The MIND's recently published autism epidemiology (Hertz-Picciotto & Delwiche, 2009) has been widely publicized (press release is here, typical media story is here), with the upshot that large increases in reported rates of autism are real rather than apparent and should be studied as such.
Bloggers have taken on several aspects of the MIND's new autism epidemiology (here, here, here, here, with more and a summary here) as well as discrepancies between the paper itself and how it has been promoted. I'm going to add just a few more points to consider.
This isn't the MIND's first crack at epidemiology. Their previous highly publicized but unrefereed report, described as "A Comprehensive Pilot Study," was shown, in a refereed paper (Gernsbacher et al., 2005), to suffer from "unwarranted conclusions" about the effect of changes in how autism is diagnosed on reported rates of autism.
So what's new, this time around? Hertz-Picciotto and Delwiche use census data to account for migration into California, so as to generate what they call incidence (rather than prevalence) figures. Also, they've included individuals identified as autistic, for the purpose of receiving services, before age three.
But the basis for the MIND's new epidemiology remains the same: the California DDS data. Turning DDS figures into incidence data doesn't alter the problem that DDS-type administrative numbers should not be used as epidemiology. The pitfalls of using service-based numbers have been explored formally in the literature (e.g., Laidler, 2005; Gernsbacher et al., 2005) and expressed less formally by an NIH official:
"I'll call a kid a zebra if it will get him the educational services I think he needs."Hertz-Picciotto and Delwiche's inclusion of children identified in service records as autistic before age 3 raises the problem that early diagnosis of autism may be very unstable. In Turner and Stone (2007), only 53% of children diagnosed autistic at an average age of 29 months kept their autism diagnosis at an average age of 53 months. Diagnostic instability was related to earliness of diagnosis (the earlier the diagnosis, the less stable it was) but unrelated to interventions received.
Like Hertz-Picciotto and Delwiche's conclusions about changes in diagnostic criteria (pulled from an unrelated Finnish prevalence study; Kielinen et al., 2000), their conclusions about effect of including other autistic spectrum diagnoses (Asperger's and PDD-NOS, which the authors presume are "milder cases") do not arise from the incidence data they report. Instead, figures were taken from an earlier paper involving a specific (much smaller and narrower) sample of children recruited to study environmental causes of autism (Hertz-Picciotto et al., 2006).
According to Hertz-Picciotto and Delwiche, their study involves children with the specific diagnosis of autism only. The authors refer to "an International Classification of Diseases (ICD) code of 299.0 (autistic disorder)" (except of course this is a DSM-IV, not ICD, code and category), while assuming that children diagnosed with other autistic spectrum diagnoses do not meet ADI-R or ADOS cut-offs for autism or ASD (ADI-R has autism cut-offs only). This, the premise of the authors' analysis re what they presume to be "milder cases," doesn't stand up well to scrutiny.
For example, in Baird et al. (2006), 69% of children with non-autism autistic spectrum diagnoses met autism criteria on the ADI-R. This doesn't count Asperger children, who were lumped in with autistic children, virtually all (98%) meeting ADI-R autism criteria. In work I'm involved in, virtually all Asperger individuals (as virtually all autistics) meet all ADI-R and ADOS cut-offs for autism.
Also, in using the ADOS (which has, more recently, been revised again) and ADI-R, as well as clinical experience, Baird et al. (2006) produced a range of prevalence figures for children aged 9-10, from ~25/10,000 to ~116/10,000. That's a 4.6-fold discrepancy in the same sample at the same time with the same instruments and the same group of diagnosing clinicians.
Maybe there are other ways of exploring whether children receiving DDS autism services in the past are similar to children receiving these services more recently. In their unrefereed 2002 report, the MIND compared subsamples of children born from 1983 to 1985 with children born from 1993 to 1995. In the earlier subsample, 61% were judged to be in the range of intellectual disability, whereas in the later subsample, 27% were--a discrepancy noted by Gernsbacher et al. (2005).
More data about the entire population receiving DDS autism services are available. In a 1999 DDS report, graphed data show that whereas in 1987, ~84% of those receiving DDS autism services were judged to be in the range of intellectual disability, by 1998 that figure was 58%. A further look at DDS quarterly reports shows that by the beginning of 2002, that figure was 42%, and by the end of 2007 it was 33%. Between 1987 and 2007, the proportion of individuals receiving DDS autism services and scoring in the severe and profound ranges of intellectual disability dropped from ~36% to ~6%.
So--there are a few more reasons to question Hertz-Picciotto and Delwiche's conclusions about the effect of differences in how autism is defined and diagnosed. Also, there are many reasons to question the whole enterprise of trotting out DDS numbers yet again and pretending that, with enough distracting decorations stuck on, they in fact are epidemiology. After all, Hertz-Picciotto and Delwiche are denying the existence of older autistics, and they are doing this by using poor quality data dressed up in definitive conclusions. I would be much more cautious than Dr Hertz-Picciotto and her university have been, in going out in public wearing the MIND's new epidemiology.
Irva Hertz-Picciotto, Lora Delwiche (2009). The Rise in Autism and the Role of Age at Diagnosis Epidemiology, 20 (1), 84-90 DOI: 10.1097/EDE.0b013e3181902d15