Verified by 
To recap: An epidemic is defined as an increase in incidence (the rate of occurrence of new cases of a disorder). We don't know the incidence of ASD, because nobody has figured out how to stand in the delivery room and count babies with autism-to-be as they are born (or to count the number of children who regress each year). The next best piece of information would be prevalence: the percent of children who already have ASD. The correct way to determine prevalence is to test all children by standardized methods, making sure that all examiners are adhering to the same techniques and criteria for classifying the results. Or, if testing all children is too big a job, then we need to test a large, random sample of the entire pediatric population. At least then we could be sure that we had a representative picture, rather than data that may have been biased in some way. But we don't have that either. The best we have is service data: the number of children receiving services because of an ASD diagnosis. Service numbers are a poor substitute for population-based prevalence data, and they tell us almost nothing about incidence. Why is that?
Service data are at risk for sampling bias: Perhaps people in certain ethnic, geographic, religious, or economic groups are - as a group - more likely to seek or avoid public services. These phenomena can skew the results significantly.
Service data are can never tell us how many children may have been "out there," whose families never sought services at all. This is an even bigger problem than the much-discussed dilemma of changing criteria for a diagnosis, as the DSM morphed from one set of standards to another between 1980 and 1994. If, over time, families are simply more likely to come in (because of increasing awareness of the problem), then the apparent prevalence will rise; not because of an epidemic, and not even because the criteria for a diagnosis are looser.
Perhaps there are simply more diagnostic resources than before. The presence of specialists drives up the rate of specialty procedures. We know, for example, that the more plastic surgeons there are per capita, the higher the rate of plastic surgical procedures. In a similar fashion, if more diagnostic facilities for ASD are established, then the number of identified children will probably rise! This is not necessarily a bad thing, but it boosts service figures irrespective of true prevalence (or incidence) of ASD.
Service data can never tell us how much of the "explosion" is bogus, because parents or school districts are gaming the system in order to access funds that have been earmarked for children with ASD. Just this week I evaluated a non-autistic child, who has been labeled autistic by his school district, with the willing cooperation of his parents, in order to access state-funded services, such as a 1-on-1 aide in the classroom. I wrestle with this dilemma myself, in borderline situations: If I diagnose a child with Nonverbal Learning Disability, the family will not get a penny of earmarked funds. But if I bend my clinical judgment just a little bit, and give the child a diagnosis of mild PDD-NOS, then suddenly the coffers open, and the child can access all kinds of services.
Service data are also distorted by changes in Federal service and reporting requirements. According to US Department of Education Office of Special Education Programs (OSEP), the number of children with autism receiving services in the public schools prior to 1990 was zero. Not because they weren't there, of course, but because it wasn't until 1990 that the Federal government began listing autism as a reportable and reimbursable disability on the forms each school district has to submit to the government.
A bit of history: In 1975, Congress enacted Public Law 94-142, the "Education for All Handicapped" act. Patterned after the 1954 Supreme Court decision barring racial discrimination (Brown vs. Board of Ed; "separate is inherently unequal"), PL 94-142 for the first time placed a requirement on public schools to serve children with disabilities. Prior to that 1975, public schools were able to exclude children with special needs from school altogether - and they often did. But, despite the title "All Handicapped," the law only covered a few specified disabilities, and autism wasn't one of them. It wasn't until the law was updated in 1990 (at which time it was also renamed IDEA - the Individuals with Disabilities Education Act) that autism made the list. So, prior to 1990, kids with ASD were routinely labeled something else: "Emotionally Disturbed," "Other Health Impaired," "Mentally Retarded," or one of the other Federally recognized categories of disability. The "explosion" in service data for autism coincides with the passage of IDEA.
The Figure shows the prevalence of autism based on service data, plus the changes in Federal law (OSEP = Office of Special Education Programs, of hte US Dept of Education), and revisions of the DSM. The bottom line is that service data were never meant to be used as a proxy for prevalence, and the sooner we do the basic science the right way (universal screening, or a random sample of the entire population, using standard methods), the happier everybody will be.
For more on this subject, see my book. And for a nice critique of the dangers in trying to calculate prevalence from service data, see Gernsbacher MA, Dawson M, & Goldsmith HH. Three reasons not to believe in an autism epidemic. Current directions in psychological science, 2005:14 (2), 55-58, available online here: http://www.autcom.org/pdf/Epidemic.pdf )
