Friday, December 16, 2005


I don’t write much about the emerging science looking at hypothesized biological causes of Autism Spectrum Disorders (“ASD”) for a variety of reasons. One of those reasons is that, as a lay person, I don’t want to be perceived as giving unqualified expert opinions. But even lay people are capable of understanding the basics of science, just as lay people are capable of understanding the basics of my field (law) given a little effort. Parents of autistic children need to take note of ongoing scientific developments if they want to understand what is going on with their children. A very interesting new study has been the subject of internet discussion the last couple of days that deserves mentioning.

Researchers, based at the University of Texas Health Science Center set out to determine if a link can be drawn between environmental releases of mercury and the incidence of autism. The abstract reads as follows:

The association between environmentally released mercury, special education and autism rates in Texas was investigated using data from the Texas Education Department and the United States Environmental Protection Agency. A Poisson regression analysis adjusted for school district population size, economic and demographic factors was used. There was a significant increase in the rates of special education students and autism rates associated with increases in environmentally released mercury. On average, for each 1000lb of environmentally released mercury, there was a 43% increase in the rate of special education services and a 61% increase in the rate of autism. The association between environmentally released mercury and special education rates were fully mediated by increased autism rates. This ecological study suggests the need for further research regarding the association between environmentally released mercury and developmental disorders such as autism. These results have implications for policy planning and cost analysis.

The implications of this study, scheduled for future publication in Health & Place, are obvious. If environmental mercury has a casual connection to autism, then shouldn’t the same hold true for the injection of thimerosal into our children? Critics will argue that a possible link to environmental mercury discharges diminishes the impact of thimerosal by showing an alternate causal source. (The underlying causation theory of this study is that the burning of fossil fuels discharges inorganic mercury into the air, which, after being deposited onto soil or water, is converted into toxic methylmercury by chemical reactions or exposure to bacteria.) On the other hand, common sense (there I go again, resorting to common sense) dictates that if environmentally released mercury poses a danger, wouldn’t mercury injected directly into children be even more dangerous? And assuming the danger posed by mercury in general, it’s hard to argue against thimerosal being the most pervasive source of mercury exposure, at least prior to the much ballyhooed, and debateable, cut-back of thimerosal in vaccines.

I have often stated that autism is too complex a problem for us to oversimplify explanations for a cause of the recent ASD epidemic in the United States. But a hypothesis that mercury exposure, particularly from childhood vaccines, is a big piece of the puzzle makes sense to me. So when I see a study that links mercury and autism, I take note.

The Texas study is not going to be immune from criticism. I expect that we will see some commentary decrying the use of a Poisson regression analysis. I am not a statistician, and I would certainly invite any comments from those who are better versed in that discipline than I am. That being said, my understanding is that a Poisson distribution predicts probabilities of occurrence based on average rates in a discrete time period. If I understand the Texas study, the regression was used to account for demographic differences between various school districts included in the statistics. But the use of any regression analysis opens up the study to claims of bias.

Methodology aside, the real criticism we may expect is that the study can only show association and cannot show causation. That is a criticism with which the authors of the paper agree:

. . . these results indicate that the association between mercury release and school district special education rates was completely accounted for by increased rates of autism. This indicates that, in Texas, the increase in special education rates attributable to environmental mercury can be explained by increases in autism. The results of this study are consistent with our prior nation-wide study where an association between various developmental disabilities and environmentally released mercury was observed at the state level (unpublished manuscript). However, the results of this report should be interpreted with caution for a number of reasons.

First, this is an ecological study that precludes interpretation at the individual level. We have used aggregate units in this analysis to investigate differential rates of autism as a function of pounds of mercury at the county level. While we properly addressed the potentially biasing effects of clustering (school districts nested within counties) by utilizing appropriate analytic methods (e.g. multilevel-analysis), individual data are required to make a better case for the observed associations and their interpretations. Nevertheless, ecological studies of this type are often an important first step in identifying subsequent areas of investigation.

Second, a causal association between environmentally released mercury and developmental disorders cannot be determined from this cross-sectional data. Data availability permitting, future studies could investigate this association by using longitudinal data where changes in mercury levels over time may be used as a predictor of the rate of change in developmental disorders over time.

Third, we should consider that school-based administrative autism data, such as these, are only a proxy for true community prevalence. However, these autism rates are most likely biased downward.

R.F. Palmer, et al, Environmental Mercury Release, Special Education Rates, and Autism Disorder: An Ecological Study of Texas.

My own reservation with the Texas study is the same problem that any statistical analysis has. It cannot prove causation. Indeed, I would be hypocritical if I did not acknowledge the weaknesses of a statistical analysis like this one after so often criticizing the epidemiological studies relied upon by those who deny a connection between vaccines (particularly those containing thimerosal) and the triggering of ASD because the input data can be too easily manipulated -- by either design or error -- to be a reliable indicator of causation. There is no such thing as a perfectly designed analysis as long as the conclusions depend on arbitrary decisions concerning what data to use, and how to use it.

So as enticing as the Texas results are, let’s not overstate the case. The study is no more than what the authors call it: “a necessary first step in identifying plausible contributing factors of risk for developmental disabilities.” I could certainly quibble with the use of the term “first step,” as I believe the first steps have already been taken.

On the other hand, let’s not understate the case either. The Texas study is an important step toward gaining an understanding how a combination of genetics and environmental factors can trigger ASD; it’s another piece of the puzzle. It cannot be ignored; it must be followed up.


Anonymous MAría Luján said...

Mr Rankin
Thank you for your analysis of the paper of Palmer et al. I found it very important and, as you say, the first steps have been already done.
The same drawback of the association and no causal relation the authors of this paper agree is the same that epidemiological studies DENYIng the vaccine compounds ( including thimerosal)-neurodevelopment problems link have. Epidemiological studies CAN NOT discard a causation, only to quantify a risk at the overall population level...but they have been used to DISCARD, what is especialy worrisome in my opinion.
Unfortunately, the overall picture is enough complex in my humble opinion that epi studies focused in a causal relationship only FOR ME are prone to miss the point. It has been published recently that multicausal syndromes of individual presentation are very difficult to prove. If we agree that ASD shows an important INDIVIDUAL component ( including genetics) HOW an epi study is going to show this, focused only in ONE diagnosis??
Thank you for your thoughts
María Luján

12/17/05, 1:12 PM  
Anonymous Shawn said...

This sounds like a very interesting study. I admit that my first reaction about a new study on mercury and autism was to roll my eyes. I've become jaded about discussions on mercury and autism as there always seems to be too much sensationalism around the topic. I take a view, that I believe is similar to yours, that autism is way to complex to spend all our time talking about mercury.

As usual, your take on the study avoided all the typical sensationalism around the issue. Most of the focus on linking mercury and autism has been on vaccines and most vaccines that children get now are now free of mercury. There hasn't been much said about environmental mercury outside of fish consumption. It looks like we may have a few things to learn.

Are there more details on the study available on-line other than the abstract on Pub-Med?


12/17/05, 4:12 PM  
Blogger Wade Rankin said...

Agreed, Maria. It’s a very complex issue, and epidemiology and statistics are not going to provide the answers we need.


As always, it’s great to hear from you.

A pdf of the full paper is available at the Safe Minds site.

I agree that we need to look at various environmental dangers that could contribute to the problem, but I'm not so sure that vaccines are still completely safe. Someone can correct me if I'm wrong, but I believe that even the vaccines that do not use thimerosal as a preservative still use it as an antibacterial agent in the manufacturing process. Supposedly, the thimerosal is removed leaving only "trace" amounts, but there is a controversy as to whether all of the manufacturers are effectively removing most of the thimerosal. Also, there is still some debate as to whether ANY amount of thimerosal is safe. And, of course, most flu vaccines still contain a full dose of thimerosal.

All that being said, the main point is we need to look at a lot of factors, including BUT NOT LIMITED TO, mercury exposure. The Texas study is an important step in the process, but it is only a step.

12/17/05, 5:05 PM  
Blogger Kristina Chew said...

Would be interesting to see if other states did a similar study---I'm thinking also of the CDC study report on autism in Brick Township, NJ.

Keep on instilling sense into all these topics.

12/17/05, 7:14 PM  
Anonymous John Johnson said...

I've been meaning to comment on this for a couple of weeks, and now that I'm laid up in bed for a while, I finally have a chance.

On the statistical side of the argument, Poisson regression is a fine method to use in this case. You're right -- the Poisson distribution is used to describe count data, and the number of children with autism is certainly a count data.

There are two criticisms I can see of the method. The first, and perhaps most serious, is that the data can be analyzed with a logistic regression, and logistic regression is a much more common method. The second is that the Poisson distribution has very specific assumptions, and is controlled by only one parameter. For example, if you have a mean of 1 in a Poisson distribution, then you have a standard deviation of 1 also (or else you have some other distribution).

I'll address these complaints in reverse order (because the second one is easier). Poisson regression methodology has for years taken into account the fact that we rarely run into a true Poisson distribution. It is common these days to adjust for "overdispersion," which is what the authors stated they did (and they clearly had overdispersion according to their summary tables).

To the first complaint, I'll say that they took district size into account and looked at relative rates. While I'd be curious to see a logistic regression performed also, I don't think it will show something too different from what they've seen here.

The second comment I'd like to make is that a study designed to show causation between mercury and neurological damage (ASD or otherwise) would be unethical and immoral. (And impractical.)

In short, ecological studies are going the be the best statistically-based tools we have.

That said, I do think that if they performed a similar but smaller observational study (with a random sample) where they check individual mercury levels, I think that would address their questions with even stronger evidence.

12/30/05, 8:47 AM  

Post a Comment

Links to this post:

Create a Link

<< Home