You’ve probably already heard–the CDC’s estimate of the autism rate was announced as 1% today. This was published in an MMWR came out today, with the new CDC prevalence estimate of 1 in 110. I made some notes on the conference call, but I thought it worthwhile to make a few comments on the document itself.
This is the third in the autism prevalence MMWR’s. The first was
Prevalence of Autism Spectrum Disorders — Autism and Developmental Disabilities Monitoring Network, Six Sites, United States, 2000
and
This new one is titled
In each they estimate the prevalence for 8 year olds. So, the 2000 study was on children born in 1992 and the 2002 study was on children born in 1994. They found prevalences of 1 in 166 (200) and 1 in 150 (2002), respectively.
The new prevalence is 1 in 110 for 2006 (8 year old children born in 1998). This is a big jump, no doubt. It is worth taking very seriously. Unfortunately, it isn’t as simple as saying, “the number of autistic children has increased”.
As the authors note in the introduction:
Since the early 1990s, the number of persons receiving services for ASDs has increased substantially. However, identifying children for services for autism might not be equivalent to using consistent diagnostic standards to identify persons in the population because services within communities are not available uniformly to all persons with ASDs. For this reason, studies that rely exclusively on single-source administrative datasets (e.g., disability service records or annual reports of special education counts) most likely underestimate ASD prevalence and might not adequately capture changes in the ASD population over time
Most of the states reporting in the ADDM use more than a single source. Most use medical and educational records. Those that use a single source (e.g. Florida) show lower prevalence estimates.
Let’s get back to that 1% estimate. This is a big jump in the prevalence–with the previous prevalence being 1 in 150 (about 0.67%). Yes, it should be taken seriously, but at the same time 1% is not a surprising number. The authors list 5 studies that show a prevalence >1%, using other methods.
Note that this is for the 1998 birth cohort, children born 4 years after those studied in the previous MMWR. I bring this up because the “buzz” before the MMWR was discussing this as the 1996 cohort. The studies are not spaced out evenly in time–2000, 2002, 2006.
Methodology
The methodology is through a record screen, not through direct observation of children:
Children aged 8 years with a notation of an ASD or descriptions consistent with an ASD were identified through screening and abstraction of existing health and education records containing professional assessments of the child’s developmental progress at health-care or education facilities.
If on review of the records, the child is deemed to meet the DSM-IV criteria for autism, he/she is counted. This is a good way to get an estimate, but there is a lot of room for children to fall through the cracks, in my opinion. As such, it is and has always been, an underestimate. In some cases, it is probably a very large underestimate. As will be shown below, there is a big variation by state–a factor of 3. I don’t think anyone really believes that the autism prevalence in Florida, or the prevalence amongst Hispanics is much lower than in other states or other ethnicities.
It is interesting to note that the ADDM network does find a significant number of children whom they (the ADDM) categorize as autistic even though they haven’t been given that diagnosis before. About 23% of the autisistics counted in the 1% value were not identified before the ADDM did their review.
That’s worth repeating–about 23% of 8 year olds identified as autistic were mislabeled as non-autistic by their schools, parents and doctors. That’s an interesting fact for those who claim that autism is easily identified.
The CDC study is not capable of explaining whether part of the increase represents a “real” increase in the number of autistic kids. It certainly can’t go into whether or not any specific proposed cause is valid. However, one can use the data to test hypotheses. This type of study just isn’t capable of making claims about causation.
States Reporting to the ADDM
This new study covers children born in 1994 living in 11 states:
Alabama, Arizona, Colorado, Florida, Georgia, Maryland, Missouri, North Carolina, Pennsylvania, South Carolina, and Wisconsin.
Note that this is different from the last MMWR, which looked at kids born 2 years earlier (1996) and included sites in:
Alabama, Arizona, Arkansas, Colorado, Georgia, Maryland, Missouri, New Jersey, North Carolina, Pennsylvania, South Carolina, Utah, West Virginia , and Wisconsin.
The new study added Florida, but took out Arkansas, New Jersey, Utah and West Virginia. These changes do make it more difficult to compare the prevalence data. They do not explain the increase. By that I mean, it isn’t as though the low prevalence states were removed from the previous MMWR study and high prevalence states were added.
I would have been very interested what happened to the prevalence in New Jersey. The last two MMWRs had New Jersey at about 1%. A good comparison would be if that state had the same autism prevalence or a higher value.
Prevalence by state
Data in Table 2 from the MMWR. Data from 11 states were reported. The prevalence ranged from 4.2 (Florida) to 12.1 per 1,000 (Missouri and Arizona). That’s a big spread, a factor of 3. The highest prevalence was for males in Missouri–19.1 per 1,000 (1.9%, or 1 in 52).
There was some variation in the male:female ratio, from 3.2 (Alabama) to 7.6 (Florida). Again, this shows that there is a big variation by state in how autistics are identified.
Variation by ethnicity.
There is a large difference in the prevalence by ethnicity. This also varies by state. For example, there is a 9.4 White-to-Hispanic ratio in Alabama, but the same ratio is only 0.7 in Florida. In general, Whites are identified more often than either Hispanic or African Americans.
We have a lot of work to do in making sure minorities are correctly identified and getting appropriate services and supports.
Earlier age of identification.
The median age of autism diagnosis is about 4.5 years. This is a big jump down from the 2002 data, which had a median of 5 years. That is a good thing.
This is worth a couple of comments. First, there is still room for improvement. Most of the children had development concerns noted in their records before age 2. The obvious question is whether these children could have been identified sooner. This was one comment stressed by the CDC in the conference call.
It is also worth noting that autism isn’t “obvious” or “easy” to identify autistics. 4.5 years old is the median age.
It is also worth noting that the big jump down in median age of identification is a very clear indicator that, yes, people are getting better at identifying and diagnosing autistics.
The range of age of identification is as low as 2 months (!) to as high as 106 months. The low end is pretty remarkable. I don’t think there are any tests qualified for such young children. At the high end, that is nearly 9 years old. Again, this goes to the idea that autism is not obvious.
One complaint that is always used in these discussions is “what, are you saying that no one noticed these kids?”. No. They likely had other diagnoses or that some concerns were noted about development for them.
Number of kids receiving special education services.
Table 3 of the report is quite interesting. They show for six states what fraction of identified autistic kids are getting special education, and under what primary label.
In Colorado, 76% of autistic students are in special education, while in Arizona it is 96%. Is there really such a big difference in the needs of these kids from state to state?
Colorado has only 34% of the autistic students under the primary category of autism. Compare that to Maryland with 76%. Again, it is pretty clear that there is a lot of variation in how special education services are granted by state.
Cognitive Functioning
From the MMWR:
Data on cognitive functioning are reported for sites having IQ test scores available on at least 70% of children who met the ASD case definition. The proportion of children with ASDs who had test scores indicating cognitive impairment (IQ ?70) ranged from 29.3% in Colorado to 51.2% in South Carolina (average: 41%)
I find this information rather interesting. Most of the children with IQ data were not cognitively impaired.
2004 Prevalence data.
There was also a small appendix released today, with information on 2004 prevalence data (note that the study noted above is for 2006). Those data show a prevalence of 8 in 1,000, or 1 in 125. This is a “smaller scale effort”
It is interesting to put the data from the various studies into a timeline:
2000 originally 1 in 166, revised to 1 in 150 (6.7 per 1,000) (1992 cohort)
2002 1 in 152 (6.6 per 1,000) (1994 cohort)
2004 1 in 125 (8.0 per 1,000) (1996 cohort)
2006 1 in 110 (9.0 per 1,000) (1998 cohort)
If you recall, David Kirby claimed the new CDC data would support his hypothesis that the Hepatitis B vaccine caused an upsurge. Mr. Kirby mistakenly assumed that the 1% figure would be for the 1996 birth cohort. Mr Kirby then correlated the jump from 1 in 150 to 1 in 100 with the uptake of HepB vaccination in that birth cohort. He stated (incorrectly, we now know):
But according to the CDC’s National Immunization Survey (which also includes parental telephone interviews), only 8% of infant children received the Hep B vaccine in 1992, when that birth cohort showed an ASD rate of 67-per-10,000.
By 1994, the number of children receiving Hep B vaccine had reached just 27% — and the cohort showed a similar ASD rate.
But the Hep B coverage rate had risen to 82% by 1996, when that cohort’s ASD rate rose to around 100-per-10,000.
So, there was a smaller jump that correlates with the uptake in HepB in the 1994 cohort–the rate increased from 0.66% (1994) to 0.80% (1996). Then, two years later, there is a similar jump, from 0.80% (1996) to 0.90%. But, this time without the big increase in HepB vaccination.
Not so dramatic a story with the real data. Just a constant rise.
Final words
I hope this highlights some of the information in the study. I wish I had the time to spend to make this summary a bit easier of a read. But, there are other posts to write. Like one talking about Autism Speaks and the way they are capitalizing on the new prevalence numbers with a very deceptive graph.
Left Brain Right Brain 
Sullivan,
I completed my own, much more basic, analysis with reference to the situation in the UKhere and came to the same conclusion.
Autism Speaks’ behaviour over this data is absolutely despicable – deliberately deceptive scaremongering.
Since the advent of the hubble telescope we have seen the prevalence of exoplanets (planets in orbit around suns other than our own) grow exponentially.
To take Autism Speaks logic, that would mean that a new one is being created every few days, scary or what.
(I exagerate a little to make my point, but hey fight fire with fire)
To take the extreme mercurians logic, it is not so much that we have to worry about the spontaneous creation of exo planets, we have to fear being invaded by them because of a conspiracy of astronomers opening channels through the universe for this to happen.
It is interesting that those who comment on the CDC stats have felt it neccessary to include the phrase “we cannot rule out a real increase” that is obvious, the unknown can never be absolutely ruled out, as indeed we cannot ever by scientific means refute the philosophy of Bishop Berkeley or discount that the unobserved Universe spontaneously comes into being whenever we see a bit more of it. However most sober astronomers do not find it necessary to add such riders, the fact that the CDC does says more about there social and political location than anything else about figures.
This whole talk of prevalence should be quashed to begin with, what they are talking about is a hypothesised prevalence extrapolated from increasing incidence (or better reporting) it is not controlled in any way for the whole population, it does not concern itself with adults.
Is there any sense or any science in it at all? All I see is politics, and a week is a long time in politics.
While being mainly in agreement with Larry’s argument I think we should not rush to dismiss the concept of prevalence. Accurate studies of incidence are very hard to do because of the difficulties with accurate screening and diagnosis. Studying prevalence amongst older children partly overcomes this by ensuring that fewer are missed and we need a reasonable guide to numbers if we are to plan suficient provision to meet needs.
There is a perpetual problem with a prevalence paradigm for autism, and that is the shifting definition.
Change the subject, let us try a prevalence survey for people with an IQ above 120. Well what datum do we start from.
I would wager with the flyn effect that is something you could never do, my present IQ is bound to be greater than Einsteins measured IQ, because of that cos if he were alive and tested today, well who knows because it is all an artefact.
We still do not know what it is we are measuring to take a base line from. I don’t give a fig for gold standard ADOS, because my generation is not todays generation and culture undlerlines it all, we don’t even know what tommorows disorders will be other than that they will be.
Until we can find a bio marker it is always going to be speculative. I don’t think the biomarker is there myself, autism is something that you know when you see it, but you can’t properly define.
We do need to look at a bigger picture, because the very construction of autism is multifaceted, and each of those facets leads to another set of possibly linked conditions.
We don’t even agree what language is yet in order to define a language disorder as a subset from it.
Science is still in it’s infancy, and unfortunately the direction of science is toward many threads rather than the link between them, it is increasingly difficult for any specialist to have sufficient of a general education in science to even know what others are doing.
Prevalence means nothing unless we are looking at other “co-morbities” to use the jargon, and then we could plot some very pretty diagrams.
Gillberg perhaps was on the right track years ago (heck Hawkwind has just entered my mind, I’ll be off on one soon)
Gillberg is quite a smart cookie really and a pioneer of this whole epidemiology business.
The earliest >1% actual prevalence finding is Kadesjö et al (1999). The “children” from this Swedish study were born in 1985.
Lorna Wing actually had suggested prevalence might be closer to 1% even earlier, but this was basically an extrapolation from various different findings. Her intuition proved correct.
About incidence vs. prevalence, see the Wikipedia article for an explanation of the concept. The “incidence” of autism is probably 1 in 10,000 per year. Actual incidence is indeed difficult to measure, simply because you don’t know date of onset. You just know date of diagnosis or date of inclusion in a database.
But the issue of undercounting is not an incidence vs. prevalence issue. That’s a misconception. There can be administrative prevalence and administrative incidence, both inaccurate.
Among the outcomes is that 40% of the children carrying a suspected diagnosis for an ASD actually lost the diagnosis over time.
A new interesting study looked at genetic syndromes associated with ASD (Fragile X, Tuberous Sclerosis, Angelman Syndrome, PKU among them) and found that the ASD symptoms are superficially related to ASD and cautioned about how to interpret ASD rates within the genetic syndromes as the rates of ASD in genetic syndromes are directly related to the severity of mental retardation.
http://www.ncbi.nlm.nih.gov/pubmed/19708861?
Dignostic substitution is not a new phenomena. Kanner in 1965 also reported on diagnostic substituion that had occurred over a 20 year period since the publication of his original article in 1943.
He wrote in 1965:
“Moreover, it became a habit to dilute the original concept of infantile autism by diagnosing it in many disparate conditions which show one or another isolated symptom found as a part feature of the overall syndrome. Almost overnight, the country seemed to be populated by a multitude of autistic children, and somehow this trend became noticeable overseas as well. Mentally defective children who displayed bizarre behavior were promptly labeled autistic”
While Kanner used politically incorrect labels, the comparison with current conceptions of ASD definition is the same.
http://neurodiversity.com/library_kanner_1965.html
Both then and now, a superficial association with ASD symptoms explains diagnostic substitution, or as Kanner would have called it ‘Misdiagnosis’.
Among the outcomes is that 40% of the children carrying a suspected diagnosis for an ASD actually lost the diagnosis over time
wrong study RAJ. You are thinking of the pediatrics paper that came out a few months ago.
While Kanner used politically incorrect labels, the comparison with current conceptions of ASD definition is the same.
For the first two decades after Kanner’s first paper, people considered autism and intellectual disability to be distinct. So, people with ID were not diagnosed autistic. How is that the same as today?
“For the first two decades after Kanner’s first paper, people considered autism and intellectual disability to be distinct. So, people with ID were not diagnosed autistic. How is that the same as today”?
You obviously didn’t bother to read Kanners paper. His complaint that was that the “mentally defective children with bizarre behaviors” were being diagnosed as ‘Autistic’. Diagnostic substitution was going on then as it is now.
The paper by Patricia Howlin’s group demontrates how diagnostic substitution is still alive and well since the association with ASD in the genetic syndromes have only a superficial association with ASD because they may share a number of isolated secondary symptoms with other people with structural brain anomolies.
Adult stroke patients have also been found as a group to have social-communication impairments with Theory of Mind and executive function impairments. Are adult stroke patients also on the spectrum?
The number of studies which have found that Romanian Orphans abandoned at birth and suffering horrendous conditions in the snake-pit orphanages in Romania meet diagnostic criteria for ASD using Gold Standard diagnosis tools (AUT-R):
http://www.ncbi.nlm.nih.gov/pubmed/16167089?
Rutter has called these children ‘Quasi-Autistic’ children. This is nothing new. In his 1965 paper noted that Crewel coined a new term ‘pseudo-autism’ just like Rutter has in coining the term ‘Quasi-autism’ in reference to the Romanion children. And it can’t be forgotten that the Romanian orphans met diagnostic criteria for autism using Gold Standard diagnostic tools, which says a lot more about the validty of Gold Standard diagnostic tools than it does about ‘Quasi-autism’ or the superficial ASD symptoms present in genetic syndromes.
Kanner’s comments on ‘pseudo-autism’:
To complicate things further, Crewel, in the hope of avoiding confusion between true autism and other conditions with autistic-like features, suggested the term pseudo-autism for the latter. Even this term came to be employed haphazardly, and conditions variously described as hospitalism, anaclitic depression, and separation anxiety were put under the heading of pseudo-autism.
If you consider Romanian orphans to be on the ‘Autism Spectrum’ then you might agree that severe early emotional deprivation can cause ‘Autism’ which should bring Bettleheim back into the conversation. The Gold Standard diagnostic tools that are admired so much says so, doesn’t it?
What would reading Kanner’s first paper have to do with this? Did he predict how diagnostic trends would go for the next two decades?
The statement I made was based on an old review paper which, in turn, cited a paper from 1966.
If you want to take this up with the author of the review, or with Eisenberg, feel free.
Hi RAJ –
Regarding orphanages, I’m not so sure that extreme social deprivation during critical timeframes couldn’t have profound and real neurological defects; probably in some cases resulting in behaviors consistent with autism. Several animal models of social deprivation show neurological, behavioral, and neuroimmune similarities to what is seen in autism. Conversely, also in animal models, an enriching environment has been shown to have protective effects at a neurological level.
Lets assume that what we call autism today is way too broad and is only useful because we cannot apply sufficient subtlty to our observations to avoid over generalizations. What then? What are your thoughts on helping to alleviate this problem?
– pD
There is discussion that ASD is increasing and there is discussion that ASD is being diagnosed more often and there is discussion that more children are being tested, resulting in more being counted.
There is discussion that ASD is being found at earlier ages and there is discussion that ASD is manifesting itself at earlier ages.
Nowhere have I found a discussion concerning the changes in our cultural treatment of children during the early school ages. It is well known that adult ASD is very hard to diagnose because adults have learned to camouflage their symptoms well over the years. I feel that this camouflage technique reaches much farther down the age scale than many realize.
However, our cultural emphasis on social conformity has shifted in the past few decades. Might this be one of the reasons we are recognizing more ASD youngsters? They aren’t being pushed to conform (camouflage) as much as were those in the 1950 through 1980 generations.
Just a thought from an ASD, in his 70th decade.
Pops