Food & Water Watch, Inc. et al v. Environmental Protection Agency et al

Filing 445

FINDINGS OF FACT AND CONCLUSIONS OF LAW. Signed by Judge Edward M. Chen on 9/24/2024. (emclc2, COURT STAFF) (Filed on 9/24/2024)

1

UNITED STATES DISTRICT COURT

2

NORTHERN DISTRICT OF CALIFORNIA

3
4

FOOD & WATER WATCH, INC., et al.,
Plaintiffs,

5
6
7
8

UNITED STATES ENVIRONMENTAL
PROTECTION AGENCY, et al.,
Defendants.

10

United States District Court
Northern District of California

FINDINGS OF FACT AND
CONCLUSIONS OF LAW

v.

9

11

Case No. 17-cv-02162-EMC

I.

INTRODUCTION

In 2016, Congress amended the Toxic Substances Control Act (“TSCA”), empowering

12

United States citizens to petition the Environmental Protection Agency (“EPA”) to consider

13

whether a chemical presents an unreasonable risk of injury to health. See Pub. L. No. 114-182,

14

114th Congress (Frank R. Lautenberg Chemical Safety for the 21st Century Act) (the “Act”). The

15

Act addresses the modern day reality that “human beings and the environment are being exposed

16

each year to a large number of chemical substances and mixtures,” 15 U.S.C. § 2601(a)(1), and

17

that, “among the many chemical substances and mixtures which are constantly being developed

18

and produced, there are some whose manufacture, processing, distribution in commerce, use, or

19

disposal may present an unreasonable risk of injury to health or the environment,” id. §

20

2601(a)(2).

21

To this end, under TSCA, as amended by the Act (“Amended TSCA”), a citizen is entitled

22

to judicial review of the EPA’s denial of the citizen’s petition, wherein a court considers whether

23

the chemical poses an unreasonable risk de novo, i.e., without deference to the EPA’s decision.

24

See id. § 2620(b)(4)(B). Amended TSCA sets up a system of judicial review that is remarkably

25

different from the usual scope of judicial review of administrative actions under the

26

Administrative Procedure Act, which confers substantial deference to administrative agencies.

27

See id. Under Amended TSCA, the Court owes no deference to the EPA in assessing the risk

28

posed by chemical substances. See id. If the Court finds anew that the chemical at issue presents

United States District Court
Northern District of California

1

an unreasonable risk, it then orders the EPA to engage in rulemaking regarding the chemical. See

2

id. The EPA is afforded in the first instance the authority to respond; regulatory actions can range

3

from requiring a mere warning label to banning the chemical. See id. § 2605(a)(1)-(7). The EPA,

4

in short, has options. See id.

5

The issue before this Court is whether the Plaintiffs have established by a preponderance

6

of the evidence that the fluoridation of drinking water at levels typical in the United States poses

7

an unreasonable risk of injury to health of the public within the meaning of Amended TSCA. For

8

the reasons set forth below, the Court so finds. Specifically, the Court finds that fluoridation of

9

water at 0.7 milligrams per liter (“mg/L”) – the level presently considered “optimal” in the United

10

States – poses an unreasonable risk of reduced IQ in children. It should be noted that this finding

11

does not conclude with certainty that fluoridated water is injurious to public health; rather, as

12

required by the Amended TSCA, the Court finds there is an unreasonable risk of such injury, a

13

risk sufficient to require the EPA to engage with a regulatory response. This order does not dictate

14

precisely what that response must be. Amended TSCA leaves that decision in the first instance to

15

the EPA. One thing the EPA cannot do, however, in the face of this Court’s finding, is to ignore

16

that risk.

17

A.

Context
Water fluoridation has a long history in the United States and has been a source of political

18
19

discord, at times. See, e.g., Dkt. No. 429-3, Trial Ex. 13 at 15.1 In 1975 the EPA recommended

20

adding fluoride to water, with an optimal level up to 1.2 mg/L for its dental health benefits. Id. at

21

16. Between 1981 and 1984, fluoride’s association with adverse effects including osteosclerosis,

22

enamel fluorosis, and psychological and behavioral problems was contested. Id. at 17-18. Still, as

23

of 1986, up to 1.2 mg/L water fluoridation was considered optimal, and the maximum level was 4

24

mg/L. Id. at 14-18. After evidence increasingly established fluoride’s connection to adverse

25
26
27
28

1

Controversy over fluoridation of drinking water has even found its way into Hollywood. See DR.
STRANGELOVE (Columbia Pictures 1964) (General Ripper characterizing fluoridation as a threat to
our “precious bodily fluids” and “the most monstrously conceived and dangerous communist plot
we’ve ever had to face”).
2

1

effects, including severe enamel fluorosis, risk of bone fracture, and potential skeletal fluorosis,

2

recommended levels were lowered in 2006. Id. at 10. Community water fluoridation has since

3

continued at levels believed to be safe for its dental health benefits. At present, fluoride is added

4

to tap water in the United States, with an optimal level of 0.7 mg/L.
However, scientific evidence has increasingly identified a link between fluoride exposure

United States District Court
Northern District of California

5
6

and adverse cognitive effects in children (reduced IQ). Accordingly, Plaintiffs exercised their

7

power under Amended TSCA and petitioned the EPA to consider whether fluoride in drinking

8

water presents an unreasonable risk of injury to human health. Notwithstanding the growing and

9

robust body of evidence indicating an association between fluoride intake and cognitive

10

impairment in children, the EPA denied Plaintiffs’ petition. Plaintiffs filed suit in this Court,

11

arguing that the EPA was wrong and that community water fluoridation at 0.7 mg/L (the

12

“condition of use”) poses an unreasonable risk of injury to human health.

13

B.

Summary

14

To succeed in a suit brought under the Amended TSCA, Plaintiffs must prove, by a

15

preponderance of the evidence, that a risk of injury to human health is present and that such risk is

16

unreasonable. For a risk to be present, Plaintiffs must show that some segment of the United

17

States population is exposed to the chemical at issue at levels that either exceed, or are too close to

18

the dosage at which the chemical presents a hazard.2 The reasonableness of the risk is informed

19

by several factors, including inter alia, the size and susceptibility of impacted populations,

20

severity of the harm at issue, and the frequency and duration of exposure.
There is little dispute in this suit as to whether fluoride poses a hazard to human health.

21
22

Indeed, EPA’s own expert agrees that fluoride is hazardous at some level of exposure. And ample

23

evidence establishes that a mother’s exposure to fluoride during pregnancy is associated with IQ

24

decrements in her offspring. The United States National Toxicology Program (“NTP”) – the

25

federal agency regarded as experts in toxicity – undertook a systematic review of all available

26

literature near the time of publication considering whether fluoride poses cognitive harm,

27
28

2

The level at which the chemical presents a hazard is known as the “hazard level.” The level at
which human populations are exposed to the chemical is known as the “exposure level.”
3

1

reviewing 72 human epidemiological studies considering this question. The NTP concluded that

2

fluoride is indeed associated with reduced IQ in children, at least at exposure levels at or above 1.5

3

mg/L (i.e., “higher” exposure levels). And notwithstanding inherent difficulties in observing

4

effects at lower exposure levels, explained in further detail below, scientists have observed a

5

statistically significant association between fluoride and adverse effects in children even at such

6

“lower” exposure levels (less than 1.5 mg/L).

United States District Court
Northern District of California

7

Notwithstanding recognition by EPA’s expert that fluoride is hazardous, the EPA points to

8

technicalities at various steps of the risk evaluation to conclude that fluoride does not present an

9

unreasonable risk. Primarily, the EPA argues the hazard level and the precise relationship between

10

dosage and response at lower exposure levels are not entirely clear. These arguments are not

11

persuasive.

12

Importantly, the chemical at issue need not be found hazardous at the exposure level to

13

establish that a risk is present under Amended TSCA. Instead, the EPA requires a margin exist

14

between the hazard level and exposure level to ensure safety; if there is an insufficient margin then

15

the chemical poses a risk. The trial evidence in this case establishes that even if there is some

16

uncertainty as to the precise level at which fluoride becomes hazardous (hazard level), under even

17

the most conservative estimates of this level, there is not enough of a margin between the accepted

18

hazard level and the actual human exposure levels to find that fluoride is safe. Simply put, the risk

19

to health at exposure levels in United States drinking water is sufficiently high to trigger

20

regulatory response by the EPA under Amended TSCA.

21

To this end, as mentioned previously, the NTP compiled and analyzed all relevant studies it

22

could find and concluded that, at least at dosages of 1.5 mg/L or higher, fluoride is associated with

23

reduced IQ in children. Subsequently, toxicology experts endeavored to put a finer point on the

24

impact of fluoride on children’s IQ at “lower” exposure levels, i.e., those below 1.5 mg/L, and

25

conducted a pooled benchmark dose analysis to define the precise hazard level of fluoride. For

26

reasons described below, this pooled benchmark dose analysis benefited from increased statistical

27

power relative to the NTP’s assessment due to its methodology (i.e., the benchmark dose analysis

28

used individualized, continuous data, while the NTP assessment did not, due to quantity and variety
4

United States District Court
Northern District of California

1

of studies the NTP reviewed in that assessment). The pooled benchmark dose analysis concluded

2

that a 1-point drop in IQ of a child is to be expected for each 0.28 mg/L of fluoride in a

3

pregnant mother’s urine. This is highly concerning, because maternal urinary fluoride levels for

4

pregnant mothers in the United States range from 0.8 mg/L at the median and 1.89 mg/L

5

depending upon the degree of exposure. Not only is there an insufficient margin between the

6

hazard level and these exposure levels, for many, the exposure levels exceed the hazard level of

7

0.28 mg/L.

8

The EPA challenges, for a variety of reasons, whether this 0.28 mg/L hazard level

9

(measured in maternal urinary fluoride) is appropriate for this risk evaluation. The EPA argues,

10

among other things, that the hazard and exposure levels should not be expressed in maternal

11

urinary fluoride because that metric reflects total fluoride exposure – not just exposure resulting

12

from drinking fluoridated water from one’s community. Fluoride may also be ingested through,

13

e.g., tea, fish, toothpaste, and commercial food and beverage made with fluoridated water.

14

Nonetheless, the risk analysis should consider the additive effect of the chemical under the

15

subjected condition of use (here, fluoridated community drinking water), especially where, as here,

16

the fluoridated drinking water is a significant (and likely primary) contributor to aggregate

17

exposure to fluoride. Indeed, the Amended TSCA, expressly contemplates that the aggregate

18

exposure to a chemical will be considered when conducting a risk assessment. See 15 U.S.C. §

19

2605(b)(4)(F). In this sense, maternal urinary fluoride is not just an acceptable metric, it is highly

20

useful in assessing the real-world end result of exposure from drinking fluoridated water along

21

with other sources.

22

Even if urinary fluoride were not the appropriate metric in assessing health risk, or even if

23

the toxicologically determined hazard level of 0.28 mg/L were deemed insufficiently

24

substantiated, evidence in the record still establishes with little doubt that fluoridated drinking

25

water presents a risk of injury to health. Using a highly conservative estimate of the hazard level

26

of 4 mg/L measured in drinking water fluoride (well above the 1.5 mg/L identified as hazardous to

27

children by the NTP) based on the consistent and repeated observation of adverse effects

28

summarized in the NTP’s assessment, a risk is present. There is little dispute that there is a
5

1

statistically significant association between IQ decrements in children and fluoride concentration

2

levels at 4 mg/L.

United States District Court
Northern District of California

3

The EPA’s default margin of error requires a factor of 10 between the hazard level and

4

exposure level due to variability in human sensitivities. Put differently, only an exposure that is

5

below 1/10th of the hazard level would be deemed safe under Amended TSCA, given the margin

6

of error required. Here, an even greater margin (100x) is owed because the methodology (which

7

yields the 4 mg/L hazard level) uses the lowest observed adverse effect level (“LOAEL”); this

8

methodology adds an additional level of uncertainty (and hence the application of a 100x rather

9

than 10x margin). But even if only the default 10x margin is required, the safe level of fluoride

10

exposure would be 0.4 mg/L (4 mg/L (hazard level) divided by 10). The “optimal” water

11

fluoridation level in the United States of 0.7 mg/L is nearly double that safe level of 0.4 mg/L for

12

pregnant women and their offspring.

13

In all, there is substantial and scientifically credible evidence establishing that fluoride

14

poses a risk to human health; it is associated with a reduction in the IQ of children and is

15

hazardous at dosages that are far too close to fluoride levels in the drinking water of the United

16

States. And this risk is unreasonable under Amended TSCA. Reduced IQ poses serious harm.

17

Studies have linked IQ decrements of even one or two points to e.g., reduced educational

18

attainment, employment status, productivity, and earned wages. Indeed, the EPA recognizes that

19

reduction of IQ poses a serious community health issue. Moreover, highly susceptible populations

20

are impacted, including over two million pregnant women and babies, a number far exceeding

21

population size the EPA has looked to in determining whether regulatory action was warranted in

22

other risk evaluations (i.e., 500 people or less).

23

Thus, the Court finds Plaintiffs have met their burden in establishing, by a preponderance

24

of the evidence, that community water fluoridation at 0.7 mg/L presents an unreasonable risk of

25

injury to health under Amended TSCA and that the EPA is thus obliged to take regulatory action

26

in response. The Court does not in this order prescribe what that response should be.

27
28
6

1
2

II.
A.

Factual and Procedural Background
1. Section 6(a) of the Toxic Substances Control Act (“TCSA”) requires Defendant United

3

United States District Court
Northern District of California

BACKGROUND

4

States EPA3 to regulate the use of certain chemical substances that it determines pose an

5

unreasonable risk to health or the environment. 15 U.S.C. § 2605(a).

6

2. The TSCA was initially passed in 1976, codified at 15. U.S.C. § 2601 et seq. Congress

7

enacted the original TSCA, motivated by findings that “human beings and the environment are

8

being exposed each year to a large number of chemical substances and mixtures,” 15 U.S.C. §

9

2601(a)(1), and that, “among the many chemical substances and mixtures which are constantly

10

being developed and produced, there are some whose manufacture, processing, distribution in

11

commerce, use, or disposal may present an unreasonable risk of injury to health or the

12

environment,” id. § 2601(a)(2).

13

3. On June 22, 2016, the Frank R. Lautenberg Chemical Safety for the 21st Century Act was

14

signed into law. See Pub. L. No. 114-182, 114th Congress. The Act amended the TSCA. See id.
4. Amended TSCA requires the EPA to regulate the use of certain chemical substances that

15
16

pose an unreasonable risk of harm to health or the environment. 15 U.S.C. § 2605(a). If a

17

chemical substance poses a risk of unreasonable harm, the EPA must promulgate a rule imposing

18

one or more of a wide range of possible requirements. See id. § 2605(a)(2). Specifically, the rule

19

adopted by the EPA must impose one or more of the following: a prohibition, restriction, or

20

limitation of the amount of such substance that may be manufactured, processed, or distributed in

21

commerce, id. § 2605(a)(1); a prohibition, restriction, or limitation upon such manufacture,

22

processing, or use in connection with “a particular use” or “a particular use in a concentration in

23

excess of a level specified by the Administrator,” id. § 2605(a)(2); labeling requirements for such

24

substance, id. § 2605(a)(3); record-keeping requirements for manufacturers or processors of the

25

substance, id. § 2605(a)(4); commercial-use regulations, id. § 2605(a)(5); disposal requirements,

26
27
28

3

Scott Pruitt, Administrator of the EPA is also named as a Defendant in his official capacity. Dkt.
No. 372 (Supplemental Complaint (“FAC”)) ¶ 1.
7

1

id. § 2605(a)(6); and/or notice requirements, id. § 2605(a)(7). The EPA may limit the application

2

of such requirements to “specified geographic areas.” Id. § 2605(a).
5. After the Act’s amendment to TSCA, there are three pathways to obtain a Section 6(a) rule

United States District Court
Northern District of California

3
4

regulating a chemical: (1) an EPA’s sua sponte designation of a chemical as “high priority,”

5

resulting in a finding that it presents an unreasonable risk,4 15 U.S.C. § 2605(c)(1); (2) an EPA

6

risk evaluation of a chemical at the request of a manufacturer, see id. § 2605(b)(4)(C)(ii), which

7

results in a finding of unreasonable risk; or (3) a successful Section 21 “citizen petition,” see id. §§

8

2620(a), (b)(3).

9

6. A Section 21 citizen’s petition to the EPA to initiate Section 6(a) rulemaking is to be

10

granted if the petitioner demonstrates a chemical substance poses an unreasonable risk of harm.

11

Id. § 2620(a). Amended TSCA provides judicial review of a denial of such a petition to the EPA.

12

Id. § 2620(b)(4). In contrast to the typical standard of judicial review under the Administrative

13

Procedure Act, in considering a Section 21 citizen’s petition, the Court considers the issue de

14

novo; no deference is owed under to the EPA’s denial of the petition. See id. § 2620(b)(4)(B).

15

7. Plaintiffs in the instant suit are non-profit advocacy organizations and associations and

16

individuals suing on behalf of themselves and their children. FAC ¶ 1.5

17
18
19
20
21
22
23
24
25
26
27
28

4

To elaborate, Section 6(b) requires the EPA to perform its own evaluations of the risks posed by
certain chemical substances. 15 U.S.C. § 2605(b)(4)(A). To this end, the EPA is required by
Amended TSCA to designate chemical substances as “high-priority” or “low-priority” based on a
risk screening process. See id. § 2605(b)(1). “High-priority” chemicals are those that “may
present an unreasonable risk to health or the environment because of potential hazard and a
potential route of exposure under the conditions of use.” Id. § 2605(b)(1)(B)(i). A “low-priority”
substance, in contrast, is one that the Administrator “concludes, based on information sufficient to
establish . . . does not meet the standard” to be designated a high-priority substance. Id. §
2605(b)(1)(B)(ii). Once the EPA has designated a chemical substance “high-priority,” it must
initiate a Section 6(b) “risk evaluation.” Id. §§ 2605(b)(3)(A), (4)(C)(i). A risk evaluation is not
required for a “low-priority” substance. Id. § 2605(b)(1)(A). The EPA must pursue these risk
evaluations at a minimum pace established by statute: within 6 months, risk evaluations must be
underway on at least 10 substances drawn from the 2014 TSCA Work Plan for Chemical
Assessments, id. § 2605(b)(2)(A); within three and a half years, risk evaluations must be
underway on “at least 20 high-priority substances,” id. § 2605(b)(2)(B); a new high-priority
substance must be designated anytime a risk evaluation has been completed (other than those
commenced at the request of a manufacturer), id. § 2605(b)(3)(C); and, generally, the EPA must
continue designating substances and conducting evaluations “at a pace consistent” with its ability
to meet the 3-year deadline to complete each risk evaluation, id. § 2605(b)(2)(C).
5

Specifically, Plaintiffs are Food & Water Watch, Fluoride Action Network, and Moms Against
Fluoridation (“Organizational Plaintiffs”), and Audrey Adams individually and on behalf of Kyle
8

1

8. On November 22, 2016, a group of organizations and individuals including Plaintiffs

2

petitioned the EPA under Section 21 of Amended TSCA to regulate the fluoridation of drinking

3

water supplies under Section 6(a). Dkt. No. 117-1, Ex. 1. Plaintiffs asserted that the ingestion of

4

fluoride poses an unreasonable risk of neurotoxic harm to humans including IQ loss and other

5

neurotoxic effects, particularly for infants, young children, and other subpopulations standing at

6

elevated risk. Id.

7
8

United States District Court
Northern District of California

9

9.

On February 17, 2017, the EPA denied Plaintiffs’ petition. Dkt. No. 28-1; 82 Fed. Reg.

11,878 (Feb. 27, 2017).
10. After the EPA denied Plaintiffs’ petition, Plaintiffs filed this suit seeking judicial review of

10

the EPA’s denial pursuant to 15 U.S.C. § 2620. Dkt. No. 1 (Complaint (“Compl.”)) ¶¶ 106-07.

11

11. Beginning on June 8, 2020, after the parties engaged in fact and expert discovery, the

12

Court held a seven-day bench trial, which included expert testimony regarding the state of the

13

scientific research on fluoride neurotoxicity (“Trial Phase 1”). See Dkt. Nos. 219, 238.

14

12. On August 10, 2020, the Court stayed the case due to concerns about Plaintiffs’ standing

15

and developments in scientific literature regarding fluoride. See Dkt. No. 262. The Court

16

explained that the stay would allow EPA to consider new scientific studies published after EPA’s

17

denial of Plaintiffs’ administrative petition and allow the Court to consider the imminent

18

publication of the NTP systematic review “Monograph on the Systematic Review of Fluoride

19

Exposure and Neurodevelopmental and Cognitive Health Effects.” Id. at 3-5.

20
21

13. Thereafter, Plaintiffs filed a supplemental administrative petition for reconsideration to the
EPA. Dkt. No. 270.

22

14. EPA again denied the petition. Dkt. No. 278.

23

15. On October 28, 2022, the Court granted Plaintiffs’ motion to lift the stay and take the case

24

out of abeyance, finding that Plaintiffs had standing and that there was new evidence that scientific

25

developments had changed, including the fact that the aforementioned NTP’s systematic review

26
27
28

Adams, Kristen Lavelle individually and on behalf of Neal Lavell, and Brenda Staudenmaier
individually and on behalf of Ko Staudenmaier and Hayden Staudenmaier (“Individual Plaintiffs”)
(collectively “Plaintiffs” or “FWW”). FAC ¶ 1.
9

1

had since undergone three additional rounds of peer review resulting in a near-final version of the

2

document. See Dkt. No. 319 at 2-5.
16. Beginning on January 31, 2024, the Court held a second, ten-day bench trial (“Trial Phase

3
4

2”) which included expert testimony regarding the updated state of the scientific research on

5

fluoride neurotoxicity. See Dkt. Nos. 407-413, 422-424.

6

B.

7

17. Plaintiffs contend that the addition of fluoridation chemicals to drinking water at levels

8

recommended in the United States (0.7 mg/L) presents an unreasonable risk of neurological harm

9

when assessed under the risk evaluation framework that EPA uses under the Amended TSCA.

10

Dkt. No. 378 (Joint Pretrial Conference Statement (“PTC Statement”)) at 1-2.
18. Plaintiffs seek a declaration that fluoridation of water at 0.7 mg/L presents an unreasonable

11
United States District Court
Northern District of California

Relief Requested

12

risk of injury to health and injunctive relief requiring the EPA to initiate the rulemaking

13

proceeding requested by Plaintiffs in their Petition to the EPA. PTC Statement at 2. Specifically,

14

Plaintiffs seek an order requiring the EPA to “initiate a proceeding for the issuance of a rule,” but

15

the order would not “prescribe the content of a rule or the outcome of such a proceeding.” Id. In

16

short, rulemaking would be left in the first instance to the EPA.

17

19. Plaintiffs also seek recovery of their costs of suit and reasonable fees for attorneys and

18

expert witnesses, as permitted by 15 U.S.C. § 2620(b)(4)(C), and such further relief that the Court

19

may deem just and proper. PTC Statement at 2.

20

C.

21

Statutory Standard and Burden
20. Plaintiffs bear the burden of proving, by a preponderance of the evidence, that the

22

chemical substance at issue presents an “unreasonable risk of injury to health or the environment,

23

without consideration of costs or other nonrisk factors, including an unreasonable risk to a

24

potentially exposed or susceptible subpopulation under the conditions of use.” 15 U.S.C. §

25

2620(b)(4)(B)(ii). The Court considers the issue de novo; no deference is owed under TSCA to

26

the EPA’s denial of the petition. Id. § 2620(b)(4)(B).

27

21. If the Court determines that petitioner has met its burden, demonstrating unreasonable risk

28

by a preponderance of the evidence, the Court “shall order the Administrator to initiate the action
10

1

requested by the petitioner.” Id. Specifically, EPA would be directed to engage in rulemaking

2

pursuant to Subsection 6(a) of TSCA wherein the EPA would consider applying one or more

3

methods to neutralize the risk, ranging from requiring a notice be provided to the public of risks

4

(i.e., utilizing a warning label or disseminating a public advisory), see id. § 2605(a)(7), to

5

prohibiting manufacturing or distributing the chemical at issue, see id. § 2605(a)(1).

6

D.

United States District Court
Northern District of California

7

Standing
22. The Court previously held, in lifting its stay on proceedings and allowing the case to

8

proceed to phase two of trial, that Plaintiffs had standing. Dkt. No. 319 at 2-3. The Court

9

reaffirms this finding. At a minimum, Organizational Plaintiff FWW has standing in a

10

representative capacity. An association has standing to sue on behalf of its members when: “(1)

11

its members would otherwise have standing to sue in their own right; (2) the interests it seeks to

12

protect are germane to the organization’s purpose; and (3) neither the claim asserted nor the relief

13

requested requires the participation of individual members in the lawsuit.” Am. Unites for Kids v.

14

Rousseau, 985 F.3d 1075, 1096 (9th Cir. 2021) (citing Hunt v. Wash. State Apple Advert. Comm’n,

15

432 U.S. 333, 343 (1977)). Each prong is satisfied:

16

a. In its previous order, the Court found that Jessica Trader, a member of FWW, has

17

standing. Dkt. No. 319 at 2-3. Article III standing requires: (1) an injury-in-fact that is concrete

18

and particularized and actual or imminent, (2) a causal connection between the injury and the

19

conduct complained of, and (3) probable redressability. Id. (citing Lujan v. Defs. of Wildlife, 504

20

U.S. 555, 560–61 (1992)). Ms. Trader became pregnant in November 2020 and gave birth in

21

August 2021 (during the pendency of this lawsuit) and testifies that she plans to have several more

22

children; she has taken steps to effectuate this goal including discontinuing her use of birth control

23

medication. Dkt. No. 430-18, Trial Ex. 66 (Declaration of Jessica Trader) ¶¶ 5-8 & Ex. A. Ms.

24

Trader has incurred costs and taken measures to avoid fluoridated water during her first pregnancy

25

and continues to do so to protect her future children. Id. ¶¶ 9-16. As the Court previously

26

explained, neurodevelopmental harm from fluoride exposure to Ms. Trader’s child and future

27

children is concrete and imminent; there is a credible causal connection between that

28

neurodevelopmental harm and EPA’s regulation of fluoride exposure or lack thereof; and the harm
11

1

would likely be redressed if EPA were to pass a rule prohibiting the addition of fluoridation

2

chemicals to public drinking water supplies. Dkt. No. 319 at 2-3. Moreover, the EPA has

3

conceded that standing would be satisfied by “someone who is an expectant parent who – who

4

could be consuming fluoridated water, and, and – that could have potential effects on the baby

5

she’s carrying in utero. It could be a potential – a parent, someone with very young children.” Id.

6

(quoting Dkt. No. 133 at 14:9-17). Ms. Trader is such an individual. Thus, the first prong is

7

satisfied; a member has standing.

United States District Court
Northern District of California

8

b. As for the second prong, there is no dispute that FWW’s mission is to ensure

9

“clean, safe water for drinking” which it views as a “fundamental right that should be afforded to

10

all people,” and to “advocate for more government responsibility in protecting our drinking water

11

resources.” Dkt. No. 430-8, Trial Ex. 52 (Second Amended Declaration of Scott Edwards, Co-

12

Director of FWW) ¶¶ 4, 6. Thus, the interests at stake in this suit – regulation of water

13

fluoridation to protect public health – are germane to the organization’s purpose. See, e.g., Am.

14

Unites for Kids, 985 F.3d at 1097 (explaining that where there is a close connection between the

15

organization’s mission and the interests of others it seeks to represent, organizational standing is

16

appropriate); G.G. by & through A.G. v. Meneses, 638 F. Supp. 3d 1231, 1241 (W.D. Wash. 2022)

17

(finding nonprofit disability rights organization had associational standing to bring claims on

18

behalf of disabled members as rights of people with developmental disabilities was an interest the

19

organization sought to protect).

20

c. The third prong is a “judicially fashioned and prudentially imposed” question, as

21

opposed to a constitutional requirement of standing. Or. Advocacy Ctr. v. Mink, 332 F.3d 1101,

22

1109 (9th Cir. 2003). This suit is appropriately brought by a representative plaintiff; analysis

23

under Amended TSCA focuses on scientific evidence substantiating the alleged risk to public

24

health rather than focusing upon anecdotal evidence from plaintiffs. See ¶¶ 26-95; accord

25

Laborers Int’l Union Loc. 261 v. City & Cnty. of San Francisco, 2022 WL 2528602, at *6 (N.D.

26

Cal. July 6, 2022) (explaining that unlike claims seeking damages which require individualized

27

proof, claims seeking injunctive relief are well-suited for adjudication by organizational plaintiff)

28

(citing Comm. for Immigrant Rts. of Sonoma Cnty. v. Cnty. of Sonoma, 644 F. Supp. 2d 1177,
12

1

1194 (N.D. Cal. 2009)). The harm redressable herein is precisely the kind of harm that Amended

2

TSCA is designed to address. For these reasons, the Court reaffirms its finding that requirements

3

of standing have been satisfied.

4
5

FINDINGS OF FACT

23. To discern whether a chemical substance presents an unreasonable risk of injury to health

6

or the environment, without consideration of costs or other non-risk factors, including an

7

unreasonable risk to a potentially exposed or susceptible subpopulation, under the conditions of

8

use, under TSCA section 6, the EPA engages in a TSCA risk evaluation process. 15 U.S.C. §

9

2605(b)(4); 82 Fed. Reg. 33,726 (July 20, 2017); Dkt. No. 434-18, Trial Ex. 544.

10

United States District Court
Northern District of California

III.

24. The TSCA risk evaluation is comprised of a risk assessment and risk determination. See

11

Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 653:22-655:11 (Barone). The National Research Council

12

(NRC, 1983) has defined risk assessment as including the following components: (1) hazard

13

assessment (including hazard identification and quantitative dose response analysis); (2) exposure

14

assessment; and (3) risk characterization. A risk evaluation under the Amended TSCA includes

15

the three aforementioned steps of a risk assessment, as well as a fourth and final step: (4) a risk

16

determination. See id. The “risk assessment” is the scientific technical evaluation, encompassing

17

the first three parts of this process, resulting in an unbiased, transparent, and reproducible

18

description of the risk. See id. The “risk determination” is the final step of the risk evaluation

19

process, where EPA summarizes its findings and determines whether a chemical does or does not

20

present unreasonable risk. See id.

21
22
23

25. The following is a summary of the risk evaluation steps. See id.; accord 15 U.S.C. §
2605(b)(4)(F)(i)-(v).
a. At step 1 (hazard assessment) the EPA determines if a chemical is considered

24

hazardous and if so, the EPA endeavors to determine the point at which the chemical becomes

25

hazardous (“point of departure” or “hazard level”). See Dkt. No. 400, Feb. 5, 2024, Trial Tr. at

26

653:22-655:11 (Barone); accord 15 U.S.C. § 2605(b)(4)(F)(i)-(iii), (v).

27

b. At step 2 (exposure assessment) the EPA determines the level at which populations

28

are exposed to the chemical. See Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 653:22-655:11 (Barone);
13

1

accord 15 U.S.C. § 2605(b)(4)(F)(i)-(iii), (v).

2

c. At step 3 (risk characterization), the EPA compares the point of departure with the

3

exposure level to determine if a risk is present. See Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 575:8-

4

583:13 (Barone). Because of uncertainty in data, the EPA establishes a margin between the point

5

of departure and the community’s exposure level. There must be a sufficient margin to find

6

absence of risk. See id. The appropriate margin varies based upon how much uncertainty there is

7

in the chosen point of departure. See id. The appropriate or required margin is referred to as the

8

benchmark margin of exposure (“benchmark MOE”). See id. The actual margin is the actual

9

margin of exposure (“actual MOE”). If there is an insufficient margin, i.e., the actual MOE is less

10

than the benchmark MOE, a risk has been identified. See id.
d. At step 4 (risk determination) if a risk is identified, the EPA will then determine if

United States District Court
Northern District of California

11
12

that risk is unreasonable, considering various factors such as the type of harm at issue and number

13

of people exposed. See Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 653:22-655:11 (Barone); accord

14

15 U.S.C. § 2605(b)(4)(F)(iii)-(v). Each step of the risk assessment is discussed in turn below.6

15

A.

Step 1: Hazard Assessment
26. The Hazard Assessment step is comprised of three subparts: (a) hazard identification; (b)

16
17

weight-of-the-scientific evidence; and (c) dose-response assessment. See Dkt. No. 400, Feb. 5,

18

2024, Trial Tr. at 654:19-655:11 (Barone). Each are addressed in turn below.
1.

19

Step 1A: Hazard identification
a.

20

Framework

27. The first component of the hazard assessment is hazard identification. Dkt. No. 417, Feb.

21
22

2, 2024, Trial Tr. at 489:11-17 (Barone), 656:8-661:16 (Barone). At the hazard identification step

23

of the risk evaluation framework, the reviewer determines if an adverse effect is associated with a

24

chemical exposure. See Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 489:11-17 (Barone), 656:8-661:16

25

(Barone).

26
27
28

6

The evaluation of fluoridation chemicals under TSCA follows the same standards for
demonstrating hazard and risk that EPA uses for its evaluations of other industrial chemicals under
TSCA; there is no justification for holding fluoridation chemicals to a higher burden. See Dkt.
No. 401, Feb. 6, 2024, Trial Tr. at 742:25-743:8 (Barone).
14

1

28. Proof of causation is not required to establish a hazard of neurotoxicity, only association

2

between the chemical exposure and the adverse effect is required for a hazard to be identified. See

3

id. at 490:1-5.
29. At this stage of the process EPA reviews, searches, screens, and evaluates all studies

4
5

related to different hazards to determine whether the data are sufficient or insufficient for

6

identified adverse effects. Id. at 492:24-494:9.
b.

7

30. The hazard identification step of the hazard assessment here is satisfied; exposure to the

8
9
10

chemical fluoride is associated with the adverse effect of reduced IQ in children, and particularly
in boys.
c.

11
United States District Court
Northern District of California

Key finding

Underlying findings

12

31. The NTP is headquartered within the National Institute of Environmental Health Sciences

13

(“NIEHS”). Dkt. No. 440, Feb. 13, 2024, Trial Tr. at 1425:23-1426:8 (Barone). By May of 2022,

14

the NTP completed its systematic review of fluoride, titled NTP Monograph on the State of the

15

Science Concerning Fluoride Exposure and Neurodevelopmental and Cognitive Health Effects: A

16

Systematic Review (hereafter “NTP Monograph”). Dkt. No. 431-1, Trial Ex. 67. See also Dkt.

17

No. 440, Feb. 13, 2024, Trial Tr. at 1427:5-8 (Barone); Dkt. No. 400, Feb. 4, 2024, Trial Tr. at

18

535:15-21 (Berridge). In August 2024, the NTP Monograph was formally published. See Dkt No.

19

442 (letter from parties recognizing publishing of document). The parties agree that there are no

20

material differences between the published Monograph and the pre-publication version that was

21

the subject of testimony and argument at trial (i.e., Trial Exhibit 67). Id.7

22
23
24
25
26
27
28

7

The parties originally filed a letter agreeing that the published version of the NTP Monograph
was the same in all material respects as the Monograph this Court reviewed at trial. Dkt. No. 442.
Subsequently, Plaintiffs filed a letter suggesting that certain aspects of the published NTP
Monograph were modified in a way that lends additional support for their case. See Dkt. No. 443.
In particular, Plaintiffs assert:
Page 101 of the now-published version of the NTP Monograph summarizes the findings of
the “in-press” meta-analysis as follows:
The group-level meta-analysis of 59 studies (n = 20,932 children)
used SMD as the effect measure and reported statistically significant
15

1
2

specific question and uses explicit, pre-specified methods to identify, select, assess, and

3

summarize the findings of similar but separate studies.” Dkt. No. 255 (EPA Proposed Findings of

4

Fact, Trial Phase 1) at 15 (citing 82 Fed. Reg. at 33,734). Moreover, “[t]he goal of systemic

5

review methods is to ensure that the review is complete, unbiased, reproducible, and transparent.”

6

Id. The EPA explains that a systematic review is pertinent and is ideal in conducting a risk

7

assessment under TSCA. See id. at 14-19 (arguing that during the first phase of trial, before the

8

NTP Monograph was finalized, that Plaintiffs failed to meet their burden because they did not

9

conduct a systematic review).

10
11
United States District Court
Northern District of California

32. According to the EPA, a systematic review is “a scientific investigation that focuses on a

33. The NTP Monograph is a systematic review as the EPA has defined that term. The NTP
Monograph is a scientific investigation, focusing on a specific question using explicit, pre-

12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

inverse associations between fluoride exposure measures and
children’s IQ. There was also a significant dose response
relationship between group-level fluoride exposure and IQ. In
stratified dose-response meta-analyses of the low risk-of-bias
studies, the direction of association remained consistent when
group-level exposure was restricted to <4mg/L, <2 mg/L, and
<1.5 mg/L fluoride in drinking water and <4 mg/L, <2 mg/L,
and <1.5 mg/L fluoride in urine. The regression slopes metaanalysis of 13 studies (n = 4,475 children) with individual-level
measures of fluoride found a significant decrease in IQ of 1.63
points (95% CI: −2.33,−0.93; p-value <0.001) per 1-mg/L
increase in urinary fluoride. In subgroup analyses of both grouplevel and individual level data, the direction of the association
remained inverse when stratified by study quality (high versus low
risk of bias), sex, age group, outcome assessment, study location,
exposure timing, and exposure metric.
Dkt. No. 443 (citing NTP Monograph on the State of the Science Concerning
Fluoride Exposure and Neurodevelopment and Cognition: A Systematic Review,
National Toxicology Program (August 2024),
https://ntp.niehs.nih.gov/sites/default/files/2024-08/fluoride_final_508.pdf
(emphases added)).
The EPA disputes whether the post-trial version of the NTP Monograph is properly considered by
this Court. See Dkt. No. 444. Because the Court finds in Plaintiffs favor based upon the version
of the NTP Monograph that the Court reviewed at trial, and because neither party suggests the
aspects of the NTP Monograph that the Court reviewed therein have changed in a way that
undermines Plaintiffs’ case, the Court need not resolve this dispute. Instead, the Court bases its
finding upon the version of the NTP Monograph reviewed at trial (Trial Exhibit 67), though noting
that it has since been published formally, and that if it were considered, it would find the published
Monograph even more supportive of the decision reached herein.
16

United States District Court
Northern District of California

1

specified methods. Namely, the objective of the NTP Monograph was “[t]o conduct a systematic

2

review of the human, experimental animal, and mechanistic literature to evaluate the extent and

3

quality of the evidence linking fluoride exposure to neurodevelopmental and cognitive effects in

4

humans.” NTP Monograph at xii (Abstract). Regarding the methods: “[a] systematic review

5

protocol was used following the standardized OHAT [referring to the Office of Health Assessment

6

and Translation] systematic review approach for conducting literature-based health assessments.

7

This monograph presents the current state of evidence associating fluoride exposure with

8

neurocognitive or neurodevelopmental health effects and incorporated predefined assessments of

9

study quality and confidence levels. Benefits of fluoride with respect to oral health are not

10

addressed in this monograph.” Id. Ultimately, the NTP Monograph analyzed all available studies

11

assessing impacts of fluoride, including seventy-two human studies that assessed the association

12

between fluoride exposure and IQ in children and integrated the findings in the studies to draw

13

conclusions about the impact of fluoride to neurodevelopmental and cognitive effects in humans.

14

Id. at xii-xiii. Moreover, the NTP Monograph’s protocol underwent multiple rounds of peer

15

review. Id. at G-1. And the Monograph’s substance underwent multiple rounds of peer review,

16

including assessment of technical accuracy, and the sufficiency of evidence supporting the NTP

17

Monograph’s conclusion. Id. at x. The peer review panel includes professors from Brown

18

University School of Public Health, Columbia University Medical Center, Johns Hopkins

19

Bloomberg School of Public Health, and other epidemiological experts. See id. The EPA does

20

not dispute that the NTP Monograph is likely to have captured all relevant studies that were in

21

existence as of the Monograph’s literature cutoff date analyzing human data regarding

22

neurodevelopmental impacts of fluoride. Dkt. No. 421 at 12-13. Even before the NTP

23

Monograph was formally published, the EPA agreed that the NTP Monograph “followed the rules

24

that have been developed by NTP for conducting systematic reviews” and utilized a “rigorous

25

approach to assembling the evidence,” “clearly defined rules for identifying and evaluating

26

studies,” and “a well-defined protocol for drawing inferences” from the studies. Id.8 Indeed,

27
28

8

Plaintiffs submitted evidence indicating that the delay in publication was highly irregular, and
perhaps politically motivated. See Dkt. No. 385 at 12-13. The Court excluded evidence regarding
17

United States District Court
Northern District of California

1

EPA’s expert, Dr. Barone agreed that the NTP Monograph is a “high quality review.” Dkt. No.

2

440, Feb. 13, 2024, Trial Tr. at 1427:2-4 (Barone). Accordingly, the Court finds that the NTP

3

Monograph is probative and afforded significant weight in the risk evaluation analysis.

4

34. The NTP Monograph concludes that the majority of the 72 epidemiological studies on

5

fluoride and IQ that had been published by April 2021 found an association between fluoride and

6

reduced IQ in children, including 18 of the 19 studies the NTP Monograph deemed “high quality”

7

and “low-risk-of-bias” as well as 46 of the 53 lower-quality studies. NTP Monograph at xii (NTP

8

Monograph Abstract describing 46 of the 53 low-quality studies found an association between

9

higher fluoride exposure and lower IQ in children and 18 of 19 high-quality studies reported an

10

association between higher fluoride exposure and lower IQ in children including 3 prospective

11

cohort studies and 15 cross-sectional studies); accord Dkt. No. 428-1, Trial Ex. 69 at 65 (NTP

12

Board of Scientific Counselors Working Group Report agreeing that low-risk-of-bias studies were

13

“consistent,” meaning generating results in the same direction, in finding a negative association

14

between fluoride exposure and children’s IQ); Dkt. No. 396, Feb. 1, 2024, Trial Tr. at 313:25-

15

314:5 (Grandjean) (summarizing and agreeing with NTP Monograph’s finding that higher

16

fluoride exposure (at or above 1.5 mg/L) was found to be associated with lower IQ scores in

17

children in the majority of both low- and high-quality studies the NTP Monograph reviewed); Dkt.

18

No. 414, Feb. 9, 2024, Trial Tr. at 1197:2-15 (Savitz) (expressing confidence in NTP’s literature

19

search strategy and its ability to identify all relevant studies on fluoride exposure published prior

20

to the closing date of April 21, 2021, and confirming that the “vast majority of studies” that NTP

21

reviewed identified an association between fluoride and reduced IQ), 1114:24-1115:1 (describing

22

NASEM critique of adequate definition of the term “consistent” in NTP Monograph, but not

23

disagreeing with characterization of NTP Monograph finding association between IQ and

24

fluoride). The NTP Monograph explained its key finding regarding the impact of fluoride on

25

children’s IQ as follows:

26
27
28

partisanship relating to publishing of the Monograph, in large part because the EPA did not argue
the Monograph be afforded less weight for its draft status. Id. at 17. Eventually, the NTP
Monograph was published, in August 2024. See Dkt. No. 442.
18

In summary, the high-quality studies (i.e., studies with low
potential for bias) consistently demonstrate lower IQ scores with
higher fluoride exposure [e.g., represented by populations whose
total fluoride exposure approximates or exceeds the WHO
Guidelines for Drinking-water Quality of 1.5 mg/L of fluoride
(WHO 2017)]. The consistency in association is observed among
studies of varying study designs, exposure measures, and study
populations. Although some studies that conducted multiple
analyses observed within-study variations in results (e.g.,
differences between subsets of IQ tests), these variations were
unique to individual studies and did not detract from the overall
consistency in the findings that higher fluoride is associated with
lower IQ scores.

1
2
3
4
5
6
7
8
9
10

United States District Court
Northern District of California

11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

Trial Ex. 67 at 47 (emphasis added).
35. To come to this conclusion: the NTP Monograph identified 19 studies as being highquality (i.e., low risk-of-bias); all but one identified an association between fluoride and reduced
IQ in children: Bashash et al. 2017; Choi et al. 2015; Cui et al. 2018; Ding et al. 2011; Green et al.
2019; Rocha-Amador et al. 2007; Saxena et al. 2012; Seraj et al. 2012; Sudhir et al. 2009; Till et
al. 2020; Trivedi et al. 2012; Wang et al. 2012; Wang et al. 2020b; Xiang et al. 2003a; Xiang et al.
2011; Yu et al. 2018; Zhang et al. 2015b. NTP Monograph at 40, 29-39 (Table 6). To summarize
these high-quality studies:
a.

Bashash (2017): This study evaluated 211 mother-child pairs that were participants

in The Early Life Exposures in Mexico to Environmental Toxicants Project (“ELEMENT
Cohort”)9 and concluded that higher prenatal fluoride exposure was associated with statistically
9

Bashash (2017) (like Green (2019) and Till (2020), discussed in subparagraphs (b) and (c)), is a
longitudinal cohort study, evaluating fluoride in the urine of pregnant mothers. In such a cohort
study design:
[A] healthy group of people is assembled and followed forward in
time and observed for the development of dysfunction. Such studies
are invaluable for determining the time course for development of
dysfunction (e.g., follow-up studies performed in various cities on
the effects of lead on child development). This approach allows the
direct estimate of risks attributed to a particular exposure, since
toxic incidence rates in the cohort can be determined. Prospective
study designs also allow the study of chronic effects of exposure.
One major strength of the cohort design is that it allows the
calculation of rates to determine the excess risk associated with an
exposure. Also, biases are reduced by obtaining information before
the disease develops. This approach, however, can be very timeconsuming and costly. In cohort studies information bias can be
introduced when individuals provide distorted information about
their health because they know their exposure status and may have
been told of the expected health effects of the exposure under study.
19

United States District Court
Northern District of California

1

significant lower scores on tests of cognitive function in offspring at ages 4 and 6-12 years; an

2

increase in maternal urine fluoride of 0.5 mg/L predicted a 3.15 lower General Cognitive Index

3

(“GCI”) score and 2.50 lower IQ score of the offspring. Dkt. No. 432-2, Trial Ex. 106 at 1.

4

ELEMENT collected urinary samples from women during pregnancy and from their children

5

when the children were 6-12 years old (299 mother-child pairs) recruited from hospitals caring for

6

low to moderate income populations in Mexico City. Id. at 1-2. The mean urinary fluoride in

7

mothers and children was 0.90 mg/L (mothers) and 0.82 mg/L (children). Id. Child intelligence

8

was measured via GCI for children at age 4 and IQ and from the Wechsler Abbreviate Scale of

9

Intelligence (“WASI”) at ages 6-12. Id. Fluoride exposure derived from fluoridated salt and

10

naturally occurring fluoride in drinking water in Mexico City, ranging from 0.15 to 1.38 mg/L. Id.

11

at 2. A second morning void (“spot”) urine sample was targeted for collection during each

12

trimester of pregnancy from mothers and the offspring children at time of measurements of

13

intelligence. Id. A total of 1,484 prenatal samples was measured; after controlling for, e.g.,

14

quality, duplicates, covariates, and outliers, 877 urine samples adjusted for creatinine were

15

retained, stemming from 512 unique mothers. Id. at 3. A total of 287 mother-child pairs had

16

complete data on exposure and outcome for children at 4 years and 211 for children at 6-12 years.

17

Dkt. No. 434-27, Trial Ex. 656 (Savitz Summary of Methods in Key Studies of Fluoride Exposure

18

and Neurodevelopment).

19

b.

Green (2019): Green et al. (2019) studied mother-child pairs in Canada that were

20
21
22
23
24
25
26
27
28

More credence should be given to those studies in which both
observer and subject bias are carefully controlled (e.g., double-blind
studies). A special type of cohort study is the retrospective cohort
study, in which the investigator goes back in time to select the study
groups and traces them over time, often to the present. The studies
usually involve specially exposed groups and have provided much
assistance in estimating risks due to occupational exposures.
Occupational retrospective cohort studies rely on company records
of past and current employees that include information on the dates
of employment, age at employment, date of departure, and whether
diseased (or dead in the case of mortality studies). Workers can then
be classified by duration and degree of exposure.
Dkt. No. 429-7, Trial Ex. 17 at 17-18. Moreover, “[p]ositive or negative results
from a properly controlled prospective study should weigh heavily in the risk
assessment process.” Id. (emphasis added).
20

United States District Court
Northern District of California

1

participants in the Maternal-Infant Research on Environmental Chemicals program (“MIREC

2

Cohort”) and found a statistically significant, negative association between fluoride exposure and

3

IQ in boys, but not girls. Dkt. No. 432-5, Trial Ex. 109 at 940, 944. The study concluded that 1

4

mg/L increase in maternal urinary fluoride was associated with a 4.49-point lower IQ score in

5

boys and 1 mg higher daily intake of fluoride among pregnant women was associated with a 3.66

6

lower IQ score in boys and girls. Id. MIREC collected urinary spot samples and estimates of

7

daily fluoride intake from water consumption for pregnant women recruited from cities across

8

Canada (Vancouver, Montreal, Kingston, Toronto, Hamilton, Halifax). Id. at 941-942. Urinary

9

samples from the women were collected across each trimester of pregnancy; the mean maternal

10

urinary fluoride of mothers was 0.42 mg/L in fluoridated communities and 0.27 mg/L in non-

11

fluoridated communities. Id. at 944. The mean estimated intake of water fluoride concentration

12

was 0.39 mg/day; 0.43 mg for women in communities with fluoridated drinking water and 0.26 for

13

those living in communities without fluoridated drinking water. Id. Children were between ages 3

14

and 4 years at testing. Id. at 940. Data on exposure and outcome was complete for 512 mother-

15

child pairs measuring exposure through maternal urinary fluoride and 400 mother-child pairs

16

estimating water fluoride intake. Id.
c.

17

Till (2020): Till (2020) studied samples taken from 398 mother-child pairs that

18

participated in the MIREC Cohort project (the cohort studied in Green (2019)), to evaluate IQ of

19

children that were breastfed compared to formula-fed as infants in areas that had fluoridated and

20

non-fluoridated water. Dkt. No. 432-19, Trial Ex. 123 at 1. This study found that an increase in

21

fluoride intake from infant formula corresponded to an 8.8 decrement in performance IQ which

22

was statistically significant, including after controlling for fetal fluoride exposure. Id.10
d.

23

Cross-sectional studies11 of children in China found significant inverse association

24
25
26
27
28

10

Till (2020) and Green (2019) exemplify how the same samples from one cohort may be
analyzed in multiple studies to either confirm results from a previous study or to extract different
information from the same samples from a given cohort.
11

Cross-sectional studies are afforded less weight than cohort studies. As the EPA guidelines
explain:
In cross-sectional studies or surveys, both the disease and suspected
risk factors are ascertained at the same time, and the findings are
21

United States District Court
Northern District of California

1

between fluoride and children’s IQ score: Xiang (2003a) (finding significant inverse correlation

2

between IQ and urinary fluoride; significant association of fluoride on IQ score based on drinking

3

water levels); Ding (2011) (significant association between urinary fluoride and decrease in IQ

4

score); Xiang (2011) (significant association between serum (blood-derived sample) fluoride and

5

reduced IQ score in children); Wang (2012) (significant correlation between total fluoride intake

6

and reduced IQ); Zhang (2015b) (significant correlation between reduced IQ score and children’s

7

serum fluoride, and urinary fluoride), Cui (2018) (significant association between IQ score and

8

urinary fluoride); Yu (2018) (significant difference in mean IQ scores in high water fluoride areas

9

compared to normal water fluoride areas); and Wang (2020b) (significant negative association

10

between IQ and water and urinary fluoride and IQ in boys and girls). NTP Monograph at 29-33

11

(Table 6). One study, Cui (2020) identified a directionally negative, though not statistically

12

significant decrease in mean IQ score with increasing fluoride levels. Id. at 32.

13
14
15

e.

Rocha-Amador (2007), a cross-sectional study of children in Mexico found

significant associations between fluoride and IQ scores. Id. at 33.
f.

Cross-sectional studies of children in India found significant association between

16

fluoride and intellectual impairment: Sudhir (2009) (found a significant increase in proportion of

17

children with intellectual impairment with increasing drinking water fluoride levels); Saxena

18

(2012) (significant correlations between reduced IQ and water fluoride and urinary fluoride

19

levels); Trivedi (2012) (found significantly lower mean IQ scores in high fluoride villages

20

compares to low-fluoride villages for boys and girls combined and separately). Id. at 38.

21
22

g.

Siraj (2012), a cross-sectional study of children in Iran found a significant negative

association between water fluoride and IQ score. Id. at 39.

23
24
25
26
27
28

useful in generating hypotheses. A group of people are interviewed,
examined, and tested at a single point in time to ascertain a
relationship between a disease and a neurotoxic exposure. This
study design does not allow the investigator to determine whether
the disease or the exposure came first, rendering it less useful in
estimating risk. These studies are intermediate in cost and time
required to complete compared with case reports and more complex
analytical studies, but should be augmented with additional data.
Dkt. No. 429-7, Trial Ex. 17 at 16.

22

1
2
3

Soto-barreras (2019), a cross-sectional study of children in Mexico 9-10 years of

age did not find a significant association between fluoride and IQ levels. Id. at 34.
36. In addition to the studies that the NTP Monograph deemed “high-quality,” and thus most

4

relevant to understanding impact of fluoride, the NTP Monograph explains that 46 of the 53

5

studies deemed low-quality by the NTP Monograph also found an association between fluoride

6

exposure and reduced IQ in children. NTP Monograph at xii.

7

37. Several studies published after the NTP Monograph literature cut-off date (April 2021), see

8

NTP Monograph at 5-12, 12 n.8, B-2, C-2-C-44, also found negative association between fluoride

9

and IQ, and acutely, for boys – bolstering the NTP Monograph’s finding of a negative association

10
11
United States District Court
Northern District of California

h.

between IQ in children and fluoride exposure:
a. Goodman (2022a): studied samples from the ELEMENT cohort and concluded that

12

an increase in maternal urinary fluoride predicated an average 2.12-point decrease in GCI scores

13

of 4-year-olds and a 2.63 decrease in performance IQ of 6- to 10-year-olds. Dkt. No. 432-11, Trial

14

Ex. 115 at 1-2. The study also found a marginal association with maternal urinary fluoride and

15

verbal IQ across time. Id. at 2. The study concluded that visual-spatial and perceptual reasoning

16

ability may be more impacted by prenatal fluoride exposure as compared to verbal abilities. Id.

17

b. Cantoral (2021): studied 103 mother-child pairs from the Programming Research in

18

Obesity, Growth, Environment and Social Stressors (“PROGRESS Cohort”) program. Dkt. No.

19

432-6, Trial Ex. 110 at 2. The PROGRESS Cohort collected data regarding dietary fluoride intake

20

from mothers (via food and beverage) during pregnancy and neurodevelopmental testing from

21

their offspring for 948 mother-child pairs from Mexico City. Id. at 2. Dietary fluoride intake was

22

measured via food frequency questionnaires from mothers in trimesters two and three of

23

pregnancy and children’s cognitive, motor, and language outcomes were measured at 12 and 24

24

months. Id. at 1. Cantoral (2021) studied data from 103 mother-child pairs from the PROGRESS

25

Cohort to understand if dietary fluoride intake during pregnancy is associated with toddlers’

26

neurodevelopment. Id. The study found a statistically significant association between maternal

27

fluoride intake and cognitive outcome in 24-month-old boys (0.5 mg/day increase in overall

28

dietary fluoride intake associated with 3.5-point lower cognitive outcome). Id. There was no
23

United States District Court
Northern District of California

1

statistical association for girls or boys at 12 months of age. Id. Averaging across the entire age

2

group, a 0.5 mg/day increase was associated with a 3.46-point lower cognitive outcome in boys,

3

which was statistically significant. Id. The study concludes: “[t]hese findings suggest that the

4

development of nonverbal abilities in males may be more vulnerable to prenatal fluoride exposure

5

than language or motor abilities, even at levels within the recommended intake range.” Id.

6

c. Godebo (2023): this study assessed the association between chronic exposure to

7

naturally occurring fluoride and drinking water and cognitive function in school-aged children,

8

measured by two distinct assessments: a drawing test with familiar objects and the Cambridge

9

Neuropsychological Test Automated Battery, Paired Associate Leaning (“CANTAB PAL”)12 test.

10

Dkt. No. 432-14, Trial Ex. 118 at 15-16. The population studied was recruited from eight

11

communities exposed to chronic fluoride ranging from 0.41 to 15.5 mg/L fluoride in water

12

sources. Id. at 15. The study reported adverse associations of fluoride exposure in drinking water

13

with children’s drawing and CANTAB task performance, with the most significant negative

14

impacts observed for more challenging drawing tasks (i.e., drawing a donkey rather than a house

15

or a person). Id. at 16. The study concluded that this may be indicative of a greater challenge

16

“accessing working memory for this task.” Id.
d. Adkins (2022): this study evaluated data collected from the Cincinnati Childhood

17
18

Allergy and Air Pollution Study (“CCAAPS”). Dkt. No. 432-8, Trial Ex. 112 at 1. CCAAPS

19

collected urine samples from children at 12 years of age and collected Behavior Assessment

20

System for Children-2 which evaluates internalizing symptoms such as anxiety depression and

21

somatization. Id. at 2. The study found that higher children’s urinary fluoride concentrations were

22

significantly associated with increased somatization, but not depression or anxiety. Id. The study

23

found that male participants exhibited higher internalizing and somatization behaviors relative to

24

female participants. Id. at 6. The study concluded that “[d]espite males and females having

25

comparable urinary fluoride concentrations, males may be at greater risk for adverse effects of

26
27
28

12

The tests present patterns and shapes on a screen and ask children to touch and recount the
patterns to assess spatial memory and learning. Dkt. No. 432-14, Trial Ex. 118 at 10-11. Spatial
memory and learning are linked to the medial temporal lobe e.g., hippocampus, which the study
reports is the brain region thought to be most affected by fluoride toxicity. Id. at 5.
24

1

fluoride exposure as the association between fluoride concentrations and internalizing symptoms

2

was more robust among males.” Id. at 9.
e. Risk Sciences International (“RSI”), under contract with Health Canada, also

United States District Court
Northern District of California

3
4

conducted an extensive systematic review of the fluoride neurotoxicity literature: Taher (2024).

5

Dkt. No. 433-4, Trial Ex. 129; Dkt. No. 433-6, Trial Ex. 131 (Taher (2024) Supplementary

6

Materials). Taher (2024) came to a similar conclusion as the NTP Monograph, finding a

7

“moderate to strong magnitude (strength) of association between fluoride and neurocognitive

8

effects with consistent evidence across studies for the impact on childhood IQ.” Dkt. No. 433-4,

9

Trial Ex. 129 at 21; Dkt. No. 433-6, Trial Ex. 131 at 1516 (“The overall evidence identified to date

10

strongly suggests that fluoride can affect cognitive outcomes in children (specifically, reduction in

11

IQ scores), at levels close to those currently seen in North American drinking water.”).13
38. Other post-NTP Monograph studies did not find fluoride was associated with adverse

12
13

cognitive outcomes in children:
a. Ibarluzea (2021): the study evaluated data from 316 to 248 mother-child pairs from

14
15

the Infancia y Medio Ambiente cohort project (“INMA Cohort”). Dkt. No. 432-10, Trial Ex. 114

16

at 1. The INMA Cohort draws on data from mothers and children in Gipuzkoa, Spain (Basque

17

Country) living in fluoridated and non-fluoridated water communities that supplied water with the

18

mean fluoride level of 0.81 mg/L. Id. at 1, 3. The INMA study collected maternal urinary

19

fluoride levels in the first and third trimesters of pregnancy, and children’s cognitive domains and

20
21
22
23
24
25
26
27
28

13

Unlike the NTP Monograph, Taher (2024) considered evidence relating to multiple endpoints
(i.e., a particular adverse effect, see Dkt. No. 434-15, Trial Ex. 535 at 43) aside from reduced IQ to
decide which endpoints need be accounted for by regulators; endpoints considered included
kidney dysfunction, sex hormone disruptions, and dental fluorosis, see Dkt. No. 433-4, Trial Ex.
129 at 21-23. Taher (2024) concluded that dental fluorosis and reduced IQ are critical endpoints;
evidence supported the association between fluoride and those two adverse effects. See id. at 27.
Taher (2024) did find that dental fluorosis should be the primary endpoint used by regulators
because data regarding the association between dental fluorosis and fluoride was more certain than
evidence regarding the association between IQ reduction and fluoride. Id. However, Taher (2024)
explained that both dental fluorosis, and separately, IQ reduction in children should be considered
by regulatory bodies, including the United States EPA, when assessing regulation of fluoride. Id.
To this end, the review recommended that fluoride at 1.56 mg/L be deemed hazardous, explaining
that this level should be utilized by regulators in its calculations to protect the public against both
dental fluorosis and IQ reduction. See id. Thus, the findings of Taher (2024) are consistent with
the NTP Monograph’s finding that fluoride is associated with reduced IQ, particularly at exposure
levels above 1.5 mg/L.
25

1

intelligence indexes, evaluated used the Bayley Scales (age 1) and McCarthy Scales (age 4). Id. at

2

1. The study concluded that per unit of maternal fluoride across the pregnancy was associated

3

with a sizeable increase in IQ scores (15-point increase) and an increase in verbal, performance,

4

numeric, and memory domains in boys. Id. For girls, there was no significant association

5

between maternal fluoride and cognitive score. Id.

United States District Court
Northern District of California

6

b. Dewey (2023): This study compared data collected from maternal-child pairs in

7

Calgary, Canada pre- and post-May 19, 2011, when the city stopped fluoridating its drinking water

8

(with a recommended level of 0.7 mg/L) to discern if fluoridated drinking water was associated

9

with children’s intelligence and executive function at 3-5 years of age. Dkt. No. 432-13, Trial Ex.

10

117 at 1. The study compared data from maternal-child pairs that were either fully exposed to

11

fluoridated drinking water throughout pregnancy, exposed for part of the pregnancy, and those not

12

exposed to fluoridated drinking water. Id. The study found no adverse associations between

13

maternal exposure to fluoridated drinking water for intelligence. Id. at 7. The study observed that

14

maternal exposure to fluoridated drinking water was associated with poorer executive function in

15

preschool aged children and, particularly, girls. Id.

16

c. Do (2022): This study collected additional data from participants in Australia’s

17

National Child Oral Health Study 2012-14, which gathered data from children aged 5-10 years,

18

and collected additional data from them again 7-8 years later but before the children turned 18

19

years of age. Dkt. No. 432-9, Trial Ex. 113 at 1. The study estimated lifetime exposure to

20

fluoridated water based upon residential history and postcode-level fluoride levels in public tap

21

water and measured children’s emotional and behavioral development and executive functioning

22

using questionnaires. Id. The study concluded that exposure to fluoridated water during the first 5

23

years of life (post-birth) was not associated with altered measures of child emotional and

24

behavioral development and executive functioning by 18 years of age. Id.

25

39. For several reasons, the studies that did not find a negative association between fluoride

26

and IQ, or that observed the association in some groups (boys) but not others (girls) do not

27

undermine the significant evidence finding such an association, reflected in the NTP Monograph

28

and studies published after the Monograph. The Court affords less weight to these studies finding
26

1
2

United States District Court
Northern District of California

3

lack of an association due to various characteristics of those studies:
a. The reliability of Ibarluzea (2021) is questionable in several respects:
i.

This study found that per one unit increase in the mg/L maternal urinary

4

fluoride, there was an association with a 15-point increase in the IQ of boys associated with

5

maternal urinary fluoride. Dkt. No. 432-10, Trial Ex. 114 at 1. Dr. Savitz, EPA’s expert, agrees

6

that this finding is an outlier and unexpected, insofar as no other study has reported a positive

7

association between fluoride exposure upon IQ, and does not meaningfully support that fluoride is

8

beneficial. See Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1067:2-1069:11 (Savitz) (“Again, based on

9

what I know, I would doubt that that is an accurate reflection of the causal impact of fluoride on

10

IQ.”). Experts also testified that they were not aware of any other chemical known to increase the

11

IQ of humans by 15 points. Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 372:14-16 (Grandjean); Dkt.

12

No. 395, Jan. 31, 2024, Trial Tr. at 111:4-6 (Hu). This association appears scientifically

13

implausible and raises questions about the overall reliability of this study.

14

ii.

Further, the 15-point increase in IQ disappeared to reflect a null finding

15

when the maternal urinary fluoride was not adjusted for creatinine. Dkt. No. 395, Jan. 31, 2024,

16

Trial Tr. at 109:5-11 (Hu). Adjusting maternal urinary fluoride for creatinine is standard practice,

17

and results from creatinine-adjusted urinary fluoride are considered the informative and reliable

18

results of a study. Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 108:7-10 (Hu); Dkt. No. 414, Feb. 9,

19

2024, Trial Tr. at 1089:5-17 (Savitz), 1090:24-1091:2 (Savitz). However, adjusting for creatinine

20

is expected to sharpen results, because the adjustment countervails for urinary dilution which

21

might introduce noise into a study; the adjustment is not, however, expected to have any

22

significant impact on the direction of results of the study. Dkt. No. 395, Jan. 31, 2024, Trial Tr. at

23

108:11-22 (Hu); Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 372:25-373:22 (Grandjean), 376:15-

24

378:24 (Grandjean). The results in the Ibarluzea (2021) study, which transitioned from a

25

significant positive association to a null finding when urinary fluoride was adjusted for creatinine,

26

was considered surprising and not a plausible result. Dkt. No. 395, Jan. 31, 2024, Trial Tr. at

27

109:13-110:7 (Hu); Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 372:25-373:22 (Grandjean), 376:15-

28

378:24 (Grandjean). Plaintiffs’ experts credibly testified that this discrepancy suggests there was
27

1

an error when matching fluoride and creatinine data. Dkt. No. 417, Feb. 2, 2024, Trial Tr. at

2

372:25-373:22 (Grandjean). EPA’s experts at trial could not explain or account for this aspect of

3

the study. Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1091:3-1093:8 (Savitz).

United States District Court
Northern District of California

4

iii.

Another concern with the Ibarluzea (2021) study is that it did not adjust for

5

seafood as a covariate in the analysis of fluoride and IQ. Dkt. No. 397, Feb. 2, 2024, Trial Tr. at

6

453:12-17 (Grandjean). Seafood is both high in fluoride content and omega 3 fatty acids. Dkt.

7

No. 395, Jan. 31, 2024, Trial Tr. at 110:20-23 (Hu). This is problematic because omega 3 fatty

8

acids have beneficial effects on cognition, and thus seafood may be a confounding factor, skewing

9

results of a study if the population has a high seafood ingestion rate. Dkt. No. 395, Jan. 31, 2024,

10

Trial Tr. at 110:20-111:3 (Hu). The study did adjust for cord blood mercury levels, which could

11

operate as an adjustment for fish consumption because fish often contain mercury. Dkt. No. 414,

12

Feb. 9, 2024, Trial Tr. at 1073:20-1074:14 (Savitz). However, the bigger the fish, the more likely

13

the accumulation of mercury; conversely, the smaller the fish, the less likely the accumulation of

14

mercury. Id. at 1076:20-1078:9. Yet, in coastal Spain where the study was conducted, sardines

15

and anchovies are popular, which are small fish that are lower on the food chain and accordingly

16

low in mercury. See Dkt. No. 417, Feb. 2, 2024 at 458:23-459:17 (Grandjean); Dkt. No. 414, Feb.

17

9, 2024, Trial Tr. at 1269:24-1270:12 (Savitz). Thus, it is not clear that the adjustment for cord

18

blood mercury levels is a sufficient proxy for seafood consumption. To this end, Dr. Savitz agreed

19

that it is a reasonable hypothesis that fish consumption accounted for the beneficial results

20

associated with IQ observed in the Ibarluzea (2021) study. Dkt. No. 414, Feb. 9, 2024, Trial Tr. at

21

1069:23-1070:18 (Savitz).

22

iv.

Taher (2024) likewise concluded that Ibarluzea (2021) does not overcome

23

evidence linking fluoride to reduced IQ in children. Namely, Taher (2024) concluded that “[t]he

24

available evidence demonstrated a moderate to strong magnitude (strength) of association between

25

fluoride and neurocognitive effects with consistent evidence across studies for the impact on

26

childhood IQ at fluoride exposures relevant to current North American drinking water levels.”

27

Dkt. No. 433-4, Trial Ex. 129 at 21. This is because, “[f]ocusing on high quality cohort studies,

28

most of the evidence suggests a reduction in childhood IQ scores associated with fluoride levels,
28

1

though results from one 2023 study in Spain (Ibarluzea et al. 2022) documented an improvement

2

in specific cognitive domain scores in boys.” Id.

United States District Court
Northern District of California

3

b. Dewey (2023) is not strong evidence regarding the association between fluoride

4

and reduced IQ because of the design of this study. The study attempted to take advantage of

5

what was thought to be a naturally occurring cohort with an exposure contrast (i.e., one cohort

6

exposed to fluoride and one not exposed to fluoride) to see if there was a meaningful difference in

7

cognitive outcomes amongst the two groups. Dkt. No. 397, Jan. 31, 2024, Trial Tr. at 368:22-

8

369:7 (Grandjean). Specifically, the study looked at individuals from a Canadian community that,

9

for a long time, fluoridated its water and stopped fluoridating the water; the study compared the

10

cognition of children in fluoridated and non-fluoridated groups to discern the impact of fluoride.

11

Dkt. No. 397, Jan. 31, 2024, Trial Tr. at 368:22-369:7 (Grandjean). However, the study did not

12

collect data on the urinary fluoride levels of the mother or assess how long pregnant mothers lived

13

in the area prior to their pregnancy. Dkt. No. 397, Jan. 31, 2024, Trial Tr. at 368:22-369:18

14

(Grandjean). This is relevant because women who live in a fluoridated area throughout their lives

15

will have fluoride which accumulates in her bones from consumption of fluoridated water, along

16

with other sources; for several years after cessation of fluoride exposure she is likely to release

17

accumulated fluoride from her bones into blood due to skeletal breakdown. Dkt. No. 397, Jan. 31,

18

2024, Trial Tr. at 370:6-371:12 (Grandjean); Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 932:16-20

19

(Thiessen). This skeletal breakdown is particularly present during pregnancy, as the maternal

20

skeleton dissolves itself to provide calcium to the growing fetal skeleton. Dkt. No. 395, Jan. 31,

21

2024, Trial Tr. at 121:10-20 (Hu). Accordingly, the group that was considered non-fluoridated in

22

the study, thus creating an exposure contrast between the two groups allowing for a potential

23

association to be observed, may have in fact exposed the child to fluoride during pregnancy if she

24

lived in a fluoridated area prior to the study (a phenomenon that is not reported or considered by

25

the study). This could lessen the exposure contrast and calls the results of the study into question.

26

See Dkt. No. 397, Jan. 31, 2024, Trial Tr. at 368:22-369:18 (Grandjean). EPA’s expert witnesses

27

did not account for this concern regarding the study design. Thus, the Dewey study is accorded

28

diminished weight.
29

United States District Court
Northern District of California

1

c. Do (2022) assessed primarily behavioral outcomes rather than impact on IQ in

2

children and, as Dr. Savitz testified, “doesn’t stand out as definitive or more persuasive,” relative

3

to other studies directly on point to association of fluoride on the IQ of children. Dkt. No. 414,

4

Feb. 9, 2024 Trial Tr. at 1106:22-1107:10 (Savitz). Plaintiffs’ experts also expressed concerns

5

with the study. The study utilized the “SDQ” test to measure impact of fluoride on children in

6

Australia, which is a test that, for certain cultural or linguistic reasons, has been determined to be

7

unreliable for Australians by another study conducted by the co-author of Do (2022). Dkt. No.

8

397, Jan. 31, 2024, Trial Tr. at 364:8-14, 365:15-366:4 (Grandjean). EPA’s expert witness did not

9

rebut evidence that there were significant problems with the validity of the SDQ test in Australia.

10

Dkt. No. 415, Feb. 12, 2024, Trial Tr. at 1240:1-6 (Savitz). Further, the value of this study is

11

weakened because it did not analyze individualized data, but instead measured exposure based on

12

residence of the child and community-wide data on fluoride in that area. See Dkt. No. 396, Feb. 1,

13

2024, Trial Tr. at 240:17-19 (Lanphear) (explaining that individualized data is generally a strength

14

of a study); Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 366:5-367:4, 367:15-368:4 (Grandjean). Lack

15

of individualized data can lead to exposure imprecision, creating “noise” in the data, which may

16

bias results toward the null, i.e., noise makes it less likely to show an association between the

17

chemical and a result. Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 106:18-107:16 (Hu); Dkt. No. 396,

18

Feb. 1, 2024, Trial Tr. at 281:14-17 (Lanphear), 281:24-282:3 (Lanphear), 317:16-24 (Grandjean);

19

Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1176:4-17 (Savitz) (agreeing with a statement made in his

20

textbook that in general exposure misclassification tends to produce results with a bias towards the

21

null). Thus, this study is not particularly probative evidence as to association between fluoride

22

and IQ of children.

23

40. EPA experts agreed, in line with the NTP Monograph’s conclusion, that fluoride is

24

associated with adverse IQ in children at “higher” levels of exposure. Namely, Dr. Barone

25

testified that he agreed that there is “something going on” at higher-dose levels, though unclear

26

about where the threshold is. Dkt. No. 415, Feb. 12, 2024, Trial Tr. at 1372:9-1373:9 (Barone).

27

Dr. Barone agreed that, at 4 mg/L of fluoride exposure and above, there is more data to support a

28

finding of an adverse effect associated with fluoride. Id. at 1373:1-9 (Barone). Dr. Barone further
30

United States District Court
Northern District of California

1

testified: “I agree with the NTP’s conclusions that at some level above 1.5 mg/L that there is

2

moderate evidence to support an association between fluoride and developmental IQ decrements.”

3

Dkt. No. 416, Feb. 12, 2024, Trial Tr. at 1428:4-11 (Barone).14 The primary concern presented by

4

EPA’s experts relates to lack of clarity as to whether lower exposure levels of fluoride (below 1.5

5

mg/L) results in an adverse outcome and the precise relationship between dose and response. See

6

Dkt. No. 415, Feb. 12, 2024, Trial Tr. at 1357:9-1360:10 (Barone). For example, Dr. Savitz

7

(EPA’s expert witness) did not opine that the NTP Monograph’s main conclusion that fluoride is

8

presumed to be a cognitive neurodevelopmental hazard to humans was incorrect, though

9

expressing concerns as to a previous draft of the Monograph regarding whether its conclusion was

10

well explained and qualified. Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1113:16-1115:23 (Savitz)

11

(“Whether [a committee reviewing a draft of the NTP Monograph] agreed with [the NTP

12

Monograph’s conclusion] was not the issue. It was – the story that gets to the punchline at the end

13

we did not find persuasive.”). Indeed, Dr. Savitz explained that he does not have a basis to

14

challenge the NTP’s conclusion that, with moderate confidence, there is an association or appears

15

to be an association between neurological decrements in fluoride concentrations above 1.5 mg/L.

16

Id. at 1140:10-19 (Savitz) (“I don’t have any reason to challenge [this conclusion], but I haven’t

17

corroborated it by going through the dozens of studies one-by-one to make my own assessment.”).

18

Dr. Savitz likewise made clear he did not undertake a complete review of the NTP Monograph,

19

but testified his primary concern was the Monograph’s “inferences regarding lower levels of

20

fluoride exposure.” Id. at 1129:11-1131:3 (Savitz).
41. The robust body of scientific literature systematically assessed by the NTP Monograph

21
22

(described above, ¶ 35) and literature published after the NTP Monograph cutoff date (described

23

above, ¶ 37), even considering some countervailing scientific literature (described above, ¶¶ 38-

24

39) establishes by a preponderance of the evidence that fluoride is associated with reduced IQ in

25
26
27
28

14

Dr. Barone testified that the NTP Monograph was helpful but not complete and thus insufficient
to satisfy the hazard identification prong of TSCA hazard assessment. Dkt. No. 440, Feb. 13,
2024, Trial Tr. at 1428:22-1429:3 (Barone). That testimony is not credible because it directly
contradicts Dr. Barone’s prior testimony during his deposition that the literature the NTP reviewed
through April 2021 was sufficient to satisfy the human evidence standard for identifying a hazard
under the EPA’s TSCA guidelines. Id. at 11-21.
31

1

children – at least at “higher” concentration levels, i.e., above 1.5 mg/L (measured in either water

2

fluoride levels or urinary fluoride levels). At the hazard identification step, the EPA does not

3

require showing that an adverse effect is present at the level akin to the exposure in the community

4

(i.e., 0.7 mg/L) or require the establishment of a dose-response relationship of the chemical at

5

“lower” levels. Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 493:16-495:12 (Barone). The evidence

6

regarding the “higher” exposure levels is sufficient to satisfy the hazard identification step of the

7

analysis.

8
9

United States District Court
Northern District of California

10

42. Regardless, scientific literature in the record also indicates there is an association between
fluoride and reduced IQ in children even at “lower” levels of exposure (i.e., below 1.5 mg/L).
43. Two of the three high-quality studies that evaluated the effects of “lower” levels of fluoride

11

exposure (below 1.5 mg/L) did observe an association between fluoride and reduced IQ in

12

children or boys. Namely: (1) Bashash (2017), studied mother-child pairs from the ELEMENT

13

Cohort (Mexican population) and observed a statistically significant decrement of 3.15 GCI score

14

and 2.5 IQ score of offspring per an increase of 0.5 mg/L of maternal urinary fluoride where the

15

mean maternal urinary fluoride in mothers was 0.9 mg/L, Dkt. No. 432-2, Trial Ex. 106 at 1; and

16

(2) Green (2019) studied mother-child pairs in the MIREC Cohort (Canadian population) and

17

found a statistically significant decrement of 3.66 IQ score in boys only (3.66 IQ score decrement

18

per a 1 mg/L per day increase in maternal urinary fluoride) where the mean maternal urinary

19

fluoride of mothers was 0.42 mg/L, Dkt. No. 432-5, Trial Ex. 109 at 1-3, 5.

20

44. Another program collected samples from 837 mother-child pairs from the Odense

21

municipality in Denmark: the Odense Child Cohort (“OCC Cohort”). Dkt. No. 432-15, Trial Ex.

22

119 at 1. The OCC Cohort measured maternal urinary fluoride during pregnancy and the IQ of

23

school-aged offspring of those mothers. Id. The maternal urinary fluoride concentrations

24

averaged at 0.58 mg/L per day. Id. at 2. The study, when accounting for covariables did not

25

observe a statistically significant association between maternal urinary fluoride and child Full-

26

Scale IQ score, with no clear interaction between sex and fluoride exposure. Id.

27

45. The result of the OCC Cohort does not negate the findings regarding the MIREC and

28

ELEMENT cohorts. It is inherently more difficult to observe an adverse effect of a chemical at
32

United States District Court
Northern District of California

1

lower exposure levels because of reduced exposure contrast15 at those levels. Dkt. No. 395, Jan.

2

31, 2024, Trial Tr. at 113:2-25 (Hu), 114:8-14 (Hu); Dkt. No 396, Feb. 1, 2024, Trial Tr. at 213:5-

3

25 (Lanphear); Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 525:9-526:13 (Berridge). EPA’s expert,

4

Dr. Savitz, agreed. Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 1009:7-23 (Savitz) (“[Y]ou could think

5

of the worst cases, if we all had the exact same value, everybody in the population had the same

6

exposure, you could not do an informative study of the association of exposure with a disease.

7

And if it’s very narrow, of course, you’re only able to study – if you’re only able to study, let's

8

say, the contrast of, you know, .4 and .5 milligrams per liter fluoride, you're going to have a tough

9

time, even if there were an effect, it's going to be difficult to find because you have a very limited

10

contrast. As you spread that out more, of course, you are – you have a larger contrast and you're

11

able to address a more informative range of exposure.”). It is particularly difficult to observe

12

effects of fluoride at lower exposure levels because of challenges in finding a control group with

13

zero or very little fluoride exposure. Dkt. No 396, Feb. 1, 2024, Trial Tr. at 212:7-213:25

14

(Lanphear). This is because fluoride exposure is prevalent. Some common sources aside from

15

fluoridated water include naturally occurring fluoride in food and beverage, fluoride in food and

16

beverage made with fluoridated water, and other products, like toothpaste. Id. at 212:10-19

17

(Lanphear). Thus, it is difficult to find a control group without any fluoride exposure; the “noise”

18

created by background fluoride exposure tends to obscure the contrast between those who

19

consume fluoridated water and those who do not. Id. at 212:19-23 (Lanphear) (“And so if we

20

wanted to ask a question . . . is there a difference in children who are unexposed to fluoride? Well,

21

we really can’t find children who are unexposed to fluoride versus kids who have levels in a

22

nonfluoridated community or a fluoridated community.”). It is thus more challenging to observe

23
24
25
26
27
28

15

Exposure contrast refers to the difference between exposure of a chemical in one group (a
control group) and another group (the group exposed to the chemical). Dkt. No. 395, Jan. 31,
2024, Trial Tr. at 113:6-22 (Hu). For example, an observer would compare a group with less or
no fluoride exposure to a group with more exposure to determine if there is a meaningful
difference in the group with more exposure. See Dkt. No 396, Feb. 1, 2024, Trial Tr. at 212:10-23
(Lanphear). When trying to observe effects of a chemical at lower levels, there is less “exposure
contrast” between the control group and exposed group. See id. at 212:10-213:25. Dr. Hu
provided an illustration: “It’s sort of like looking at, you know, a picture and trying to determine
whether this shade is different from that shade. If you increase the contrast, it’s easier to see.”
Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 114:12-14 (Hu).
33

1

effects at lower concentration levels of fluoridated water. Id. at 212:24-213:25 (Lanphear).

2

Accordingly, the Court finds convincing and credible the expert testimony that studies analyzing

3

the OCC Cohort are not inconsistent with studies analyzing the ELEMENT and MIREC Cohorts;

4

the lower exposure levels account for some difficulty in repeating observed effects. Dkt. No. 395,

5

Jan. 31, 2024, Trial Tr. at 116:24-117:4 (Hu).16 In short, the association between intake of water

6

at lower fluoridated levels and IQ is likely harder to detect. Inconsistent results between studies

7

are not unexpected. The two high-quality studies which detected such an association at lower

8

concentration levels of fluoride remain significant and are not undermined by the OCC Cohort

9

study.
46. In conclusion, Plaintiffs have established by a preponderance of the evidence that exposure

United States District Court
Northern District of California

10
11

to fluoride is associated with the adverse effect of reduced IQ in children, and particularly, young

12

boys. Hence, the hazard identification step of the analysis is satisfied.
2.

13

Step 1B: Weight of the scientific evidence
a.

14

Framework

47. Once a hazard has been identified, the EPA assesses the weight of the scientific evidence,

15
16

wherein the risk assessor considers the weight of that evidence, determining which adverse effects

17

(endpoints) are to be assessed, and which studies are appropriate for use in quantifying the

18

relationship between the dose of the chemical and adverse effect(s) (response) at issue (the “dose-

19

response” assessment). Dkt. No. 400, Feb. 4, 2024, Trial Tr. at 661:18-666:14 (Barone). To this

20

end, not all studies are appropriately utilized in the dose-response assessment. See Dkt. No. 417,

21

Feb. 2, 2024, Trial Tr. at 494:17-495:12 (Barone). Rather, the EPA identifies the studies from the

22

hazard identification step that are generally of high or medium quality, and thus are deemed

23

permissible to use in the dose-response assessment. Id. at 494:17-495:12; Dkt. No. 421 at 5

24

(undisputed fact).

25
26
27
28

16

Expert witnesses also testified credibly that there are some possible explanations for the
differing study results; for example, it is possible that Denmark has higher iodine consumption,
accounting for the discrepancy, as iodine deficiency is theorized to be an aggravating factor for
impacts of fluoride on neurodevelopment. See Dkt. No. 396, Feb. 1, 2024, Trial Tr. at 248:10250:3 (Lanphear).
34

United States District Court
Northern District of California

1

48. The parties disagree as to precisely how the weight-of-the-scientific evidence analysis

2

intersects with the subsequent step of the analysis: the dose-response assessment wherein a point

3

of departure17 is identified (Step 1C, discussed in Section III.A.3.). See Dkt. No. 421 at 22-23.

4

Plaintiffs assert that the weight-of-the-scientific evidence analysis is a distinct, qualitative

5

characterization of the evidence regarding a “chemical’s potential to produce neurotoxicity,”

6

separate from the quantitative dose-response assessment wherein a point of departure is calculated

7

(Step 1C, discussed in Section III.A.3). Id. The EPA asserts that there is not a clear distinction

8

between the qualitative and quantitative dose-response assessment. See id. Dr. Barone, EPA’s

9

expert does recognize that risk evaluation includes a “quantitative track wherein the agency is

10

doing a quantitative measurement, deriving a point of departure, and a qualitative track where [the

11

assessor is] assessing whether that evidence is appropriate for that purpose.” See Dkt. No. 400,

12

Feb. 4, 2024, Trial Tr. at 666:9-14 (Barone). Moreover, Dr. Barone stated that: “in this weight of

13

the scientific evidence evaluation . . . [we ask] how much data do we actually have for that

14

particular endpoint or that particular outcome, and are there a series of outcomes that are related to

15

neurotoxicity that we should consider as an example or reproductive toxicity. So we may have

16

multiple endpoints to consider and multiple studies within that, that can be carried forward to

17

dose response.” Id. at 662:2-19 (emphasis added). This testimony intimates that the weight-of-

18

scientific-evidence analysis occurs prior to, and separately from, the quantitative dose-response

19

assessment wherein a point of departure is calculated. See id. However, to avoid any doubt, the

20

Court assesses the weight-of-the-scientific evidence both as a standalone, qualitative issue,

21

characterizing the weight of the evidence assessing the association between the chemical and

22

endpoint (in this section of the analysis (Section III.A.2., as Step 1B)) and also assesses the

23

weight-of-the-scientific-evidence, as part of the quantitative dose-response assessment wherein a

24

point of departure is identified (Section III.A.3, as Step 1C, discussed below).
b.

25

Key finding

49. The weight of the scientific evidence regarding fluoride’s association with reduced IQ is

26
27
28

17

As explained in depth in Section III.A.3., the point of departure represents the level at which the
chemical at issue becomes hazardous.
35

1

sufficient to proceed to the dose-response assessment; the evidence in the record is appropriate for

2

use in calculating a point of departure.

3
4

Underlying findings

50. The term “weight of the scientific evidence” is supported by EPA’s systematic analysis of

5

the related information to support the Agency’s findings. Id. at 651:22–652:5; accord 40 CFR

6

702.33. The assessor uses the “best available science,” in the analysis, which means that TSCA

7

risk evaluations need to be unbiased and objective, and the methodologies employed must be

8

transparent and reproducible and generally peer reviewed. Dkt. No. 400, Feb. 5, 2024, Trial Tr. at

9

652:6-16 (Barone); accord 40 C.F.R. 702.33.

10

United States District Court
Northern District of California

c.

51. In the weight-of-the-scientific-evidence analysis, generally, high- or medium-quality

11

studies are adequate to move to the dose-response determination. Dkt. No. 417, Feb. 2, 2024,

12

Trial Tr. at 494:17-495:12 (Barone); Dkt. No. 421 at 5 (undisputed fact). Still, the EPA

13

sometimes carries over low-quality studies into the dose-response analysis as well. Dkt. No. 417,

14

Feb. 2, 2024, Trial Tr. at 494:21-495:1 (Barone). In this weight-of-the-scientific-evidence

15

analysis, some or all factors referred to as the “Bradford Hill” factors may be considered. Dkt.

16

No. 400, Feb. 5, 2024, Trial Tr. at 626:8-24 (Barone). The nine Bradford Hill factors are: (1)

17

strength of the association, (2) consistency of the association; (3) specificity of the association; (4)

18

temporality of the association; (5) biological gradient (i.e., dose response) of the association; (6)

19

plausibility of the association; (7) coherence of the association, (8) experimental support for the

20

association, and (9) analogies for the association. See Dkt. No. 198-3, Grandjean Trial Decl. ¶¶

21

111-125. However, there is no mandate that each of the Bradford Hill factors be considered in the

22

weight-of-the-evidence assessment in a non-cancer TSCA risk evaluation such as this one. See

23

Dkt. No. 437-1, Trial Ex. 96 (hereinafter “PCE Risk Evaluation”) at 326 (considering only

24

consistency of association factor); Dkt. No. 437-7, Trial Ex. 102 (hereinafter “Methylene Risk

25

Evaluation”) at 285-95 (considering some, but not all, of the Bradford Hill factors).

26

52. As discussed previously, not every epidemiological study on fluoride has found

27

associations with reduced IQ in children. See ¶¶ 35, 38. However, the evidence at issue is overall

28

consistent as to the finding that fluoride is associated with reduced IQ in children, and there is a
36

1

United States District Court
Northern District of California

2

vast amount of experimental support for the association:
a. The NTP Monograph studied a robust amount of literature regarding fluoride’s

3

impact on children’s IQ: 72 epidemiological studies – 19 of which were deemed “high quality”

4

and “low-risk-of-bias,” and 53 lower-quality studies – a large majority of which identified an

5

association between fluoride and reduced IQ. NTP Monograph at xii (describing that 46 of the 53

6

low-quality studies found an association between higher fluoride exposure and lower IQ in

7

children and 18 of 19 high-quality studies reported an association between higher fluoride

8

exposure and lower IQ in children including 3 prospective cohort studies and 15 cross-sectional

9

studies). Indeed, when narrowing evidence to view only 19 studies that are high quality and low

10

risk-of-bias, all but one identified an association between fluoride and reduced IQ: Bashash et al.

11

2017; Choi et al. 2015; Cui et al. 2018; Ding et al. 2011; Green et al. 2019; Rocha-Amador et al.

12

2007; Saxena et al. 2012; Seraj et al. 2012; Sudhir et al. 2009; Till et al. 2020; Trivedi et al. 2012;

13

Wang et al. 2012; Wang et al. 2020b; Xiang et al. 2003a; Xiang et al. 2011; Yu et al. 2018; Zhang

14

et al. 2015b). NTP Monograph at 29-40 (Table 6).

15

b. The findings of the NTP Monograph are properly afforded substantial weight. The

16

NTP is headquartered within NIEHS, which is “is one of the premier environmental health

17

sciences research institutions in the world.” Dkt. No. 440, Feb. 13, 2024, Trial Tr. at 1425:23-

18

1426:2 (Barone). The EPA does not dispute this fact. Dkt. No. 421 at 10. Even before the NTP

19

Monograph was formally published, the EPA agreed the NTP Monograph is a high-quality review,

20

followed rules that have been developed by NTP for conducting systematic review, had a

21

“rigorous approach to assembling the evidence,” “clearly defined rules for identifying and

22

evaluating studies,” and “a well-defined protocol for drawing inferences” from the studies. Dkt.

23

No. 440, Feb. 13, 2024, Trial Tr. at 1427:9-21 (Barone), 1427:2-8 (Barone).

24

c. Though there were some critical peer review comments on earlier drafts of the NTP

25

Monograph, the core conclusion of the NTP Monograph regarding the high-quality studies was

26

not called into question by reviewers. See, e.g., Dkt. No. 438-1, Trial Ex. 69 at 65 (NTP Board of

27

Scientific Counselors Working Group Report agreeing that low-risk-of-bias studies were

28

“consistent,” meaning generating results in the same direction, in finding a negative association
37

United States District Court
Northern District of California

1

between fluoride exposure and children’s IQ); Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1114:24-

2

1115:1 (Savitz) (describing NASEM critique of adequate definition of the term “consistent” in

3

NTP Monograph, but not disagreeing with characterization of NTP Monograph finding

4

association between IQ and fluoride). Indeed, EPA’s experts at trial expressed confidence in the

5

NTP Monograph’s methodologies. Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1197:2-15 (Savitz)

6

(expressing confidence in NTP’s literature search strategy and its ability to identify all relevant

7

studies on fluoride exposure published prior to the closing date of April 21, 2021, and agreeing

8

that the “vast majority of studies” that NTP reviewed identified an association between fluoride

9

and reduced IQ). See also Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1140:10-19 (Savitz) (“I don't

10

have any reason to challenge [this conclusion], but I haven’t corroborated it by going through the

11

dozens of studies one-by-one to make my own assessment.”). Further, Dr. Savitz, the expert

12

called by the EPA herein, acknowledged he is not an expert in conducting risk assessment, and

13

particularly not under Amended TSCA. Dkt. No. 415, Feb. 9, 2024, Trial Tr. at 1264:2-6 (Savitz).

14

Formal publication of the NTP Monograph affirms its quality. See also ¶ 33.

15

d. As explained previously, studies published after the NTP Monograph’s literature

16

cut-off date likewise observed a negative association between fluoride and children’s cognition:

17

Goodman (2022(a)), Cantoral (2021), Godebo (2023), and Adkins (2022)). See ¶ 37.

18

e. Further, notwithstanding difficulties in observing effects of a chemical at lower

19

levels, see ¶ 45, adverse outcomes have even been observed at those levels with statistical

20

significance: Green (2019) and Bashash (2017), ¶¶ 42-43.

21

f. As explained previously, some studies have not observed an association between

22

fluoride and reduced IQ: Soto-barreras (2019), ¶ 35(h); Ibarluzea (2021), ¶ 38(a); Dewey 2023, ¶

23

38(b); Do (2022), ¶ 38(c); and the OCC Cohort, ¶ 44. However, complete consistency amongst

24

studies is not expected. Dkt. No. 414, Feb. 9, 2024, Trial Tr. at 1172:23-1173:6 (Savitz). To this

25

end, various co-factors or susceptibilities can influence the impact or manifestation of

26

neurotoxicants, and as such, it is to be expected that there will be some variability in results across

27

studies of different populations. See id. What may appear to be a discrepant result may, in fact,

28

reflect unmeasured differences in cofactors that influence the course of a chemical’s
38

1

neurotoxicity. See Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 102:22-104:24 (Hu); Dkt. No. 417, Feb. 1,

2

2024, Trial Tr. at 242:21-243:9 (Lanphear), 328:14-23 (Grandjean). And, as also explained

3

previously, particular characteristics of these studies finding null outcomes render them less

4

probative here. See ¶ 39. Namely, Ibarluzea (2021) found an unrealistic 15-point IQ benefit,

5

included unexplained and implausible results regarding creatinine adjustments, and failed to

6

control for seafood, ¶ 39(a); Dewey (2023) did not account for previous residence of mothers or

7

continued excretion of fluoride from skeletal breakdown during pregnancy in the control group, ¶

8

39(b); Do (2022) utilized an unreliable IQ test and did not analyze individualized data, ¶ 39(c);

9

and the OCC Cohort measured lower exposure levels which makes it more difficult to observe

United States District Court
Northern District of California

10

adverse effects, ¶ 45.

11

53. Though not definitive, there is additional evidence that supports the plausibility of the

12

association by assessing potential mechanisms for fluoride to impact IQ. Specifically, studies

13

have endeavored to consider explanations for the observed association between fluoride and IQ

14

and hypothesize that thyroid disruption may be the mechanism by which fluoride impacts

15

cognitive function:

16

a. Goodman (2022b) studied samples from the MIREC Cohort to assess the three-way

17

interplay between prenatal fluoride exposure, maternal iodine status, and child IQ. Dkt. No. 432-

18

12, Trial Ex. 116 at 1, 8. The study found that the negative association between fluoride exposure

19

and IQ observed in Green (2019) was exacerbated by low maternal iodine in pregnancy among

20

boys. Id. The study hypothesized that change in thyroid function may be a mechanism by which

21

fluoride impacts cognition; iodine impacts thyroid function. Id. at 1-2.

22

b. Hall (2023): studied samples from the MIREC Cohort and concluded that fluoride

23

in drinking water was associated with increased risk of hypothyroidism in pregnant women, and

24

that thyroid disruption may contribute to developmental neurotoxicity of fluoride. Dkt. No. 432-

25

16, Trial Ex. 120 at 1-2.

26

54. A lack of a dose-response relationship in the data may suggest that the effect is not related

27

to the putative neurotoxic effect or that the study was not appropriately controlled. Dkt. No. 429-

28

7, Trial Ex. 17 (Guidelines for Neurotoxicity Risk Assessment, Fed. Reg. 63(93):26926-26954
39

1

(hereinafter “EPA Guidelines”))18 at 50. As discussed in the next section regarding the dose-

2

response assessment, there is some lack of clarity as to the precise dose-response relationship at

3

lower exposure levels of fluoride. However, evidence indicates that there is no threshold by

4

which fluoride and adverse IQ cease to be associated. See ¶¶ 42-43.

5

55. In conclusion, this evidence is sufficient to proceed to the dose-response assessment of the

6

analysis. Cf. Methylene Risk Evaluation at 262 (conducting dose-response analysis for Methylene

7

under Amended TSCA based upon one animal study).
3.

8

a.

9

Framework

56. The point at which the chemical ceases to be safe is known as the “point of departure” (i.e.,

10

United States District Court
Northern District of California

Step 1C: Dose-response assessment

11

“POD”) or “hazard level.” See Dkt. No. 429-20, Trial Ex. 38 at 1; Dkt. No. 417, Feb. 2, 2024,

12

Trial Tr. at 495:9-14 (Barone); Dkt. No. 421 at 5. To this end, the dose-response assessment

13

describes the relationship between dosage of the chemical and a response, and endeavors to

14

identify the dosage at which a chemical is safe, and conversely, becomes hazardous; this is the

15

point of departure. See EPA Guidelines at 57. See also Dkt. No. 429-20, Trial Ex. 38 at 1

16

(describing that the objective of the dose-response assessment is to “document the relationship

17

between dose and toxic effect”).
57. There are different points of departure that can be used in a risk assessment. EPA

18
19

Guidelines at 57-58. The first approach is the NOAEL/LOAEL approach. See Dkt. No. 429-19,

20

Trial Ex. 38 at 3-4. A No-Observed-Adverse-Effect Level (“NOAEL”) is the “highest exposure

21

level at which no statistically or biologically significant increases are seen in the frequency or

22

severity of adverse effect between the exposed population and its appropriate control population.”

23

Id. at 4. In cases in which a NOAEL cannot be identified, the term lowest-observed-adverse-effect

24

level (“LOAEL”) is used, which is the lowest dose tested at which an adverse effect is detected.

25
26
27
28

18

These Guidelines were published in April 1998 and are the currently applied guidelines for EPA
Neurotoxicity Risk Assessment according to the EPA’s website. See Guidelines for Neurotoxicity
Risk Assessment, UNITED STATES ENVIRONMENTAL PROTECTION AGENCY (last visited September
12, 2024), https://www.epa.gov/risk/guidelines-neurotoxicity-risk-assessment.
40

United States District Court
Northern District of California

1

Id. at 4. Alternatively, when possible, the benchmark dose (“BMD”) approach can be used to

2

arrive at a point of departure. Id.

3

58. The BMD approach is preferred over the NOAEL/LOAEL approach, and use of a NOAEL

4

is preferred over the LOAEL. Id. See also Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 495:23-496:25

5

(Barone); EPA Guidelines at 2-3, 57-58; Dkt. No. 421 at 5 (undisputed fact). The

6

NOAEL/LOAEL approach derives the point of departure from a dosage and corresponding

7

response in subjects that was actually observed. See EPA Guidelines at 57-59. See also Dkt. No.

8

400, Feb. 5, 2024, Trial Tr. at 672:1-11 (Barone) (“So generally a NOAEL or LOAEL, as we

9

described earlier, comes directly from what is the observed concentration for an effect or no effect.

10

So it’s directly coming from the study of where that threshold for non-cancer – generally gets a

11

threshold – where does that concentration occur. And that’s describing, generally speaking, a

12

single dose. It’s within the dose continuum of how many doses were employed in the study, what

13

concentration did they measure an effect.”). See also EPA Guidelines at 57-59. The

14

NOAEL/LOAEL is thus limited to only dosages observed in the study. See EPA Guidelines at 57-

15

59. Other limitations of the NOAEL/LOAEL approach include that this approach is highly

16

dependent upon sample size of a study (e.g., where a sample size is limited, it might present a

17

higher point of departure than the true point of departure), and it does not account for the shape of

18

the dose-response curve from the experiment at issue. Id. Because of these limitations, the BMD

19

approach is preferred if the data set is appropriate for such modeling. See Dkt. No. 429-20, Trial

20

Ex. 38 at 4; Dkt. No. 400, Feb. 5, 2024, Trial Tr. at at 479:14-580:9 (Barone).

21

59. In utilizing the BMD approach, a benchmark dose, i.e., BMD or benchmark concentration

22

(“BMC”) is identified. See Dkt. No. 429-20, Trial Ex. 38 at 4. The BMD/BMC is the dose of a

23

substance that produces a “predetermined change in the response rate of an adverse effect.” Id.

24

The benchmark dose level (“BMDL”) or benchmark concentration level (“BMCL”) is the lower

25

end of the statistical confidence limit on the dose that produces the selected response. Id. In other

26

words, there is a statistical confidence interval on either side of the BMD/BMC; the

27

BMDL/BMCL is the point at the lower side of that confidence interval. See id. Like the

28

NOAEL/LOAEL, the BMCL/BMDL can be used as the point of departure. Id.
41

1

Key findings

2

60. 0.28 mg/L, or alternatively, 0.768 and/or 1.536 mg/L measured in maternal urinary

3

fluoride is associated with a 1-point decrease in IQ of girls and boys and is a legitimate point of

4

departure (BMCL) to use in this risk evaluation.

5
6
7

61. Alternatively, 4 mg/L measured in either urinary fluoride or water fluoride, is a legitimate,
conservative point of departure (LOAEL) to use in the risk evaluation.
62. Regarding the weight of the scientific evidence, the quality and weight of the evidence in

8

the record substantiates points of departure derived from either BMD modeling of the data or from

9

a LOAEL/NOAEL approach.

10
11
United States District Court
Northern District of California

b.

12
13

c.

Underlying findings
(a)

POD: 0.28 mg/L BMCL (Grandjean (2023)) or in the alternative,
0.768 mg/L and/or 1.536 mg/L BMCL (Grandjean (2022))

63. Dr. Philippe Grandjean (“Grandjean”) was the lead author of two pooled BMCL analyses,

14

one published in 2022 and another in 2023. Dkt. No. 432-20, Trial Ex. 124. (hereinafter

15

“Grandjean (2022)”); Dkt. No. 432-15, Trial Ex. 119 (hereinafter “Grandjean (2023)”).

16

64. Dr. Grandjean and his co-authors are well-regarded for their benchmark dose analyses. To

17

this end, EPA cited a pooled benchmark dose analysis authored by Grandjean as an example of

18

how to perform such an analysis in its Benchmark Dose Technical Guidance Manual, and EPA

19

has relied upon the authors’ benchmark dose analysis work in its assessment of other chemicals.

20

Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 287:16-288:18 (Grandjean), 479:25-5 (Grandjean); Dkt.

21

No. 401, Feb. 6, 2024, Trial Tr. at 748:19-750:6 (Barone).

22

65. Grandjean (2022) analyzed data from two cohorts, the ELEMENT Cohort and the MIREC

23

Cohort to conduct its BMCL analysis. Grandjean (2022) at 1-2. Grandjean (2023) analyzed three

24

cohorts: ELEMENT, MIREC, and the OCC cohorts. Grandjean (2023) at 1.

25

66. The pooled BMCL analyses of the birth cohorts sought to determine the level of fluoride in

26

maternal urine (“MUF”) that is associated with a 1-point drop in the IQ of the mothers’

27

offspring. Dkt. No. 417, Feb. 1, 2024, Trial Tr. at 339:13-23 (Grandjean). As described by RSI,

28

“[t]he choice of a BMR of 1 IQ point (corresponding to a 1% reduction from a mean IQ of 100)
42

United States District Court
Northern District of California

1

has been adopted as an appropriate benchmark on this endpoint by several regulatory bodies,

2

including the US EPA and EFSA.” Dkt. No. 433-4, Trial Ex. 129 at 27. Pooled analyses are also

3

particularly useful because a pooled analysis benefits from heightened statistical power and

4

precision that comes from large samples sizes. Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 111:9-

5

112:16 (Hu).

6

67. Grandjean (2023) concluded that “[t]he joint analysis of all three cohorts showed a

7

statistically significant association between urine-fluoride and IQ, with a BMC of 0.45 mg/L

8

(BMCL, 0.28 mg/L).” Grandjean (2023) at 1-2. Specifically, Grandjean (2023) found that the

9

BMCL associated with a 1-point decrease in IQ scores of boys and girls was 0.28 mg/L maternal

10

urinary fluoride; this BMCL was adjusted for creatinine and derived from use of a linear dose-

11

response model. Grandjean (2023) at 1-2, 9. This BMCL is a legitimate point of departure to use

12

in the risk evaluation for fluoride.

13

68. When determining whether the point of departure can be derived using the BMD or BMC

14

approach, as opposed to identifying a LOAEL or NOAEL, it is necessary to consider whether the

15

data set is appropriate for use in the BMD/BMC modeling. See Dkt. No. 400, Feb. 5, 2024, Trial

16

Tr. at 658:9-659:10 (Barone) (explaining that in identifying studies and key endpoints to “carry

17

forward to the dose-response analysis,” the assessor considers whether “are [the studies] amenable

18

to BMDS, benchmark dose modeling? Are they amenable to a LOAEL/NOAEL approach?

19

Should we use some other type of approach?”). To this end, the EPA’s technical guidance

20

provides that the following should be considered as to whether the data set is appropriate for BMD

21

modeling: (1) whether there is a statistically or biologically significant dose-related trend in the

22

selected endpoint; (2) whether a response is not only seen at a high dose; and (3) where there are

23

adequate model fits. See Benchmark Dose Technical Guidance, U.S. ENVIRONMENTAL

24

PROTECTION AGENCY (June 2012) available at https://www.epa.gov/sites/default/files/2015-

25

01/documents/benchmark_dose_guidance.pdf (hereinafter “EPA’s Benchmark Dose Technical

26

Guidance”) at 12-18.19

27
28

19

This document was not submitted as an exhibit, but the EPA’s witnesses rely on the document
for their testimony. See, e.g., Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 745:9-25 (Barone) (“Q:
43

1

69. For the reasons discussed below, the data that Dr. Grandjean analyzed is appropriate for

2

use in BMD modeling, and for similar reasons, his point of departure is supported by the weight of

3

the scientific evidence. See ¶ 51 (discussing weight-of-scientific-evidence factors). It is

4

demonstrated by a preponderance of the evidence.

5

70. As explained previously, there is a well-supported and documented, statistically significant

6

dose-related trend in the selected endpoint (reduced IQ). See ¶¶ 52-53 (discussing the robust body

7

of evidence establishing the relationship between fluoride and reduced IQ, including studies

8

observing this relationship at “lower” exposure levels).
71. Dr. Grandjean rests his BMCL analysis upon studies observing the ELEMENT, MIREC,

United States District Court
Northern District of California

9
10

and OCC cohorts. Grandjean (2023) at 1-2. These high-quality studies are appropriate for use in

11

BMD modeling, particularly because they include data regarding dose-response at “lower”

12

exposure levels, i.e., 0.9 mg/L (mean maternal urinary fluoride in ELEMENT cohort), 0.42 mg/L

13

(mean maternal urinary fluoride in MIREC cohort), and 0.58 mg/L (average maternal urinary

14

fluoride in the OCC cohort). See ¶¶ 42-44. Thus, rather than observing only a response at high

15

dosages, the data set utilized by Dr. Grandjean observes dose-response at low exposure levels.

16

The data set are thus appropriate for BMD modeling. To this end, RSI found that the MIREC and

17

ELEMENT cohorts represent a “high quality of evidence partly based on Canadian population,

18

conducted within a context relevant to Canadian drinking water fluoride exposure levels.[20] Both

19

studies included prospective data collection, with prenatal exposure assessment (maternal urine

20

collection over successive trimesters) and follow-up during the early life of the infants and

21

children.” Dkt. No. 433-4, Trial Ex. 129 at 23. And the ELEMENT and MIREC cohort studies

22
23
24
25
26
27
28

Now, moving beyond semantics, I wanted to ask you about your testimony about benchmark dose,
okay? You made comments in your testimony about Dr. Grand[j]ean's BMCL analysis, correct?
A. Yes, I did. Q. You based your comments on EPA’s BMD guidance technical
manual, correct? A. Yes, I did.”). The Court thus considers this technical guidance document.
20

The United States and Canada take a similar approach to water fluoridation; this finding is
applicable to United States drinking water fluoride exposure levels. See Tr. Ex. 129, Dkt. No.
433-4 at 16 (describing optimal water fluoride levels in Canada of 0.7 mg/L). See also Dkt. No.
396, Feb. 1 , 2024 Trial Tr. at 245:1-22 (Lanphear) (describing optimal 0.7 mg/L water fluoride
standard in Canada).
44

1

are strong for their extensive control for covariates and individualized measurements of fluoride

2

exposure during the prenatal period. Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 95:2-96:5 (Hu).
72. The model fits of the data utilized by Grandjean’s BMCL are also adequately supported.

United States District Court
Northern District of California

3
4

On this point, the EPA takes issue with the fact that Dr. Grandjean’s BMCL of 0.28 mg/L was

5

derived by applying a linear model of the dose-response curve.21 Grandjean (2023) at 1-2, 9. To

6

discern the best model fit for a set of data, a model is used to find a fit to the data, and based upon

7

that fit, an “AIC” score is generated; the lower the AIC score, the better the model fit. Dkt. No.

8

417, Feb. 2, 2024, Trial Tr. at 421:20-21 (Grandjean). To EPA’s point, Grandjean (2023) did not

9

include a published table illustrating the AIC scores for all model fits, but did so only for the linear

10

model and piece-wise model, though not the squared model. See Grandjean (2023) at 9 (Table

11

S3). The government thus argued at trial that Dr. Grandjean improperly assumed, without testing

12

the assumption, that the linear model was appropriate for the data set evaluated. However, the use

13

of the linear model in Grandjean (2023) to generate the BMCL is sufficiently justified:
i.

14

Dr. Grandjean testified, and the Court finds this testimony credible,

15

that he did not assume that the linear model was the best fit, but rather that he and his co-authors

16

compared various models and determined that the linear model was the preferred model for the

17

data. Dkt. No. 396, Feb. 1, 2024, Trial Tr. at 333:6-19 (Grandjean). Dr. Grandjean did state that

18

“[i]n my communications with the EPA, I was told that the default curve function was the linear

19

one.” Id. at 333:8-9. However, Dr. Grandjean clarified that this default was only a starting point

20
21
22
23
24
25
26
27
28

21

When a curve is linear, generally the dose and effect increase or decrease in a somewhat
uniform fashion, i.e., when the dose increases, the effect increases; when the dose decreases, the
effect decreases. See EPA’s Benchmark Dose Technical Guidance at 25-26, 77-78 (describing
linear, quadratic, and other models), 71 (defining “Linear Dose-Response Model” as “[a]
mathematical relationship in which a change in response is proportional to a fixed amount of
change in dose, e.g., Response = a + b × Dose. This is in distinction from a more general linear
mathematical model, which is a linear combination of parameters”). The shape of the doseresponse curve is relevant, particularly because it is used to extrapolate to lower levels of exposure
not observed in the study, and thus to calculate the BMCL. See id. at 5 (“The dose response
assessment under the guidelines is a two-step process: (1) response data are modeled in the range
of empirical observation — modeling in the observed range is done with biologically based or
curve-fitting models; and then (2) extrapolation below the range of observation is accomplished by
modeling, if there are sufficient data, or by a default procedure (linear, nonlinear, or both).”). The
model will thus determine the BMCL identified. See id. at 5, 25-26, 77-78.
45

United States District Court
Northern District of California

1

and that “what we’ve done in our work is to compare that to some variations and the statistical

2

methods so that you can actually compare the fit if, let’s say, curvilinear or a broken line fits

3

better. And in our case the linear was actually – was the best fit.” Id. at 333:10-14. And further,

4

Dr. Grandjean testified that he also used “nonlinear methods to assess whether the dose-response

5

relationship is linear,” id. at 333:15-19. See also id. at 339:24-340:7 (“We started out with EPA’s

6

default recommendation, namely that linear association. But we then also looked at a curvilinear,

7

for example, log 2 transformation of exposure. We also looked at broken lines of – and overall the

8

linear association was not inferior to anything. It was sometimes clearly superior.”); Dkt. No. 417,

9

Feb. 2, 2024, Trial Tr. at 440:23-419:1 (Grandjean) (“[W]e certainly did look at other models.”).

10

Dr. Grandjean and his co-authors did not simply assume that the linear model was the best fit for

11

the data. It was chosen through an analytical process.
ii.

12

Moreover, Grandjean (2022) includes a table that reports the AIC

13

scores for squared models as they fit to data from the MIREC and ELEMENT cohorts and reveals

14

comparable fit scores and supports Dr. Grandjean’s testimony as to the validity of the linear model

15

fit:

16
17
18
19
20
21
22
23
24
25
26
27
28

Grandjean (2022) at 17 (Table 2) (red annotation added). The AIC scores for the linear and
squared models were comparable, with the best fit for boys and girls individually, measuring IQ,
using a linear model (AIC 4766.7 linear compared to 4769.4 squared), and squared combined
(AIC 4768.8 squared compared to 4770.1 linear). See id. For GCI (the General Cognitive Index
46

1

score), the linear model was a better fit than the squared model for all categories. See id. Even if

2

not definitive, the comparable AIC fits for linear and squared models reflected in Grandjean

3

(2022) support that the linear model is a justifiable model to apply to the MIREC and ELEMENT

4

cohort data.
iii.

5

Dr. Grandjean’s analysis is also consistent with the NTP’s analysis.

6

The NTP Meta-analysis did not publish AIC scores for models restricted to low-risk-of bias

7

studies. See Dkt. No. 431-2, Trial Ex. 68 at 40-41 (eTable 4) (hereinafter “NTP Meta-analysis”).

8

However, it did publish AIC scores for model fit of data in all studies, as reflected in the below

9

table:

10

United States District Court
Northern District of California

11
12
13
14
15
16
17
18
19

Id. Using urinary fluoride as the exposure metric, the linear model reflected the lowest AIC score

20

unilaterally. See id. And although the linear model did not generate a statistically significant

21

inverse association at all exposure levels, the linear model generated a statistically significant

22

inverse association at <1.5 mg/L (in line with Grandjean (2023)’s finding relating to lower-

23

exposure levels as noted above), and the findings remained directionally negative at all levels

24

which also supports Grandjean (2023)’s use of the linear model. See Dkt. No. 395, Jan. 31, 2024,

25

Trial Tr. at 115:16-25 (Hu) (“In fact, epidemiology is moving away from a simple reliance on just

26

P values and saying ‘this is significant, this is not significant.’ It’s really important to also look at

27

the so-called directionality of the relationships.”). Additionally, as explained in more detail

28

below, some of the loss of association observed in the NTP Meta-analysis may be explained by the
47

United States District Court
Northern District of California

1

use of the means effect method in the Meta-analysis, which results in loss of statistical power and

2

sensitivity in the data. See ¶ 74(b). Ultimately, the authors of the NTP Meta-analysis concluded

3

that “[b]ased on the AIC and likelihood ratio tests, the best model fit was achieved when quadratic

4

or restricted cubic spline exposure levels were added to the linear models for drinking water

5

(eFigure 17); the linear model was the best fit for urinary fluoride (eFigure 18).” NTP Meta-

6

analysis at 10 (emphasis added). This further bolsters the legitimacy of Grandjean (2023)’s use of

7

a linear model to generate the BMCL, expressed in maternal urinary fluoride.

8

73. Assuming, in the alternative, that the squared model is a more appropriate fit for this data

9

set, as EPA suggested at trial, a BMCL of 0.768 mg/L and/or 1.536 mg/L is appropriately used to

10

conduct the risk assessment. Though Grandjean (2023) did not identify a BMCL using the squared

11

model, Dr. Grandjean’s 2022 BMCL analysis did identify a BMCL of 0.768 mg/L utilizing a

12

squared model. Grandjean (2022) at 17 (Table 2); Dkt. No. 417, Feb. 2, 2024, Trial Tr. at

13

423:12-21 (Grandjean). It is true that this BMCL is derived from the ELEMENT and MIREC

14

cohort data only and excludes data from the OCC Cohort. This is relevant because inclusion of

15

the OCC Cohort data is likely to increase the BMCL; when the OCC cohort data was added to the

16

BMCL analysis in Grandjean (2023), the BMCL increased by 0.08 mg/L, or forty percent22 (from

17

0.20 mg/L (MIREC and ELEMENT alone) to 0.28 mg/L (MIREC, ELEMENT and OCC cohort

18

data)). See Grandjean (2023) at 3 (“The joint BMC was found to be 0.45 mg/l (BMCL, 0.28 mg/l),

19

i.e. slightly higher than previously found (BMC, 0.33 mg/l; BMCL, 0.20 mg/L) for the two North

20

American cohorts alone.”). But a preponderance of the evidence indicates the inclusion of the

21

OCC Cohort data would not make a material difference. To be highly conservative, the BMCL of

22

0.768 mg/L can be doubled, to account for any discrepancy caused by the omission of the OCC

23

data: 1.536 mg/L (0.768 mg/L times two). This could be used conservatively as an alternative

24

point of departure implied from the data if the squared model is used. As discussed below, even

25

using this higher point of departure, the ultimate finding of an unreasonable risk would not change.

26
27
28

22

((.08 / .20) * 100).
48

74. One additional concern with Dr. Grandjean’s BMCL calculation is that it, at first glance,

1
2

appears to be in tension with the NTP Monograph’s conclusion that “[m]ore studies are needed to

3

fully understand the potential for lower fluoride exposure [i.e., below 1.5 mg/L] to affect

4

children’s IQ.” NTP Monograph at xiii.23 However, this ultimately does not undermine the

5

validity of the BMCL identified in Grandjean (2023) for the following reasons:
a. Though the authors of the NTP Monograph recognized some lack of clarity in the

United States District Court
Northern District of California

6
7

precise relationship between fluoride and reduced IQ at lower exposure levels, NTP Monograph at

8

xiii, given the strength of the association between fluoride and reduced IQ, the authors of the NTP

9

Monograph refused to limit the applicability of its findings in the systematic review to higher

10

exposure levels and made clear that its confidence assessment also considered fluoride exposures

11

“that are similar to, or lower than, those associated with optimally fluoridated water supplies in the

12

United States,” i.e., 0.7 mg/L. Dkt. No. 438-1, Trial Ex. 69 at 24-25 (comments and responses

13

from NTP Monograph authors and evaluators of the NTP Monograph).

14

b. The NTP also conducted a Meta-analysis, integrating all of the studies assessed in

15

the NTP Monograph to analyze the dose-response relationship between fluoride and reduced IQ.

16

The findings of the NTP Meta-analysis first appear to be in tension with Dr. Grandjean’s findings

17

but are, in fact, consistent with those findings because of the methodologies used. Namely, the

18

NTP Meta-analysis concluded that “the consistency of the data supports an inverse association

19

between fluoride exposure and children’s IQ.” NTP Meta-analysis at 3. However, the Meta-

20

analysis reported somewhat mixed results regarding the dose-response relationship, particularly at

21
22
23
24
25
26
27
28

23

Regarding “lower” fluoride exposure levels – both Grandjean (2023) and the NTP Monograph
analyzed data from the ELEMENT and MIREC cohorts though Grandjean (2023) also analyzed
data from the OCC Cohort, another lower-exposure level study.
49

1

lower levels of fluoride exposure:

2
3
4
5
6
7
8
9
10

United States District Court
Northern District of California

11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

Id. at 41 (eTable 4) (red annotation added). In reviewing all studies and measuring exposure of
fluoride per urinary fluoride the NTP Meta-analysis found a statistically significant inverse
association between children’s urinary fluoride exposure and IQ at <4 mg/L urinary fluoride. Id.
When restricted to <2 mg/L and <1.5 mg/L urinary fluoride, there was still an inverse association.
Id. This finding is consistent with Grandjean (2023). However, when analyses were restricted to
low risk-of-bias publications, the associations at <2 mg/L and <1.5 mg/L became smaller in
magnitude and were only statistically significant at <1.5 mg/L, but not at <2 mg/L. Id. That
finding of an adverse association at <1.5 mg/L is consistent with the conclusion in Dr. Grandjean’s
pooled benchmark dose analysis (though appearing somewhat anomalous compared to the finding
at <2 mg/L). Dr. Grandjean’s pooled benchmark analysis uses a method with more statistical
precision than the NTP Meta-analysis, and thus could account for the more specific findings as to
the relationship between fluoride and IQ at lower exposure levels. Specifically, the NTP Metaanalysis used a “means effect analysis,” which is useful for its ability to compare different types of
studies with varied methodologies and metrics (72 total and 19 low-risk-of-bias studies) – but it
loses sophistication and precision in the underlying data of each study when it converts the
findings into standard, comparable metrics. Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 469:3-471:6
(Grandjean). Specifically, so that different studies using different exposures or result metrics
could be compared, the data was grouped into buckets (e.g., high exposure, low exposure) and
analyzed. Id. at 471:6-15. Accordingly, each of the underlying studies lose some of its statistical
power when data is simplified to allow for cross-study, like-to-like comparison. See id. at 471:6473:24. On the other hand, the pooled benchmark analysis maintains individualized, continuous
50

1

data and does not simplify that data for meta-analysis comparison; the benchmark analysis

2

maintains increased sophistication and statistical sensitivity. Id. at 473:18-24. Thus, the findings

3

of the NTP Meta-analysis are not inconsistent with Dr. Grandjean’s pooled benchmark analysis.
75. Ultimately, TSCA does not require complete certainty as to the threshold level at which a

United States District Court
Northern District of California

4
5

chemical produces the hazard; indeed, such certainty is very difficult to obtain from epidemiologic

6

studies of human populations. Dkt. No. 440, Feb. 13, 2024, Trial Tr. at 1440:18-23 (Barone); Dkt.

7

No. 414, Feb. 9, 2024, Trial Tr. at 1173:7-13 (Savitz). Either BMCL of 0.28 mg/L (linear model

8

per the MIREC, ELEMENT, and OCC cohort data) or 0.768 mg/L (squared model per the MIREC

9

and ELEMENT cohort data) identified by Dr. Grandjean and his co-authors are legitimate points

10

of departure to utilize in a risk analysis. So is the implied BMCL of 1.536 mg/L (were the OCC

11

study taken into account). The Court finds, though not with absolute certainty, Dr. Grandjean’s

12

BMCLs are supported by a preponderance of the evidence.24
(b)

13

POD: 4 mg/L urinary or water fluoride (LOAEL)

76. As described previously, use of the BMD approach is preferred in identifying a point of

14
15

departure because of limitations of a NOAEL or LOAEL, but where data is not amenable to

16

benchmark dose modeling, a NOAEL or LOAEL may be utilized instead. See ¶¶ 57-59. The

17

Court thus examines this alternative approached to establishing a point of departure.
77. Again, notwithstanding the limitations of the NOAEL/LOAEL approach, this approach is

18
19

properly used, and has been used by the EPA, with the application of uncertainty factors, to

20

determine the point of departure where datasets are, for various reasons, not amenable to BMD

21

modeling. See Dkt. No. 429-20, Trial Ex. 38 at 4. For example, the EPA conducted a risk

22

evaluation of Perchloroethylene (“PCE”), pursuant to Amended TSCA, and utilized

23

NOAEL/LOAELs as PODs because it was unable to use BMD modeling. See PCE Risk

24

Evaluation at 351 (“For this risk evaluation, non-cancer PODs were all based on NOAELs and

25

LOAELs because the data for the selected endpoints was unable to be BMD modeled. This results

26
27
28

24

The government also takes issue with the use of maternal urinary fluoride (“MUF”) as the
metric of the exposure or hazard level utilized in the risk assessment analysis. The validity of
maternal urinary fluoride as a metric is taken up subsequently in Section III.B (Exposure
Assessment).
51

1

in reduced precision in POD estimates because the POD is dependent on the dose selection of the

2

study as opposed to the response rate/level for the effect of interest.”); Dkt. No. 401, Feb. 6, 2024,

3

Trial Tr. at 772:3-11 (Thiessen).

United States District Court
Northern District of California

4

78. To the extent that the BMD approach is not appropriate based upon the present data set, in

5

the alternative, 4.0 mg/L (using exposure measurement of water fluoride intake) is a legitimate and

6

highly conservative LOAEL to utilize as a point of departure to conduct a risk assessment of

7

fluoride per the findings of the NTP Meta-analysis. Utilizing 4.0 mg/L as the LOAEL is

8

especially conservative in view of the NTP Monograph’s conclusion with moderate confidence

9

that exposure to fluoride concentration in drinking water at or above 1.5 mg/L is associated with

10

lower IQ in children. One could reasonably take 1.5 mg/L as a LOAEL. Nonetheless, the Court

11

uses the more conservative 4.0 mg/L based on a close analysis of the NTP Meta-analysis which

12

establishes with consistency an association with reduced IQ at that level. Specifically, the NTP

13

Meta-analysis observed a statistically significant inverse association between fluoride and reduced

14

IQ at 4 mg/L measured in water fluoride, based on low-risk-of-bias/high quality studies (i.e., 6

15

epidemiological studies deemed high quality), which is reflected in the below table from the Meta-

16
17
18
19
20
21
22
23
24
25
26
27
28
52

1

analysis summarizing the NTP’s dose-response analysis:25

2
3
4
5
6
7
8
9
10

United States District Court
Northern District of California

11
12
13
14

Dkt. No. 431-2, Trial Ex. 68 at 39 (eTable4) (red annotation added). That value was derived from

15

a linear model which, for this group of studies, had the lowest AIC score. See id. (identifying AIC

16

of 16.1 (linear for all studies), 21.1 (quadratic for all studies), 16.9 (restricted cubic splines for all

17

studies)).

18
19
20
21
22
23
24
25
26
27
28

25

Note that where values in the parenthesis, which represent the confidence interval, are below
zero, the finding is statistically significant. See Dkt. No. 417, Feb. 2, 2024, Trial Tr. at 394:2-14
(Grandjean).
53

1

79. Further, the NTP Meta-analysis observed an association between fluoride and

2

reduced IQ at <4 mg/L measured in urinary fluoride, based on low-risk-of-bias/high-quality

3

studies (9 epidemiological studies deemed high quality):

4
5
6
7
8
9
10

United States District Court
Northern District of California

11
12
13
14
15
16
17
18
19
20
21

Dkt. No. 431-2, Trial Ex. 68 at 39 (eTable 4) (red annotation added). That value was also derived

22

from a linear model which, for this group of studies, likewise had the lowest AIC score. See id.

23

(identifying 68 (linear for all studies), 75.8 (quadratic for all studies), 73.3 (restricted cubic splines

24

for all studies)).

25

80. Even if there may be some uncertainty about the dose-response relationship below that

26

exposure level (4 mg/L), significant data supports that there is an adverse effect at or above the

27

specified level. See Dkt. No. 415, Feb. 12, 2024, Trial Tr. at 1373:1-9 (Barone) (testimony from

28
54

United States District Court
Northern District of California

1

Dr. Barone agreeing that at 4 mg/L of fluoride exposure and above there is relatively more data to

2

support a finding of an adverse effect associated with fluoride.), 1428:4-11 (Barone) (“I agree with

3

the NTP’s conclusions that at some level above 1.5 that there is moderate evidence to support an

4

association between fluoride and developmental IQ decrements.”). Again, TSCA does not require

5

absolute certainty as to the threshold level at which a chemical produces the hazard, and indeed as

6

noted above such certainty is very difficult to obtain from epidemiologic studies of human

7

populations. Dkt. No. 440, Feb. 13, 2024, Trial Tr. at 1440:18-23 (Barone); Dkt. No. 414, Feb. 9,

8

2024, Trial Tr. at 1173:7-13 (Savitz). In view of the record evidence, 4 mg/L as the lowest-

9

observed-effect-level would be a conservative point of departure to utilize in the analysis; it is

10

certainly well-supported by scientific evidence as described in the conclusion of the NTP

11

Monograph: “the high-quality studies (i.e., studies with low potential for bias) consistently

12

demonstrate lower IQ scores with higher fluoride exposure [e.g., represented by populations

13

whose total fluoride exposure approximates or exceeds the WHO Guidelines for Drinking-water

14

Quality of 1.5 mg/L of fluoride (WHO 2017)].” NTP Monograph at 47.
81. The EPA has identified a LOAEL based upon far less evidence than that in the record

15
16

before this Court. In the EPA’s risk evaluation of Methylene, conducted pursuant to Amended

17

TSCA, it used a LOAEL for developmental neurotoxicity, derived from the analysis of one study

18

conducted upon mouse pups (Fredriksson et al., 1992). See Methylene Risk Evaluation at 262.

19

Here, there are between six and nine26 high-quality, epidemiological studies of human populations

20

underlying the point of departure. Dkt. No. 431-2, Trial Ex. 68 at 39, 41 (eTable 4).
82. To restate, in conclusion, either the LOAEL of 4.0 mg/L, measured either in urinary

21
22

fluoride or water fluoride, or the BMCL of 0.28 mg/L, 0.768 mg/L, or even 1.536 mg/L measured

23

in maternal urinary fluoride, is a well-supported point of departure to utilize in the risk evaluation.

24

Each of these measures of the point of departure is supported by a preponderance of high-quality

25

evidence.

26
27
28

26

Six studies measuring fluoride exposure by way of water fluoride and nine studies measuring
urinary fluoride. Dkt. No. 431-2, Trial Ex. 68 at 39, 41 (eTable 4).
55

1

B.

Step 2: Exposure Assessment
a.

2

United States District Court
Northern District of California

3

Framework

83. At this step, the EPA conducts an exposure assessment to identify the exposure level under

4

the conditions of use for the chemical at issue. Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 567:18-

5

568:2 (Barone); 15 U.S.C. § 2605(b)(4)(F)(iv) (“In conducting a risk evaluation under this

6

subsection, the Administrator shall . . . take into account, where relevant, the likely duration,

7

intensity, frequency, and number of exposures under the conditions of use of the chemical

8

substance.”). Namely, the EPA identifies sources of exposure to the chemical (e.g., food or

9

water), estimates what the intake level of exposure is, and endeavors to understand and

10

characterize the population that is exposed. Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 694:4-695:11

11

(Barone).

12

84. To understand the level of exposure, the EPA estimates a range of exposure levels for a

13

condition of use from the central tendency exposure (e.g., 50th percentile) to high-end exposure

14

(e.g., 95th percentile). Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 649:1-650:10 (Barone), 697:15-

15

698:6 (Barone); see also EPA Guidelines at 64 (describing consideration of upper percentile

16

exposure and highest-exposed individuals in risk assessment).

17

85. As discussed in depth in the next section (Section III.C), the exposure level is important

18

because it is used to calculate whether the chemical presents a risk to humans. Specifically, in the

19

next step of the analysis (risk characterization), the exposure level is compared to the point of

20

departure to determine if a risk is present. See Dkt. No. 401, Feb. 6, 2024, Trial Tr. (Barone) at

21

705:7-706:21. At that step, the EPA determines the appropriate margin that needs to exist from

22

the point of departure (i.e., point at which the chemical becomes hazardous). See id. This is the

23

benchmark Margin of Exposure (“MOE”). See id. The benchmark MOE is calculated by

24

multiplying the point of departure by Uncertainty Factors (“UFs”) to account for assumptions or

25

uncertainty in the data. See id. The benchmark MOE is then compared to the actual MOE, i.e.,

26

the existing margin between the exposure level and the point of departure, to determine if that

27

margin is sufficient. See id.

28
56

b.

1

86. For reasons discussed below, maternal urinary fluoride is an appropriate metric to use in

2
3

conducting the risk evaluation of fluoride under the condition of use, i.e., community water

4

fluoridation at 0.7 mg/L.
87. Pregnant mothers in fluoridated communities in the United States have a median exposure

5
6

level to fluoride of 0.8 mg/L, measured in maternal urinary fluoride; at the 95th percentile,27

7

pregnant mothers have an exposure level to fluoride of 1.89 mg/L, measured in maternal urinary

8

fluoride. Approximately half of these maternal urinary fluoride levels is attributed to community

9

water fluoridation.
88. Alternatively, the exposure levels of 0.7 mg/L, or 0.56 mg/L measured in water fluoride,

10
11
United States District Court
Northern District of California

Key findings

is an appropriate exposure level to use in this risk evaluation.
c.

12

89. Two studies are highly probative in assessing exposure levels in this risk evaluation: Till

13
14

Underlying findings

(2018), and Malin (2023). To summarize these studies:
a. Till (2018) studied samples collected from the MIREC Cohort (1,566 pregnant

15
16

women in Canada) to assess the relationship between maternal urinary fluoride in pregnant women

17

and water fluoride concentrations and concluded that “[c]ommunity water fluoridation is a major

18

source of fluoride exposure” for the pregnant women studied. Dkt. No. 432-4, Trial Ex. 108 at 1.

19

Specifically, the study observed that the mean urinary fluoride values were almost two times

20

higher for pregnant women living in fluoridated regions compared to non-fluoridated regions, and

21

“significantly lower” for women living in non-fluoridated regions. Id. at 6. The median

22

concentration of fluoride in drinking water in Canada was 0.56 mg/L in fluoridated areas. Id. at 8

23

(Table 2). Given that the United States fluoridates its water levels at an optimal 0.7 mg/L (higher

24

than the median in Till (2018)), the urinary fluoride levels in this sample are lower, if anything,

25

relative to the condition of use at issue (fluoridation at 0.7 mg/L). The findings of Till (2018),

26
27
28

27

The 95th percentile reflects individuals that have exposure levels greater than 95 percent of the
population. See Dkt. No. 108 at 6. The median, on the other hand, reflects individuals at the midpoint of exposure. See id.
57

1

comparing the maternal urinary fluoride levels of pregnant women in fluoridated compared to

2

non-fluoridated reasons are exemplified in the below tables, summarizing the key results of this

3

study:

4
5
6
7
8
9
10

United States District Court
Northern District of California

11
12
13
14
15
16

Id. at 25 (Table S4) (red annotations added). This data is reflected in the below bar graph,

17

illustrating that Till (2018) found that fluoride levels were approximately two times higher in

18

fluoridated vs. non-fluoridated areas:28

19
20
21
22
23
24
25
26
27
28

28

Though not in evidence, the Court includes this demonstrative bar graph (presented to the Court
58

1
2

Los Angeles, California (i.e., samples collected from the Maternal and Developmental Risks from

3

Environmental and Social Stressors cohort (“MADRES Cohort”)) to discern if those levels of

4

American women were comparable to levels observed amongst pregnant women in Mexico and

5

fluoridated communities in Canada. Dkt. No. 432-18, Trial Ex. 122 at 9. Malin (2023) concluded

6

that the maternal urinary levels observed in Los Angeles were comparable to those found in

7

pregnant women in Mexico and Canada. Id. at 1, 9. These findings corroborate the conclusions of

8

Till (2018), and further support that water intake is an important contributor to maternal urinary

9

fluoride levels.

10

United States District Court
Northern District of California

b. Malin (2023) studied the maternal urinary fluoride levels of pregnant women in

90. Plaintiffs have shown, by a preponderance of the evidence, that a pregnant mother in the

11

United States, under the condition of use (community water fluoridation of 0.7 mg/L, which is

12

higher than the median water fluoridation levels in the Till (2018) data set of 0.56 mg/L found in

13

Canada) produces a maternal urinary fluoride concentration level of at least 0.8 mg/L for median

14

water consumption or 1.89 mg/L for 95th percentile water consumption.

15

a. As explained above, Till (2018) studied urinary fluoride levels in fluoridated areas

16

of Canada, and identified a median (specific gravity adjusted) urinary fluoride level of 0.77

17

mg/L and a 95th percentile urinary fluoride level of 1.89 mg/L. Dkt. No. 432-4, Trial Ex. 108 at

18

25-26 (Table S4); Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 118:5-20 (Hu). Malin (2023) studied

19

pregnant mothers living in Los Angeles, California, a fluoridated city, and similarly observed that

20

those mothers had a median (specific gravity-adjusted) urinary fluoride level of 0.8 mg/L,

21

and a 95th percentile level of 1.89 mg/L, in the third trimester. Dkt. No. 432-18, Trial Ex. 122

22

at 5 (Table 2); Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 124:1-16 (Hu). Dr. Hu testified credibly

23

that the Malin (2023) cohort is representative of mothers in the United States as a whole, though if

24

anything, this cohort would present lower fluoride exposure levels relative to other populations

25

because data indicates Hispanic communities have a greater distrust of tap water relative to other

26

communities, in part due to immigration from Mexico where tap water is distrusted. Dkt. No. 395,

27
28

as Plaintiff’s Demonstrative No. 4 at trial) to illustrate fully the trial testimony.
59

1

Jan. 31, 2024, Trial Tr. at 118:11-119:9 (Hu). Canada and the United States each take a similar

2

approach to water fluoridation; both countries identify 0.7 mg/L as the optimal fluoridation level.

3

See NTP Monograph at 1; Dkt. No. 396, Feb. 1, 2024, Trial Tr. at 245:1-22 (Lanphear). It follows

4

that pregnant woman in the United States, exposed to fluoride under the condition of use at issue

5

(community water fluoridation at a typical or optimal level of 0.7 mg/L) have an exposure level of

6

0.8 mg/L measured in maternal urinary fluoride (median water intake) and 1.89 mg/L

7

measured in maternal urinary fluoride (95th percentile water intake), urinary fluoride levels

8

that reflect the real world results of drinking water fluoride levels at the condition of use at issue in

9

this case.

United States District Court
Northern District of California

10

b. To be sure, maternal urinary fluoride reflects not only fluoride that a pregnant

11

woman is exposed to from drinking fluoridated water from her community (the condition of use at

12

issue), but also fluoride from other sources such as food and beverage and household items such as

13

toothpaste; it reflects aggregate exposure to fluoride. See Dkt. No. 395, Jan 31, 2024, Trial Tr. at

14

105:10-25 (Hu); Dkt. No. 416, Feb. 12, 2024, Trial Tr. at 1404:19-21 (Barone); Dkt. No. 198-1

15

(Hu Trial Decl.). The EPA argues that because maternal urinary fluoride reflects aggregate

16

fluoride exposure, rather than exposure attributed solely from community water fluoridation,

17

maternal urinary fluoride is an inappropriate metric to use in assessing the risk of community

18

water fluoridation. However, exposure level of fluoride expressed in the metric of maternal

19

urinary fluoride is properly used in this risk assessment because:

20

i. Maternal urinary fluoride, though not a perfect metric in all respects,

21

is a valuable metric in assessing risk associated with water fluoridation since it is a comprehensive

22

metric, reflecting the true aggregate exposure to the chemical at issue. As Dr. Hu explained:

23

“[T]he primary benefit [of using urinary fluoride as the metric of fluoride exposure] is that you’re

24

integrating fluoride exposure from whatever exposure source there is. So if it’s dietary, if it’s in

25

the water, it’s in the food, it’s in the food that was cooked with the fluoridated water; if you

26

happen to swallow toothpaste or if you’re using other sources of fluoride, it will integrate all of it

27

and express it in terms of what is the level of fluoride that’s circulating in your blood and then gets

28

filtered out into the kidneys. And that ultimately is the component of fluoride in the body that’s
60

United States District Court
Northern District of California

1

available to cross the blood-brain barrier to the brain and also to go to other target organs in the

2

body.” Dkt. No. 395, Jan. 31, 2024, Trial Tr. at 105:13-25 (Hu). Put differently, this metric

3

reflects that water fluoridation does not occur in a vacuum; in the real world, fluoridating water

4

means exposing women to fluoride in addition to the exposure a woman has to fluoride via other

5

sources. Because dosage matters, it makes good sense to consider other sources of exposure to

6

fluoride in deciding if adding to that exposure level presents a risk. See Dkt No. 400, Feb. 5,

7

2024, Trial Tr. at 676:12-21 (Barone) (recognizing that exposure and point of departure can be

8

expressed in urine content in a risk assessment); Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 1015:9-

9

1020:13 (Savitz) (discussing pros and cons of using urinary fluoride as a measurement of water

10

fluoridation and recognizing that urinary fluoride has a “number of positive features,” including

11

integrating exposure from different sources, that it is a measurement reflecting not just what is in

12

that body on a given day but for a longer period of time, and explaining that he has used urinary

13

fluoride as a metric in assessing another chemical, PFAS); Dkt. No. 401, Feb. 6, 2024, Trial Tr. at

14

790:8-12 (Thiessen) (“there’s no scientific reason why [the exposure level and hazard level] have

15

to be milligrams per kilogram per day. They could also be milligrams per liter in the drinking

16

water, they could also be milligrams per liter in the urine”) (emphasis added).

17

ii. The EPA permits considering the additive risk posed by a chemical

18

under the condition of use at issue when conducting a risk evaluation. To this end, Dr. Barone

19

explained that in a situation where the condition of use is additive to other background sources,

20

“you want to be able to understand, well, what’s the background, be able to subtract the

21

background; you want to be able to say what’s the dietary component and what is the actual water

22

intake component. And then if you have information on the other sources, potential sources,

23

whether it’s pharmaceutics or other inhaled or overly ingested pollutants having a similar kind of

24

exposure, additive exposures, you want to be able to capture that to the best of your ability.” Dkt.

25

No. 400, Feb. 5, 2024, Trial Tr. at 678:6-21 (Barone) (emphasis added). See also Dkt. No. 400,

26

Feb. 5, 2024, Trial Tr. (Barone) at 567:18-568:2 (“Q. And the point of the exposure assessment is

27

to identify what the human exposure level is under the specific conditions of use of the chemical

28

being evaluated, right? A. It is – it is condition-of-use specific. Q. Now, it is condition-of-use
61

1

specific, but TSCA specifically permits EPA to consider aggregate exposures to the chemical,

2

correct? A. TSCA specifically allows for consideration of aggregate exposures. It doesn’t require

3

us to quantify based upon aggregate exposures”) (emphasis added). Indeed, rather than preventing

4

a risk evaluator from considering aggregate exposure to a chemical in evaluating risk, Amended

5

TSCA expressly identifies that a risk evaluator should describe whether aggregate exposure was

6

considered and explain why, or why not. See 15 U.S.C. § 2605(b)(4)(F). Specifically, the statute

7

provides: “[i]n conducting a risk evaluation under this subsection, the Administrator shall . . .

8

describe whether aggregate or sentinel exposures to a chemical substance under the conditions of

9

use were considered, and the basis for that consideration.” 15 U.S.C. § 2605(b)(4)(F) (emphasis

10

United States District Court
Northern District of California

11

added).
iii. If water fluoridation was a minor contributor to overall exposure to

12

fluoride, then it may be less appropriate to utilize an aggregate exposure metric in assessing risk of

13

water fluoridation. If that were the case, much of the risk at issue would not derive from water

14

fluoridation but another source; regulating water fluoridation would be of little consequence to the

15

total exposure. But that is not the case. Instead, as described in depth below at ¶ 91(a), water

16

fluoridation accounts for more than half of a pregnant woman’s aggregate exposure level (i.e.,

17

maternal urinary fluoride level). To this end, Dr. Thiessen credibly testified that fluoride content

18

of the urine “will be driven by the fluoride content of the water,” as “for most individuals, the

19

intake is driven by the fluoridated water.” Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 934:18-22

20

(Thiessen). Drinking water fluoridation is highly consequential to a pregnant woman’s overall

21

exposure level and so it is wholly appropriate to use maternal urinary fluoride as the metric of

22

exposure in assessing the risk of community water fluoridation. See also Dkt. No. 401, Feb. 6,

23

2024, Trial Tr. at 790:8-12 (Thiessen) (“[T]here’s a consistent association between urinary

24

fluoride and drinking water fluoride concentrations. As the concentration of fluoride in the

25

drinking water increases, the fluoride concentration in the urine will increase.”), 792:19-2793:16

26

(“[I]n most cases, the primary driver of the total fluoride intake [is fluoride concentration in the

27
28
62

1

drinking water]. So you can still make that hazard-to-exposure comparison.”).29

2

91. To the extent that risk assessment requires determining the exposure level attributed solely

3

to the condition of use (community water fluoridation), Plaintiffs have shown, by a preponderance

4

of the evidence, that at least half of the maternal urinary fluoride levels observed, 0.4 mg/L

5

(median) (i.e., 0.8 mg/L divided by two) maternal urinary fluoride and 0.945 mg/L (95th

6

percentile) (i.e., 1.89 mg/L divided by two) maternal urinary fluoride can be attributed to the

7

condition of use (community water fluoridation):
a. As explained above, ¶ 89(a), Till (2018) observed that the maternal urinary fluoride

8

United States District Court
Northern District of California

9

levels were approximately two-times higher for pregnant women living in fluoridated regions

10

compared to non-fluoridated regions. Dkt. No. 432-4, Trial Ex. 108 at 6, 25-26 (Table S4). Dr.

11

Thiessen credibly testified that it is reasonable to conclude from Till (2018) that the 2x increase in

12

maternal urinary fluoride levels in fluoridated areas can be attributed to community water

13

fluoridation in those areas. See Dkt. 401, Feb. 6, 2024, Trial Tr. at 784:1-16 (Thiessen) (“The

14

primary difference and the only main group difference that we’re aware of is that one group is

15

fluoridated and one is not. So a difference in the urinary fluoride would be attributable to the

16

fluoride in the drinking water.”); Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 934:18-22 (Thiessen).

17

And the EPA’s expert witness agreed that the increase in maternal urinary fluoride levels can

18

largely be attributed to intake of fluoridated water. Dkt. No. 416, Feb. 13, 2024, Trial Tr. at

19

1408:10-1409:11 (Barone) (explaining that the “parsimonious” explanation as to the 2x increase

20

of maternal urinary fluoride levels observed in Till (2018) is that it is “due to intake, total intake,

21
22
23
24
25
26
27
28

29

In Thippeswamy (2021), the researchers compared fluoride concentrations in urine, serum, and
cord blood of women consuming water with designated “low” and “optimum” concentrations of
fluoride to understand the relationship of these metrics. Dkt. No. 432-7, Trial Ex. 111 at 1.
Thippeswamy (2021) did not observe a one-to-one correlation between urinary fluoride and water
fluoride concentration, but concluded that “the low/optimum fluoride concentration in drinking
water compared to urine . . . correlated significantly.” Id. The strong relationship between the
fluoride concentration in water and urinary fluoride is further corroborated by Green (2019).
Green (2019) studied samples collected from the MIREC Cohort (Canadian women and offspring)
and identified a moderate correlation between maternal urinary fluoride intake and water fluoride
concentration. Dkt. No. 432-5, Trial Ex. 109 at 1, 5 (“The MUF, was moderately correlated with
fluoride intake (r = 0.49; P < .001) and water fluoride concentration (r = 0.37; P < .001).”).
Though not a one-to-one comparison, the correlation observed in these studies further corroborates
Dr. Thiessen’s testimony as to the relationship between water fluoride and urinary fluoride.
63

United States District Court
Northern District of California

1

and that’s probably both food and water . . . [a]nd water is a significant portion . . . of that”).

2

Moreover, water fluoridation also contributes to fluoride exposure indirectly because commercial

3

food and beverages are made using fluoridated water; this is known in the scientific community as

4

the “halo effect” of water fluoridation. See, e.g., Dkt. No. 401, Feb. 6, 2024, Trial Tr. at

5

799:7:800:13 (Thiessen) (describing the “halo effect” of water fluoridation wherein individuals

6

ingest water that has been fluoridated by way of beverages such as colas, juices, beer and wine,

7

that were made using water from a fluoridated community); Dkt. No. 396, Feb. 1, 2024, Trial Tr.

8

at 212:7-23 (Lanphear) (describing the “halo effect” of communities that fluoridate water, causing

9

exposure of fluoride in surrounding areas by way of food and beverage). See also Dkt. No. 432-4,

10

Trial Ex. 108 at 6-7 (describing the “diffusion or halo effect” . . . “which refers to the extension of

11

fluoridation to residents of nonfluoridated communities as a result of foods and beverages that are

12

commercially processed in fluoridated areas and consumed in nonfluoridated communities”)

13

(citing Griffin et al. 2001; Ripa 1993). Accordingly, it is appropriate to infer conservatively that

14

approximately half of the maternal urinary fluoride observed in a pregnant woman’s urine is

15

attributed to community water fluoridation.30 Here, that is 0.4 mg/L (0.8 mg/L divided by two)

16

(median) maternal urinary fluoride and 0.945 mg/L (1.89 mg/L divided by two) (95th percentile)

17

maternal urinary fluoride.
b. One concern regarding extrapolating water intake from maternal urinary fluoride is

18
19

that fluoride intake is not necessarily equivalent with fluoride excretion; the absorption and

20

excretion process adds complexity. For example, a pregnant woman will experience the

21

breakdown of her own skeleton during pregnancy to form the fetal skeleton, releasing fluoride

22

absorbed in her bones, resulting in an increase in excretion of urine not tied to additional fluoride

23

consumption. See Dkt. No. 395, Jan. 1, 2024, Trial Tr. at 121:10-20 (Hu). To this end the EPA

24

argues that because of the complexities regarding absorption and excretion of fluoride, use of a

25

physiologically based pharmacokinetic (“PBPK”) modeling31 is necessary to convert maternal

26
27
28

30

As noted below in Paragraph 91(b)(i), the EPA allows for assumptions, including, e.g.,
absorption rates, when specific data is not available.
31

PBPK model is “a computer model that estimates concentrations of a substance in other parts of
64

1

urinary fluoride levels to estimate the fluoride intake level. Because Plaintiffs have not done

2

PBPK modeling, EPA argues, it is inappropriate to estimate exposure attributed to the condition of

3

use from maternal urinary fluoride. See Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 943:1-7

4

(Thiessen) (recognizing that PBPK models have not been identified to predict maternal urinary

5

fluoride concentrations based on drinking water exposures.). The Court rejects the EPA’s

6

argument for the following reasons.
i.

United States District Court
Northern District of California

7

While PBPK modeling may be useful and perhaps ideal, it is not

8

essential to conduct a risk evaluation. The Amended TSCA does not expressly mandate use of a

9

PBPK model, but instead affords ample discretion in the methodologies and modeling the risk

10

assessor may employ in assessing risk. See 15 U.S.C. § 2625(h) (describing factors to be

11

considered determining the methodologies or models to employ when assessing risk and omitting

12

any reference to a PBPK model).32 And the EPA Guidelines expressly recognize that

13

pharmacokinetic data may not always be available and instructs a risk assessor to be aware of

14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

the body based on physiological parameters like absorption” and is used to convert from excretion
level to intake level. See Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 943:1-7 (Thiessen).
32

This section provides in full:
In carrying out sections 2603, 2604, and 2605 of this title, to the
extent that the Administrator makes a decision based on science, the
Administrator shall use scientific information, technical procedures,
measures, methods, protocols, methodologies, or models, employed
in a manner consistent with the best available science, and shall
consider as applicable – (1) the extent to which the scientific
information, technical procedures, measures, methods, protocols,
methodologies, or models employed to generate the information are
reasonable for and consistent with the intended use of the
information; (2) the extent to which the information is relevant for
the Administrator’s use in making a decision about a chemical
substance or mixture; (3) the degree of clarity and completeness
with which the data, assumptions, methods, quality assurance, and
analyses employed to generate the information are documented; (4)
the extent to which the variability and uncertainty in the
information, or in the procedures, measures, methods, protocols,
methodologies, or models, are evaluated and characterized; and (5)
the extent of independent verification or peer review of the
information or of the procedures, measures, methods, protocols,
methodologies, or models.

15 U.S.C. § 2625(h).

65

United States District Court
Northern District of California

1

uncertainties posed by lack of such data. Specifically, the EPA Guidelines provide: “If data to be

2

used in a risk characterization are from a route of exposure other than the expected human

3

exposure, then pharmacokinetic data should be used, if available, to make extrapolations across

4

routes of exposure. If such data are not available, the Agency makes certain assumptions

5

concerning the amount of absorption likely or the applicability of the data from one route to

6

another (U.S. EPA, 1992).” EPA Guidelines at 62. This is an implicit recognition that a risk

7

evaluation can proceed without pharmacokinetic modeling when such data is not available. See

8

also EPA Guidelines at 47 (“Pharmacokinetic data may be helpful in defining the dose-response

9

curve, developing a more accurate basis for comparing species sensitivity (including that of

10

humans), determining dosimetry at sites, and comparing pharmacokinetic profiles for various

11

dosing regimens or routes of administration. The correlation of pharmacokinetic parameters and

12

neurotoxicity data may be useful in determining the contribution of specific pharmacokinetic

13

processes to the effects observed.”) (emphasis added). Dr. Barone likewise testified that the EPA

14

has conducted risk evaluations under Amended TSCA without PBPK modeling as such models are

15

not always available, explaining: “[w]e used PBPK models in five of the first [ten] risk

16

evaluations. And to varying degrees . . . In some cases we actually had the ability to . . .

17

incorporate studies that included oral exposures, inhalation exposures and dermal exposures . . . so

18

we could look at a wider range of exposures and to do that aggregation of exposures across routes.

19

That’s not always available to us, we don’t always have those kinds of models available to us.”

20

Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 675:9-676:7 (Barone). See also Dkt. No. 401, Feb. 6,

21

2024, Trial Tr. at 576:12-17 (Barone), 578:8-10 (Barone) (“Q. And in EPA’s 10 risk evaluations

22

under TSCA, EPA has only departed from using the default uncertainty factor of 10 for

23

intraspecies variability when it had an acceptable physiologically-based pharmacokinetic model

24

for the chemical, correct? A. In the first ten that is a true statement.”). Put simply, this lack of

25

PBPK modeling is not fatal to Plaintiffs’ proof.

26

ii.

Though Plaintiffs do not present a PBPK model, Till (2018) and

27

Malin (2023) provide real-world, observational data as to the exposure level of for the population

28

at issue under the condition of use at issue. See ¶ 90. See also Dkt. No. 400, Feb. 5, 2024, Trial
66

United States District Court
Northern District of California

1

Tr. at 678:6-21 (Barone) (describing that in assessing risk under a condition of use one endeavors

2

to subtract the background exposure from the water intake component to understand the risk at

3

issue, ideally through modeling, but ultimately “to the best of your ability”). And uncertainties

4

posed by lack of modeling may be accounted for in subsequent steps of the analysis (i.e., assessing

5

overall confidence in data in the risk characterization, see ¶¶ 112-13 and when determining the

6

appropriate uncertainty factor to employ when assessing the margin of exposure, see ¶ 101(b)).

7

Under the present circumstances, there is sufficient data to support the exposure levels identifies

8

notwithstanding lack of PBPK modeling.33

9

iii.

As stated above, Till (2018) observed an approximately 2x increase

10

in maternal urinary fluoride levels comparing the mothers in fluoridated relative to non-fluoridated

11

communities across three trimesters of pregnancy. See Trial Ex. 108, Dkt. No. 432-4 at 6-7, 8-9;

12

Dkt. No. 432-18, Trial Ex. 122 at 5-6 (Table 2 and Fig. 1). However, Till (2018) and Malin

13

(2023) also observed that pregnant women’s maternal urinary fluoride levels increased in both

14

fluoridated and non-fluoridated areas in the third trimester of pregnancy relative to the first

15

trimester. See Dkt. No. 432-4, Trial Ex. 108 at 8-9, Table 3; Dkt. No. 432-18, Trial Ex. 122, at 5-6

16

(Table 2 and Fig. 1). This would, at first blush, suggest that something other than fluoridated

17

water contributed to increased maternal urinary fluoride levels in the third trimester, undermining

18

the assumption that fluoridated water is a significant contributor to those levels. However, this

19

observation is well accounted for. As explained previously, the increase in maternal urinary

20

fluoride across both populations in the third trimester of pregnancy is believed to be caused by the

21

breakdown of the maternal skeleton in later trimesters of pregnancy to facilitate the formation of

22

the fetal bone – a process that releases fluoride. See, e.g., Dkt. No. 395, Jan. 1, 2024, Trial Tr. at

23

121:10-20, 121:25-123:8 (Hu). This observation thus does not undermine the probative value of

24
25
26
27
28

33

Though EPA does not bear the burden of proof in this context the Court does note that EPA has
not explained why, if PBPK modeling is necessary to understand risk associated with water
fluoridation and appropriate models are available, the EPA has not itself conducted this PBPK
modeling. This is not legally relevant given the statutory framework, and does not bear on the
Court’s findings. However, to the extent that the EPA determines that PBPK modeling is
necessary to engage in rulemaking, it may conduct this assessment to put a finer point on risk
posed by the condition of use before taking regulatory action; there is nothing preventing EPA
from doing so.
67

United States District Court
Northern District of California

1

Till (2018) and Malin (2023).

2

92. The present recommended water fluoride concentration in the United States is 0.7 mg/L

3

fluoride. NTP Monograph at 1. It follows that pregnant women living in a fluoridated community

4

in the United States are typically exposed to fluoride levels of 0.7 mg/L fluoride, measured in

5

water fluoridation. Even more conservatively, the Till (2018) median water fluoride level of 0.56

6

mg/L measured in water fluoride is also an appropriate, conservative exposure level to utilize in

7

the risk evaluation. This is because the United States and Canada (where data for Till (2018) was

8

collected) take a similar approach to water fluoridation. See Dkt. No. 433-4, Trial Ex. 129 at 16

9

(describing optimal water fluoride levels in Canada of 0.7 mg/L); Dkt. No. 396, Feb. 1, 2024 Trial

10

Tr. at 245:1-22 (Lanphear) (describing optimal 0.7 mg/L water fluoride standard in Canada).

11

Moreover, urinary fluoride levels in mothers from Los Angeles observed in Malin (2023) and Till

12

(2018) are highly similar. See Dkt. No. 432-18, Trial Ex. 122 at 1, 9.

13

93. The EPA often expresses exposure and hazard level in mg/kg/day, but this is not

14

necessary. What is vital, however, is that the exposure level and hazard level is in the same unit.

15

Dkt. No. 400, Feb. 5, 2024, Trial Tr. (Barone) at 672:22-673:4 (testifying that what matters is that

16

the “[e]xposure concentration in the denominator has to be in the same units as the hazard point of

17

departure or hazard level in the numerator[;] [t]hey have to match up”). Dr. Thiessen likewise

18

testified that “there’s no scientific reason why [the hazard and exposure levels] have to be

19

milligrams per kilogram per day. They could also be milligrams per liter in the drinking water,

20

they could also be milligrams per liter in the urine. What matters is comparison of a hazard level

21

and exposure level that are in the same units.” Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 790:18-

22

791:16 (Thiessen). Thus, the exposure and hazard level need not be expressed in mg/kg/day, but

23

the units for each must match when conducting subsequent steps of the analysis.

24
25
26

94. For the reasons stated above, and in view of the record evidence, Plaintiffs have shown by
a preponderance of the evidence that:
a. Pregnant mothers in fluoridated communities in the United States are typically

27

exposed to fluoridation of drinking water at a concentration level of 0.7 mg/L, or conservatively,

28

0.56 mg/L. They have a median exposure level to fluoride of 0.8 mg/L (measured in maternal
68

1

urinary fluoride), and at the 95th percentile have an exposure level to fluoride of 1.89 mg/L

2

(measured in maternal urinary fluoride).

3

b. To the extent that the exposure level used in this risk assessment must reflect

4

exposure attributed solely to the condition of use of the chemical, approximately half of the

5

maternal urinary fluoride levels discussed in Paragraph 87 are attributed to water fluoridation.

6

C.

United States District Court
Northern District of California

7

Step 3: Risk Characterization
a.

Framework

8

95. At this step, the EPA calculates the risk presented by the chemical at issue by comparing

9

the point of departure (i.e., hazard level) with the human exposure level. See Dkt. No. 401, Feb. 6,

10

2024, Trial Tr. at 705:7-706:21 (Barone). To ensure a risk is not present, the EPA utilizes a

11

Margin of Exposure (MOE) equation that compares a safe margin from the point of departure

12

(benchmark MOE) with the actual margin between the exposure level and point of departure

13

(MOE). See id. at 707:13-708:19.

14

96. The actual MOE is calculated by discerning the ratio of the point of departure and the

15

human exposure level, i.e., the point of departure divided by the exposure level. Dkt. No. 429-7,

16

Trial Ex. 17 at 65. The benchmark MOE (i.e., the safe or requisite margin) is the product of the

17

applicable uncertainty factors (UFs) (i.e., UF x UF). See id. at 2-3; Dkt. No. 400, Feb. 5, 2024,

18

Trial Tr. at 575:17-576:24 (Barone), 580:10-13 (Barone) (“Q. Now, the benchmark MOE is the

19

product of all uncertainty factors that are found to be applicable to a given – to a given hazard,

20

correct? A. To a given hazard, that’s correct.”), 580:24-581:19 (Barone) (“We don’t add them.

21

We multiply – if the uncertainty factor is the default of 10 for human variability, then we use that

22

and multiply is by any other uncertainty factors.”). For example, if there is an uncertainty factor

23

of 10 for intraspecies variability, and an uncertainty factor of 10 for using a LOAEL as the point

24

of departure, the benchmark MOE is 100 (10 times 10). Id. at 581:12-582:11. As another

25

example, if the first uncertainty factor is 10, and the second uncertainty factor is 3, the benchmark

26

MOE is 30 (10 times 3). Id.

27
28

97. If the actual MOE is lesser (i.e., there is a smaller margin) than the benchmark MOE, then
there is a risk present; if the actual MOE is greater (i.e., there is a bigger margin) than the
69

1

benchmark MOE then a risk is presumed not to be present. See Dkt. No. 400, Feb. 5, 2024, Trial

2

Tr. at 583:8-13 (Barone) (explaining that if the benchmark MOE exceeds the MOE between the

3

hazard and exposure level a risk is present); Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 707:20-708:9

4

(Barone) (explaining the converse).
b.

5

98. A risk is present when using the BMCL of 0.28 mg/L (maternal urinary fluoride) as the

6

United States District Court
Northern District of California

Key findings

7

point of departure, and whether calculating risk using either the median or high-end exposure

8

levels; the exposure levels exceed the point of departure.

9

99. A risk is present when using the BMCL of 0.768 mg/L or even 1.536 mg/L (maternal

10

urinary fluoride) as the point of departure, whether calculating risk using either the median or

11

high-end exposure levels; the exposure levels exceed the point of departure.
100.

12
13

Alternatively, a risk is present when utilizing the conservative 4 mg/L (water

fluoride) as the point of departure; the actual MOE is less than the benchmark MOE.
c.

14

Underlying findings
(a)

15

BMCL: 0.28 mg/L and in the alternative, 0.768 mg/L and/or 1.536
mg/L (maternal urinary fluoride)

16
101.

17

The appropriate benchmark MOE to use in calculating risk for the BMCLs

18

identified by Dr. Grandjean is 10, which includes at least one UF of 10 to account for intraspecies

19

variability:
a. A UF of 10 is utilized as a default practice in calculating risk to account for

20
21

intraspecies variability, i.e., the variability within the human species in reacting to chemicals.34

22

See Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 712:12-713:22 (Barone).

23

b. Absent use of physiologically based pharmacokinetic (PBPK) modeling to account

24

for those variabilities, which could allow for the reduction of the UF from 10 down to 3, the EPA

25

applies the UF of 10 in calculating the benchmark MOE. See id. at 712:24-713:22; Dkt. No. 401,

26
27
28

34

Intraspecies variability can be compared with interspecies variability, which accounts for
variability between different species (i.e., animals and humans) when extrapolating from
animal studies. Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 713:6-10 (Barone).
70

1

Feb. 6, 2024, Trial Tr. at 576:12-17 (Barone), 578:8-10 (Barone) (“Q. So the default uncertainty

2

factor that EPA uses to account for intraspecies variability and uncertainty is 10, correct? A. That

3

is the default. Q. And in EPA’s 10 risk evaluations under TSCA, EPA has only departed from

4

using the default uncertainty factor of 10 for intraspecies variability when it had an acceptable

5

physiologically-based pharmacokinetic model for the chemical, correct? A. In the first ten that is

6

a true statement.”).
c. A PBPK model has not been performed to assess fluoride intake in pregnant

United States District Court
Northern District of California

7
8

women. Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 943:1-16 (Thiessen); Dkt. No. 440, Feb. 13,

9

2024, Trial Tr. at 1396:17-1397:2 (Barone), 1397:20-23 (Barone) (“Q. And so in the nearly four

10

years since the first trial in this case, plaintiffs still have not performed a PBPK model to extract a

11

urinary fluoride value to an intake value, right? A. No, they haven’t.”).
d. Because there is no PBPK model utilized here, which would decrease uncertainty

12
13

and allow from a downward departure of the default UF of 10, the default UF of 10 is

14

appropriately used as the benchmark MOE in the present risk evaluation.

15
16
17

102.

The median exposure level for pregnant women measured in urinary fluoride is 0.8

mg/L, and the 95th percentile is 1.89 mg/L. See ¶ 87.
103.

The actual MOE for the BMCL of 0.28 mg/L at the median exposure level is 0.35

18

(0.28 mg/L divided by 0.8 mg/L) and 0.148 at the 95th percentile exposure level (0.28 mg/L

19

divided by 1.89 mg/L). The actual MOEs, 0.35 and 0.148, do not exceed the benchmark MOE of

20

10; thus, the MOE is below the benchmark MOE and a risk is present. See Dkt. No. 401, Feb. 6,

21

2024, Trial Tr. at 707:20-708:9 (Barone) (explaining that a risk is not present where the actual

22

MOE is higher than the benchmark MOE). Another way of looking at exposure/risk is taking the

23

BMCL and adjusting it downward for risk factors. To account for a ten-fold risk factor of human

24

variability, actual exposure should not exceed 1/10th of the BMCL of 0.28 mg/L – i.e., 0.028

25

mg/L. However, the trial evidence establishes actual exposure of levels of 0.8 and 1.89 mg/L –

26

this far exceeds that safety limit of 0.028 mg/L. See also Dkt. No. 198-4 at 75-77 (Thiessen Decl.)

27

(providing MOE calculations).

28

104.

The actual MOE for the BMCL of 0.768 mg/L at the median exposure level is 0.96
71

1

(0.768 mg/L divided by 0.8 mg/L) and 0.406 at the 95th percentile exposure level (0.768 mg/L

2

divided by 1.89 mg/L). The actual MOEs, 0.96 and 0.406, do not exceed the benchmark MOE of

3

10; thus, the MOE is below the benchmark MOE and a risk is present. See Dkt. No. 401, Feb. 6,

4

2024, Trial Tr. at 707:20-708:9 (Barone). See also Dkt. No. 198-4 at 75-77 (Thiessen Decl.)

5

(providing MOE calculations). Put differently, 1/10th of this BMCL is 0.0768 mg/L (0.768 mg/L

6

divided by 10). Both the median and upper exposure levels of fluoride found in mothers’ urine

7

exceed this amount.

8

United States District Court
Northern District of California

9

105.

Even using the higher 1.536 mg/L BMCL to account for omission of the OCC

Cohort data, see ¶ 73 (discussing exclusion of OCC Cohort data in deriving 0.768 mg/L BMCL

10

using squared model in Grandjean (2022)), a risk is present. Using this figure, the actual MOE at

11

the median exposure level is 1.92 (1.536 mg/L divided by 0.8 mg/L) and 0.813 at the 95th

12

percentile exposure level (1.536 mg/L divided by 1.89 mg/L). 1.92 and 0.813 do not exceed 10;

13

thus, the actual MOE is below the benchmark MOE and a risk is present. See Dkt. No. 401, Feb. 6,

14

2024, Trial Tr. at 707:20-708:9 (Barone). See also Dkt. No. 198-4 at 75-77 (Thiessen Decl.)

15

(providing MOE calculations). Put differently, 1/10th of this BMCL is 0.1536 mg/L (1.536 mg/L

16

divided by 10). Both the median and upper exposure levels in mothers’ urine exceed this amount.

17

106.

Even if the Court were to consider only half of the exposure level, directly

18

attributable to water fluoridation, as opposed to other sources of fluoride (0.4 mg/L (0.8 mg/L

19

divided by 2) (median) maternal urinary fluoride and 0.945 mg/L (1.89 mg/L divided by 2) (95th

20

percentile) maternal urinary fluoride, a risk is still present. Both of these figures exceed the safe

21

level using a BMCL of 0.28 mg/L (0.028 mg/L). See ¶ 103. And these figures also exceed the

22

safe level considering the margin of error if the BMCL of 0.768 mg/L or 1.536 mg/L; the safe

23

levels are 0.0768 mg/L and 0.1536 mg/L (1/10th of each BMCL), respectively. See ¶¶ 104-05.
(b)

24
25

107.

LOAL: 4 mg/L (water fluoride)

Alternatively, to the extent that the BMCLs identified previously are not

26

appropriate points of departure, or maternal urinary fluoride is not an appropriate metric, a risk is

27

present using a LOAL of 4 mg/L measured in water fluoride.

28

108.

The appropriate UF applied in the benchmark MOE analysis using the LOAEL of 4
72

1

mg/L is 100 (10 x 10):
a. The UF of 10 is appropriately applied to account for intraspecies variability.

2
3

See ¶ 101.
b. A second UF of 10 is also appropriately applied when using a LOAEL as the point

4
5

of departure. Dkt. No. 440, Feb. 13, 2024, Trial Tr. at 1425:13-17 (Barone) (“Q. Right. If we

6

were using a human study and only had a LOAEL, like was the case with PCE, you would, at that

7

point, consider an additional uncertainty factor beyond the intraspecies variability uncertainty

8

factor? A. Generally, yes. Yes, we would.”).
c. Again, the benchmark MOE is calculated by multiplying the applicable UFs. Dkt.

United States District Court
Northern District of California

9
10

No. 400, Feb. 5, 2024, Trial Tr. at 575:17-576:24 (Barone), 580:10-13 (Barone) (explaining that

11

the benchmark MOE is the product of applicable UFs), 580:24-581:19 (stating that “[w]e don’t

12

add them[;] [w]e multiply”).
109.

13

Pregnant women in “optimally” fluoridated communities in the United States have

14

an exposure level of at least 0.7 mg/L (water fluoride). See ¶ 86. Or conservatively, 0.56 mg/L

15

derived from Till (2018), in the alternative. See ¶ 89(a).
110.

16
17

The actual MOE for the LOAEL of 4 mg/L (water fluoride) is 5.71 (4 mg/L divided

by 0.7 mg/L) or 7.14 (4 mg/L divided by 0.56 mg/L).
111.

18

5.71 and/or 7.14 do not exceed 100; the actual MOE is below the benchmark MOE

19

and thus a risk is present. Dkt. No. 400, Feb. 5, 2024, Trial Tr. at 583:8-13 (Barone) (explaining

20

that if the benchmark MOE exceeds the MOE between the hazard and exposure level a risk is

21

present). See also Dkt. No. 198-4 at 75-77 (Thiessen Decl.) (providing MOE calculations).

22

Again, another way of looking at this is to take the LOAEL of 4 mg/L, and divide that by the two

23

risk factors. To this end, 4 mg/L divided by 100 equals 0.04 mg/L, reflecting the tolerable

24

concentration of exposure given the risk factors. Exposure to 0.7 mg/L in United States drinking

25

water, or conservatively 0.56 mg/L (Till (2018)),35 far exceeds that limit.

26
27
28

35

The condition of use at issue in this suit is fluoridation of water at 0.7 mg/L. However, it is
useful to consider the risk posed with the lesser exposure level of 0.56 mg/L given the findings of
Till (2018). There, subjects in Canada – which has the same optimal level of water fluoridation as
the United States – had a median community water fluoride level of 0.56 mg/L. It follows that
73

1

D.

Step 4: Risk Determination
a.

2
3

Once the risk has been identified, in the last step of the risk evaluation process the

4

assessor determines if that risk is an unreasonable one. Dkt. No. 401, Feb. 6, 2024, Trial Tr. at

5

735:11-19 (Barone).

6

United States District Court
Northern District of California

112.

Framework

113.

In making the determination of whether the risk is unreasonable, the assessor

7

considers several factors including: (1) severity of the hazard; (2) exposure-related considerations

8

(e.g., duration, magnitude, or frequency of the exposure, and size of the affected population); (3)

9

other characteristics of the population that is exposed, including the susceptibility of

10

subpopulations; (4) confidence in the information used to inform the hazard and exposure values;

11

and relatedly, the (5) overall strength of the evidence and uncertainties and assumptions included

12

throughout the risk assessment. See Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 735:11-736:19

13

(Barone); Dkt No. 437-1, Trial Ex. 96, at 500 (PCE Risk Evaluation); Dkt. No. 437-3, Trial Ex.

14

98 at 271 (1,4-Dioxane Risk Evaluation).
b.

15
16
17

114.

Based on the aforementioned factors, and in view of the record evidence, the risk at

issue – reduced IQ in children posed by water fluoridation at 0.7 mg/L – is an unreasonable risk.
c.

18
19

Key finding

115.

Underlying findings

Given the seriousness of reduced IQ, and the ample support in the record that the

20

United States population is at risk of experiencing IQ decrements of over four IQ points, the

21

severity of the hazard at issue (reduced IQ in children, see Section III.A.1.), weighs in favor of

22

finding the risk at issue unreasonable:

23

a. The EPA has recognized that cognitive deficits including reduced IQ are critical

24

chronic health effects, as exemplified by its in its risk evaluation of PCE under the Amended

25

TSCA which identified cognitive deficits as the hazard warranting regulatory action. Dkt. No.

26
27
28

some communities in the United States may have similar median water fluoridation levels. Thus,
it is worth considering if a risk is present at this lower level of exposure, to understand the risk of
setting an optimal fluoridation level of 0.7 mg/L as is the standard in the United States.
74

1

400, Feb. 5, 2024, Trial Tr. at 597:9-13 (Barone). Moreover, according to the EPA’s Clean Air

2

Science Advisory Commission, in the context of its analysis of lead: “[a] population loss of 1-2 IQ

3

points is highly significant from a public health perspective.” Dkt. No. 430-1, Trial Ex. 42 at

4

67000. To this end, a 1-to-2 point loss in IQ was the hazard that supported the identification of

5

lead as a substance posing an unreasonable risk. Id. See also Dkt. No. 433-4, Trial Ex. 129 at 27

6

(recognizing that one study found that a reduction of one IQ point “has been shown to be

7

associated with reduced educational attainment, employment status, productivity, and earned

8

wages, reflecting substantial public health concerns”).
b. In risk assessments, the EPA evaluates not only the hazard presented at median

United States District Court
Northern District of California

9
10

exposures levels, but considered the hazard posed to the 95th percentile (i.e., high exposure

11

populations). Dkt. No. 430-1, Trial Ex. 42 at 67000. And the EPA considers impact upon

12

smaller, susceptible subpopulations in assessing the risk at issue. See Dkt. No. 400, Feb. 5, 2024,

13

Trial Tr. at 587:7-18 (Barone) (testifying that the EPA considered impact on small, susceptible

14

subgroup of population in regulating lead).
c. As Dr. Grandjean explained, women in the 95th percentile exposure level to

15
16

fluoride exceed the BMCL for a 1-point loss in IQ by over a factor of four. See Dkt. No. 397, Feb.

17

2, 2024, Trial Tr. at 358:2-18 (Grandjean). Indeed, when considering high-end exposure levels,

18

relative to Dr. Grandjean’s BMCL identifying the dosage at which a 1-point IQ decrement is

19

expected, fluoride presents a risk of a decrease in IQ ranging from 2.86 to 6.75 IQ points.36
116.

20

Exposure-related considerations (e.g., duration, magnitude, or frequency of the

21

exposure, and size of the affected population) weighs heavily toward finding the risk at issue

22

unreasonable; the exposure is continuous, and nearly all Americans are affected.

23
24
25
26
27
28

36

According to Dr. Grandjean’s analysis, an increase of 0.28 mg/L of fluoride exposure (measured
in maternal urinary fluoride) is associated with a 1-point IQ loss in the mother’s offspring (boys
and girls). See Dkt. No. 432-15, Trial Ex. 119 (Grandjean (2023)) at 1-2, 9. Pregnant mothers in
fluoridated communities in the United States have a median and 95th percentile exposure level to
fluoride of 0.8 mg/L and 1.89 mg/L, respectively (measured in maternal urinary fluoride). See ¶¶
86-88; Trial Ex. 122, Dkt. No. 432-18 at, Trial Ex. 122 at 9. Thus, fluoride presents a hazard of
reduced IQ ranging from approximately 2.86 points at the median intake level,((0.8 mg/L (median
exposure level) divided by 0.28 mg/L (dosage at which 1 IQ point decrease is observed)), i.e.,
2.857) to 6.75 points at the 95th percentile ((1.89 mg/L (95th percentile exposure level) divided by
0.28 mg/L (dosage at which 1 IQ point decrease is observed)), i.e., 6.75).
75

United States District Court
Northern District of California

1

117.

The size of the affected population is vast. Approximately 200 million Americans

2

have fluoride intentionally added to their drinking water at a concentration of 0.7 mg/L. See Dkt.

3

No. 421 at 206-07 (undisputed). Other Americans are indirectly exposed to fluoridated water

4

through consumption of commercial beverages and food manufactured with fluoridated water (i.e.,

5

the “halo effect”). See, e.g., Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 799:7:800:13 (Thiessen)

6

(describing the “halo effect” of water fluoridation); Dkt. No. 396, Feb. 1, 2024, Trial Tr. at 212:7-

7

23 (Lanphear) (similar). See also Dkt. No. 432-4, Trial Ex. 108 at 6-7 (describing the “diffusion

8

or halo effect” . . . “which refers to the extension of fluoridation to residents of nonfluoridated

9

communities as a result of foods and beverages that are commercially processed in fluoridated

10

areas and consumed in nonfluoridated communities”) (citing Griffin et al. 2001; Ripa 1993).

11

Approximately two million pregnant women, and over 300,000 exclusively formula-fed babies are

12

exposed to fluoridated water. Dkt. No. 421 at 209-210. See also Dkt. No. 401, Feb. 6, 2024, Trial

13

Tr. at 815:6-816:23 (Thiessen). The number of pregnant women and formula-fed babies alone

14

who are exposed to water fluoridation each year exceeds entire populations exposed to conditions

15

of use for which EPA has found unreasonable risk; the EPA has found risks unreasonable where

16

the population impacted was less than 500 people. See Dkt. No. 400, Feb. 5, 2024, Trial Tr. at

17

588:11-15 (Barone) (testifying that under TSCA the EPA had made unreasonable risk

18

determinations for conditions of use that involve less than 500 people, and that “many are less

19

than 500 people”). See also Dkt. No. 421 at 209-210 (EPA agreeing that “the exposed population

20

for the condition of use of community water fluoridation exceeds the exposed populations of the

21

first ten risk evaluations under Amended TSCA”).

22

a. Individuals are exposed to fluoride through water intake every day; the parties do

23

not dispute that frequency of exposure for most people is several times daily (i.e., through

24

drinking tap water). Dkt. No. 421 at 207 (undisputed).

25

b. And the duration of exposure to fluoridated water is continuous with its effects

26

long-lasting. See Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 813:18-20 (Thiessen) (describing that

27

exposure to community water fluoridation is intended to be lifelong). To this end, fluoride

28

remains in the body through years; for several years after cessation of fluoride exposure a woman
76

1

is likely to release fluoride into blood due to skeletal breakdown. Dkt. No. 397, Jan. 31, 2024,

2

Trial Tr. at 370:6-371:12 (Grandjean); Dkt. No. 402, Feb. 8, 2024, Trial Tr. at 932:16-20

3

(Thiessen).

4

118.

5

issue unreasonable. It is undisputed that large numbers of susceptible individuals are being

6

exposed each year to fluoride through fluoridation, namely, approximately two million pregnant

7

women, and over 300,000 exclusively formula-fed babies. Dkt. No. 421 at 209-210. See also

8

Dkt. No. 401, Feb. 6, 2024, Trial Tr. at 815:6-816:23 (Thiessen).

9

United States District Court
Northern District of California

The susceptibility of exposed populations weighs heavily toward finding the risk at

119.

The scientific literature in the record provides a high level of certainty that a hazard

10

is present; fluoride is associated with reduced IQ. There are uncertainties presented by the

11

underlying data regarding the appropriate point of departure and exposure level to utilize in this

12

risk evaluation. But those uncertainties do not undermine the finding of an unreasonable risk; in

13

every scenario utilizing any of the various possible points of departures, exposure levels and

14

metrics, a risk is present in view of the applicable uncertainty factors that apply:

15

a. Regarding the point of departure, as discussed above, there is some uncertainty

16

regarding the appropriate point of departure to utilize. Specifically, there is lack of certainty

17

regarding the model fit to be utilized in the BMD modeling analysis, which determines the BMCL

18

to utilize as a point of departure. See ¶ 72 (discussing use of linear vs. squared model to derive

19

BMCL). However, under either scenario (whether using a linear or squared model), there is an

20

insufficient safety margin between the exposure level and hazard level; a risk is present. See ¶¶

21

102-106. Even assuming BMD modeling cannot be used for the data set and using a highly

22

conservative LOAEL of 4 mg/L, a risk remains present by a substantial margin. See ¶¶ 107-111.

23

Accordingly, the uncertainty regarding the point of departure (hazard level) is ultimately not

24

consequential to the conclusion herein. The EPA has deemed a risk unreasonable even where it

25

lacked high confidence in the hazard data. See Dkt. No. 421 at 211 (undisputed).

26

b. Regarding the exposure level, there is uncertainty presented by the fact that a

27

PBPK model was not utilized to determine the precise amount of fluoride reflected in pregnant

28

women’s maternal urinary fluoride levels that derives from fluoridated water. See ¶ 91(b).
77

United States District Court
Northern District of California

1

Uncertainty due to lack of modeling is offset by the fact that it is appropriate to view risk

2

presented by water fluoridation in context of its additive effects on aggregate exposure, which is

3

best reflected by real world maternal urinary fluoride levels. See ¶¶ 89-90. And this is

4

particularly true where, as here, water fluoridation is known to be a significant contributor to

5

maternal urinary fluoride levels, and indeed functions roughly as a 2x multiplier to those levels.

6

See id. Further, here, there is real-world observational data showing what the maternal urinary

7

fluoride levels of women that live in communities with fluoridation levels comparable to that of

8

the United States; this data makes the PBPK model less critical to the analysis. See ¶¶ 89-91. The

9

uncertainty from the lack of PBPK model weighs against finding the risk unreasonable, but not

10

strongly so due to these mitigating circumstances. Moreover, when utilizing the conservative

11

LOAEL as a point of departure, that metric is derived from water fluoride intake, and does not

12

present the same uncertainty posed by using maternal urinary fluoride levels as the metric of

13

hazard and exposure. Finally, the EPA has deemed a risk unreasonable even where it lacked high

14

confidence in the exposure data. See Dkt. No. 421 at 211 (undisputed).

15

c. There is significant certainty in the data set regarding the association between

16

fluoride and reduced IQ. Namely, there is a robust body of evidence finding a statistically

17

significant adverse association between fluoride and IQ. A large majority of the 72

18

epidemiological studies assessed by the NTP Monograph observed this relationship including all

19

but one of the 19 high-quality studies, see ¶¶ 34-36, and literature published after the NTP

20

Monograph cutoff date observed the same relationship, see ¶ 37 – and countervailing evidence, for

21

various reasons described previously, are of little impact on this repeated, and consistently

22

observed association between fluoride and reduced IQ, see ¶ 39. Moreover, complete consistency

23

amongst studies is not expected. See Dkt. No. 414, Feb. 9, 20240, Trial Tr. at 1172:23-1173:6

24

(Savitz). Notably, notwithstanding inherent difficulties in observing this association at lower

25

exposure levels, studies assessing such levels still observed a statistically significant relationship

26

between fluoride and reduced IQ. See ¶¶ 42-44. Again, to put the breadth of evidence supporting

27

this finding in perspective, the EPA has identified a LOAEL based upon far less in other contexts.

28

For instance, in the EPA’s risk evaluation of Methylene, conducted pursuant to Amended TSCA,
78

1

the EPA used a LOAEL for developmental neurotoxicity, derived from the analysis of one study

2

conducted upon mouse pups (Fredriksson et al., 1992). See Methylene Risk Evaluation at 262.

3

Compare this with 6 (water fluoride) and 9 (urinary fluoride), high-quality, epidemiological

4

studies of human populations underling the 4 mg/L LOAEL underlying the POD here. Dkt. No.

5

431-2, Trial Ex. 68 at 39, 41 (eTable 4). The scientific literature in the record provides a high

6

level of certainty that a hazard is present; fluoride is associated with reduced IQ. The qualitative

7

evidence is superior.

8

United States District Court
Northern District of California

9

120.

In sum, the first three factors weigh toward finding the risk unreasonable. Namely,

the severity of the hazard weighs toward finding the risk unreasonable. The exposure-related

10

considerations and exposure of susceptible populations weighs strongly toward finding the risk

11

unreasonable; millions of susceptible individuals are exposed to fluoride and the exposure is

12

frequent and long-lasting. The two final factors, confidence in hazard data and overall strength of

13

the evidence and uncertainties, are largely neutral. Because the first three factors weigh strongly

14

toward finding the risk unreasonable and the last two are largely neutral, the totality of the factors

15

establish that the risk is unreasonable under the Amended TSCA. The Court thus finds that the

16

Plaintiffs have established by a preponderance of the evidence that the risk at issue is

17

unreasonable.

18
19

IV.
121.

CONCLUSIONS OF LAW

Plaintiffs have proven, by a preponderance of the evidence, that water fluoridation

20

at the level of 0.7 mg/L – the prescribed optimal level of fluoridation in the United States –

21

presents an “unreasonable risk of injury to health or the environment, without consideration of

22

costs or other non-risk factors, including an unreasonable risk to a potentially exposed or

23

susceptible subpopulation under the conditions of use.” 15 U.S.C. § 2620(b)(4)(B)(ii).

24
25
26

122.

The Court thus orders the Administrator to initiate rulemaking pursuant to

Subsection 6(a) of TSCA. See id. §§ 2605(a), 2620(a).
123.

The Court defers ruling as to whether Plaintiffs are entitled to recovery of their

27

costs of suit and attorneys and expert witness fees. Parties are ordered to submit a proposed

28

supplemental briefing schedule regarding costs and fees within two weeks of the date of this order.
79

1

Defendant shall respond two weeks thereafter. The Court will take the matter under submission

2

unless it orders a hearing.

3
4

The Clerk of Court is directed to enter judgment in Plaintiffs’ favor.

5
6

IT IS SO ORDERED.

7
8

Dated: September 24, 2024

9
10

United States District Court
Northern District of California

11

______________________________________
EDWARD M. CHEN
United States District Judge

12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
80