A new study reporting the ranges of values for kidney and liver parameters in a healthy population is being actively misrepresented by anti-fluoride campaigners. The Fluoride Action Network’s (FAN) latest bulletin claims the study shows “that fluoride at commonly experienced doses can damage the kidneys and livers of adolescents.”
The study shows nothing of the sort. How could it – individuals suffering liver or kidney disease were specifically excluded from the study population. The reported parameter values are all for healthy individuals.
Readers can check for themselves – there is a free download. The paper is:
Malin, A. J., Lesseur, C., Busgang, S. A., Curtin, P., Wright, R. O., & Sanders, A. P. (2019). Fluoride exposure and kidney and liver function among adolescents in the United States: NHANES, 2013–2016. Environment International,
It is important to understand what this study really found. Not only is it being misreported by anti-fluoride activists. The University (The Mount Sinai Hospital/Mount Sinai School of Medicine) press release also appears to attribute more to the study’s findings than is warranted. This is a common problem with university public relations departments. (Readers are warned – the press release includes the disclaimer:
“AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system”
Below I list some information on the study
This is not a study about kidney or liver disease
Individuals showing such disease were specifically excluded. The study reports values for kidney and liver parameters in “generally healthy” subjects. The authors make this very clear in the discussion saying:
“this study did not aim to determine whether fluoride exposure is associated with clinical decrements in kidney function among U.S. adolescents. Rather, this study aimed to examine subclinical changes in kidney or liver parameters associated with fluoride exposure among a generally healthy population. For example, the lowest GFR estimated in this study was 84 mL/min/1.73m2, and therefore none were below the<75 mL/min/1.73m2 value considered reflective of
abnormal kidney function. Future prospective studies including participants with and without kidney disease are needed to assess clinical changes in kidney or liver function.”
So, this study just could not have identified factors causing kidney and liver disease, let alone confidently attribute a cause to the disease. So we can reject the anti-fluoride activist’s claims and their misrepresentation of the study results.
But why all this fuss about fluoride?
Because the authors have a preoccupation with fluoride they used statistical analyses to see if they could find any association between drinking water fluoride or blood plasma fluoride and the measured kidney and liver parameters. They did find a small number of very weak associations.
They do not support the claims made by anti-fluoride activists so details of their results and a critique of their results are irrelevant to the main arguments. But I do have a hangup about the way statistical analyses are used, and the way they are over-interpreted to support pet biases so will discuss their data below.
Very few associations found
The study included nine kidney and liver function test parameters. Only one of these (Blood Urea Nitrogen [BUN]) had a statistically significant relationship with water fluoride (Uncorrected p <0.001) – see figure below.
The relationship of BUN with blood plasma F was also statistically significant (Uncorrected p <0.001) – see figure below.
The Standard Reference Range of BUN for this adolescent population is 6–23 mg/dL. Only a few data points are outside that range and they mainly occur for low water F or plasma F concentrations.
The authors also reported statistically significant associations of estimated glomerular filtration rate (eGFR) and Serum uric acid (SUA) with blood plasma F. However, once adjustments were made for plasma cotinine levels (a biomarker of tobacco smoke exposure) associations were not statistically significant (uncorrected p=0.18 for eGFR) or only “marginally” statistically significant (uncorrected p=0.06 for SUA).
In effect, statistically significant associations with either water F or plasma F occurred for only one. It is not credible for FAN to use these associations as indicators “that fluoride at commonly experienced doses can damage the kidneys and livers of adolescents.”
Reported associations may be “a pure act of will”
The authors appear to place a lot of reliance, in my opinion far too much reliance, of p values as somehow providing a causal mechanism behind the reported associations. This reliance has been strongly criticised by statisticians. Recently Briggs (2019) (Everything Wrong with P-Values Under One Roof) concluded:
“P-values should not be used. They have no justification under frequentist theory; they are pure acts of will. Arguments justifying p-values are fallacious. P-values are not used to make all decisions about a model, where in some cases judgment overrules p-values. There is no justification for this in frequentist theory. Hypothesis testing cannot identify cause. Models based on p-values are almost never verified against reality. P-values are never unique. They cause models to appear more real than reality.”
He goes on to elaborate:
“a small p-value has no bearing on any hypothesis . . . Making a decision about a parameter or data because the p-value takes any particular value is thus always fallacious . . . . Decisions made by researchers are often likely correct because experimenters are good at controlling their experiments, . . . . . ., but if the final decision is dependent on a p-value it is reached by a fallacy. It becomes a pure act of will.”
I believe Malin et al., (2019) place too much reliance on the p values they obtained and should have provided more complete results from the statistical analyses. Citing and relying on p values alone is, I believe, a major deficiency in this paper.
To their credit, while not providing full statistical analysis results the authors did display individual data points in their figures 1 and 2. This enables careful readers to make some judgments about the statistical analyses which would not be possible if the figures had not been provided.
Problems with outliers
The figures show a small number of outlying data points with some of the parameters. One has to be very careful that any association found only has a low p-value because of the influence (or leverage) of these outliers. The figures above for the BUN parameter illustrate the problem – particularly for water F where 2 data point greater than 6 mg/L clearly have a lot of influence.
This problem should stand out to any informed reader of the paper. The authors claim “Cook’s distance estimates were used to test for influential data points; none were identified.” However, this does not seem credible (particularly for Water F) so it is understandable that I should ask to see the results of these estimates so I can make up my own mind. They were not provided.
The associations were extremely weak
There is a huge scatter in the data points obvious in the figures above. This tells us that the reported associations can explain only a small amount of the variance. This is one reason why p-values alone can be misleading. A low p-value for an association (or fitted line) explaining only a few percent of the variance is meaningless. Concentration on such associations means that more important ones (explaining more of the variance) may be ignored. It also ignores the fact that the risk-modifying factor (in this case water F or plasma F) may simply be acting as proxies for more important factors (see Perrott 2018 for an example of this).
Malin et al., (219) should have provided more complete statistical analyses results to help readers judge the strength of the reported association. however, the figures themselves enable us to conclude the associations are very weak.
It is misleading to use the statistical result predictively
Malin et al., (2019) appear to “predict” the effect of fluoride on liver and kidney parameters, particularly BUN. They write in their abstract:
“A 1 mg/L increase in water fluoride was associated with a 0.93 mg/dL lower blood urea nitrogen concentration (95% CI: −1.44, −0.42; p=0.007)”
“1 μmol/L increase in plasma fluoride was associated with . . . . . a 1.29 mg/dL lower blood urea nitrogen concentration (95%CI: −1.87, −0.70; p < 0.001).”
But consider going from 0 to 1 mg/L in the image above for water F. The fitted line suggests that BUN would drop from about 11 to about 10 mg/dL. Taking the 95% CI interval into account the line “predicts” a value in the range of about 9.56 to 10.58 mg/dL. But only a small number of the points scattered at about 1 mg/L F have values in that range.
[Yes, I know. The authors only refer to associations, but reports of this work in the alternative health media are using these statements as predictions and that is how activists are suing the information.]
All that the best fit line can predict are values which fit the line. As the association represented by the best-fit line explains only a very small percentage of the variance (despite the small p-value) these “predictions” are meaningless. Unfortunately, the authors do not make this clear in their paper and this deficiency only contributes to the ability of anti-fluoride activists to misrepresent the findings.
Anti-fluoride activists are misrepresenting the finding reported in this paper. The authors themselves stress that their study was not designed to determine if fluoride exposure is associated with, or causes, declines in kidney or liver health. The FAN claim that the study shows“that fluoride at commonly experienced doses can damage the kidneys and livers of adolescents” is completely incorrect.
That is all we need to know regarding the way activists are misrepresenting the study. However, a closer look at the data suggests that the associations with fluoride for healthy individuals reported in the paper are extremely weak.