Autism and paternal age: let's hear it for population-based studies
Today's Washington Post has a nice article by Shankar Vedantam about the association between increasing paternal age and risk of autism. Vedantam does a great job of laying out the context for readers: what you need to know about autism; what you need to know about paternal age; and a hint at some of the controversy over autism research, one aspect of which is whether an apparent epidemic of the disease may be in part attributable to expanding diagnostic criteria.
The beauty of this study, published in the September issue of the Archives of General Psychiatry, is that it followed the health of 98% of males, and 75% of females, born in Israel during a 6-year period in the 1980s, based on clinical army-enrollment exams. (The lower number for females is explained by Orthodox Jewish women not being required to enlist in the army.) That's a pretty complete picture. Final diagnoses were assigned at age 17 by draft-board psychiatrists, either in person or based on years of clinical data referred to the draft-board docs. The authors point out that the majority of these diagnoses were made before expanded diagnostic criteria were used, so that the numbers were not inflated - in fact, the numbers were quite small, inconveniently so for the researchers, because multivariable analyses are difficult with small numbers.
The reason that population-based estimates are more valid is that they are less susceptible to selection bias. Selection bias has the potential to leave disproportionate numbers of people with an unmeasured risk factor, or confounder, in one of the comparison groups relative to another. As a result, a measured association may be artificially inflated. In the case of this study, since the study sample for the country was so complete (everyone has to show up to the draft board), the chance that people were not included because of a diagnosis of autism or because of paternal age was essentially eliminated.