Young-Geier Autism Study: What the—? (Part 4)

This post represents the fourth (and god willing, final) installment of my review of the study by Young et al, which links exposure to Thimerosal-containing vaccines with autism. Previous posts on the subject can be found here (IRB approval), here (IRB follow-up), here (article Introduction), here (Materials and Methods), and here (Results). The current post provides commentary on statements made in the article's Discussion.

The authors begin by pointing out the favorable aspects of their study.

[T]he study design...helps to strengthen the observed results. The medical conditions examined were selected a priori as biologically or not biologically plausibly linked to Hg exposure from Thimerosal-containing vaccines administered during specific exposure windows.

Not to be too snide here, but when was it decided that foregone conclusions are study strengths?

Additionally the study design also allowed us to be certain that virtually all exposures to Hg preceded the diagnoses of the diseases examined (ie, allowing for a potential cause-effect relationship between exposure and disease). This was ensured because only children receiving a vaccine by age 3 months were examined, and as Table 2 shows the median age of initial diagnosis for the conditions examined in the present study ranged from 1.7 to 6.4 years-old.

The diagnoses in question may follow vaccine exposures, but that doesn't necessarily mean their existence does. For instance, the presence of congenital anomalies (which, by definition, exist at birth) should not be affected by subsequent exposure to vaccine-related ethylmercury. However, if Young et al were consistent in their thinking, they would propose that a 100-µg increase in ethylmercury exposure somehow confers a retroactive, protective effect against congenital anomalies (with a risk reduction of ~40%).*

[T]he methods of ensuring capture of Hg exposure from Thimerosal-containing vaccines and outcomes appear to have yielded results consistent with previous studies. For example, the US CDC previously published that the overall prevalence of ASD [autism spectrum disorder] in children, born for similar birth years as examined in the present study, was 3.4 per 1000 children in the metropolitan Atlanta, Georgia area [13]. The adjusted overall prevalence of ASDs (3.67 per 1000) in the present assessment of the VSD was consistent with the previous observation made by the US CDC.

The CDC investigators examined multiple medical or educational records from 5 Atlanta-area counties and calculated a 0.34% prevalence rate for autistic disorder, pervasive developmental disorder (NOS), or Asperger disorder (per DSM-IV criteria) among children 3-10 years of age in 1996. Young et al's overall (unadjusted) prevalence rate for ASD in the VSD was 0.293%, and the adjusted prevalence rate was 0.367%, after adding approximately 206 cases (see explanation here). Calculations for the year 1996 indicate an unadjusted rate of autism (ASD data for 1996 were not provided) of 0.17% and an adjusted rate of 0.33%, after Young et al added 80 cases. So Young et al's prevalence rates are "consistent" with the CDC's (circa 1996), after their liberal imputation of autism cases.

[T]he birth cohort years examined in the VSD help to strengthen the results observed. The birth cohort years examined from 1990 through 1996 occurred many years prior to the raising of concern about potential problems with Thimerosal in childhood vaccines by the American Academy of Pediatrics and the US Public Health Service, so that their announcement to remove Thimerosal from childhood vaccines in July of 1999 should have had virtually no impact on physicians' thoughts about Thimerosal in childhood vaccines.

While the above statement is not entirely unreasonable, it doesn't acknowledge the groundswell of anti-Thimerosal sentiment from lay and fringe medical groups during the 1990s. Therefore it is possible, as suggested in a prior post, that parents within the Kaiser system requested, or even demanded, the administration of some newly available, Thimerosal-free vaccines to their children during 1997 or later. 

Additionally, the years examined in the VSD help to ensure that changes in diagnostic criteria for outcomes such as autism that came into effect in 1994 would have minimally impacted the present study since most children examined were diagnosed post-1994 with autistic disorder.

"Minimally impacted" is a disingenuous conclusion, in my opinion. According to a recent post by EpiWonk, studies indicate that the rise in autism rates during the last 3 decades is an artifact due to broadened diagnostic criteria, diagnostic substitution, and the diagnosis of autism at a younger age. Moreover, there's no reason to assume that the rate of recognizing autism would have leveled off after 1994. In fact, the increasing prevalence of autism depicted by Young et al in Figure 1 is probably due to a continuing uptick in diagnostic acuity.

[A]nother significant strength of the present study stems from the trends in birth cohort Hg exposure and outcomes. As shown in Fig. 1 for autism, it was observed there were increasing/decreasing trends in exposure and outcomes across the birth cohort years examined, and that for the neurodevelopmental disorders there were significant associations between birth cohort mean Hg exposure and disease prevalence rates. It is important to note that the increasing/decreasing trends in Hg exposure were not simply the result of random yearly fluctuations in vaccine uptake rates or even simply the result of increasing exposure to vaccine antigens, but instead reflect known changes in the Hg content of the US childhood vaccine schedule...

It's difficult to understand how Young et al can so easily dismiss the idea that "the increasing/decreasing trends in Hg exposure were not simply the result of random yearly fluctations." Note that the largest difference in the average per-person ethylmercury exposure is pretty small, approximately 35 µg according to Figure 1 (~110 µg for the 1990 birth cohort and ~145 µg for 1992), which equals a little more than 1 vaccine dose. Also note that there are many ways to represent the data in Figure 1, so that the putative association between ethylmercury exposure and the prevalence of autism is less graphically compelling (see examples below).

[B]ecause of the ecological nature of the study design, we were not able to link vaccine exposures across individual patient records...

EpiWonk explains the important distinctions between an ecological, or group-level, study and an individual-level study. He argues that Young et al have committed the "ecological fallacy."

Another possible limitation of the present study was the potential for under ascertainment of a child's total Hg exposure...Other potential sources of Hg such as fish consumption or environmental exposure, while potentially significant to the risk of a child being diagnosed with a neurodevelopmental disorder, could not be examined...We believe that these other exposures to Hg should not have biased the effects observed. In actuality, such sources of Hg exposure would potentially minimize the significance of the effects observed.

Here Young et al appear to contradict themselves. If I'm reading correctly, they first state that non-vaccine sources of mercury "should not have biased" their results, but they follow with the conclusion that non-vaccine mercury would "potentially minimize" their results.

This study was also limited to a maximum of 4 years of follow-up time for the latest birth cohorts...

Young et al's highly dubious imputation of cases for later birth cohorts, owing to shorter follow-up periods, has been discussed by EpiWonk and myself.

Finally, the present study was not able to adjust for potential factors that might have resulted in vaccine avoidance but may have predisposed one towards neurodevelopmental disorders under study. Specifically, Fine and Chen reported that there are several social and medical attributes associated with avoidance or delay of vaccination and an increased risk of neurological adverse events, and that confounding of this sort is a general problem for studies of adverse reactions to prophylactic interventions, as they may be withheld from some individuals precisely because they are already at high risk of the adverse event [14]...As a result, the effects observed in the present study may represent an underestimate of the true effects of Hg exposure from Thimerosal-containing vaccines on the risk of neurodevelopmental disorders.

Not to sound like a broken record, but What the—? Young et al cite a 1992 CDC article on confounding (an issue discussed by EpiWonk here). But as far as I can tell, Young et al have taken the article's conclusion—that is, putative vaccine benefits may be observed because fewer high-risk children are vaccinated—and turned it on its head. (BTW, this phenonmenon may well explain the observed low rate ratio for congenital anomalies and increased ethylmercury exposure.)

Young et al then go on to support their findings by citing themselves (Geier and Geier) on 3 occasions, a "neutral" 2003 study by Verstraeten et al (who ultimately concluded that further study was needed), and a 2003 study by Stehr-Green et al, who concluded that the data do not support an association between Thimerosal exposure and autism.

Young et al also attempt to undermine the applicability of studies from the United Kingdom, Canada, Sweden, and Denmark (which did not find an association between Thimerosal exposure and autism) to the United States. They write, "In many of these countries, alternate vaccines in different vaccine schedules and different diagnostic measures were used, and many countries apparently had very different neurodevelopmental disorder prevalence rates than in the US." However, Young et al fail to note that WHO found these studies convincing and dismissed 2 US studies (at least one of which was by Geier and Geier) as "unconvincing."

Mercury exposure was also observed to be significantly associated with neurodevelopmental disorders and autism in a series of epidemiological studies...

As in their introduction, Young et al conclude by attempting to link Thimerosal (ethylmercury) exposure to environmental methylmercury from a number of sources including seafood (for instance, in the Faroe Islands) and Ecuadorian gold mines. They also cite one California Department of Public Health study, which found associations between autism spectrum disorder and a number of airborne substances, including elemental mercury.

Young et al continue to cherry-pick or outrightly misrepresent data (again, in my opinion) from other sources, including 1) a primate study, which concluded that methylmercury "is not a suitable reference for risk assessment from exposure to thimerosal-derived Hg"; 2) mouse models, which suggested that Thimerosal-induced neurotoxicity is breed dependent; and 3) a rat study, which examined the perinatal neurotoxic effects of methylmercury.

Other studies are cited, including 3 more by Geier and Geier, which attempt to link elevations in mercury concentrations and "potential biochemical and genomic susceptibility factors to Hg poisoning" with autism. One of the Geiers' articles alleges that urinary porphyrins are a marker for heavy metal exposure, a favorite hypothesis featured on YouTube.

Summarizing the drawbacks of Young et al's study and write-up is a considerable challenge, frankly because there are so many. Nevetheless, I'll attempt to list the most salient here:

• The a priori association between microgram doses of ethylmercury and autism. This association really isn't anymore biologically plausible than an association between any trace substance and autism.
• The loose association between methylmercury and ethylmercury, two substances with considerably different pharmacokinetic properties.
• The liberal addition of neurodevelopmental cases, particularly for the later birth cohorts of 1995 and 1996.
• The absence of controlling for confounders, such as birthdate, and the alleged commission of "ecological fallacy" (per EpiWonk). Controlling for birthdate in a study of autism is essential, because of time-sensitive changes in the diagnosis of autism.
• The possible overestimation of ethylmercury exposure from vaccines, particularly among later birth cohorts.
• The presentation of astonishingly high rate ratios (which have to be taken on faith) that have very wide confidence intervals, for neurodevelopmental disorders. These rate ratios suggest, in some cases, several-hundred-fold increased risks of disorders due to very small increases in ethylmercury exposure—a dubious proposition, in my opinion.
• A rate ratio that suggests (although this is not addressed by the authors) that ethylmercury exposure actually reduces the risk of congenital anomalies by approximately 40%.*
• The use of the authors' previous questionable studies to support the current study.
• The authors' cherry-picking or outright mispresentation, in my opinion, of other references to support their beliefs.

*This observation may well be the result of confounding, as discussed by Fine and Chen (which Young et al ironically cite for other reasons, later in their Discussion).

Photo: iStockPhoto.