Thimerosal II: The autism epidemic

Okay, here goes. As anyone who has visited this site knows, I'm not doing this because I have the least trust in the pharmaceutical industry or federal regulatory agencies. But the thimerosal controversy (which is now largely a one-person show, driven by the celebrity of a single champion), is a good opportunity to think about some of the complexities, difficulties, and also the strengths of epidemiology. It's also a chance for us to think about the nature of evidence in general.

To people who tend to share my general perspective on politics, Robert Kennedy Jr. is a sympathetic figure. He has written a compelling polemic (Crimes Against Nature) about the environmental catastrophe facing the planet. He is an attorney for the Natural Resources Defense Council, a well-established crusading Washington lobbying group. (I should point out, however, that the NRDC has not endorsed Kennedy's theories about thimerosal and autism, despite a history of being nearly paranoid about toxins getting into humans.) In his Rolling Stone article, Kennedy quite pointedly pits himself against Bill Frist, who is trying to protect Eli Lilly, the manufacturer of thimerosal, against lawsuits. While that is wrong in principle -- the proper place to resolve such disputes is in the courts -- that Bill Frist is a champion of Eli Lilly does not constitute evidence that thimerosal causes autism. Frist would be on the side of his campaign contributors no matter what.

First, if we're going to look for the cause of a disease entity, we need to define it. This is actually quite problematic for many kinds of disorders, but particularly so for ones that manifest as behavior.

The word autism actually applies to three major diagnostic entities, which are commonly referred to as "autistic spectrum disorders." This seems to imply that they represent degrees of severity of the same entity, but there is actually no evidence for this. They could have unrelated etiologies (underlying causes). Indeed, it is not at all certain that any of the three major ASDs are indeed single entities. No two people diagnosed with autism have exactly the same symptoms.

The popular image of autism is strongly colored by an entity called Asperger syndrome. People with Asperger syndrome often have above average or even superior intellectual functioning as measured by IQ. They may have superior verbal fluency and they often have strong, but unusually focused intellectual interests. However, they have impaired social talents. They are not intuitively able to read other people's feelings or detect and respond to social cues, as most people can do without even thinking about it or noticing what they are doing.

Classic autism is another matter. Children with this diagnosis usually have highly impaired verbal ability, and many do not talk at all, or merely echo what they hear. They are profoundly withdrawn, typically are very intolerant of changes in routine or strong stimuli, have great difficulty in making any sort of social contact, and often engage in repetitive, sometimes self-injurious behaviors. They typically score very low on IQ tests. But the severity of this disorder is quite variable and many children, with patient, intensive guidance and a highly structured environment do make progress in communication and social engagement. Somewhere in between is so-called Pervasive Developmental Disorder Not Otherwise Specified, which is a garbage can category for children who have autism-like symptoms but don't really fit the definition.

In his Rolling Stone article, Kennedy writes that autism "was unknown until 1943, when it was identified and diagnosed among eleven children born in the months after thimerosal was first added to baby vaccines in 1931." He obviously wants us to think that autism did not exist before that time. This is just unconscionable intellectual dishonesty. Essentially all of the psychiatric diagnoses we use today emerged during the 20th Century, and many of them have been extremely controversial. At one time, physicians did not distinguish between what we today call schizophrenia, and tertiary syphillis. That does not mean that schizophrenia did not exist prior to the 20th Century, or that it was actually syphillis. It just means that people used to think that schizophrenics were possessed by demons, or they had various other theories about people who behaved bizarrely, but they didn't call it schizophrenia or define a single set of symptoms that corresponded to the diagnostic criteria we use today.*

Many suspect that Isaac Newton and Albert Einstein may have had Asperger syndrome. There are many tales from before the 20th Century of people who may have been autistic, but a the time they were thought to have been raised by wild animals, or had their souls stolen by supernatural beings, and so on. Once diagnostic criteria for autism had been established, naturally cases began to be identified. And as always happens in such situations, as the disorder became more widely known and clinicians learned how to ascribe the label of autism to individuals, the number of cases identified tended to grow.

Once we have a method of assigning a diagnostic label, how do we establish the prevalence of a condition? Actually, it is not easy. Physicians are required to report some diagnoses, such as TB and HIV infection, but for the vast majority of diseases, there is no surveillance system.

Kennedy claims there has been an explosive epidemic of autism, coinciding more or less with the increased use of thimerosal in vaccines. However, the statistics used to establish this come from the U.S. Department of Education, and are based on reports from school districts. As should be obvious, school districts identify children as autistic when they qualify for special education services. In the most recent issue of Pediatrics, James Laidler and colleagues report that these statistics are not a useful estimate of the prevalence of autism at all. From 1993 to 2003, it appears from U.S.D.E. data that there was a shocking increase in the prevalence of autism, from 5 children in 1,000 to more than 25/1,000. But of course that isn't so. It is simply that more children are being identified with autism by schools. As Laidler points out, the states use varying definitions. Hence the prevalence in Washington is 1/3 the prevalence in Oregon.

Dr. Laidler has personal experience with this: a teacher told him that his own son was autistic, but he is not. The school district's critiera for assigning a label of autism are different from the medical criteria. Educational assessments are done for the purpose of establishing eligibility for services, not for establishing medical diagnoses.

Special studies conducted during the 1980s and early 1990s found lower rates of autism than more recent studies, so there is evidence that autism has become more prevalent in the United States, although there is not proof because it is difficult to be sure that these studies are really comparable. In any case, this obviously doesn't tell us anything about the cause. Kennedy is very enthusiastic about studies by Geir and Geir which correlate, over time, thimerosal exposure in the population with autism prevalence as derived from the Department of Education data. But as we have already seen, the USDE data is not a useful measure of the true prevalence of autism. Furthermore, as Sarah K. Parker and colleagues, writing in Pediatrics in Sept. 2004 show, Geir and Geir's estimates of average thimerosal exposure in the population are also unreliable.

This type of study is called an "ecological study." That has nothing to do with the most familiar meaning of the term, of the systematic interactions of biological species. Rather, it means that two or more factors are measured on average over groups of people, rather than being measured directly for individuals. Of all epidemiological methods, it is probably the weakest for inferring causation. Even if Geir and Geir's data were reliable, their findings would be merely suggestive. Many other things have changed in children's environment over the same time period, and this method gives us no particular reason to believe that thimerosal is responsible. It could serve to rule out the hypothesis, but not to confirm it. But in fact, the data they used are completely inappropriate for the purpose, and the study (actually they published the same results three different times) is worthless.

Next: Methodologically stronger studies; and the conspiracy theory.

*BTW: This is an excellent example of what social scientists mean by "the social construction of reality." It doesn't mean we are non-materialists, or we don't believe in the existence of a neumenon. Kennedy's mistake, in fact, is to confuse social construction with reality itself - a very lawyerly sort of mistake, in my view.