Baron-Cohen et al. (2009) propose a world where there are 157/10,000 autistics. The authors used what they call a "novel approach" to autism epidemiology and this is worth a closer look.
The study is based on questions sent to schools about the UK SEN (Special Education Needs) register, and a survey, including a screening questionaire, sent to parents of students aged 5 to 9 years. Many schools, 66 out of 162, refused to participate. Of schools that agreed, 79 of 96, representing 8824 out of 11,635 children, provided SEN register information. Of the surveys sent to parents, 29%, representing 3373 out of 11,700 children, were responded to.
Participating schools located 83 children identified as autistic spectrum according to the SEN register. Of these, 10 children were assigned the specific diagnosis of autism (I'll call them "specific-autism"), while the rest were assigned other autistic spectrum diagnoses ("other-spectrum"). None of these diagnoses was verified by Baron-Cohen et al. (2009) in any way. These 83 SEN children constitute 94/10,000 of the proposed 157/10,000.
In the survey, parents were asked whether their child had an existing autistic spectrum diagnosis. The authors "confirmed," to their standards, through information from parents only, diagnoses for 19 children. Four others attended special schools and were assumed to be "confirmed" as autistic without any verification, for a total of 23. No information about how many of these children are specific-autism or other-spectrum is given. The 23 children were expanded into a prevalence of 99/10,000--impressively close to the SEN register figure.
The screening questionaire sent to parents was the recently re-named CAST (Childhood Autism Spectrum Test). All children reported by parents as scoring above one CAST threshold, and a minority of children scoring above a lower CAST threshold, and who were not already reported by the parent survey to be autistic, were invited for assessment. Eventually 77 children underwent assessment by ADOS, ADI-R, and clinical judgment. Eleven children were given autistic spectrum diagnoses--four specific-autism, seven other-spectrum.
So there are three sets of autistic children here: 83 SEN children, 23 diagnosis survey children, and 11 CAST children. The CAST children are considered to represent "unknown" or "undetected" autistics in schools, while the other two sets of children (SEN and diagnosis) represent equivalent samples of "known" autistics who have been located in two different ways from very nearly the same cohort.
Working from the diagnosis and CAST samples, Baron-Cohen et al. (2009) come up with a ratio of total number of cases (known plus unknown) to known cases, which yields a multiplier of 1.67. This in turn is applied to the SEN-based prevalence of 94/10,000, resulting in that grand total of 157/10,000.
And this all looks quite reasonable. Even the Neuroskeptic thinks so. But here are a few things I noticed.
The authors report no means of verifying whether the SEN and diagnosis samples are equivalent, an assumption on which the study depends. There is a dearth of demographic information, including the absence of any reported demographic measure common to both the SEN and diagnosis samples.
And for neither "known" sample (SEN and diagnosis) were any of the diagnoses verified either through direct assessment or direct access to records. Other recently reported UK epidemiology has not arrived at such a tidy agreement between prevalence of existing autism diagnoses and prevalence of autistics on the SEN register, nor at such a high prevalence using information from one or both sources.
Before they started sending out surveys, Baird et al. (2006) found that about 15% of 255 autistics with a "known" autism diagnosis (according to various health and clinical records) were not on the SEN register--under any diagnosis at all. This percentage would likely be higher if SEN autistic spectrum diagnoses only were considered.
Whether the 255 "known," pre-survey diagnoses from Baird et al. (2006) would be considered "confirmed" according to the standards of Baron-Cohen et al. (2009) is unclear to me, but in Baird et al. (2006), they add up to a prevalence of ~45/10,000. This is less than half the prevalence of what should be the comparable "known" diagnosis sample in Baron-Cohen et al. (2009).
Then there is Williams et al. (2008), where the authors had access to both medical and SEN records for an entire 14,062-child cohort. Using both kinds of records, they located a total of 86 "known" autistics for a prevalence of ~62/10,000. Of these, 36% were found in medical records only; 17% were found in SEN records only; and 47% were found in both.
Here again the "known" diagnosis total prevalence, derived from direct examination of complete records rather than selective responding to questions directed at schools or parents, is a much lower figure than found via both the SEN and diagnosis samples in Baron-Cohen et al. (2009). And Williams et al. (2008) show that diagnoses appearing in medical records, and diagnoses recorded via SEN, do not produce the same samples of autistic children--at least, not when there is direct access to these records for a complete cohort.
The original contribution of Baron-Cohen et al. (2009) is, according to the authors, the attempt to identify school-aged autistics who haven't been spotted as atypical. This is claimed to set their study apart from Baird et al. (2006). But 8 of the 77 children chosen for direct assessment through CAST screening in Baron-Cohen et al. (2009) were reported to be from "special schools" which, according to the authors, "require extensive assessment of individual special needs." Yet the authors suggest the CAST-screened sample was, in their words,
...quiet and cooperative at school and not difficult to manage and therefore teachers may not be aware that they have difficulties.I don't know whether the CAST-screened special school children (who may or may not have been in the final 11) were on the SEN register, but doesn't this seem likely? Assuming for now (because you can only assume) that this was not under an autism diagnosis, in fact these were exactly the children screened for autism by Baird et al. (2006)--all children on the SEN register for any reason.
Further in this direction, Baron-Cohen et al. (2009) surprisingly report no way for the authors to verify whether the CAST-screened children were or were not on the SEN register, under any diagnosis--including autism. But the study hinges on the two samples, the SEN and CAST samples, being independent from each other.
One more thing to note is the actual composition of the 157/10,000. These numbers come from the reported data. Here goes:
11/10,000: "known" specific-autism children (SEN register)
23/10,000: "unknown" specific-autism children (CAST screening)
83/10,000: "known" other-spectrum children (SEN register)
40/10,000: "unknown" other-spectrum children (CAST screening)
If you add up the figures, you get the 157/10,000 total, of which 34/10,000--well within the range of previous findings--represents prevalence for the specific diagnosis of autism. The authors checked ratios of known to unknown autistic males vs females, and found no difference, showing that female autistics were not more likely than male autistics to be undetected. But the authors didn't check ratios of known to unknown autistics for specific-autism vs other-spectrum diagnoses.
If they did, they would find that for every known specific-autism child, there are about two unknown. The opposite is true for other-spectrum children, where about two are known for every unknown child. While only one-third of all specific-autism children are known, two thirds of other-spectrum children are known. Does this mean that children who meet criteria for the specific diagnosis of autism are much harder to spot, or are doing much better in school, than children who are popularly regarded as having "less severe" autistic spectrum diagnoses?
Or maybe the data are just very difficult to interpret. As usual, if I've made any factual errors, please let me know.
References:
Baird, G., Simonoff, E., Pickles, A., Chandler, S., Loucas, T., Meldrum, D., & Charman, T. (2006). Prevalence of disorders of the autism spectrum in a population cohort of children in South Thames: the Special Needs and Autism Project (SNAP). Lancet, 368, 210-215.
Baron-Cohen, S., Scott, F., Allison, C., Williams, J., Bolton, P., Matthews, F., & Brayne, C. (2009). Prevalence of autism-spectrum conditions: UK school-based population study The British Journal of Psychiatry, 194 (6), 500-509 DOI: 10.1192/bjp.bp.108.059345
Williams, E., Thomas, K., Sidebotham, H., & Emond, A. (2008). Prevalence and characteristics of autistic spectrum disorders in the ALSPAC cohort Developmental Medicine & Child Neurology, 50 (9), 672-677 DOI: 10.1111/j.1469-8749.2008.03042.x
