More on autism
Hurrah! I have a present for you: it's a free, full-text article from the journal Pediatrics, specifically a critical review of the epidemiologic evidence regarding the putative link between mercury and autism, by Parker and colleagues at the University of California Health Sciences Center (2004).
In this article you will read about the highest quality studies available on the topic, prospective cohort studies comparing children exposed to thimerosal-containing vaccines to similar children who received vaccines without thimerosal. These studies showed that the two groups had similar risk of being diagnosed with autism or autistic spectrum disorders (ASD). Prospective studies are extremely important for diseases whose diagnostic criteria for a disease are fuzzy, or changing, or both; much of the apparent increase in autism, as I mentioned in my post earlier this week, does appear to be due largely to expanded diagnostic criteria.
The studies that didn't pass muster were studies that compared changes in rates of autism diagnoses in a population over time to changes in mercury content of vaccines. These are ecologic studies, and while they sound compelling, and can even be used to generate hypotheses, they have a fatal design flaw: the lack of individual-level information on other factors that are associated with exposures or disease. (Another example is dietary fat and breast cancer: countries with high fat consumption also have high breast cancer incidence, but there could be 100 other things that explain this apparent association, and without individual-level information, you can't control for them.)
I read some other interesting comments from scientists along the way; one went so far as to suggest that the pathology of autism suggested that postnatal environmental risk factors are "unlikely" (Taylor B, Child Care Health Dev 2006; 32:511-519). (Paternal age would be a good example of a prenatal risk factor, and not just because prenatal, parental, and paternal just happen to be anagrams of one another... oh, never mind.) Certainly the expanded diagnostic criteria should be taken into account in any epidemiologic study. The fact that we don't have good incidence and prevalence data on autism makes the aforementioned ecologic studies that much less reliable.
Also check out the extensive coverage of this issue in the blogosphere at Respectful Insolence (and I mean extensive; I've not gotten through all of it myself).