., non-pregnant, non-lactating, 15-44 years of age (0.032 L/kg/day). This decision is consistent with the analysis used in deriving an RSC, which was performed using food consumption information for a population of women of childbearing age from NHANES. The 90th percentile is chosen to account for variability in drinking water ingestion rates, but also adds another layer of health protection for 90% of women (Table III-3).

The EPA did not use water intake data for pregnant women because the sample sizes were too small to be statistically stable. The use of the drinking water intake for 15-44 year old women is consistent with the analysis used in deriving an RSC w (described below), which was performed using food consumption information for a population of women of childbearing age from NHANES. The EPA acknowledges there is a difference in the age range defining women of childbearing age used to develop the drinking water ingestion rate and that used to develop the RSC (20-44 years of age). The age range used to develop the

The age range used for women of childbearing age in the BBDR model fits within the age range used to develop the ingestion rates provided in the Exposure Factors Handbook. Thus, the Agency believes the difference in the age ranges will have minimal impact on the resulting MCLG analysis

g water ingestion rate and that used to develop the RSC (20-44 years of age). The age range used to develop the

The age range used for women of childbearing age in the BBDR model fits within the age range used to develop the ingestion rates provided in the Exposure Factors Handbook. Thus, the Agency believes the difference in the age ranges will have minimal impact on the resulting MCLG analysis.

Table III-3—Consumers-Only Estimated Direct and Indirect Community Water Ingestion Rates From Kahn and Stralka (2008) [L/kg/day] Female population categories Sample size Mean 90th Percentile 95th Percentile Pregnant 65 a 0.014 a 0.033 a 0.043 Lactating 33 a 0.026 a 0.054 a 0.055 Non-pregnant, non-lactating, 15 to 44 years of age 2,028 0.015 0.032 0.038 a The sample size does not meet minimum reporting requirements to make statistically reliable estimates as described in the Third Report on Nutrition Monitoring in the United States, 1994-1996 (FASEB/LSRO, 1995). Individuals are exposed to perchlorate through ingestion of both food and drinking water (ATSDR 2008, Huber et al., 2011). In calculating the MCLGs, the EPA applies a relative source contribution (RSC) to the RfD to account for the percentage of the RfD remaining for drinking water after other sources of exposure to perchlorate have been considered. Thus, the RSC for drinking water is based on the following equation where “Food” is the perchlorate dose from food ingestion:

EP26JN19.015

To estimate the dose of perchlorate for women of childbearing age coming from food, the EPA implemented a data integration methodology that combined demographic variables, food consumption estimates, and perchlorate contamination estimates in food from multiple sources (USEPA, 2019c)

water is based on the following equation where “Food” is the perchlorate dose from food ingestion:

EP26JN19.015

To estimate the dose of perchlorate for women of childbearing age coming from food, the EPA implemented a data integration methodology that combined demographic variables, food consumption estimates, and perchlorate contamination estimates in food from multiple sources (USEPA, 2019c). These sources include:

• The NHANES data available from the Centers for Disease Control and Prevention's (CDC) National Center for Health Statistics (NCHS) including the What We Eat in America (WWEIA) 24-hour food diary data (CDC & NCHS, 2007, 2009, 2011); and

• The Food and Drug Administration's (FDA's) Total Diet Study (TDS) (U.S. Food and Drug Administration (FDA), 2015), which analyzes contaminants in about 280 kinds of food and beverages commonly consumed by the U.S. population.

The NHANES data provided individual food consumption profiles for female participants age 20-44 (the women of childbearing age range used for the BBDR model). The EPA matched TDS perchlorate concentrations with each food consumed by a participant and calculated each participant's daily perchlorate dose (μg/kg/day) from food using the participant's body weight. The EPA estimated each participant's perchlorate dose using both mean and 95th percentile perchlorate concentrations in food. The details of these assumptions are explained on page 5-5 of the Technical Support Document: Deriving a Maximum Contaminant Level Goal for Perchlorate in Drinking Water (USEPA 2019c). Specifically, the EPA calculated both the mean and the 95th percentile of the perchlorate levels in each food based on the 20 samples included in the TDS data. In order to estimate the 95th percentile from the 20 samples, the EPA used the second-highest test result for each food to represent the 95th percentile concentration

m Contaminant Level Goal for Perchlorate in Drinking Water (USEPA 2019c). Specifically, the EPA calculated both the mean and the 95th percentile of the perchlorate levels in each food based on the 20 samples included in the TDS data. In order to estimate the 95th percentile from the 20 samples, the EPA used the second-highest test result for each food to represent the 95th percentile concentration. While simple, this method avoids the need to assume a distributional shape for the samples, and has been used in recent publications of TDS data for iodine (Carriquiry et al., 2016). The aforementioned method for identifying the 95th percentile concentration of perchlorate from food was selected over other, more “statistically based” methods for estimating percentiles as it avoids the need to assume a distributional shape for the samples. The EPA determined that it was more reliable to assume the empirically derived distribution as the basis for selecting the 95th percentile ( i.e., assuming the distribution was equal to the distribution of samples collected in the TDS), as opposed to forcing a distributional shape, such as normal or log-normal, onto the data that may not necessarily be appropriate. With the chosen method, we can at least be sure that the distributional shape is appropriate for the data at hand, whereas by choosing the alternative that assumes a distributional shape, in many instances we would not even be certain of that. The EPA used these individual bodyweight-adjusted perchlorate doses from food to calculate distributions of perchlorate dose from food for the population of women age 20-44.

Table III-4 presents the mean and selected percentiles of the distribution of perchlorate dose from food for women ages 20-44, for both mean and 95th percentile perchlorate concentrations in food based on the TDS. To calculate the RSC, the EPA selected the 90th percentile dose of perchlorate from food, assuming a scenario where the food contained the 95th percentile perchlorate concentration

44.

Table III-4 presents the mean and selected percentiles of the distribution of perchlorate dose from food for women ages 20-44, for both mean and 95th percentile perchlorate concentrations in food based on the TDS. To calculate the RSC, the EPA selected the 90th percentile dose of perchlorate from food, assuming a scenario where the food contained the 95th percentile perchlorate concentration. This corresponds to a perchlorate dose for food of 0.45 μg/kg/day. The EPA chose to use the 90th percentile bodyweight-adjusted perchlorate consumption from food using the 95th percentile TDS results to estimate the perchlorate RSC from drinking water. The EPA believes this is the most appropriate value for perchlorate consumption from food to ensure the protection of potentially highly exposed individuals. Given the range of perchlorate concentrations in food, and that food is the only other exposure source being considered in the RSC analysis, the EPA believes it is sufficiently protective to estimate the MCLG for drinking water using the 90th percentile bodyweight-adjusted perchlorate consumption based on the 95th percentile perchlorate food concentrations in TDS. This assures that highly exposed individuals from this most sensitive population are considered in the evaluation of whether perchlorate is found at levels of health concern.

Table III-4—Perchlorate Dose From Food (μg/kg/day) in U.S. Women Ages 20-44 Using the Mean and 95th Percentile TDS Results 1 Level of bodyweight adjusted perchlorate consumption from population distribution Perchlorate dose from food (μg/kg/day) Based on mean concentrations of perchlorate in food Based on 95th percentile concentrations of perchlorate in food Mean 0.09-0.12 0.23-0.24 50th Percentile 0.08-0.10 0.17-0.19 90th Percentile 0.18-0.21 0.45 99th Percentile 0.33-0.38 1.16-1.17 1 Ranges are due to various approaches for handling values level of detection. If no range is presented all approaches resulted in the same value. Bolded value represents the selected value

perchlorate in food Based on 95th percentile concentrations of perchlorate in food Mean 0.09-0.12 0.23-0.24 50th Percentile 0.08-0.10 0.17-0.19 90th Percentile 0.18-0.21 0.45 99th Percentile 0.33-0.38 1.16-1.17 1 Ranges are due to various approaches for handling values level of detection. If no range is presented all approaches resulted in the same value. Bolded value represents the selected value. The EPA used the drinking water intake and perchlorate dose from food to calculate MCLGs for the three RfD values. Table III-5 shows the RSC values for the three RfD values and the corresponding MCLGs calculated using the EPA's standard equation.

EP26JN19.016

IV. Maximum Contaminant Level Goal and Alternatives

Section 1412(a)(3) of the SDWA requires the EPA to propose a maximum contaminant level goal (MCLG) simultaneously with the NPDWR. The MCLG is defined in Section 1412(b)(4)(A) as “the level at which no known or anticipated adverse effects on the health of persons occurs and which allows an adequate margin of safety.” The EPA is proposing an MCLG of 56 μg/L based on the rationale and methology described in Section III above. The derivation of the proposed MCLG uses a point of departure based upon a two percent decrease in IQ for offspring of hypothyroxinemic women of child bearing age have with low iodine intake. The EPA selected a 2 percent decrease in IQ for the proposed

As described in Section III, the EPA has selected model parameters and other factors for the derivation of the MCLG that are health protective, including the focus on the most sensitive life stage. The EPA believes that the selection of the combination of protective parameters and this point of departure assures no known or anticipated adverse effects on the health of the most sensitive subpopulation and allows for an adequate margin of safety. The EPA also acknowledges the uncertainties in the derivation of the proposed (and alternative) MCLGs

ing the focus on the most sensitive life stage. The EPA believes that the selection of the combination of protective parameters and this point of departure assures no known or anticipated adverse effects on the health of the most sensitive subpopulation and allows for an adequate margin of safety. The EPA also acknowledges the uncertainties in the derivation of the proposed (and alternative) MCLGs. The EPA acknowledges in particular the challenge associated with selecting the decrement of IQ that represents an adverse effect at the population level and the uncertainties in predicting the dose of perchlorate that may result in a particular IQ decrement given the absence of robust human epidemiological data directly linking perchlorate exposure to IQ decrements. The Agency seeks comment on the alternative MCLG values of 18 μg/L and 90 μg/L, which the EPA derived using the methodology described in Section III based on a one percent and three percent decrease in IQ, respectively.

V. Maximum Contaminant Level and Alternatives

Under section 1412(b)(4)(B) of the SDWA, the EPA must establish a maximum contaminant level (MCL) as close to the MCLG as is feasible. The EPA evaluated available analytical methods to determine the lowest concentration at which perchlorate can be measured and evaluated the treatment technologies for perchlorate that have been examined under field conditions (USEPA 2018a, 2019b). The EPA determined that setting an MCL equal to the proposed MCLG of 56 μg/L is feasible given that the approved analytical method for perchlorate for UCMR 1 has a minimum reporting level (MRL) of 4 μg/L (USEPA 1999, 2000c) and that available treatment technologies can treat to concentrations well below 56 μg/L (USEPA, 2018c). Therefore, the EPA is proposing to set the MCL for perchlorate at 56 μg/L.

Because the EPA is taking comment on alternative MCLG values of 18 μg/L and 90 μg/L the Agency evaluated the feasibility of setting an MCL at these levels

s a minimum reporting level (MRL) of 4 μg/L (USEPA 1999, 2000c) and that available treatment technologies can treat to concentrations well below 56 μg/L (USEPA, 2018c). Therefore, the EPA is proposing to set the MCL for perchlorate at 56 μg/L.

Because the EPA is taking comment on alternative MCLG values of 18 μg/L and 90 μg/L the Agency evaluated the feasibility of setting an MCL at these levels. The EPA determined that the proposed MCL of 56 μg/L is feasible, therefore a higher MCL alternative such as 90 μg/L is also feasible. The EPA has concluded that analytical methods are capable of measuring perchlorate at 18 μg/L and that treatment technologies have been demonstrated to achieve this level under field conditions (USEPA 2018a, 2019b). Therefore, the EPA is requesting comment on the feasibility of the proposed MCL of 56 μg/L as well as the feasibility of the alternative MCLs of 18 μg/L and 90 μg/L.

As the occurrence analysis in section VI demonstrates, there is infrequent occurrence of perchlorate at 18 μg/L, 56 μg/L, or 90 μg/L. Therefore, the EPA did not evaluate alternative MCL values greater than the corresponding MCLG values. The purpose for evaluating alternative MCL values is to determine whether there is an MCL at which benefits justify the costs of setting an MCL. Given infrequent occurrence, the majority of the costs associated with establishing an NPDWR for perchlorate are for administrative and initial monitoring activities (see section XI.B), which will not be significantly affected by MCL values greater than corresponding MCLG values

ng alternative MCL values is to determine whether there is an MCL at which benefits justify the costs of setting an MCL. Given infrequent occurrence, the majority of the costs associated with establishing an NPDWR for perchlorate are for administrative and initial monitoring activities (see section XI.B), which will not be significantly affected by MCL values greater than corresponding MCLG values.

When proposing an MCL, the EPA must publish, and seek public comment on, the health risk reduction and cost analyses (HRRCA) of each alternative MCL considered (SDWA Section 1412(b)(3)(C)(i)), including: The quantifiable and nonquantifiable health risk reduction benefits attributable to MCL compliance; the quantifiable and nonquantifiable health risk reduction benefits of reduced exposure to co-occurring contaminants attributable to MCL compliance; the quantifiable and nonquantifiable costs of MCL compliance; the incremental costs and benefits of each alternative MCL; the effects of the contaminant on the general population and sensitive subpopulations likely to be at greater risk of exposure; any adverse health risks posed by compliance; and other factors such as data quality and uncertainty. The EPA provides this information in section XII in this preamble. The EPA must base its action on the best available, peer-reviewed science and supporting studies, taking into consideration the quality of the information and the uncertainties in the benefit-cost analysis (SDWA Section 1412(b)(3)). The following sections, as well as the health effects discussion in section III document the science and studies that the EPA relied upon to develop estimates of benefits and costs and understand the impact of uncertainty on the Agency's analysis.

VI. Occurrence

The UCMR 1 is the primary source of occurrence data the EPA relied on to estimate the number of water systems (and associated population) expected to be exposed at levels of perchlorate which could potentially exceed the proposed and alternative MCL levels

PA relied upon to develop estimates of benefits and costs and understand the impact of uncertainty on the Agency's analysis.

VI. Occurrence

The UCMR 1 is the primary source of occurrence data the EPA relied on to estimate the number of water systems (and associated population) expected to be exposed at levels of perchlorate which could potentially exceed the proposed and alternative MCL levels. Since UCMR 1 data was first used to inform the Agency actions on the 2008 preliminary regulatory determination and the 2011 final regulatory determination, the Agency has modified its analysis of the UCMR 1 data set in response to concerns raised by stakeholders regarding the data quality and to represent current conditions at some States that have enacted perchlorate regulations since the UCMR 1 data was collected. Despite these updates, the EPA continues to rely on the UCMR 1 data because they are the best available data collected in accordance with accepted methods from a census of the large water systems (serving more than 10,000 people) and a statistically representative sample of small water systems that provides the best available, national assessment of perchlorate occurrence in drinking water.

In 1999, the EPA developed the first round of the UCMR program in accordance with SDWA requirements to provide national occurrence information on unregulated contaminants (USEPA, 1999, 2000b). The UCMR 1 required sampling from systems in all 50 States, the District of Columbia, four U.S

ter systems that provides the best available, national assessment of perchlorate occurrence in drinking water.

In 1999, the EPA developed the first round of the UCMR program in accordance with SDWA requirements to provide national occurrence information on unregulated contaminants (USEPA, 1999, 2000b). The UCMR 1 required sampling from systems in all 50 States, the District of Columbia, four U.S. territories, and tribal lands in five EPA Regions including:

• All 3,097 large (serving more than 10,000 people) CWSs and NTNCWSs, which analyzed either four quarterly samples collected at 3-month intervals (surface water sources), or two samples collected 5 to 7 months apart (ground water sources); and

• a statistically representative selection of 800 small CWSs and NTNCWSs, which analyzed either four quarterly samples collected at 3-month intervals (surface water sources) or two samples collected 5 to 7 months apart (ground water sources).

Water systems submitted UCMR 1 sampling results to the EPA from 2001 until 2005. Water systems were required to analyze samples for 26 contaminants including perchlorate. The EPA established a minimum reporting level of 4 μg/L for perchlorate in the UCMR.

The EPA conducted a data quality review of the UCMR 1 data submitted by systems prior to analyzing the occurrence data for the 2011 perchlorate regulatory determination. The UCMR 1 dataset used by the EPA included 34,331 samples with 637 measurements of perchlorate above the minimum reporting level from 3,865 systems.

In September of 2012, the EPA received a “Request for Correction” i.e., the occurrence of perchlorate in drinking water) used by the EPA in its 2011 determination to regulate perchlorate. The U.S. Chamber of Commerce letter stated that the EPA relied upon: (1) Data that did not comply with data quality guidelines and (2) data that was not representative of current conditions.

In response 12 to the U.S

tember of 2012, the EPA received a “Request for Correction” i.e., the occurrence of perchlorate in drinking water) used by the EPA in its 2011 determination to regulate perchlorate. The U.S. Chamber of Commerce letter stated that the EPA relied upon: (1) Data that did not comply with data quality guidelines and (2) data that was not representative of current conditions.

In response 12 to the U.S. Chamber of Commerce, the EPA conducted a detailed assessment of the source water sample detections and determined that it was most appropriate to exclude the source water sample detections from the UCMR 1 perchlorate data set when those samples had appropriate follow-up entry point samples that were included in the UCMR 1 perchlorate data set. In contrast, any source water sample perchlorate detections for which no follow-up entry point sampling was conducted by PWSs were retained in the UCMR 1 perchlorate data set. As a result of the assessment, the EPA removed 199 source water samples (97 detections) that could be paired with a second follow-up sample located at the entry point to the distribution system. Following this convention, the resulting UCMR 1 data set contains 34,132 perchlorate samples from 3,865 systems with a total of 540 detections from 149 PWSs.

12 See the EPA response letter at https://www.epa.gov/sites/production/files/2017-08/documents/12004-response_0.pdf .

Table VI-1 shows sample distribution by system size category and measurement status. It also shows the number of entry points and systems where perchlorate measurements were reported. The entry point estimates differ from the system estimates because many water systems have more than one entry point. For example, a ground water system with two wells that has separate connections to the distribution system has two entry points.

In response to the U.S. Chamber of Commerce request, the EPA has also reassessed the UCMR 1 data in light of the adoption of regulatory limits in two states

ntry point estimates differ from the system estimates because many water systems have more than one entry point. For example, a ground water system with two wells that has separate connections to the distribution system has two entry points.

In response to the U.S. Chamber of Commerce request, the EPA has also reassessed the UCMR 1 data in light of the adoption of regulatory limits in two states. Massachusetts promulgated a drinking water standard for perchlorate of 2 μg/L in 2006 (MassDEP, 2006), and California promulgated a drinking water standard of 6 μg/L in 2007 (California Department of Public Health, 2007). Systems in these states are now required to keep perchlorate levels in drinking water below their state limits, which are lower than the proposed MCL and alternative MCLs. Therefore, the UCMR 1 sampling results from systems in these states do not reflect the current occurrence and exposure conditions. For the purpose of estimating the costs and benefits of the proposed rule, the EPA assumed that no additional monitoring and treatment costs would be incurred by the systems in the States of California and Massachusetts. Systems in California account for some of the perchlorate measurements reported below. The notes in the tables below indicate whether results include or exclude systems in California and Massachusetts.

To update the occurrence data for systems sampled during UCMR 1 from the States of California and Massachusetts, the EPA identified all systems and corresponding entry points which had reported perchlorate detections in UCMR 1

for some of the perchlorate measurements reported below. The notes in the tables below indicate whether results include or exclude systems in California and Massachusetts.

To update the occurrence data for systems sampled during UCMR 1 from the States of California and Massachusetts, the EPA identified all systems and corresponding entry points which had reported perchlorate detections in UCMR 1. Once the systems and entry points with detections were appropriately identified, the EPA then used a combination of available data from Consumer Confidence Reports (CCRs) and perchlorate compliance monitoring data from California ( https://sdwis.waterboards.ca.gov/PDWW/ ) and Massachusetts ( https://www.mass.gov/service-details/public-water-supplier-document-search ) to match current compliance monitoring data (where available) to the corresponding water systems and entry points sampled during UCMR 1.

Out of the 540 detections previously described the EPA updated data for 321 detections (320 from California systems and 1 from a Massachusetts system). The convention used by the EPA to accomplish the substitution of data was to match entry points with compliance data for active entry points based on most recently reported compliance monitoring data, if more than one data point was reported for an entry point, the assigned value is an average of the annual monitoring results at the entry point. In cases were the EPA could not find updated entry point data, then the original data from UCMR 1 for such entry point was kept

o match entry points with compliance data for active entry points based on most recently reported compliance monitoring data, if more than one data point was reported for an entry point, the assigned value is an average of the annual monitoring results at the entry point. In cases were the EPA could not find updated entry point data, then the original data from UCMR 1 for such entry point was kept.

Table VI-1—UCMR 1 Data Summary Statistics Item Small system sample Large system census Sum Total samples 3,295 30,837 34,132 Sample measurements ≥4 μg/L 15 525 540 Sample measurements >18 μg/L 1 16 17 Sample measurements >56 μg/L 0 2 2 Sample measurements >90 μg/L 0 1 1 Total entry points 1,454 13,482 14,936 Entry points at which measurements ≥4 μg/L 8 328 336 Entry points at which measurements >18 μg/L 1 16 17 Entry points at which measurements >56 μg/L 0 2 2 Entry points at which measurements >90 μg/L 0 1 1 Total systems 797 3,068 3,865 Systems at which measurements ≥4 μg/L 8 141 149 Systems at which measurements >18 μg/L 1 14 15 Systems at which measurements >56 μg/L 0 2 2 Systems at which measurements >90 μg/L 0 1 1 Source: (USEPA, 2019b). The total row counts and counts of measurements ≥4 μg/L identify all instances where perchlorate was detected at or above the minimum reporting level, including water systems in California and Massachusetts, which account for 537 systems in total and 51 systems at which measurements ≥4 μg/L. The instances where perchlorate measurements equal or exceed either 18 μg/L, 56 μg/L, or 90 μg/L exclude results from California and Massachusetts because water systems in these States must meet limits below 18 μg/L. The small system counts reflect sample results that have not been extrapolated to small systems nationwide. Table VI-2 shows the service populations that correspond with the occurrence summary in Table VI-1

chlorate measurements equal or exceed either 18 μg/L, 56 μg/L, or 90 μg/L exclude results from California and Massachusetts because water systems in these States must meet limits below 18 μg/L. The small system counts reflect sample results that have not been extrapolated to small systems nationwide. Table VI-2 shows the service populations that correspond with the occurrence summary in Table VI-1. The entry point population estimates reflect the assumption that system population is uniformly distributed across entry points; e.g., the entry point population for a system with two entry points is one-half the total system population.

Table VI-2—UCMR1 Data Service Population Summary Statistics Item Small system sample Large system census Sum Total entry point population 2,760,570 222,853,101 225,613,671 Population served by entry points at which measurements ≥4 µg/L 9,484 4,281,937 4,291,420 Population served by entry points at which measurements >18 µg/L 2,155 618,406 620,560 Population served by entry points at which measurements >56 µg/L 0 32,432 32,432 Population served by entry points at which measurements >90 µg/L 0 25,972 25,972 Total system population 2,760,570 222,853,101 225,613,671 Population served by systems at which measurements ≥4 µg/L 13,483 16,159,082 16,172,565 Population served by systems at which measurements >18 µg/L 4,309 696,871 701,180 Population served by systems at which measurements >56 µg/L 0 64,733 64,733 Population served by systems at which measurements >90 µg/L 0 25,972 25,972 Source: (USEPA, 2019b). The populations for entry points/systems with measurements ≥4 µg/L identify all instances where perchlorate was detected at or above the minimum reporting level, including water systems in California and Massachusetts, which account for 39.6 million of the 225.6 million total population in UCMR 1, and 1.9 million of the 4.3 million population served by entry points at which measurements ≥4 µg/L

). The populations for entry points/systems with measurements ≥4 µg/L identify all instances where perchlorate was detected at or above the minimum reporting level, including water systems in California and Massachusetts, which account for 39.6 million of the 225.6 million total population in UCMR 1, and 1.9 million of the 4.3 million population served by entry points at which measurements ≥4 µg/L. The instances where perchlorate measurements equal or exceed either 18 µg/L, 56 µg/L, or 90 µg/L exclude results from California and Massachusetts because water systems in these States must meet limits below 18 µg/L. The small system counts reflect sample results that have not been extrapolated to small systems nationwide. As shown in the tables, 149 systems serving 16.2 million people had measured levels of perchlorate greater than the minimum reporting level. However, many of these systems have several entry points with no measured levels of perchlorate greater than the minimum reporting level; at the entry point level, the exposed population is approximately 4.3 million people served by 336 entry points. Because the uniform population distribution assumption may over or underestimate the service population of any particular entry point, the entry point estimates are uncertain. The system population estimates serve as upper bounds on exposure.

The EPA used entry point maximum measurements to estimate potential baseline occurrence and exposure at levels that exceed the proposed MCL and alternative MCLs. The maximum measurements indicate perchlorate levels that occurred in at least one quarterly sample among surface water systems and at least one semi-annual sample among ground water systems.

Table VI-3 through Table VI-5 show the occurrence and exposure estimates based on the 56 µg/L, 18 µg/L MCL, and 90 µg/L values, respectively. Each table provides estimates of the entry points at which the maximum perchlorate concentrations exceed the MCL value

hat occurred in at least one quarterly sample among surface water systems and at least one semi-annual sample among ground water systems.

Table VI-3 through Table VI-5 show the occurrence and exposure estimates based on the 56 µg/L, 18 µg/L MCL, and 90 µg/L values, respectively. Each table provides estimates of the entry points at which the maximum perchlorate concentrations exceed the MCL value. The tables also report the system-level information for these entry points.

Table VI-3—Estimated Perchlorate Occurrence and Exposure: Entry Point Max Exceeds 56 µg/L Affected entity Small systems Large systems Total systems Entry points 0 2 2 Population served 0 32,432 32,432 Water systems 0 2 2 Population served 0 64,733 64,733 Source: (USEPA, 2019b). Table VI-4—Estimated Perchlorate Occurrence and Exposure: Entry Point Max Exceeds 18 µg/L Affected entity Small systems 1 Large systems Total systems Entry points 1 16 17 Population served 2,155 618,406 620,560 Water systems 1 14 15 Population served 4,309 696,871 701,180 Source: (USEPA, 2019b). 1 The values shown in the table are estimates based on the UCMR 1 data. The EPA also applied the statistical sampling weights to the results to extrapolate results to national results. The entry point at which a measurement exceeds 18 µg/L is one of 20 in its sample stratum; no other sample in the stratum had a measurement of perchlorate greater than the minimum reporting level. The entry point population of 2,155 represents 5.31% of the total population served by the six UCMR 1 systems in the stratum (40,574). Currently, the stratum population of 774,780 accounts for 1.32% of the 58.7 million national population served by small systems. Thus, the UCMR 1 results indicate that 0.07% (5.31% × 1.32%) of small system customers (approximately 41,100) may be exposed to perchlorate greater than 18 µg/L

on of 2,155 represents 5.31% of the total population served by the six UCMR 1 systems in the stratum (40,574). Currently, the stratum population of 774,780 accounts for 1.32% of the 58.7 million national population served by small systems. Thus, the UCMR 1 results indicate that 0.07% (5.31% × 1.32%) of small system customers (approximately 41,100) may be exposed to perchlorate greater than 18 µg/L. Table VI-5—Estimated Perchlorate Occurrence and Exposure: Entry Point Max Exceeds 90 µg/L Affected entity Small systems 1 Large systems Total systems Entry points 0 1 1 Population served 0 25,972 25,972 Water systems 0 1 1 Population served 0 25,972 25,972 Source: (USEPA, 2019b). In summary, the perchlorate occurrence information suggests that at an MCL of 56 µg/L, two systems (0.004% of all water systems in the U.S.) would exceed the regulatory threshold. One of these two systems would exceed the alternative MCL of 90 µg/L. In addition, at an MCL of 18 µg/L, there would be 15 systems (0.03% of all water systems in the U.S.) that would exceed the regulatory threshold.

VII. Analytical Methods

The SDWA directs the EPA to set a contaminant's MCL as close to its MCLG as is “feasible”, the definition of which includes an evaluation of the feasibility of performing chemical analysis of the contaminant at standard drinking water laboratories. Specifically, the SDWA directs the EPA to determine that it is economically and technologically feasible to ascertain the level of the contaminant being regulated in water in public water systems (Section 1401(1)(C)(i)). NPDWRs are also to contain “criteria and procedures to assure a supply of drinking water which dependably complies with such [MCLs]; including accepted methods for quality control and testing procedures to insure compliance with such levels.” (Section 1401(1)(D))

nologically feasible to ascertain the level of the contaminant being regulated in water in public water systems (Section 1401(1)(C)(i)). NPDWRs are also to contain “criteria and procedures to assure a supply of drinking water which dependably complies with such [MCLs]; including accepted methods for quality control and testing procedures to insure compliance with such levels.” (Section 1401(1)(D)).

To comply with these requirements, the EPA considers method performance under relevant laboratory conditions, their likely prevalence in certified drinking water laboratories, and the associated analytical costs. The EPA has developed five analytical methods for the identification and quantification of perchlorate in drinking water that meet these criteria. The proposed EPA methods for perchlorate are: 314.0, 314.1, 314.2, 331.0, and 332.0. A detailed description of these methods is presented in the Perchlorate Occurrence and Monitoring Report (USEPA, 2019b).

The EPA Methods 314.0, 314.1, 314.2, 331.0, and 332.0 underwent the EPA's analytical method development and validation processes. The validation process includes a protocol for modifications to any existing EPA-approved analytical methods and a protocol for new determinative techniques. Both validation protocols are rigorous and consider many technical aspects of analytical method performance, including: Detection limits; instrument calibration; precision and analyte recovery; analyte retention times; evaluation of blanks; development of Quality Control acceptance criteria; analysis of field samples; and other technical aspects of sample analysis and data reporting. All of the proposed EPA analytical methods provide performance data to demonstrate their capability to reliably and consistently measure perchlorate in drinking water at the proposed and alternate MCLs.

EPA Method 314.0, “Determination of Perchlorate in Drinking Water Using Ion Chromatography” (Revision 1.0, USEPA, 1999a) has a method detection limit (MDL) of 0.53 µg/L

alysis and data reporting. All of the proposed EPA analytical methods provide performance data to demonstrate their capability to reliably and consistently measure perchlorate in drinking water at the proposed and alternate MCLs.

EPA Method 314.0, “Determination of Perchlorate in Drinking Water Using Ion Chromatography” (Revision 1.0, USEPA, 1999a) has a method detection limit (MDL) of 0.53 µg/L. Single-laboratory mean percent recovery in various aqueous matrices range from 86% to 113% with Relative Standard Deviations (RSDs) of 1.0% to 12.8%. A minimum reporting level (MRL) is not specified in the method; however, a range of 3.0 to 5.0 µg/L is cited as a benchmark range for quality assurance/quality control (QA/QC) procedures. The MRL is to be established as either a concentration that is greater than three times the laboratory MDL or at a concentration that yields a response greater than a signal to noise ratio of five. In either case, the MRL must not be below the lowest instrument calibration standard (USEPA, 1999a). Method 314.0 was widely adopted as the standard perchlorate method.

After the EPA published Method 314.0, the Agency adopted additional method development goals for the analysis of perchlorate in drinking water including: (1) Reducing MRL to less than 1 µg/L through the application of sample concentration techniques, microbore analytical columns, and advanced detection systems ( i.e., mass spectrometry), (2) further increasing the tolerance for high ionic strength matrices, and (3) enhancing measurement selectivity.

EPA Method 314.1, “Determination of Perchlorate in Drinking Water Using Inline Column Concentration/Matrix Elimination Ion Chromatography with Suppressed Conductivity Detection” (Revision 1.0, USEPA, 2005b) documents the EPA single-laboratory Lowest Concentration Minimum Reporting Levels (LCMRLs) of less than 0.2 µg/L (DL = 0.03 µg/L) using online sample pre-concentration

easurement selectivity.

EPA Method 314.1, “Determination of Perchlorate in Drinking Water Using Inline Column Concentration/Matrix Elimination Ion Chromatography with Suppressed Conductivity Detection” (Revision 1.0, USEPA, 2005b) documents the EPA single-laboratory Lowest Concentration Minimum Reporting Levels (LCMRLs) of less than 0.2 µg/L (DL = 0.03 µg/L) using online sample pre-concentration. The method uses matrix diversion to handle high ionic strength matrices (up to 1,000 mg/L TDS) and added confirmation analysis using a second analytical column (USEPA, 2005b).

EPA Method 314.2, “Determination of Perchlorate in Drinking Water Using Two-Dimensional Ion Chromatography with Suppressed Conductivity Detection” (USEPA, 2008c) documents the EPA single-laboratory LCMRLs of less than 0.1 µg/L (DLs <0.02 µg/L) using large volume injection. The method uses 2-D chromatography to handle high ionic strength matrices (up to 1,000 mg/L total dissolved solids [TDS]) and eliminates the need for separate confirmation analysis (USEPA, 2008c).

EPA Method 331.0, “Determination of Perchlorate in Drinking Water by Liquid Chromatography Electrospray Ionization Mass Spectrometry” (Revision 1.0, USEPA, 2005c) documents the EPA single-laboratory LCMRLs of less than 0.1 µg/L (DLs <0.01 µg/L), applied multiple analytical advancements to a liquid chromatography (LC) analysis including a perchlorate selective LC column (AS-21), mass spectrometry (MS) or MS/MS detection for selectivity and sensitivity, and a custom labeled internal standard (Cl 18 O 4 − ) (USEPA, 2005c)

trometry” (Revision 1.0, USEPA, 2005c) documents the EPA single-laboratory LCMRLs of less than 0.1 µg/L (DLs <0.01 µg/L), applied multiple analytical advancements to a liquid chromatography (LC) analysis including a perchlorate selective LC column (AS-21), mass spectrometry (MS) or MS/MS detection for selectivity and sensitivity, and a custom labeled internal standard (Cl 18 O 4 − ) (USEPA, 2005c).

EPA Method 332.0, “Determination of Perchlorate in Drinking Water by Ion Chromatography with Suppressed Conductivity and Electrospray Ionization Mass Spectrometry” (USEPA, Revision 1.0, 2005d) documents the EPA single-laboratory LCMRL of 0.1 µg/L (DL = 0.02 µg/L), applied multiple analytical advancements in an IC analysis including suppressed conductivity IC, MS or MS/MS selectivity and sensitivity, and a custom labeled internal standard (Cl 18 O 4 ) (USEPA, 2005d).

VIII. Monitoring and Compliance Requirements

A. What are the proposed monitoring requirements?

The EPA is proposing to require CWS and NTNCWSs to monitor for perchlorate in accordance with the standardized monitoring framework set out in 40 CFR 141 Subpart C (Standardized Monitoring Framework). Public water systems must sample entry points to the distribution system consistent with requirements in 40 CFR 141.23(a).

Under the Standardized Monitoring Framework, the monitoring frequency for a public water system is dependent on previous monitoring results and whether a monitoring waiver has been granted. The EPA is proposing that consistent with the standardized monitoring framework water systems would be initially required to monitor quarterly for perchlorate. The EPA is also proposing that based upon the monitoring results States would be able to reduce the monitoring frequency to annually, once every three years or once every nine years if the State concludes that the system is reliably and consistently below the MCL

at consistent with the standardized monitoring framework water systems would be initially required to monitor quarterly for perchlorate. The EPA is also proposing that based upon the monitoring results States would be able to reduce the monitoring frequency to annually, once every three years or once every nine years if the State concludes that the system is reliably and consistently below the MCL. If a water system exceeds the perchlorate MCL, the system is in violation and triggered into quarterly monitoring for that sampling point in the next quarter after the violation occurred (40 CFR 141.23(c)(7)). The state may allow the system to return to the reduced monitoring frequency when the state determines that the system is reliably and consistently below the MCL. However, the state cannot make a determination that the system is reliably and consistently below the MCL until a minimum of 2 consecutive ground water or 4 consecutive surface water samples below the MCL have been collected (40 CFR 141.23(c)(8)). All systems must comply with the sampling requirements, unless a waiver has been granted in writing by the state (40 CFR 141.23(c)(6)).

B. Can states grant monitoring waivers?

Under this proposal, water systems may apply to the state, and states may grant, a 9-year monitoring waiver for perchlorate if the conditions described in 40 CFR 141.23(c)(3)-(6) are met. A state may grant a waiver for surface water systems after three rounds of annual monitoring with results less than the MCL and for groundwater systems after conducting three rounds of monitoring with results less than the MCL. One sample must be collected during the nine-year compliance cycle that the waiver is effective, and the waiver must be renewed every nine years.

C. How are system MCL violations determined?

Under this proposal, violations of the perchlorate MCL would be determined in a manner consistent with 40 CFR 141.23(i)(3). Compliance with the perchlorate MCL would be determined based on one sample if the level is below the MCL

cted during the nine-year compliance cycle that the waiver is effective, and the waiver must be renewed every nine years.

C. How are system MCL violations determined?

Under this proposal, violations of the perchlorate MCL would be determined in a manner consistent with 40 CFR 141.23(i)(3). Compliance with the perchlorate MCL would be determined based on one sample if the level is below the MCL. If the level of perchlorate exceeds the MCL at any entry point in the initial sample, a confirmation sample is required within two weeks of the system's receipt of notification of the analytical result of the first sample, in accordance with 141.23(f)(1). Compliance shall be determined based on the average of the initial and confirmation samples.

D. When must systems complete initial monitoring?

Pursuant to Section 1412(b)(10), this rule would be effective three years after promulgation. To satisfy initial monitoring requirements, CWS serving populations greater than 10,000 persons must collect 4 quarterly samples for perchlorate during the second compliance period of the fourth compliance cycle (January 1, 2023- December 31, 2025) of the Standardized Monitoring Framework. NTNCWS and CWSs serving 10,000 persons or less must collect 4 quarterly samples during the third compliance period of the fourth compliance cycle (January 1, 2026-December 31, 2028) of the Standardized Monitoring Framework.

E. Can systems use grandfathered data to satisfy the initial monitoring requirements?

As proposed today, systems would be allowed to use grandfathered perchlorate data collected after January 1, 2020, to satisfy the initial monitoring requirements

ples during the third compliance period of the fourth compliance cycle (January 1, 2026-December 31, 2028) of the Standardized Monitoring Framework.

E. Can systems use grandfathered data to satisfy the initial monitoring requirements?

As proposed today, systems would be allowed to use grandfathered perchlorate data collected after January 1, 2020, to satisfy the initial monitoring requirements. To satisfy initial perchlorate monitoring requirements, a system with appropriate historical monitoring data for each entry point to the distribution system could use the monitoring data from the compliance monitoring period between January 1, 2020, and December 31, 2022, for CWSs serving greater than 10,000 persons and between January 1, 2023, and December 31, 2025, for NTNCWs and for CWSs serving 10,000 or fewer persons.

IX. Safe Drinking Water Act Right to Know Requirements

A. What are the Consumer Confidence Report requirements?

A community water system must prepare and deliver to its customers an annual Consumer Confidence Report (CCR) in accordance with requirements in 40 CFR 141 Subpart O. A CCR provides customers with information about their local drinking water quality as well as information regarding the water system compliance with drinking water regulations. Under this proposal CWSs would be required to report perchlorate information in their CCR.

B. What are the public notification requirements?

All public water systems must give the public notice for all violations of NPDWRs and for other situations. Under this proposal, violations of the perchlorate MCL would be designated as Tier 1 and as such, public water systems would be required to comply with 40 CFR 141.202

CWSs would be required to report perchlorate information in their CCR.

B. What are the public notification requirements?

All public water systems must give the public notice for all violations of NPDWRs and for other situations. Under this proposal, violations of the perchlorate MCL would be designated as Tier 1 and as such, public water systems would be required to comply with 40 CFR 141.202. As described in Section III of this proposal, fetuses of first trimester pregnant women with low iodine are the most sensitive subpopulation, therefore, per 40 CFR 141.202(b)(1), notification of an MCL violation should be provided as soon as practicable but no later than 24 hours after the system learns of the violation under this proposal.

X. Treatment Technologies

Systems that exceed the perchlorate MCL will need to adopt new treatment or another strategy to reduce perchlorate to a level that meets the MCL. When the EPA establishes an MCL for a drinking water contaminant, Section 1412(b)(4)(E) of the SDWA requires that the Agency “list the technology, treatment techniques, and other means which the Administrator finds to be feasible for purposes of meeting [the MCL],” which are referred to as best available technologies (BAT). These BATs are used by states to establish conditions for source water variances under Section 1415(a). Furthermore, Section 1412(b)(4)(E)(ii) requires that the Agency identify small system compliance technologies (SSCT), which are affordable treatment technologies, or other means that can achieve compliance with the MCL (or treatment technique, where applicable). The lack of an affordable SSCT for a contaminant triggers certain additional procedures which can result in states issuing small system variances under Section 1412(e) of the SDWA.

The Agency solicits public comment on the choice of available treatment technologies discussed in this section.

A

, or other means that can achieve compliance with the MCL (or treatment technique, where applicable). The lack of an affordable SSCT for a contaminant triggers certain additional procedures which can result in states issuing small system variances under Section 1412(e) of the SDWA.

The Agency solicits public comment on the choice of available treatment technologies discussed in this section.

A. What are the best available technologies?

The Agency identifies the best available technologies (BAT) as those meeting the following criteria: (1) The capability of a high removal efficiency;

• Ion exchange;

• biological treatment; and

• centralized reverse osmosis.

There are also non-treatment options that might be used for compliance in lieu of installing and operating treatment technologies. These include blending existing water sources, replacing a perchlorate-contaminated source of

This text is long and has been trimmed here. Open the source document for the complete record.