Does Fluoridated Water During Pregnancy Lower a Child’s IQ? What the 2026 ECHO Study Actually Found

It is not established that fluoridated water during pregnancy lowers a child’s IQ by 3.36 points. The 2026 U.S. ECHO study behind that viral claim did not measure IQ.

What it did find is important: among 2,514 U.S. children, higher estimated fluoride concentrations in public drinking water during pregnancy were associated with lower scores on an NIH Toolbox measure of fluid cognition. In one statistical model, the association became steeper above a fitted change point of 675 micrograms per liter, or 0.675 mg/L. Above that point, each additional 100 µg/L of fluoride was associated with a 0.67-point lower fluid-cognition score.

That works out to about 3.36 fluid-cognition scale points per additional 500 µg/L. It does not mean that 0.7 mg/L of fluoride caused children to lose 3.36 IQ points.

Nor did the study prove that 0.675 mg/L is a biological toxicity threshold. A different statistical model in the same paper showed the clearest inverse association above 1.107 mg/L, illustrating how the apparent point at which the relationship becomes detectable depends partly on how the exposure-response curve is modeled.

The responsible conclusion is therefore more consequential—and more uncertain—than either “fluoride lowers IQ at 0.7 mg/L” or “there is nothing to worry about.”

The study provides a credible new U.S. signal that prenatal fluoride exposure at concentrations relevant to American water systems may be associated with some aspects of cognition. It does not establish causation, quantify an IQ loss, or prove that the current 0.7 mg/L fluoridation target is unsafe.

What the 2026 ECHO cognition study actually did

The study, published August 14, 2026, in the American Journal of Epidemiology, pooled data from 2,514 children born from 2006 through 2019 across 17 ECHO cohort sites spanning 23 states.

Researchers did not collect a maternal blood or urine sample and determine each woman’s total fluoride dose. Instead, they reconstructed an estimate of prenatal public-water exposure by linking mothers’ residential addresses during pregnancy to census-tract-level public drinking-water fluoride measurements.

Estimated prenatal water fluoride ranged from less than 1 µg/L to 1,940 µg/L, with an average of about 397 µg/L. Children later underwent NIH Toolbox cognitive testing between ages 3 and 17.

Researchers separated cognition into two broad categories.

Fluid cognition covers abilities involved in processing new information, including attention, working memory, episodic memory, processing speed and executive functions such as cognitive flexibility and inhibitory control.

Crystallized cognition reflects knowledge accumulated through learning and experience, including vocabulary and reading-related abilities. NIH Toolbox research treats these as distinct composites; its broader total cognition composite, not the fluid score alone, has been described as conceptually akin to a general or Full Scale IQ measure.

The ECHO researchers found a nonlinear inverse association with fluid cognition. They did not find the same consistent association with crystallized cognition.

That distinction is central to understanding why the viral “IQ” framing is wrong.

Where did the “3.36 points” number come from?

The number itself was not invented.

The paper’s linear change-point model identified 675 µg/L as the fluoride concentration providing the best statistical fit for a change in the slope of fluid-cognition scores.

Above 675 µg/L, the study estimated:

0.67 lower fluid-cognition points for every additional 100 µg/L of water fluoride.

Multiplying the underlying coefficient across an additional 500 µg/L produces approximately 3.36 points, which is how Columbia and ECHO summarized the result.

The mistake occurs when those fluid-cognition points become “IQ points.”

Children’s Health Defense’s The Defender, for example, reported that every 500 µg/L increase was associated with an average “3.36-point drop in IQ.” The article elsewhere correctly acknowledges that the researchers used the NIH Toolbox Cognition Battery, but substitutes “IQ” when describing the numerical result.

That is not what the study measured.

Being at 0.7 mg/L does not mean a 3.36-point difference

There is another important numerical distinction that much of the coverage misses.

The U.S. Public Health Service target of 0.7 mg/L equals 700 µg/L. The study’s fitted change point was 675 µg/L.

Those levels differ by only 25 µg/L, not 500 µg/L.

Using the published rounded coefficient of 0.67 points per 100 µg/L, the model-implied difference between 675 and 700 µg/L is only about 0.17 fluid-cognition points.

The 3.36-point figure corresponds to a much larger 500 µg/L difference within the post-change-point portion of the model—for example, roughly 675 versus 1,175 µg/L—not merely to being exposed to water at the U.S. 0.7 mg/L target.

That calculation does not prove that 0.7 mg/L is harmless. It shows why saying the study found a “3.36-point loss at 0.7 mg/L” would badly misrepresent its dose-response estimate.

Is fluid cognition basically another name for IQ?

No.

The concepts overlap, but they are not interchangeable measurements.

The NIH Toolbox Cognition Battery contains separate tests of executive function, episodic memory, processing speed, working memory and language. Five tests contribute to its fluid-cognition composite, while vocabulary and oral reading form its crystallized composite.

NIH Toolbox standard scores can also use a familiar mean of 100 and standard deviation of 15, which makes them numerically resemble many IQ scales. That does not make a point on one test equivalent to an IQ point on another. The scoring convention is the ruler; the underlying cognitive construct being measured is what matters.

In fact, NIH Toolbox validation work distinguishes the fluid and crystallized composites from its broader total-cognition measure, describing the total composite as the measure more analogous to Full Scale IQ.

The ECHO study specifically reported the association for fluid cognition.

Calling 3.36 fluid-cognition points “3.36 IQ points” therefore implies a precision and equivalence the study did not establish.

Is 0.675 mg/L a newly discovered fluoride toxicity threshold?

No.

The number deserves attention because it is close to the U.S. Public Health Service’s recommended fluoridation concentration of 0.7 mg/L, but it should not be interpreted as a line separating “safe” from “toxic” water.

The researchers tested nonlinear relationships using more than one statistical approach.

In the linear change-point model, 675 µg/L minimized the model’s residual error and provided the best-fitting inflection point.

But the study’s restricted cubic spline analysis showed statistically significant inverse associations with fluid cognition at concentrations above approximately 1,107 µg/L, or 1.107 mg/L.

If 675 µg/L were an independently established biological threshold, different reasonable models should not be expected to identify noticeably different points at which the evidence becomes clearest.

The defensible interpretation is narrower: the exposure-response relationship did not look simply linear, and one model identified a change in slope near 0.675 mg/L.

That is scientifically relevant. It is not proof that toxicity begins at exactly that concentration.

And the reverse is also true: failure to detect an association below a statistical change point does not prove every lower concentration is harmless.

What about the claim of “up to 13.5 IQ points” in children under 7?

The study did find that the association appeared substantially stronger among children whose cognition was assessed before age 7. Columbia and ECHO explicitly reported that age pattern but cautioned that more research is needed to understand it.

A secondary analysis of the study reports a post-change-point coefficient of about 2.70 fluid-cognition points per 100 µg/L among the younger children. Multiplying that by five produces 13.5 points, explaining the number circulating online.

But two caveats matter.

First, those would still be fluid-cognition score points, not IQ points.

Second, subgroup findings are generally less secure than the overall result, particularly when the researchers themselves say the age difference requires additional study. The official ECHO summary describes the younger-child association as larger but does not elevate 13.5 points into the study’s principal result.

So “children under 7 lost 13.5 IQ points” is not a defensible description of the evidence.

Why the study still matters

Correcting the viral claim should not become an excuse to dismiss the underlying research.

The study addresses a genuine evidence gap.

The National Toxicology Program’s major fluoride review concluded with moderate confidence that higher estimated fluoride exposures—such as drinking-water concentrations above roughly 1.5 mg/L—are consistently associated with lower IQ in children. But NTP also emphasized that evidence becomes much less certain at lower drinking-water concentrations typical of U.S. fluoridation.

The ECHO study is unusually relevant because it examines a large U.S. population and includes exposures around the concentrations encountered in American public-water systems.

It also used address histories across pregnancy rather than assigning exposure from one citywide or statewide fluoridation label. That makes the exposure estimate more granular than many ecological studies.

And the association was not simply a result of comparing extremely fluoridated communities with places having almost none: the modeled signal extended into a concentration range relevant to current U.S. policy.

That is why the correct response is not “the study is meaningless because it did not measure IQ.”

It measured cognition, found a significant association in one cognitive domain and raised a legitimate question about prenatal exposure at lower concentrations than much of the older literature examined.

What remains unresolved is whether fluoride itself caused that association and how accurately public-water concentration represents each mother’s actual biological exposure.

The biggest limitation: the researchers did not measure how much fluoride each mother actually absorbed

The exposure variable is often described as “prenatal fluoride exposure,” but technically it was an estimate based on fluoride concentration in public water serving a mother’s residential area.

That is not the same thing as individual dose.

Two pregnant women could live in neighborhoods with the same 0.7 mg/L public-water concentration while having very different fluoride exposures. One might drink several liters of tap water each day. Another might primarily drink bottled water. Someone else could drink substantial amounts of tea or consume other fluoride-containing foods and beverages.

The study did not directly measure total fluoride from toothpaste, tea, food, supplements, bottled water or private wells. Nor did it record each participant’s exact water consumption.

The researchers did adjust for important potential confounders including parental education and age, prenatal tobacco exposure, child sex, season of conception and neighborhood characteristics. That strengthens the analysis but cannot eliminate residual confounding or exposure error in an observational study.

Most importantly, an association between the estimated exposure and a later outcome cannot by itself demonstrate that fluoride caused the outcome.

The study’s authors explicitly say this.

A new JAMA ECHO study makes the picture more complicated

On September 3, 2026—less than three weeks after the cognition paper—a separate ECHO analysis was published in JAMA Network Open.

It studied 5,520 children from 20 ECHO sites across 34 states and used a very similar method to estimate prenatal public-water fluoride exposure from residential addresses.

The outcome, however, was different: caregiver-reported internalizing and externalizing behavioral problems, measured with the Child Behavior Checklist.

Across the study’s main analyses, researchers found no overall association.

For each additional 500 µg/L of estimated prenatal water fluoride, the adjusted difference was:

• −0.15 points for internalizing problems, with a 95% confidence interval from −0.85 to 0.56; and
• 0.06 points for externalizing problems, with a 95% confidence interval from −0.44 to 0.55.

The study likewise did not detect a meaningful difference when comparing children above and below the 700 µg/L Public Health Service level.

But a sensitivity analysis complicates the null result.

A linear change-point model identified an inflection near 625 µg/L for externalizing problems. Above that level, each additional 500 µg/L was associated with a 1.59-point higher externalizing-problem score.

The authors correctly kept the overall conclusion cautious: the main analyses were null, while smaller associations appeared under particular change-point and subgroup analyses.

This second study does not refute the cognition paper, because behavioral symptoms and fluid cognitive performance are different outcomes.

Nor does its exploratory 625 µg/L result independently prove a toxicity threshold.

Taken together, the two studies illustrate why the current evidence should be described as mixed and unresolved, not converted into a simple “safe” or “dangerous” verdict.

An invited JAMA Network Open commentary reached essentially that methodological conclusion: area-level fluoride studies can be useful, but they cannot determine how much fluoride each participant actually consumed, and current U.S. evidence does not yet establish whether chronic low-level exposure harms children’s developing brains.

Did the new study “confirm” the National Toxicology Program’s fluoride-IQ finding?

Not in the way some coverage suggests.

The NTP’s 2024 systematic review concluded with moderate confidence that higher fluoride exposure is associated with lower IQ in children. Its clearest evidence involved exposures such as drinking-water fluoride above approximately 1.5 mg/L.

NTP’s current summary is unusually explicit about the lower-dose question:

There were insufficient data to determine whether 0.7 mg/L of fluoride in drinking water affects children’s IQ.

It also notes that the review evaluated total fluoride exposure from all sources and was not designed to determine the effects of artificially fluoridated drinking water alone. None of the IQ studies included in its meta-analysis was conducted in the United States.

The 2026 ECHO paper therefore does something valuable: it supplies new U.S. evidence in a range much closer to 0.7 mg/L.

But it measured a different outcome—fluid cognition rather than IQ—and still used an area-level water concentration rather than individual total exposure.

Calling that a “confirmation” of NTP collapses important differences between the studies.

A more accurate description is:

The ECHO result is directionally consistent with concerns raised by the NTP review and extends the evidence into a U.S. population at lower water concentrations, but it does not resolve the NTP’s uncertainty about whether 0.7 mg/L drinking water causes measurable IQ loss.

So is 0.7 mg/L fluoride safe during pregnancy?

The most defensible answer today is: a neurodevelopmental effect at 0.7 mg/L has neither been convincingly demonstrated nor definitively ruled out.

That may be less satisfying than a yes-or-no answer, but it reflects the evidence.

The U.S. Public Health Service currently recommends 0.7 mg/L for community water systems that fluoridate. The recommendation was designed to maximize prevention of tooth decay while limiting dental fluorosis. CDC’s currently posted guidance continues to support that concentration.

That is different from EPA’s drinking-water limit.

EPA’s enforceable maximum contaminant level for fluoride is 4 mg/L, with a secondary standard of 2 mg/L associated primarily with protection against dental fluorosis. Those regulatory numbers should not be interpreted as proof that fetal neurodevelopment is unaffected at every concentration below them.

It is also worth noting the chronology: CDC’s main fluoridation recommendation page is dated May 15, 2024, while the ECHO cognition study was published in August 2026 and the JAMA behavioral study in September 2026.

In other words, the existence of an unchanged government webpage should not be mistaken for evidence that public-health agencies have already reviewed and resolved the implications of these new results.

At the same time, one observational study is not enough to justify declaring an established prenatal hazard at 0.7 mg/L.

The next decisive research would ideally combine accurate public-water records with individual measures of total fluoride exposure, actual drinking-water consumption, repeated measurements during pregnancy, better characterization of non-water sources, standardized neurodevelopmental outcomes and preregistered exposure-response analyses.

Should pregnant women filter fluoride from their drinking water?

The 2026 study alone does not establish a medical recommendation that pregnant women should stop drinking fluoridated tap water.

For someone concerned about exposure while the science is being resolved, however, there are several factual points worth knowing.

First, determine what is actually in the local water. CDC says the local water utility is the best source, and utilities provide annual Consumer Confidence Reports containing fluoride information. CDC’s My Water’s Fluoride database also covers participating systems.

Second, an ordinary pitcher filter should not automatically be assumed to remove fluoride. Brita, for example, explicitly says its standard filters retain fluoride rather than removing it.

CDC says reverse-osmosis systems may reduce fluoride, although consumers should verify the particular system’s contaminant-reduction claims rather than assuming every RO unit performs identically.

Reducing fluoride exposure can also reduce fluoride’s cavity-prevention benefit, so a person choosing to alter drinking-water exposure should consider the broader dental-health picture rather than treating fluoride as having only one biological effect.

The bottom line

The viral story contains a real scientific finding wrapped in claims that are more certain than the evidence.

Verified: A large 2026 U.S. ECHO study found that higher estimated prenatal public-water fluoride was associated with lower fluid-cognition scores, with a nonlinear pattern that included a best-fitting change point near 0.675 mg/L.

Misleading: The study did not find a “3.36-point IQ loss.” The 3.36 number refers to points on a standardized fluid-cognition scale per additional 500 µg/L in the portion of one statistical model above the change point.

Not established: The study does not prove that 0.675 or 0.7 mg/L causes neurological harm, and its fitted change point should not be treated as a biological toxicity boundary.

Also not established: Existing evidence does not prove that prenatal exposure at 0.7 mg/L is harmless. The NTP has explicitly identified that exposure range as an unresolved evidence gap, and the new U.S. cognition finding makes that gap more important, not less.

The scientifically defensible position is therefore neither reassurance by default nor a declaration that fluoridation has now been proven to lower children’s IQ.

There is now a credible U.S. signal at policy-relevant concentrations that deserves serious replication with better individual exposure measurement. What we do not yet have is proof that fluoridated water at 0.7 mg/L causes a measurable loss of children’s IQ.

References and Further Reading

Primary research

• Simon et al. — Prenatal exposure to fluoride in public water and child cognition in the US ECHO cohort, 2006–2019, American Journal of Epidemiology (2026) — The primary study behind the viral cognition claim. Published August 14, 2026. At publication time, Oxford labels the online version an accepted manuscript pending final copyediting/typesetting.

• Columbia University Mailman School of Public Health — Higher Fluoride Levels in Public Drinking Water During Pregnancy May Be Linked to Differences in Children’s Cognition Scores — Official research summary specifying that 3.36 refers to the standardized fluid-cognition scale, describing the younger-child finding and detailing exposure limitations.

• Hernandez et al. — Prenatal Water Fluoride Exposure and Children’s Behavioral Problems, JAMA Network Open (2026) — Separate ECHO study of 5,520 children finding no overall association with internalizing or externalizing behavior, while detecting an externalizing association in a change-point sensitivity model.

• Malin — Ecological Fluoride and Neurodevelopment Studies—An Important Starting Point, JAMA Network Open (2026) — Invited commentary explaining why area-level fluoride studies remain vulnerable to exposure misclassification and why low-level effects remain unresolved. This is expert commentary, not an independent systematic review.

Evidence reviews and measurement

• National Toxicology Program — Fluoride Exposure: Neurodevelopment and Cognition — NTP’s current summary of its systematic review and meta-analysis. It reports moderate confidence in an association at higher exposures while explicitly stating that evidence is insufficient to determine whether 0.7 mg/L drinking water affects children’s IQ.

• National Toxicology Program — NTP Monograph on Fluoride Exposure and Neurodevelopment and Cognition — Primary federal evidence review published in 2024.

• Casaletto et al. — Demographically Corrected Normative Standards for the English Version of the NIH Toolbox Cognition Battery — Explains the NIH Toolbox’s fluid, crystallized and total cognition composites and their standardized scoring.

• Akshoomoff et al. — NIH Toolbox Cognitive Function Battery: Composite Scores of Crystallized, Fluid, and Overall Cognition — Validation paper showing what the fluid composite measures and distinguishing it from the broader total cognition composite.

Current U.S. guidance and practical information

• CDC — Community Water Fluoridation Recommendations — Current posted federal public-health guidance supporting a 0.7 mg/L Public Health Service target for fluoridated systems.

• U.S. EPA — Drinking Water Regulations and Contaminants — Current federal regulatory information, including fluoride’s 4 mg/L primary maximum and 2 mg/L secondary standard.

• CDC — My Water’s Fluoride — Explains how consumers can find the fluoride concentration reported for their local water system.

• CDC — About Home Water Treatment Systems — Current guidance explaining that reverse osmosis may reduce fluoride and that consumers should verify the claims of individual treatment systems.

• Brita — Replacement Filter FAQ — Manufacturer documentation stating that Brita’s standard filtration products retain fluoride. Included only to answer the specific consumer question about Brita filters.

Source of the viral claim

• Children’s Health Defense, The Defender — Babies Whose Mothers Drank Fluoridated Water During Pregnancy Scored Lower in Problem-Solving — Advocacy publication that described the study’s fluid-cognition coefficients as “IQ” losses and characterized the 0.675 mg/L modeled change point more causally than the underlying paper does. Included to document the claim being evaluated, not as neutral scientific evidence.

Editorial currency note: Regulatory and public-health guidance was checked through September 6, 2026. The two principal ECHO studies were published in August and September 2026, so agency recommendations may change as these and subsequent studies are reviewed.