Swimming in a (Fluoridated) Sea of Inadequate Data
On the history of fluoride and the new NTP Fluoride and IQ Meta-analysis
A new meta-analysis published in the JAMA Pediatrics suggesting that fluoride is harmful for children’s IQ is making waves. This post takes a longer look at the history of water fluoridation for dental health, the long-time opposition to fluoridation, the evolving but weak evidence on fluoride’s potential hazards, and the failure of the USA research establishment to seriously fund research on environmental exposures throughout key developmental periods.
I did my PhD in Nutrition at Cornell in the city of Ithaca and got my first and only cavity within a few years of living there. While I considered the cavity may have resulted from changing dietary habits (hello, gradschool stress-relieving sugar excursions), I also learned that Ithaca’ water was not fluoridated. I decided to look into the history of water fluoridation in Ithaca and discovered that Clive McCay, a famous nutrition researcher at Cornell (who is credited as one of the first showing that caloric restriction impacts lifespan in rodents) was an OG fluoride skeptic of sorts in the 1950s and 60s when national debates about water fluoridation were ongoing - he expressed concern on a radio program in Ithaca that he was not convinced of fluoride’s benefits to teeth and was worried that there were long-term effects on the thyroid and kidney. Decades later, Ithaca still didn’t have fluoride in the water and I had my first cavity. Thanks, Clive.
The history of fluoride begins with another McKay (though with a K, not C) - Fred McKay was an early 20th century dentist in Colorado Springs who noted that there were brown stains on many of his patients teeth (mottled teeth, now referred to as dental fluorosis) and he suspected it was from something in the water. Others across the globe would report a similar “Rocky Moutain Mottled Teeth” phenotype. McKay would ultimately gather together samples from communities with a high degree of tooth mottling and found that all were high in naturally occurring fluoride , confirming findings in rodents (1931) that was able to reproduce a mottled enamel with added sodium fluoride . The mottling was initially seen as a problem and researchers sought to find alternative water sources and ways to remove fluoride from the water; however, in the course of investigating the cause of the mottling, a dentist and researcher in the UK named Norman Ainsworth would note that in areas where teeth were mottled, they also had less dental decay - providing the first indirect hint that fluoride exposure might be protective against tooth decay. These observations would ultimately lead to a famous ecological investigation called the “21-City Study” conducted by HT Dean, a dentist appointed by the US Public Health Service - the 21 City Study was conducted in 4 states (indiana, Illinois, Ohio and Colorado) across 21 cities where the fluoride and other mineral contents of the water naturally varied. The data from this study was some of the first to indicate that protective effects of fluoride could occur at levels that resulted in minimal mottling (note: this evidence is considered relatively weak ecological data by modern standards)
These early observations would launch a broader effort by health authorities to pilot artificial water fluoridation (i.e. bringing areas with naturally low levels of fluoride up to levels ~1 ppm or 1g/L). Foundational trials in 4 cities receiving fluoridation (vs 4 neighboring ‘control’ cities) were initially selected for a 15 year trial, and 2 additional cities with natural fluoride ~1ppm were also selected to compare artificial vs natural fluoridation to. In 1945, Grand Rapids Michigan would become the first city in the world to artificially fluoridate its water supply (its neighbor city, Muskegon, was its control). By the late 1940s and early 1950s, results from the foundational trials showing lower numbers of cavities in children would roll in, including the Grand Rapids-Muskegon comparisons; the planned 15 yr comparison wouldn’t ultimately happen when Muskegon opted to withdraw from the trial and fluoridate its water. Other trials were also undertaken outside the USA, including ones in Canada, the Netherland, New Zealand, the UK and Ireland, all reporting reductions in tooth decay in artificially fluoridated areas and no major change in control towns. In 1950, The American Dental Association and US Public Health Service endorsed water fluoridation and a large number of cities would adopt this practice (by 1969, 43.7% of Americans were estimated to have access to fluoridated tap water). Optimal fluoride levels for dental health were initally set in the range of 0.7ppm to 1.2ppm, though notably, at this time, fluoridated dental products weren’t quite mainstream, so these levels were established with the idea that water fluoridation was the major source of fluoride alongside the smaller amounts that naturally occurs in foods (Crest technically released the first fluoridated toothpaste in 1956 but these wouldn’t become quite mainstream until the mid 1960s-1970s).
The history of community-level water fluoridation is typically considered a public health success - relatively clear health benefits for dentition were established through quasi-experimental studies and water fluoridation was rapidly taken up (note again: this data is not considered as high quality from modern perspectives; large, randomized cluster trials randomizing many communities to fluoridated water and modeling the data with more advanced statistical methods would be ideal). The success of water fluoridation didn’t occur without some opposition, well beyond Clive McCay’s modest concerns noted above. By 1957, Donald McNeil had written “The Fight for Fluoridation”, detailing the history of fluoride in the USDA and its various opponents, including layfolk, chiropractors, and religious figures. The reasons for opposition for varied, including libertarian style thinking about freedom of choice to consume unfluoridated water, to health concerns with some thinking fluoride is a poison being added to water, with the latter having a subgroup of activists who perceied water fluoridation as a communist plot to take over the country (truly, history loves to repeat itself). Gretchen Ann Reilly wrote an excellent chapter on this titled “Not A So-Called Democracy” in the “The Politics of Healing: Histories of Alternative Medicine in Twentieth Century North America”.
Fluoridation has continued to be a contentious topic, with virtually any and all ails that can affect the human body being ascribed to fluoride since its addition to the water supply. Complaints of fluoride causing teeth to fall out were reported to health officials in Grand Rapids, before the fluoride was actually added to the water. Strident (also phantom) claims around fluoride have existed well into recent history, with a community (Kuopio) in Finland in 1992 complaining of symptoms from water fluoridation during a period that the water fluoridation had unknowing been stopped - a great examples of how the human psyche can influence the perception of health effects and how readily we can be primed to find associations that aren’t causal.
Not all of the concerns about fluoride, however, are total crackpot theories. Before we dive into thinking about safety, let’s lay some ground work. Flouride is the ionic (F-) form of the element, fluorine, and its present nearly ubiquitously (soil, water, plants, animals)- i.e. it is something that is virtually impossible to not be exposed to. In debates about fluoride’s health effects, I’ve witnessed a general feeling of “I don’t want this in my body” in response to discussions of artificial water fluoridation, but it’s not really possible to not be exposed to fluoride because its everywhere - the scientific discussion is about how much you’re exposed to and through what sources. While there is concern about the ‘artificial’ nature of water fluoridation, it’s important to note that levels of naturally occuring fluoride in the water vary and you can readily find mineral waters with higher levels than artificially fluoridated water (the EU sets a max of 5mg/L and there must be labeling for mineral waters sold with >1.5mg/L) - as we’ll get to below, the signal for health concerns in human studies comes almost entirely from studies comparing naturally high vs low fluoride levels, not from community water fluoridation. The ubiqutiy of fluoride in the environment actually contributed to research in rodent experiments that indicated impaired growth with low levels of flouride diets and resulted in its consideration as an essential nutrient in 1974; however, while flouride is found naturally within the body from our exposure to it in the environment and accumulates in the skeleton, it is currently not considered essential for life ( there is no known essential physiological process which it supports and the evidence for its requirement for normal growth is not reproducibly seen in animal models).
In general, we think of fluoride as a ubiquitous exposure with potential health benefits (i.e. reduced tooth decay/cavities), but not required for life. This model is important, because for exposures without health benefits, we typically take a purely toxicological approach to its regulation: identifying the lowest levels associated with detrimental effects (LOAEL - lowest observable adverse effect level) and the level associated with no detrimental effect (NOAEL), divide the NOAEL by a safety factor (commonly 10-1000x), and set this as an upper limit of exposure to ensure a low likelihood of acute or chronic toxicity. However, given fluoride’s ubiquitous exposure and potential benefit, there is a need to identify a broad dose-response relationship to ensure levels of exposure are both safe and effective; it is expected that fluoride can be hazardous (i.e. there will be some level of exposure where harm is observed) but the goal is to assess risk (i.e. the concentration of fluoride exposure where that harm occurs) - we’ve long known that negative effects can occur from fluoride exposure because this all started with enamel mottling. The goal is to make sure that total exposure from dental products and water fluoridation (natural or artificial) provides the maximum benefit that can be provided while not negatively impacting any other health outcome.
When it comes to toxic effects of fluoride (or any compound for that matter), we typically think about acute toxicological outcomes (e.g. vomiting, respiratory or cardiac problems, death), chronic exposure outcomes (eg developing cancer) as well as reproductive and developmental outcomes (e.g. does exposure influence the ability to get pregnant or carry pregnancies to term; does it alter the development of organ systems and their functions). There is a reasonable literature base that has investigated fluoride exposure in relation to a number of acute, chronic, reproductive and developmental outcomes in rodents and identified NOAEL and LOAELs with varying but reasonable degree of confidence for specific outcomes; fortunately, the data largely demonstrates that total human fluoride exposures - not just from fluoridated water but food and dental products - at the high end of exposures (between 50-120ug fluoride/kg body weight per day) are still well below the NOAEL and leave a margin of error ~20-50 fold (i.e. humans with very high exposures from highly fluoridated water ~4mg/L still do not reach the doses on a per kilogram basis that still caused no effects in mice). Much of this toxicological evidence was reviewed by the National Acadmies of Sciences in 2006 when reviewing the EPA’s maximum contaminant levels for fluoride in water (ie not determining the beneficial amount but the maximum amount where there is not apparent risk). The committee would ultimately recommend to reduce the maximum contaminant level goal for flouride to less than its current standard at the time of 4mg/L - there wasn’t consistent convincing evidence for concern for any health outcome apart from the risk of dental flourosis (mottling) from children with high exposures. The impetus for this report was not entirely artificial fluoridation levels (where levels are raised to 0.7-1.2mg/L) - it was primarily driven by concerns for vulnerable subpopulations including over a million people in the USA with natural fluoride concentrations between 2-4+mg/L that are also receiving exposure from use of fluoridated products. The committee was ultimately concerned that some children were at risk for enamel mottling (and potentially brittle bones more at risk of fracture from high life long exposure) with maximims of 4 mg/L and thus, recommended lowering this level.
The National Academies committee reviewed the literature on many other health outcomes at the time, using both animal models of exposure to fluoride as well as epidemiological associations of fluoride exposure and child health outcomes - these human studies are non-experimental/non-randomized, and in medicine we typically think of them as much lower quality, but given the ethical issues in randomizing individuals to fluoride to look for negative health outcomes, the risk assessment relies heavily on these imperfect lines of evidence. The committee found, for example, little concern for reproductive effects due to adverse effects only occurring at very high exposure levels unlikely to be encountere by anyone in the USA. It noted, however, that emerging data out of China, where children regularly exposed to naturally highly fluoridated drinking water (2.5-4+mg/L), reporting IQ deficits, though the quality of these studies were limited. Animal studies available to the committee at the time were limited, frequently used high doses (>10mg/L) and found changes in anatomical and molecular features in the brain but only small effects on behaviors - the committee noted the need for study relevant concentrations on a wide range of cognitive domains and better understand mechanisms. The results of this 2006 report are frequently cited by activist groups like the Fluoride Action Network as evidence of concern for fluoride and IQ and were cited in the National Toxicology Program (NTP)’s 2015 impetus for undertaking a broader effort to assess fluoride in relationship to health outcomes like neurocognition, cancer and endocrine disruption. Thus, we arrive at the modern controversy.
The road to the current NTP’s JAMA Pediatrics meta-analysis is quite long and its actually just a small part of the story on fluoride and neurocognition, a story that will continue to evolve as the NTP assesses other health outcomes like endocrine disruption and cancer. It is generally a good and reassuring thing that our federal agencies devoted to Toxicology are trying to find whether fluoride is hazard for specific outcomes and at what doses these occur - but the story is not without some controversy. The saga for fluoride and neurocognition spans over 9 years, with NTP having already drafted a systematic review of the animal research by the time it proposed evaluating flouride and IQ in 2015. For those unfamiliar, 9 years is a pretty long-time for all of this work (even the government). The JAMA Pediatrics meta-analysis is one component of a broader monograph put together by NTP on fluoride’s relationship to neurocognition and development.
Part of the reason for the lengthy process was that the monograph drafts that were put together, containing much of the data in the JAMA Pediatrics meta-analysis but also including other neurocognitive measures as well as animal data, were reviewed not once but twice by the National Academies of Science - the committee that reviewed the monographs took issue with aspects of the systematic review process, the methods and interpretation by NTP. I recommend reading the NAS reports because the commentary was pretty brutal and, while constructively responded to by NTP, ultimately revealed errors that are relatively sophomoric and cast a grey shroud of doubt about the whole process (and explains the long delays) - the NAS took issue with a lot in the NTP monographs, particularly for the human data: the protocols for the systematic reviews not being clearly being stand alone documents that could be transparently and reproducibly replicated; the double counting of studies from the same population in the meta-analysis (a big no no); inconsistency & a lack of transparency in the risk of bias assessment and how study quality was determined (e.g. whether studies were downgraded for specific characteristics, such as a lack of adjustment for other neurotoxic exposures like lead and arsenic). The NAS committee also pointed out issues with the methods of assessment of fluoride exposure and cognitive outcomes in the literature and how well the NTP was accounted for these limitations in its assessment. Probably the most relevant criticism that gets to the heart of the matter is that the NAS noted that the NTP monograph’s effort was only ever set out to identify a Hazard (i.e. can fluoride be neurotoxic), not a risk (i.e. what is the dose response relationship between fluoride exposure and cognition and at what does does this risk emerge) - indeed, the NAS committee noted that the overwhelming majority of the little comes from communities with high fluoride in the water (>2.5mg/L) and is not relevant to exposures <1.5mg/L (recall: water fluoridation levels in the USA are recommended by the CDC to be 0.7mg/L). Broad conclusions from NTP not clearly distinguishing between hazard and risk are very likely to confuse the public - we see this constantly and i’ve talked about it before as an area where the WHO’s IARC gets frequently critiqued in my post on aspartame.
The finalized monograph, accounting for the NAS criticisms, was finally published in 2024 - despite some quiet chatter online and advocate groups citing it as a slam dunk against water fluoridation, the monograph directly states that it can’t be used to say anything about fluoridation at typical water exposure levels in the USA. While the monograph looked at many outcomes, there was very limited data for human studies with outcomes besides IQ, and thus, we arrive at the JAMA Pediatrics meta-analysis that reports these findings.
The meta analysis relied on 64 cross-sectional and 10 prospective cohorts that assessed both group-level measures of flouride exposure and IQ, as well as individual level. Group level exposures here rely on fluoride levels in the water supply and individual level exposures rely on urinary fluoride measures (though there are concerns about interpreting these as truly individual level, since an individual’s urinary fluoride levels are not independent of the group, since common water sources strongly determine urinary fluoride levels). At the group level, averaging across individual studies that compared higher vs lower fluoride exposures demonstrated a very small effect on IQ (-0.45 mean difference), with extremely high heterogenity across studies (notably this heterogeneity went unexplained in all of the authors’ attempts to, and raises concerns about why the effect of fluoride differs so much from study to study — and whether its really fluoride that’s causing these effects).
Effects were generally even smaller when considering low risk of bias studies - NTP has a broad risk of study quality criteria to determine risk of bias, including adequate reporting of study details, consideration of key covariates (e.g. socioeconomic status, arsenic and lead exposure), the use of sophisticated sampling techniques (to ensure a broad capturing of sub-groups within the population) and more rigorous statistical methods - again, these studies are all observational and assessing group-level exposures, so you want to have a literature that tries to account for various sources of bias in their designs & statistically as best as possible.
Studies with urinary measures of flouride also showed an inverse relationship with IQ (note the unit of exposure here is per 1mg/L):
The NTP attempted to get at dose-response relationships as well and things start to get a bit more interesting. In the previous analyses, individual studies are comparing higher vs lower exposed groups, but we can’t really place a number of the exposure level. This is hard to get at in these types of studies that are being summarized but we can start to get a feel for whether higher vs lower fluoride exposure matters depending on the water fluoride levels reported in individual studies - we can see that there’s not really strong dose response relationships (e.g. <4 is not consistently a bigger effects than <2) and the effects are weakened substantially (and in the case of water fluoride, not statistically significant) for studies in areas with water fluoride levels <1.5mg/L (again, CDC recommends 0.7mg/L).
What do we make of this data? You can search twitter and find divergent results, with both science communicators trying to contextualize them in an ever politicized era of science, and activists groups cheering on that they were right all along about fluoride. You can see the differences quite clearly in 2 accompanying editorials to the meta-analysis as well: one that is somewhat championing the hazard posed by fluoride and the need to re-assess safety thresholds and public policy, and one that dissects the limitations of this literature base and its limited relevance to the USA.
Here’s my take:
USA Relevance: Not a single study in this meta-analysis comes from the United States, leaving the relevance to our society/policy pretty suspect.
Low data quality: These data are messy, to say the least. The majority of the data are cross-sectional in nature - we don’t expect randomized trials here but there are only a handful of prospective studies (where fluoride exposure is assessed prior to the measurement of IQ) with the majority at a higher risk of bias. There is deep uncertainty about total fluoride exposure levels, relying on water levels as a proxy, and we can’t say much about whether these exposures occurred during key developmental stages for brain development. Ultimately, this weaker observational nature of the evidence-base relying on a non-random ecological exposure has a much higher risk of confounding than we’d like. The risk of confounding is a very legitimate concern and not hard to find potential reasons for confounding that are nearly impossible to adjust for due to their collinear relationship with fluoride levels - for example, more affluent areas can filter fluoride out of the water supply and more affluent areas typically have a higher average IQ (for a multitude of reasons from high heritability of IQ to various other nutritional and environmental effectors of IQ). Higher quality prospective data from New Zealand with more relevant fluoride exposures, comparing much more similar populations than what is done in much of these data in this meta-analysis drawing from China, and employing more rigorous adjustment for confounding variables has failed to find any relationship between water fluoridation and IQ. There’s a real issue here that many many low quality studies put into the same meta-analysis yield a forest plot that looks overwhelming - but juxtaposing a bunch of low quality studies together does not obviate their limitations and can drown out the seeming importance of a single, high quality well done studies.
Dose-response: You can see how limited the relevance is to the US when you consider the water fluoridation dose-exposure categorizations, where you rapidly drop to just a few studies when you consider those with fluoride levels below 1.5mg/L and even fewer with a lower risks of bias. The impacts on IQ are either non-existent or extremely small at these levels and challenging to interpret when comparing to relevant USA settings - for example, if a community with low flouride (0.2 mg/L) wants to go up to protective levels (0.7mg/L), is there really a detectable impact on neurocognition? This meta-analysis doesn’t allow us to directly estimate the effect but it would be small if at all existent.
Biomarkers: The urinary fluoride measures here might look like a strength as an objective biomarker, rather than using community water levels as a proxy of exposure, but they’re actually quite weak. Biomarkers are highly imperfect - they’re subject to alterations in absorption and distribution in the body, the rate of excretion, and impacted by other exposures that may be confounders. For fluoride, we know that in kids about half of ingested fluoride is found in the urine (~75% for adults, probably cause kids retain more in their still-growing skeletons) and that the amount can vary considerably from day to day, being impacted by fluid intakes as well as other factors such as the pH of the urine - more alkaline urine increases the excretion of fluoride and factors like animal protein intake (which reduces pH), medications, altitude and disease can all impact urine pH and fluoride excretion rates. These are all confounders in their own right potentially for IQ, and indicate that you’d need several days worth of complete, 24 hour urines to start to approximate fluoride exposure. The reality of this literature base, however, is that it nearly entirely relies on a single spot urine sample (ie just a single void, not a 24 hr sample and thus subject to lots of variation in how much ingested fluoride ended up in that spot sample) and are not always adjusted for creatinine (this would account for how dilute the sample is, although creatinine can itself be a confounding variable). Experts actually recommend against using a 24hr urine for the estimation of individual fluoride intakes due to the wide confidence intervals around the intake/excretion relationship - the single spot urine is only worse and is a measure that is close to uninterpretable.
The issue of the biomarker data in the paper gets even worse - the IQ deficits associated with increasing urinary fluoride levels look pretty big in the paper but this is due to the unit they use (expressed per 1mg /L of urinary fluoride). This is a very large increment to consider and overestimates the theoretical IQ deficits that a population risks, especially trying to translate this to an appropriately fluoridated water supply in the USA. The WHO places optimal urinary fluoride concentrations at 0.8-1.2 mg/L, and nationally representative data from the USA place mean urinary fluoride levels at 0.67 mg/L with a standard deviation of 0.54 mg/L. Considering a 1mg/L increment’s effect on IQ from an inappropriate biomarker is vastly overstating the impacts of more subtle changes in fluoride intake and differences between groups in the population (with the exception of comparing very naturally low and very naturally high fluoridated areas, the only setting where this sort of data even begins to suggest some concern for me).
Where is the high quality data? - Arguably the most damning part of this paper is what is not present - large, nationally representative cohort data, reflecting the variety and ranges of fluoride exposures in America, that repeatedly measures 24-hr urinary fluoride exposure during critical developmental windows (during pregnancy, throughout early infancy, childhood and adolescence) and assessed a battery of cognitive domains. Such a cohort could be mined for not only fluoride exposure but for many environmental and nutritional exposures, and meaningfully inform USA policy. The fact we don’t have such a cohort is a massive failure of the USA biomedical research establishment - this is in part due to our largely funding discovery and basic sciences but there actually was a large study similar to what I’ve theoretically described here called the National Children’s Study - the NCS was mandated by the Children’s Health Act of 2000 and aimed to provide more human data behind a lot of the basic science and animal models that indicated that there are developmentally sensitive windows (e.g. pregnancy, childhood) where environmental insults (e.g. toxicants, poor nutrition) can severely impact adult health. Unfortunately, the study was shut down in 2014 after 1.2 billion dollars was spent and there’s little to be shown for it - the reasons for its failures are numerous . Despite the flop that NCS was, the scientific rationale for such a cohort is still valid and we still lack a national representative cohort to ask key questions about exposures and devleopment; national data like those from NHANES are only cross-sectional and have limited ability to answer questions about fluoride exposure across development and its impact on childhood cognition.
The current meta-analysis, at the very best, points to a signal of a small hazard from particularly high fluoride exposures on IQ; at its worst, it’s a classic ‘garbage in, garbage out’ meta-analysis that throws a bunch of low quality data with limited ability for causal inference and pretends that compiling non-rigorous research makes it more rigorous. It mostly points to deep uncertainty about what levels of fluoride exposure might actually be risky, with the evidence within not entirely exonerating current low level recommendations but also being light-years away from being useful to change policies that have demonstrated benefits for dentition. I look forward to more rigorous, prospective cohort data on this topic, as well as more comprehensive animal data (see the NTP’s August 2024 monograph where they found existing animal data to be inadequate for neurocognitive outcomes and are planning follow up experiments to fill gaps).
As a final point, you’ll note in the public and in the editorials, there is an immediate discussion of the relevance of this meta-analysis to water fluoridation efforts, despite none of the studies even remotely supporting that a community fluoridating its water up to 0.7mg/L is risky. Indeed, the vast majority of the meta-analysis, if taken completely uncritically, would suggest that naturally occurring high levels of fluoride are risky. Why is the focus on critiquing government policy and human efforts to ‘artificially’ add fluoride to water to a controlled level and not shouting from the rooftops about adding water filtration to local communities with high fluoride levels (>1.5mg/L) to purportedly protect child brain development? There’s little more that highlights the inherently politicized nature and underlying biases in this conversation.
Don’t miss the StatNews coverage of this issue (I gave some quotes).