Another Perspective on the “Autism Epidemic”
October 26th, 2007
Over the years, we have heard a lot about the “autism epidemic” and about how it will bankrupt the nation and lead to a “tsunami” of adults needing long-term residential care. You only need to look at a graph of the autism prevalence from 1993 to 2006 (USDE data) to see the terrifying news (Fig. 1):
[click on the graph to see it larger; click the "back" button on your browser to return to the post]
Figure 1.
But, there’s more to the story than meets the eye (especially if the eye is only looking at the above graph). What about a context for these numbers? What happens if we look at autism in the context of all IDEA disability categories? The graph below shows autism, mental retardation and all disabilities (all thirteen IDEA categories combined) in children ages 6 - 17 years, from 1993 to 2006 (Fig. 2):
Figure 2.
Autism is the red line ‘way down at the bottom of the graph. From this perspective, the “autism tsunami” looks more like a ripple. Another interesting thing I’d like to point out is the way the prevalence of “mental retardation” is trending downward at the same time that “autism” is trending upward. Lets look at that a bit closer (Fig. 3):
Figure 3.
It almost looks as though the two curves are mirror images of each other. In order to check that hypothesis, a little mathematical manipulation is necessary.
First, I’ll add the numbers for “autism” and “mental retardation” together and then, to correct for the gradual rise in “all disabilities” (see Fig. 2), I’ll divide that number by the total of all IDEA disability categories and express the number as a percentage of all disabilities (Fig. 4):
Figure 4
Well, look at that! As a percentage of all children with IDEA categorized disabilities, the sum of “autism” and “mental retardation” has hardly changed a bit from 1993 to 2006. This would suggest that one of the following things is happening:
[1] Children who would previously have been categorized as “mentally retarded” are now being classified as “autistic”. As Roy Grinker has pointed out in his book, Unstrange Minds, this would generally be a good thing, since the “conventional wisdom” is that autistic children have a better prognosis, vis a vis learning, than children with mental retardation.
[2] Whatever is causing the “autism epidemic” is also reducing the prevalence of mental retardation. In fact, it seems that for every child who becomes autistic, one child is spared from developing mental retardation. (see: Zero-sum game)
Of course, the second scenario is extremely unlikely, as there is nothing I can discover which could both increase the prevalence of autism and decrease the prevalence of mental retardation.
So, it appears that the “hidden horde” and all of the autistic children that people didn’t see when they were in school may have been hiding in plain sight. After all, before autism became so well known (and morphed into what it is today), the most likely category for them to be placed in was “mental retardation”.
Of course, I realize that many of the people crying “Epidemic!” will persist (or should I say perseverate?) in asserting that there must be something - besides a shift in classification - driving the rise in autism prevalence.
They will continue to insist that something - something in the vaccines, something in the environment, something in the milk, something in the whatever, but something - is causing this rise in autism prevalence.
For them I have some good news:
Whatever that something is, it is simultaneously reducing the prevalence of mental retardation by an equal amount.
Well, that’s all for now.
Prometheus
Filed under: Autism Science, Critical Thinking
October 27th, 2007 at 12:27 pm
Although I’m not necessarily keen on the IDEA data for telling us anything about true prevalence (and not that you suggested that it does either), this makes a pretty good case for diagnostic substitution. Great post.
October 27th, 2007 at 3:00 pm
I bet that if you exclude 6 and 7 year olds from the “All Disabilities” graph, it’s becomes pretty flat. I’ve looked at 8 and 9 year olds in isolation, and it’s flat. I suspect the difference is due to earlier identification.
I think there’s something causing the rise in diagnoses all right. I just don’t think it’s biological to a significant extent.
October 27th, 2007 at 7:48 pm
Love this. The power of good math! I’ll be sharing it.
October 28th, 2007 at 2:49 am
This is an interesting observation that I have heard made before (though never with the numbers matching us so nicely). I am left with a few questions. Firstly, why are the autism prevalence numbers (42 per 10,000 in 2006) so much lower than the 166 number we are used to hearing? Secondly, since a major shirt in the diagnosis of autism has been the introduction of the spectrum which includes AS, and it is unlikely that AS kids were previously given an MR label, relabling alone is does not provide a complete explanation. In other words, if AS kids are being added to the ASD group and, and the same time, kids are being removed from the MR group, then we should probably assume that at least as many kids who were previously labeled MR are leaving the MR+ASD group as there are AS kids entering this group. That’s not impossible (it could, for example, be attributable to in increase in early intervention) but it detracts from the strength of your argument.
I think that there almost certainly are some kids who are being given ASD labels today, where they would have been given MR labels in times gone by. But I am far from convinced that this change in labeling accounts for all of the reported increase in ASD.
October 28th, 2007 at 12:34 pm
I probably should have made it more clear in my post, but the USDE data, like the California DDS data, is not valid for following autism prevalence. The website for the California DDS, in fact, explicity states that the data cannot be used to compare one year to another.
VAB seems to be making the common mistake of assuming that “diagnostic substitution” means changing the diagnoses of any individual child. In fact, the term - as it is commonly used - means that some children who would have been diagnosed as “mentally retarded” in the past are now being categorized as “autistic”.
VAB also makes the point that the USDE data show an autism prevalence less than that found in the latest CDC studies. The “autism” number is an average of all the states and the states all have different criteria for categorizing autism, not to mention different motivations for categorizing students.
The “166″ number VAB mentions is, I presume the 166 per 10,000 (or one in 60) number that was erroneously reported by The Observer. That number has been “debunked” by a number of sources, including The Observer’s supposed source, Professor Simon Baron Cohen.
In the end, the real “take home” message is that the “data” sources most commonly used to support the “autism epidemic” show more than meets the eye. The reality is that we don’t have a good data source to easily show what is going on with autism prevalence.
That just leaves old-fashioned scientific research, which doesn’t seem to give unambiguous support to an “autism epidemic”.
Prometheus
October 28th, 2007 at 12:54 pm
VAB wrote:
Because the 42 per 10,000 is not based on descriptive epidemiology, it’s based on education data (services provided). The current descriptive epidemiology estimates the prevalence for as ASDs at about 65 per 10,000.
It’s likely that the education data is underrepresentative of actual prevalence.
Perhaps it has to do with the possibility that schools lag behind with the provision of services because realizing funding and creating appropriate programs does not happen overnight, and not all kids who would qualify for a diagnosis may necessarily qualify for current available special education services.
What “reported increase” are you referring to? The CDC’s? where the CDC states:
October 28th, 2007 at 3:08 pm
Perhaps VAB got the 166 figure from the old prevalence figure of 60 in 10000 which is sometimes reported as 1 in 166. This was based on epidemiological studies of by Fombonne and Chakrabarti. They acknowledged that they were probably missing a lot of children with Asperger Syndrome because AS tends to be diagnosed in children older than their study group. In the UK a consensus is gathering around 1 in a 100 for all autistic spectrum disorders.
October 28th, 2007 at 6:12 pm
The “1 in 100″ figure comes from one UK study of children born in 1990-1991 (Baird et al., 2006; prevalence of autism plus “other ASD” of 116/10,000). In this study, almost no possible Asperger syndrome cases were found and none was diagnosed. This was in children age 9-14yrs.
October 28th, 2007 at 9:08 pm
The 166 number is from a CDC study has been very widely quoted. It’s 1 one in 166 as a prevalence rate for US kids under 10. There is even a blog on the Hub called “Club 166.” It is for ASD, including PDD-NOS and AS. I don’t know if it is accurate, but those who have found fault with it generally point out that it covers the spectrum while previous numbers only accounted for classic autism, rather than disagreeing with the number itself. As for the 1 in 60 from the Observer, that’s not what I was thinking of, but it’s worth noting that, while the authors of the study in question did not back the 1 in 60 number, the number they stuck with is the common number for the UK which is about 1 in 100. Anyway, it doesn’t matter, but I just noticed that the chart gave a 42/10000 (1 in 238) which was not a number I had seen before, and which is quiet a bit lower than 1/166 or 1/100, so I wondered where it came from.
I did actually understand what “diagnostic substitution” means. What I did not do is explain my point well. I started to explain what I meant here, and then it got long, so I’ll post it on my blog in a bit.
October 28th, 2007 at 9:42 pm
If you look at 3-5 year olds only it’s 25 in 10,000 as I recall. Clearly, IDEA is under-recognizing autism. The older the kids, the greater the odds they’ll be put in the “right” category. But then again, if they are too old, from a time when autism was very much under-recognized, they probably won’t be. As I showed recently in my rebuttal of Gallup & Yazbak, the IDEA prevalence increases considerably between the ages of 6 and 11. Still, IDEA is considerably behind the epidemiology, at any age, and this is probably because not all autistic children are in special education, plus some must still be in the “wrong” categories.
October 28th, 2007 at 10:21 pm
VAB,
When you said “…why are the autism prevalence numbers (42 per 10,000 in 2006) so much lower than the 166 number we are used to hearing?”, I assumed you meant 166 per 10,000, since that was the form my numbers were in. I see where it came from now.
USDE and - if I remember correctly - the California DDS both underestimate the autism prevalence compared to studies using better methodologies. This - to me - is curious, since the catergorization criteria used by many states are so loose as to be nearly meaningless.
A number of articles in peer-reviewed journals (Newschaffer et al 2005, Laidler 2005, Shattuck 2006 and others) have addressed the problems inherent in using educational or social services client data as a surrogate for autism prevalence. This post is just illustrating another example of how these data are worthless for monitoring autism trends over time.
Joseph,
I’d be doubly cautious about using USDE data for children under age 6, since there is no mandate that children under that age be evaluated by the schools. I suspect that the data for children under 6 (and over 18, when children are no longer mandated to be in school) are more worthless than the rest of the USDE data.
If you look at the IDEA “autism” data state-by-state, you’ll see that the two states with the highest prevalence are Minnesota and Oregon ( 95 and 93 per 10,000, respectively). The surrounding states, which would be expected to share in any “toxic plumes” from China or elsewhere, have much lower prevalences:
Minnesota - 95 / 10,000
Surrounding states:
Wisconsin - 52 / 10,000
Iowa - 23 / 10,000
North Dakota - 33 / 10,000
South Dakaota - 34 / 10,000
Oregon - 93 / 10,000
Surrounding states:
Washington - 44 / 10,000
Idaho - 38 / 10,000
California - 44 / 10,000
Even New Jersey (49 / 10,000) and Louisiana (26 / 10,000) - states with reputations for toxic waste problems - have lower autism prevalence than Minnesota and Oregon - according to USDE data.
Anyway, if the true autism prevalence is 65 / 10,000 (for example), we should expect to see a continued rise in the reported autism prevalence until it reaches the real number.
Prometheus
October 29th, 2007 at 9:44 am
I have hear the 1 out of 166 number for a while now, but I’m recently beginning to hear “1 out of 150″. What’s the deal w/ that?
October 29th, 2007 at 12:40 pm
AuntieAuntie,
Different studies - even well-designed and well-executed studies - come up with slightly different results when looking at autism prevalence. One major cause of this variation is how the study defines autism.
Most recent studies looking at autism, including the CDC study, have come up with a prevalence of about 60 - 65 per 10,000. This works out to between 1 in 166 and 1 in 154. Some sloppy rounding on the part of certain “advocacy” organizations has led to a “1 in 150″ prevalence (which would be 67 per 10,000) being tossed about.
This is a bit of a mathematical “scare game”. When somebody says “1 in 150″, it sounds a lot more impressive than 67 per 10,000 (or 0.7%).
It also sounds more ominous when you say that the prevalence has gone from 1 in 166 to 1 in 150, even though that only means that they’ve found 7 more autistic children out of every 10,000 children in that age range.
Let’s do some more “scary math”.
The US Census Bureau estimates that there are a bit over 66 million children between the ages of 2 and 17 years. If the CDC autism prevalence of 65 per 10,000 is applied, this means that there are almost 430,000 autistic children (between 2 and 17 years) in the US right now.
430,000 children with autism sounds even scarier than “1 in 154″ or “65 per 10,000″ because we just don’t have much experience with large numbers. But they’re all the same number.
Given the plasticity of the criteria used to “diagnose” autism, it comes as no surprise that a wide range of prevalence numbers are being tossed about in the press and on the Internet. One respected researcher, Professor Simon Baron Cohen, has shown that if the most lax criteria are used, the autism prevalence could be as high as “1 in 58″ (172 per 10,000).
So far, the best number for prevalence - based on the quality of the studies and the consistency of the finding - is around 65 per 10,000 (1 in 154). This could change if new data becomes available.
Prometheus
October 29th, 2007 at 1:04 pm
1 in 150 is still an underestimate, I’m afraid. What I’m afraid of is that inevitably it will rise, and this doesn’t help the discourse environment. The CDC prevalence, for example, can only tell us about diagnosed cases. When you do a whole population screening, the prevalence seems to be closer to 1% as I argue here.
October 29th, 2007 at 5:52 pm
Just wanted to let you know that I have added a link to your blog on my site. You bring a refreshing dose of clarity to my daily thinking. Thank you.
October 30th, 2007 at 1:54 pm
Hi Promotheus -
I’ve been thinking about this a bit; something has been confusing me. Perhaps you’ve got an answer though.
Let us assume that mental retardation used to be diagnosed more frequently than autism, but additional awareness has reversed this situation. How can we explain the difference in male/female ratio between mental retardation and autism? It seems MR is more common in males than females, but most of what I’ve seen indicates the ratio is 1.2:1 - 1.9:1, certainly nothing like 4:1 or in some instances, reported at 6:1.
I’m open to the notion that we’ve been missing, or misdiagnosing people for a while. But how, I wonder, could we have been so wrong about the ratio of males to females?
I’m also curious on your take on the newschaffer paper where he did not find a reduction in other diagnosis relative to the increase in autism diagnosis.
Take care!
- pD
October 30th, 2007 at 4:52 pm
pD,
Strangely enough, 2006 was the first year that data on IDEA enrollment by gender was available publicly. In 2006, 67% of all students 6 - 21 years of age served by IDEA were male. Unfortunately, they do not yet have that broken down by disability, age or state.
However, since autism eligibility is but a small fraction (3.7% in 2006) of all the students served by IDEA, it would be reasonable to assume that it’s not the autism numbers alone that are driving the gender skew.
Note: comparing real epidemiological studies to IDEA or California DDS data is like comparing oranges to orangutans.
Interestingly, as awareness about autism has increased, so has the gender skew reported by social and educational sources. It’s almost as though the evaluators (many of whom are not medically trained) are telling themselves, “Autism is more common in boys, so a girl with the same behaviors probably doesn’t have autism.”
Just a thought.
The Newschaffer et al (2005) paper looked at cohorts and did not see as large a difference between age cohorts for mental retardation as they did in autism. There are two reasons for this:
[1] Their data set included 1992 to 2001. If you look on my graphs, you’ll see that the big downward trend in mental retardation seems to start in 2000.
[2] Since the number of students categorized as mentally retarded was (is) much larger than those categorized with autism, cohort data will not show as big a change in mental retardation prevalence compared to autism prevalence, even if there is a one-to-one shift in diagnosis.
When I look at the IDEA data by birth-year cohorts (as Newschaffer did), the trend to lower prevalence in younger cohorts is clear. It was not so clear from the data set (1992 - 2001) Newschaffer used. What a difference five years makes!
Prometheus
October 30th, 2007 at 7:10 pm
Dear Prometheus:
In somewhat less precise terms (to be expected, since the target audience is different than yours), essentially the same points were made in
THE NUMBERS GUY
By CARL BIALIK
How Many Kids Have Autism?
November 30, 2006
The Wall Street Journal Online
http://online.wsj.com/public/article/SB116481159830835726-lgTHSWJTEnIlIC0s0Azh2KtDbuI_20061230.html?mod=tff_article
You concluded: “Of course, I realize that many of the people crying “Epidemic!” will persist (or should I say perseverate?) in asserting that there must be something - besides a shift in classification - driving the rise in autism prevalence.”
There is a clear difference between questioning the available evidence and refusing to believe it.
October 31st, 2007 at 1:15 pm
WFJAG,
Thanks for the information!
Carl Bialik hits the nail squarely on the head in this article and I recommend it to anybody wondering about the wide range of autism “numbers” thrown about in the media and on the Internet.
Carl brings up a point I’ve been meaning to get to for some time: the “1.7 million autistic people in America” scare-tistic.
This number is often tossed out to scare people into “doing something” - usually opening their cheque-books. It is telling that the spokesperson for the Autism Society of America states that:
“The one in 166 has been the biggest boon to awareness we’ve had.”
For “awareness”, insert “fundraising”.
Or am I just being cynical?
But back to the “1.7 million” number.
That only works if you extend the prevalence in the worst case prevalence in children (of the studies reviewed by the CDC) to the entire US population, from zero to 100+ years or age.
Now, if the prevalence in increasing, as the Autism Society of America believes, then you would expect to find a lower prevalence in older age groups. In that case, the total number of people with autism would be significantly lower than “1.7 million” - or even the “1 to 1.5 million” that the Autism Society of America asserts.
It is also interesting - and perhaps Freudian - how the Autism Society of America (through its spokesperson) stands on the “autism epidemic”:
“…other evidence, including reports from her group’s members and from overtaxed state agencies, suggests a very real increase in incidence. ‘That’s what we believe, and what we see,’ Ms. Colston said.”
“[T]hat’s what we believe, and what we see.”
Not, “That’s what we see, and what we believe.”
Over and over we have seen how people’s beliefs shape what they see.
Curious.
Prometheus
October 31st, 2007 at 2:27 pm
Dear Prometheus:
If I’m reading your Figure 2 correctly, then the r squared for the decrease in MR/increase in autism rates looks very close to 1 — which is an almost unheard of result outside of incredibly well designed and controlled experiments.
I’m asking that as a clarification point. If I’m reading Figure 2 correctly, then the conclusion is that the “increase” in autism is almost fully explained by the “decrease” in MR. That would gut any “increase in autism” argument, supporting the conclusion that there isn’t any real increase in autism, but rather a shift in the diagnosis categories — and, possibly that is due to educational services available under IDEA and state IDEA laws for persons with ASD diagnoses.
I find that http://www.stats.org contains many excellent articles explaining what the numbers thrown about in many articles in the press actually mean, and criticizing their (mis)reporting in the press and by advocacy groups. It also contains links to a number of excellent sites discussing statistics.
“scare-tistic” — Excellent!!! I’ll remember that term. It deserves to be a word.
October 31st, 2007 at 2:42 pm
WFJAG,
The R-squared value is essentially one (0.9999).
However, I must point out that I don’t find the USDE IDEA data to be reliable for following trends of either autism or mental retardation. We simply don’t have a valid source of historical data at this point.
My purpose in showing this curious parallel trend is to give those who do believe that the USDE or California DDS data give meaningful information about historical trends in autism prevalence something to chew on. If you believe in the “autism epidemic”, as “supported” by USDE or California DDS data, then you would also have to believe that whatever is causing the “autism epidemic” is also causing a …. well, whatever the inverse of “epidemic” might be in the prevalence of mental retardation.
What this does do is show yet more evidence that the USDE and California DDS data are not reliable for following historical trends.
Prometheus
October 31st, 2007 at 4:27 pm
Thanks. I hadn’t considered that as the explanation, which makes a lot more sense. Outside of the physical sciences I’ve never seen a r squared anywhere close to 1, which is why I questioned it. It reminds me of the stork population/decreasing birth rate decreasing in Europe example one of my stat profs used to drill into our heads about the importance of ensuring that the underlying data and analysis made sense: “The stork population is decreasing; the birth rate is decreasing; therefore the one-to-one correlation of these facts shows. . .”
October 31st, 2007 at 8:54 pm
My answer to pD would be that it’s not a random sample of MR kids who are now being labeled as autistic. It’s basically the MR kids who *are* autistic, and they probably have the male-female ratio of autistic kids. Basically autism is being recognized more frequently in the MR population. It’s not an issue of misdiagnosis per se.
November 2nd, 2007 at 8:22 pm
Joseph: you make a good point in that IDEA requires kids to be reported in one and only one service category, even though the categories cover conditions that aren’t mutually exclusive. Therefore, shifts in choice of which category is primary can affect the trends. Yet another reason why administrative prevalence doesn’t always track true prevalence.
AS is probably underestimated because lots of AS kids aren’t in special ed (and may not need to be). I seem to remember one Canadian study (can’t remember if it was Mottron or Fombonne) that actually showed a higher administrative prevalence of AS in girls than boys. My first thought was that AS traits may “stick out” more in girls than boys and therefore girls would be more likely to receive assessments. Still another administrative/true distinction.