One More Air-Pollution Study

In the last post I confused this study (PDF) with an earlier one by the same group of researchers; I wrote about the earlier one, but linked to a post on Paul Whiteley's blog about the more recent one, which was published just last November.

(I also started describing this one, and then switched to describing the earlier one; in my last post, only one of the studies I mentioned used air-pollution data from the EPA's air-monitoring stations. The earlier study by these authors only used proximity to high-traffic roadways as their variable indicating pollution exposure.)

This paper combined the methods of the two studies I wrote about in the last post; it used the same pool of children born in California between 1997 and 2006 and drew on two sources of data on air pollution at the time and place those children were born: the EPA's air-quality data that I wrote about yesterday, and a computer model of average traffic flow, and exhaust emissions, along California's major roadways.

To some extent, you could see it as a more geographically dispersed version of the study I described yesterday that looked at prenatal exposure to air pollution in just Los Angeles County. 

(Weirdly, the LA-County-only study, though restricted to a smaller geographic area, involved way more people than the two traffic-related studies: 7,603 autistic and 75,782 control subjects, as opposed to this study's 279 autistic and 245 control subjects.)

But the relatively straightforward EPA data, which are direct measurements of the concentration of various pollutants at regular intervals, and which are only abstracted from each child's actual prenatal exposure in that 1) they do not measure what concentration of those pollutants actually got into the mothers' bodies, much less the fetuses', and 2) they were taken at sites some distance away from where the children actually live. 

So it's not a perfect data source, but it's still a lot more directly reflective of reality than this computer model seems to be:

The principal model inputs are roadway geometry, link-based traffic volumes, period-specific meteorological conditions (wind speed and direction, atmospheric stability, and mixing heights), and vehicle emission rates. Detailed roadway geometry data and annual average daily traffic counts were obtained from Tele Atlas/Geographic Data Technology in 2005. These data represent an integration of state-, county-, and city-level traffic counts collected between 1995 and 2000. Because our period of interest was from 1997 to 2008, the counts were scaled to represent individual years based on estimated growth in county average vehicle-miles-traveled data. Traffic counts were assigned to roadways based on location and street names. Traffic volumes on roadways without count data (mostly small roads) were estimated based on median volumes for similar class roads in small geographic regions. Meteorological data from 56 local monitoring stations were matched to the dates and locations of interest. Vehicle fleet average emission factors were based on the California Air Resource Board's EMFAC2007 (version 2.3) model. Annual average emission factors were calculated by year (1997-2008) for travel on freeways (65 mph), state highways (50 mph), arterials (35 mph), and collector roads (30 mph) (to convert to kilometers, multiply by 1.6). We used the CALINE4 model to estimate locally varying ambient concentrations of nitrogen oxides contributed by freeways, nonfreeways, and all roads located within 5 km of each child's home. Previously, we have used the CALINE4 model to estimate concentrations of other traffic-related pollutants, including elemental carbon and carbon monoxide, and found that they were almost perfectly correlated (around 0.99) with estimates for nitrogen oxides. Thus, our model-based concentrations should be viewed as an indicator of the traffic-related pollutant mixture rather than of any pollutant specifically.

So, to arrive at an estimate of how much of a certain category of air pollution (traffic-related air pollution) each mother and child in their study had been exposed to, they used a computer model to come up with average emissions for vehicles all over the state, traveling at various average speeds corresponding with their various categories of roads, for each year in their study. Then they entered that, along with all the other types of data mentioned above (winds, atmospheric conditions, traffic volume, road layout) into another computer model to arrive at the final answer.

I'm not criticizing their model; it actually seems like a pretty good one to my untrained eye. But my point is that there's a lot of extrapolating, averaging, assuming that what's true for location x will also be true for location y, and other things that make the model work but aren't grounded in direct observation and thus might not actually be true. 

That will be the case for any model, and this one has a few serious gaps in its data pool. They're missing eight years of traffic data from their eleven-year "period of interest," so they have to guess at what those numbers might be based on expected growth in traffic volumes. They're also missing traffic counts for some roads, so they estimate them based on the counts for other, similarly-sized roads.

It bears repeating that this model was not their only source of data on pollution exposure; they also used direct measurements taken by the EPA air-monitoring station(s) nearest to study participants' houses throughout the study period.

For traffic-related air pollution --- the type of pollution exposure they modeled rather than measured directly --- they found a difference between the highest- and lowest-exposure groups (with the former three times as likely to develop autism as the latter), but no difference between the lowest-exposure group and the two groups in the middle.

For the specific pollutants measured at EPA air-monitoring stations --- coarse and fine particulate matter, nitrogen oxides, ozone --- they found an increased likelihood of autism associated with greater exposure to particulate matter and nitrogen oxides, but not ozone. This effect was strongest during the third trimester of pregnancy. 

Unlike the other study I described that used the EPA air-quality data, this one did not find any change in the pattern when they adjusted for sociodemographic variables like child's sex, race/ethnicity, parents' educational level, mother's age, or mother's smoking during pregnancy.

Volk, H., Lurmann, F., Penfold, B., Hertz-Picciotto, I., & McConnell R. (2013). Traffic-Related Air Pollution, Particulate Matter, and AutismAir Pollution, Particulate Matter, and Autism JAMA Psychiatry, 70 (1) DOI: 10.1001/jamapsychiatry.2013.266

Bizarre Things Purported to Cause Autism: Early Exposure to Air Pollution

A quick note: my (very) occasional "Bizarre Things..." series was never intended solely as a crank roundup. No, in my mind I resolved to cover every autism hypothesis*, however well- or ill-founded, plausible or implausible, that I ever heard of and thought "well, that's weird!"

Indeed, I have tagged (though not titled) a post discussing a pretty rigorous, well-thought-out study investigating something that's been common wisdom for a long, long time with the "bizarre hypotheses" label just because its tentative conclusion --- that there may be something more to the relationship between autism diagnoses and social class than the fact that rich people can afford to get their children seen by specialists and poor people can't --- surprised me.

So sometimes bizarreness is in the eye of the beholder, I guess.

Anyway, on to the business at hand.

Paul Whiteley has written a couple of posts on the idea that exposure to air pollution might play a role in determining whether a kid develops autism; he talks specifically about these two studies working with the same two data sets: the state of California's Department of Developmental Services' records of all children diagnosed with autism in the state, and where they lived at , and also data from the federal Environmental Protection Agency's Ambient Air Monitoring Program.

I've written about the former data set before, but not the latter. The EPA measures six major outdoor air pollutants (I'm sure they do a lot more, too, but these six are the ones that affect air-quality indices): two sizes of "particulate matter" (dust, ash, soot, smog), the "coarse" particles measuring between 2.5 and 10 micrometers in diameter or length, and "fine" particles smaller than 2.5 micrometers; carbon monoxide; nitrogen oxides; sulfur dioxide; lead; and ozone. They have monitoring stations set up all over the country, particularly thick on the ground near big, sprawling cities. How often the stations record measurements varies with what kind of equipment is being used to take them; some pollutants can only be measured daily, or once every few days, though some can be measured hourly. Either way, it's a huge volume of data.

The authors of the more recent study (PDF) restricted their analysis to Los Angeles County, so I can actually tell you how many monitoring stations' data that would encompass.

Map showing the locations of all the EPA air monitoring stations in LA County --- made by me!

This PDF lists, among lots and lots of other things, all the EPA air-monitoring stations in California and where they are located. Within Los Angeles County, it looks like there are nineteen: one in Commerce, one in Azusa, one in Burbank, one in Industry, one in Compton, one in Glendora, one at the Los Angeles International Airport (LAX), two in Long Beach, two in Los Angeles, one in Pasadena, one in Pico Rivera, one in Pomona, one in Vernon, one in Reseda, one in Santa Clarita, one in Santa Fe Springs, and one at the Van Nuys Airport. So I guess that's technically five that are somehow part of LA or attached to it. Not all of them measure every one of the six pollutants, either, but you can see which of them track what in the PDF I linked earlier.

The earlier study** focused on families living near highways throughout the state of California, and while I might theoretically be able to put together a list of all the cities and towns that have highways running through them, and cross-check that with the EPA's records of where their monitoring stations are (assuming they publicize them all), it sounds like more armchair detective work*** than I want to do. So I can't tell you how many stations are contributing their data to this study, but I'd guess that it's more than the LA County-only study used.

Anyway, both studies involved matching the geographic location of each autistic child in the study with some other spatial variable: for the earlier study, this other variable was distance between where they were born and a freeway or major road; for the more recent study, it was the EPA monitoring station nearest to where they were born. In that study, the researchers looked at the measurements for each pollutant recorded nearest to each child's birthplace averaged over each trimester of gestation.

Both studies compared their autistic subjects to same-age, same-sex peers from the same general area (LA County for the one study, not specified for the other); the ratio was 1:1 in the smaller, older study and 10:1 (control:autism) in the bigger, newer one. So the idea was, I guess, to check whether kids living in the same county, city, suburb, or whatever as a kid without autism tended to have lower prenatal exposure to air pollution (new study) or live somewhat further away from the closest high-traffic road (old study), than the kid with autism.

What did they find? It's complicated; in the newer, just-LA-county study, they found that exposure to some of the six pollutants was actually associated with a slightly lower likelihood of having an autism diagnosis, at least in the (more****) raw data. The only pollutant to show a significant increase in odds ratio (a measure of how much more likely a child exposed to a given pollutant is to develop autism than an unexposed child; if it's less than 1, it means less likely, more than 1 means more likely) before logistic regression was ozone, which gave a 1.19 odds ratio. (That's also the biggest increase found anywhere in this study, regression or no).

After adjusting for a bunch of variables they had expected to correlate with pollution exposure (maternal age, maternal education, race/ethnicity, gestational age and others), the odds ratio for ozone went down while the odds ratios for the other pollutants (nitrogen oxides, carbon monoxide and particulate matter) went up. Where they had been sitting at about 0.8 or 0.9 (i.e., maybe ten, fifteen or twenty percent less likely?), they moved to about 1.05. 

I have only the faintest notion of what a logistic regression actually does, but my understanding of it is that, when researchers have a lot of variables closely intertwined with the variable they're trying to study, they use a logistic regression to "correct for" those other variables. It's like a way to try and zero in on the one strand in the snarl that you're trying to trace.

Anyway, they also did another regression, this time by maternal education only, and compared odds ratios for autism risk associated with each pollutant among three groups: mothers with less than a high-school education, mothers with a high-school education, and mothers with more than a high-school education. Except for ozone, the odds ratio for each pollutant went up slightly as maternal education increased, with the biggest differences between the least- and most-educated categories. 

Somehow, the mother's level of education affects how strongly exposure to these airborne pollutants predicts whether her children develop autism.

I have absolutely no clue what to make of that, if I'm even reading it right.

The roads study's outcome was less brain-twisting: they found a correlation between living near a freeway (i.e., a state or interstate highway) and getting a diagnosis of autism, but they found no such correlation for other high-traffic roads. Since it's probably not the case that different kinds of pollutants are being spewed out by vehicles on highways vs. other big, heavily-traveled roads*****, I'm going to follow David Gorski's lead and write off the freeway association as an artifact of data-dredging.

Becerra, T., Wilhelm, M., Olsen, J., Cockburn, M., & Ritz, B. (2012). Ambient Air Pollution and Autism in Los Angeles County, California Environmental Health Perspectives DOI: 10.1289/ehp.1205827

Volk, H., Hertz-Picciotto, I., Delwiche, L., Lurmann, F., & McConnell, R. (2010). Residential Proximity to Freeways and Autism in the CHARGE Study Environmental Health Perspectives, 119 (6), 873-877 DOI: 10.1289/ehp.1002835

*Yes, I have yet to do posts on the big ones, like mercury, thimerosal (really part of the mercury one, but some people propose mercury-based explanations that don't mention thimerosal, so they probably deserve separate posts --- anyway the thimerosal one would be long enough even if it weren't folded into a larger post on mercury compounds), MMR. It's less a matter of not knowing what I want to say than it is a matter of figuring out whether, or how, to marshal all the available evidence.

**I belatedly find out that this study is actually a precursor to the one Whiteley is writing about, not the same one. The one he writes about takes an approach that looks to me like a blend of the two I'm writing about now: looking at both air-pollution data and proximity to high-traffic roads.

***Actually, it's not even armchair detective work, as I am not in an armchair when I use the computer. It's armless-hard-wooden-chair detective work, which is somewhat more grueling than armchair detective work. 

****I'm not sure how much, if any, of the data in this paper can be considered "raw" when you consider how much tinkering they did with the air-pollution data.

*****You could perhaps make an argument that it could be the case, what with there probably being more trucks on the highways than not, and with trucks using diesel instead of gasoline. I am unsure if that would make a difference in how much particulate matter or nitrogen oxides are produced, and also quite skeptical that such a difference would show up at all in a study like this.

Two Quick Links about Why the Fight over Evolution Matters

So, apparently Kevin Drum, a political commentator I don't read but who is apparently somewhat well-known, has written an article for Mother Jones declaring the fight over how whether biology is taught in public schools an irrelevant sideshow. 


I was nonplussed by this. Obviously, it's not the most important issue in the world --- it doesn't pose an existential threat the way climate change, peak oil/peak everything, overpopulation and hunger do, and it doesn't have the dire social consequences (or moral urgency) of, say, restricting access to contraception, the mass incarceration of people of color and poor people, increasing economic inequality or the deliberate demolition of the welfare state --- but I still felt like it had some importance that went beyond mere cultural border-policing*.


Amanda Marcotte at Pandagon and PZ Myers at Pharyngula both responded to Drum's article, and each of them made points that rang true for me, and helped explain why his statement bothered me.


Amanda:

I have to admit, I'm boggled at Kevin Drum's reaction to the news that nearly half of Americans are young Earth creationists ... . He completely misreads the situation and frankly does so in a way that I personally felt thrown under the bus.

... 

Kevin takes it as a given that fights over what's taught in high school are strictly about symbolism and have no real importance. I suspect that's a much easier view to take if you're the beneficiary of a good public school in a blue area or lucky enough to have gone through or been able to put your own kids through private school. For someone who went to a rural high school in Texas, the notion that high school doesn't matter strikes me as ridiculous in the extreme. 

The reason conservatives target high schools (and junior high schools and elementary schools) isn't because they're playing for peanuts. On the contrary, conservatives understand something liberals don't, which is that if you get people while they're young, you usually have them for life. This is also, incidentally, why conservatives pay more attention to pop culture than liberals. Liberals are great people --- I'm one of them! --- but  we have a tendency towards preening individualism and therefore discount the importance of things like what's in the classroom and what's on TV because we personally feel we're iconoclasts who aren't affected by it. Which can, in this case, cause us to neglect to remember that in fact this is the air that most people breathe, and the quality of that air matters. 

I also had strong feelings of no you're wrong when I read the paragraph in Drum's article where he dismisses "a 10th grade understanding of evolution" as something so piddling as to be dispensed with entirely, so much does it pale in comparison to the understanding of evolution one gets (presumably) in college, or in books on evolutionary biology. PZ, along with a slew of other biology professors in his comments and at Mother Jones, takes him to task for this; they write that, no, actually, they have noticed a change in how readily their students grasp what they have to teach. Almost as if high school curricula were geared toward preparing the students for college ... 


Anyway, what felt unfair to me was his assumption that, if people needed the knowledge, they could always pick it up elsewhere. This made me angry, because for many people there is no "elsewhere," or at least not one that is immediately accessible without guidance. 


(Who has tried to teach themselves a completely novel subject by checking out books on it from the library? Isn't it hard to figure out which books will be the ones you need? Technical books are often overly specific and assume you've already got a grounding in the subject, and are interested in exploring a particular question within it, and books for laypeople might not be thorough enough, and also might not be trustworthy). 


As for college, well, not everyone goes there, most of the people who go there will never encounter evolution in the classroom --- I didn't, and I majored in a biological science! --- and, again, college professors have enough to do without having to burn up lecture time with remedial material and correcting misinformation. Plus, not all colleges are created equal; some 


To top it off, what you know before college helps determine where you go, whether you go at all, and what you study! College is expensive, and for many prospective students (I know I felt this way when I was applying to colleges) the size of the scholarship you can get makes the decision for you. Scholarships for merit, as opposed to need (and there will always be more people who can't really afford college than there are people officially deemed too poor to afford college on their own), depend on how one performs on standardized tests. And what does standardized test performance depend on? What you learned in high school.


There's more: as Amanda also points out, what you're exposed to in high school shapes your interests and aspirations:

[M]y high school biology course didn't teach evolution. Without evolution, biology actually doesn't make sense, and instead it's just an anatomy class. ... I had no idea how fascinating biology actually was until I was an adult, and long past any chance of starting on that as a career path. Not that I think I would, but you can easily see someone like me making that choice as a young woman, but not really being able to because I was never offered that option in a realistic sense to begin with. 

(As someone who did study biology, and who loves Amanda's writing about science- and skepticism-related topics as much as I do her feminist writing, I can see it. Her interest in the subject is obvious, as is her commitment to find out what's true and how we know it's true. While she won't --- can't --- go into the kind of technical details I sometimes do here, her grasp of general, foundational principles is firm enough that I can easily forget she does not possess a biology degree herself.)


To shift to another aspect of why Drum's assessment is so horribly, horribly wrong, I quote PZ:

[T]his is going to be the century of dependence on the sciences. Climate change is going to hit us all; environmental crises are going to rise up all over the place; we're going to face shortages of energy and fresh water; emerging diseases will be a major concern; new biomedical technologies will cause cultural shocks; the whole world is going to change. Most people, I agree, will not be doing the research that leads to changes, and most of these problems will require political and social changes to correct, but how are you going to convince people to, for instance, change their fuel consumption habits when they're in complete denial of basic facts? How can you expect people to appreciate the importance of ecology and global interactions when you tell them that evolution doesn't matter? How will you get them to make rational decisions to control pandemics when they can't comprehend probability, epidemiology, and viral/bacterial evolution on even the most basic level? 

Most importantly, though, this utilitarian attitude that all that matters is what people can directly use in their day-to-day life is a denial of the Enlightenment and principles on which our country was founded. It's a rejection of the liberal idea that human beings should be well-rounded and informed individuals --- the informed citizenry that should be the foundation of a democracy. We can't expect everyone to be biologists or poets or political scientists, but we should expect that one outcome of a public education is an appreciation of the breadth of human endeavor, and at least a smattering of the fundamentals of a wide range of subjects, sufficient that, to make it practical again, students can make informed career decisions and understand a basic argument from evidence from an expert. We lack that now. And to wave away a simple but essential starting fact about our existence as unimportant is deeply offensive.     

I would add to his first paragraph that you don't have to wait for some futuristic nightmare scenario for these things to affect people's lives. Disease-causing bacteria are already becoming resistant to a whole lot of the antibiotics we use to fight them; lots and lots of species have disappeared or are disappearing; sensitive ecosystems are already being stressed, some to the breaking point by human activity, and humans are already failing to see why they should care***.

The picture he paints in that first paragraph has already come to pass, in the way that the proverbial camel is already in the tent, he just needs to take a few steps forward before it becomes obvious to everyone.

I would also add something else that I think denying evolution does: it makes it easier to think of humans as separate from, and above, the natural world and all those other animals. It allows us to think that what we do to it, to them, will never get back to us. 

(You don't have to deny evolution to get to that mindset, though: I see it arising just as readily from the erroneous view of evolution as a teleological process of development from simple to complex organisms, with humans sitting triumphantly at the apex. I've posted before about my annoyance with this picture of evolution.)

*This is what Andrew Sullivan, another famous political commentator whom I know to be obtuse, thanks to driftglass, an obscure political blogger who is quite acute**, has called belief in creationism a "cultural signifier" that describes the group membership of the believer rather than what that person actually believes about reality. That whole line of reasoning makes me queasy, but even if he is right that no one actually believes the Earth is six thousand years old and every animal species living on it was molded from clay and animated with the breath of a god, if that is what they decide shall be taught to children as fact, in place of the full history of life on Earth (as much of it as we know, anyway) and an explanation of how life on Earth has changed as Earth itself has changed, then those children grow up with a much poorer understanding of the world around them. An idea can have consequences in the real world whether people truly believe it or not; all they need to is act as if they believe it. Also, Andrew Sullivan is dumb.

**Hee. I made a geometry pun.

***This is not, at all, meant to be an exhaustive list, or even a primer on any of those subjects. I just put a recent news/research story in every link to make the point that, yes, this stuff is happening, and it's not just one or two isolated incidents. 

Autism Diagnoses Spread through Social Networks

EXECUTIVE SUMMARY: A little more than a year ago, a group of sociologists analyzed data from the state of California --- birth records and records from California's Department of Developmental Services --- to look for patterns in where autistic children were most likely to be found. They determined that children living near (defined here as being within 500 meters) a family where one of the children had been diagnosed with autism in the previous year were somewhat (1.26 times) more likely to receive an autism diagnosis themselves. They also found that 1) the effect was stronger for the less severe cases of autism, and also for children who were three years old when they were diagnosed (as opposed to 4, 5, or 6); and 2) children living very close to a family with a recently-diagnosed autistic child were about 28% less likely to be diagnosed with intellectual disability without a concurrent diagnosis of autism. The researchers concluded that parents who know other parents who've had a child diagnosed with autism are more likely to have their own children evaluated, and thus more likely to get a diagnosis for their own children.
___________________________________________

An article published last year in the American Journal of Sociology argues that knowing someone in your neighborhood who has an autistic child makes it likelier that your child, too, will be diagnosed with autism.

The authors' explanation for this is not that autism is "catching," or that people with autistic children are likelier to live in certain places (they investigated both of those possibilities, along with the possibility that some environmental contaminant makes some neighborhoods have more autistic children than others); it's that parents talk to each other about their children, and if one family has an autistic child, and has gotten that child diagnosed and enrolled in services, those parents will tell other parents about that process, and thereby make it easier for other parents to do the same thing. (They may also, by describing their autistic children's behavior, make it more likely that other parents will consider the possibility that their child may be autistic).

What these researchers --- sociologists Peter Shawn Bearman, Marissa D. King (who has conducted a lot of research into autism demographics) and Ka-Yuet Liu --- did to derive this explanation was look at data on all children born in California since 1997, with younger siblings also born in California. (They needed to restrict their sample to children with younger siblings to get information on where the firstborn children spent their early years --- if the family's residence is the same for every child born to that family, the researchers could assume that the firstborn lived in the same place during the years between hir own birth and the birth of hir youngest sibling. That's the information they wanted --- where families lived during their children's early years). This gave them a pool of 533,244 children whose families lived at the same address when they were born as when their younger siblings were born. Then, they used data from the California Department of Developmental Services (I've written about this data set before, in a different context --- it's been used in a lot of studies investigating autism prevalence) to determine which of those children got an autism diagnosis, or a diagnosis of intellectual disability*, between the ages of 2 and 6.

They created a statistical model to see if they could accurately predict which children would end up with a diagnosis of Autistic Disorder**. Their model, which is supposed to predict the odds of any given child being diagnosed with autism in a given year, t, uses the physical proximity of a child diagnosed with autism in the previous year, t - 1, as its predictive variable. They divide this continuous measure --- how far away does [other child with autism] live? --- into six discrete groups: 1 - 250 m; 251 - 500 m; 501 m - 1 km; 1001 m - 2 km; 2001 m - 5 km; and all distances over 5 km. (They used a continuous measure of distance, too). They controlled for a slew of other variables --- sex, maternal age, socioeconomic status (measured by mother's educational attainment and whether she received state assistance with pregnancy- and birth-related medical expenses), proximity to autism-related healthcare services, neighborhood population density, and neighborhood median income --- and predicted that the odds of a child being diagnosed with autism would increase the closer they lived to a child already bearing that diagnosis, and that the odds of a child being diagnosed with intellectual disability independent of any other condition would decrease with increasing proximity to a child diagnosed with autism. They also predicted a pattern in what kind of autistic children would show the greatest change in likelihood of diagnosis: they thought that the "high[est]-functioning" children, who scored the lowest on their measure of autism severity***, would show the greatest effect, along with the youngest children (age 3 or younger). Their reasoning is that, since these are the most ambiguous cases, whether or not their parents knew someone else with an autistic child, and thus got it into their heads to consider whether their child might also be autistic, would make the most difference, since there are less likely to be really obvious, textbook signs of autism. (Or, if signs are present, say, in the very young children, knowing another family with an autistic child might predispose the parents to have their child evaluated sooner, rather than waiting to see if, say, the child is just a late talker).

Sure enough, they found the patterns they'd predicted.

This graph shows the clustering of actual data points around their theoretical curve; you can see it's a pretty close fit.

(Figure 2A, in Bearman, King and Liu, 2010. The x-axis shows distance, in kilometers, from the nearest home with a child with an autism diagnosis, and the y-axis shows the probability that any given child living at that distance will be diagnosed with autism within the next year).

Here's a different representation of the same data:

(Figure 2B, in Bearman, King and Liu, 2010. The x-axis still shows distance, but the y-axis, instead of showing probability of an autism diagnosis in the next year, now shows the odds ratio between each distance category and the reference category, which is 501 m - 1 km. The five different dots in each distance category represent the five different statistical models that were used. You can see that, compared to the middle distance category, the children in the two shorter-distance categories were more likely to receive autism diagnoses, while the children in the three longer-distance categories were less likely to be diagnosed.)

From the Results section:

[W]e also report the effects of proximity as a categorical variable. All odds ratios (ORs) are relative to the reference category of 501 meters - 1 kilometer. ... In figure 2B, we can observe that residing in close proximity to a child diagnosed with autism increases one's chance of being diagnosed with autism in the subsequent year. Compared with children who are 501 meters - 1 kilometer away from their nearest neighbor with autism, those in close proximity (1 - 250 meters) to a child with autism have a 42% higher chance of being diagnosed with autism in the subsequent year. Proximity of 201 - 500 meters increases the chance by 22%. In contrast, being farther away from a child with autism reduces the chance of a diagnosis. Although the last three categories were all significantly associated with the decrease (-21%, -36%, and -49%), there were no statistically significant differences among these three categories. This is consistent with the results in figure 2A, which show that the effect of proximity is strongest within one kilometer, followed by a flat tail.

And here are the graphs showing the effects of proximity to a family with an autistic child on 1) a child being diagnosed with autism (with intellectual disability) rather than with intellectual disability alone ...

(Figure 2C, in Bearman, King and Liu, 2010. The x-axis shows distance; the y-axis shows odds ratios --- compared, again, with the middle-distance "reference" category of 501 m - 1 km --- of being diagnosed with intellectual disability by itself. You can see that living closer to a family with a recently-diagnosed autistic child made it less likely that a given child would be given a sole diagnosis of intellectual disability, although it is only at the farthest distances, 5 kilometers or more, that children become more likely to receive a diagnosis of intellectual disability independent of any other condition).

... and 2) being diagnosed with autism at varying levels of severity:

(Figure 3, in Bearman, King and Liu, 2010. The x-axis shows the percent increase in odds of receiving an autism diagnosis associated with proximity to a recently-diagnosed autistic child; the y-axis shows the child's level of disability as assessed by the state's Client Development Evaluation Reports. The highest-scoring, or least impaired, 20% are compared to the bottom, or most seriously impaired, 20%, and to the middle 60%. You can see that the "highest-functioning" group shows a stronger effect of proximity to another autistic child on their chances of being diagnosed than either the middle group or the "lowest-functioning" group, whose chances of being diagnosed with autism were least affected by proximity to another recently-diagnosed autistic child.)

The researchers were pretty well satisfied that the effect they found for proximity wasn't an artifact of shared environmental contamination because they saw the same pattern in a large, diverse sample of neighborhoods. They looked at rural, urban and suburban communities, which would all have their own distinct toxicological profile. (For instance, pesticide exposure would be a likely risk factor in rural areas, and a possible one in suburban areas, but pretty unlikely in urban areas. Similarly, air pollution is probably only a serious factor in dense urban areas.) There were other patterns that would not be explained by shared environmental risk factors or viral transmission, but that do fit with the social-diffusion hypothesis: first, a strong tendency of people coming in for diagnostic evaluations to have been referred there in the way as their nearest neighbor with an autistic child, and second, for people living on or near a border between school districts, the proximity effect was only seen for children living nearby who also attended school in the same district.

So, I find all this pretty supportive of their idea that social networks play a role in determining who is and who is not diagnosed with autism; what is less clear to me is how much of a role they play compared with other factors, like socioeconomic status, parents' ages, male gender, or having an autistic relative.

What Drs. Bearman, King and Liu did to try and answer that question is calculate the population attributable fractions (PAFs) --- the difference any given risk factor makes in the number of people who develop a certain condition --- for proximity to another autistic child, and also for each of three sociodemographic indicators: mother's age at the time of her child's birth, mother's educational attainment, and whether the state of California paid for her prenatal and obstetric care. By this metric, they found a greater role for proximity than for any other factor: living within 500 meters of a family with a recently-diagnosed autistic child raised children's likelihood of being diagnosed with autism in a given year by 16%.

While this may be a larger effect than any of the other, more-established risk factors, it's not larger by all that much, and in absolute size it's also not very big. I would expect that, though --- most children don't have autism, so the most a social diffusion of autism awareness could do is make the parents of those children who do have autism take them in to be evaluated sooner. (Or, maybe for a few families, at all --- but with schools getting more and more proactive about identifying special-needs children, it's not very likely anymore that a child with autism could get all the way through K-12 without being flagged somewhere along the line).

Liu KY, King M, & Bearman PS (2010). Social influence and the autism epidemic. AJS; American journal of sociology, 115 (5), 1387-434 PMID: 20503647

*They tracked this category as well to look for social/spatial patterns of diagnostic substitution --- are children who are more likely to be diagnosed with autism also less likely to be diagnosed with general (i.e., non-autism-associated) intellectual disability?

**They only counted children in this category, and not any other pervasive developmental disorder, because only Autistic Disorder is a category eligible for state services. The researchers felt that parents' main motivation in pursuing a diagnosis would be to gain access to services, so they didn't bother tracking the diagnoses that wouldn't be useful toward that end.

***Their severity measure involved nine items from the California DDS's Client Development Evaluation Report (CDER) on each child. (Here is a PDF of the complete CDER form, if you want to go and look at what exactly these items are measuring). The overall measure was an equally-weighted average of scores on three "scales" --- communication (made up of 3 CDER items: item # 58, word usage; item # 61, receptive language; and item # 62, expressive language), social interaction (made up of 5 CDER items: item # 27, peer interaction; item # 28, interaction with nonpeers; item # 29, friendship formation; item # 30, friendship maintenance; and item # 31, participation in social activities), and a single CDER item, # 42, measuring repetitive and stereotyped behaviors. (I find it a bit odd that they didn't include self-injurious behavior as part of their severity measure --- there are two items on the CDER addressing that, and the presence or absence of self-harm seems to me a more important measure of how "severe" someone's problems are than, say, being able to interact "normally" with one's peers.)

1 in 38?

That's the proportion of school-aged kids found to meet diagnostic criteria for an ASD by this team of mostly Korean researchers --- Drs. Young-Shin Kim, Bennett L. Leventhal, Yun-Joo Koh, Eric Fombonne (who has done a lot of research on the prevalence of autism), Eugene Laska, Keun-Ah Cheon, Soo-Jeong Kim, Young-Key Kim, Dong-Ho Song, and Roy Richard Grinker (the Unstrange Minds guy), along with Eun-Chung Lim and HyunKyung Lee --- when they screened elementary-school children in one district of a fairly large city (Goyang City, pop. 1,073,069; Ilsan district, pop. 488,590) near Seoul, South Korea.

Part of their reasons for doing this study (full text here) was their thought that previous estimates of the prevalence of autism rely too much on documentation from official sources --- national disability offices, special-education registries, autism-specific service providers, etc. --- that wouldn't include children who have not yet been evaluated for ASDs.

To try and get more of a cross-section of Korean children, the researchers recruited from both mainstream elementary schools (41, of which 30 agreed to participate), special-education schools (3), and the city's Disability Registry (all children between 7 and 12 listed as having either ASD or intellectual disability). Children from participating institutions were initially screened for autism using the Autism Spectrum Screening Questionnaire (ASSQ), which parents and/or teachers filled out about each child.

Not surprisingly, the general-population group, recruited from the elementary schools, was much larger than the developmentally-disabled group recruited through the Disability Registry or from special-education schools: in all 30 participating elementary schools, there were 36,592 children between the ages of 7 and 12. Of these, the parents of 23,234 completed the ASSQ and sent it in to the researchers; of the 294 students attending special schools and/or listed in the Disability Registry, 103 had forms submitted for them.

Anyway, following this initial screening, the students who scored high enough on the ASSQ (with a teacher-rated score of at least 10, or a parent-rated score in the top five percentiles*) were offered further diagnostic evaluation. Among the elementary-school students, 1,742 met those screening criteria, and 234 completed the evaluation. All of the special-education students were considered to meet screening criteria, and 52 of them completed the evaluation. Each child was evaluated by two teams, and each team included both Korean- and American-trained practitioners. The evaluators used Korean versions of the Autism Diagnostic Interview (ADI-R), Autism Diagnostic Observation Schedule (ADOS), and two cognitive tests, the WISC-III and a revised version of the Leiter International Performance Scale.

Of the 286 children who were evaluated, 201 met criteria for an ASD**. Just over half of these (101) met criteria for Autistic Disorder; just under half of those (48) had come from the special-education group. All but one of the students in that group who were diagnosed with an ASD were diagnosed with Autistic Disorder, while most of the students in the general-population group were diagnosed with some other ASD, either Asperger's syndrome or PDD-NOS**.

There is one thing I found particularly interesting in their results: the ratio of boys to girls among their ASD "positives" is significantly lower than the 4:1 most commonly cited in the autism literature. These researchers found a ratio of 5.1:1 among the special-education students (who were mostly boys to begin with --- of the 103 special-education students who were screened, 84 were male, so even if *all* of the girls in that group had autism, the ratio would still be skewed in the boys' favor unless relatively few of the boys had it) and a ratio of 2.5:1 among the mainstream elementary-school students.

That number adds some support to the hypothesis that autism remains underdiagnosed among girls --- the lower ratio was only seen in the students who were not flagged as disabled or special-ed, not receiving any services. Among the students known to be developmentally disabled, or already diagnosed with autism (i.e., those registered as ASD in the Disability Registry), the ratio was closer to the current conventional-wisdom ratio of 4:1.

(It also seems like the forms of autism that are most likely to go undiagnosed in girls are the "milder" forms, like Asperger's and PDD-NOS. Very few girls in the general-population sample met criteria for Autistic Disorder; of the 27 children meeting those criteria, only five were female.)

Anyway, back to the number making up the title of this post.

How they got that number is complicated. They divided the number of autistic students they extrapolated there to be based on the number they found (more about this later) by the total number of seven-to-twelve-year-old children living in the Ilsan district of Goyang City, which is 55,266.

So much for the easy part.

There were several steps involved in correcting for the large proportion of nonresponders; first, and easiest, they guesstimated the likely full scores of the people who turned in partial ASSQs. (The simplest method they had for doing this was simply to take the average single-item score on a form in which only some of the items are answered and multiply it by the number of items on the full questionnaire, which is 27). Just doing that gave them 264 more subjects meeting initial screening criteria for an ASD. They also added in all the children listed in the Disability Registry as having ASD who did not participate in this study; this would add 101 subjects to the confirmed-ASD pool. (They left out all the non-participating children listed as having intellectual disability).

They also did some statistical manipulations that I don't understand (I had minimal coursework in statistics, and not even good old Google can tell me what a "weigh-back procedure"*** is) to correct for their only having given some of the people meeting their minimum screening criteria the opportunity for furthere evaluation. (See the Methods section and this supplemental section for their discussion of these techniques).

At the end of all of this, they ended up with an adjusted prevalence figure of 2.64%, or the 1 in 38 referred to in the post title. There are several ways in which this might be an overestimate: chief among them, the researchers assumed that the proportion of children meeting screening criteria who also meet diagnostic criteria for an ASD would be the same for the nonresponders as it was for the responders --- i.e., about 70%. Their basis for this assumption is a logistical regression model they created to try and predict parental consent to, and participation in, diagnostic evaluations. They found that since none of the co-variates they'd chosen for this missing "latent variable" --- child's age, sex and ASSQ score --- predicted whether a given child would end up with an ASD diagnosis or not, that parental willingness to have more tests done did not vary with the likelihood that their child is autistic. But all of those variables are somewhat crude proxies --- even the ASSQ score, when you consider that all of the "screen-positive" children had scores clustered within the same narrow window, and thus variation between them would be minuscule. So I think it's reasonable to think that maybe the model was wrong, and the parents who chose to have their children evaluated might have a higher proportion of autistic children than the parents who opted out.

However, there are just as many reasons to suspect that their numbers are an underestimate --- after all, they did write off all the nonparticipants labeled intellectually disabled, even though their own results show a substantial proportion of intellectually-disabled responders having autism. The authors also suspect underreporting of possible autism by both parents and teachers, because autism is stigmatized in South Korea (thus making parents less likely to want to admit that their child has it) and because the South Korean school day is so rigorous and structured (making teachers less likely to notice the social dysfunction that the ASSQ tests for, because there is so little socializing during a typical school day in Korea).

But the need to extrapolate so much data to make up for low response rates is a serious flaw in this study, though, as Prometheus points out, even when you restrict your gaze to the data that they actually have, you still see that the majority of confirmed ASD cases were in the general-population group, among children who had never seen a mental-health professional before.

Other bloggers posting about this study: Steven Novella at Science-Based Medicine; Sullivan at Left Brain/Right Brain and the Autism Science Foundation blog; Rose at Hard Won Wisdom; Michelle Diament at Disability Scoop; Paul Whiteley at Questioning Answers; Polly Palumbo at Momma Data; Aspie Editorial; Raphael Fraser at Music, Medicine & the Mind; Prometheus at A Photon in the Darkness; and MJ at Autism Jabberwocky (who has a detailed explanation of the ASSQ and its inherent biases)

*Only some of these children were referred for further evaluation: all the children scoring in the top two percentiles, half the children (chosen randomly, not by score) in the percentile third from the top, and one-third of the children in the percentiles fourth and fifth from the top.

**There's another interesting asymmetry within the general-population-derived ASD cases: a much greater proportion of those whose final diagnosis was Autistic Disorder had histories of psychiatric or psychological treatment. A majority (26 of 48) of students attending mainstream elementary schools who both met criteria for an ASD and had seen a psychiatrist or psychologist in the past ended up being diagnosed with Autistic Disorder, while the ASD students from mainstream elementary schools who had never seen a mental-health professional were mostly diagnosed with the other ASDs (27 of 104 with Autistic Disorder; 77 of 104 with either Asperger's or PDD-NOS).

***Perhaps it is how Mr. Peabody travels back in time?

Kim, Y., Leventhal, B., Koh, Y., Fombonne, E., Laska, E., Lim, E., Cheon, K., Kim, S., Kim, Y., Lee, H., Song, D., & Grinker, R. (2011). Prevalence of Autism Spectrum Disorders in a Total Population Sample American Journal of Psychiatry DOI: 10.1176/appi.ajp.2011.10101532

Too Many, Too Soon?

EXECUTIVE SUMMARY: One of the misgivings about current early-childhood vaccination recommendations is that they involve exposing infants to so many infectious agents. While there's already been research into the effect of multiple vaccines on the immune system, Smith and Woods (2010) are the first to look at whether getting lots of vaccinations very early in life might have any effect on the nervous system.

Using the vaccination records and neuropsychological data on a group of children born between 1993 and 1997 that were gathered and made publicly available for an earlier vaccine-safety study, they did two different group comparisons between children who had received all the recommended vaccines on time and children who hadn't. They found no ill effects of exposure to lots of different vaccines within the first seven months of life; the only effects of earlier vaccination on neuropsychological development that they found were beneficial ones, but these mostly did not survive controlling for other demographic variables.
_____________________________________________

In an interview with The Thinking Person's Guide to Autism editor Shannon Des Roches Rosa, Autism's False Prophets and Deadly Choices author Dr. Paul Offitt mentioned a study published last May in Pediatrics that addresses the "too many, too soon" version of the vaccines-cause-autism hypothesis.

Some parents of very young children, while they might not believe that any single vaccine is dangerous, notice that a lot more vaccines are recommended within the first six years of life now than there used to be, and wonder whether it might damage a child's immune system to expose that child to so many antigens within such a short period of time*.

While there was already evidence that the current vaccine schedule is safe from an immunological standpoint, Smith and Woods (2010) are the first researchers to look at its potential impact on the developing nervous system.

Their study re-analyzes data gathered for an earlier study looking at neuropsychological outcomes in 7-to-10-year-old children exposed to thimerosal in vaccines; those researchers recruited children from four HMOs participating in the CDC's Vaccine Safety Datalink project, tested them on 42 measures of neuropsychological outcomes (which broadly fell into seven domains --- speech and language, verbal memory, fine motor coordination, attention/executive functioning, behavioral regulation, tics, and general intelligence --- and also included two other tests, one of visuospatial ability and another of letter and word identification; see Table C, or pages 9-11 of this document, for a complete listing), and looked for relationships between a child's performance on those tests and the amount of thimerosal that child had been exposed to, as estimated from vaccination records.

That study didn't find much of an effect for early thimerosal exposure --- it found a small scattering of statistically significant effects, but those effects were 1) often not the same in boys and girls, or in children belonging to different HMO groups, 2) evenly divided between higher mercury exposure --> better outcome and higher mercury exposure --> worse outcome, and 3) very small. That suggests to me that many of those associations might simply have been noise.

What's more important in terms of whether the "too many, too soon" hypothesis holds water is the raw data those researchers collected --- complete vaccination records and neuropsychological test results for 1,047 children. The existence of those data allowed Drs. Michael J. Smith and Charles R. Woods to look for relationships between timing of vaccination in early infancy and later neuropsychological outcomes.

They compared outcome data for the 491 children who got all the recommended vaccines on time (which they defined as within 30 days of the recommended age, given in months, on the CDC's vaccination schedules for 1993-1997) with the data for the rest of the children (who fell into two categories: those who had gotten all of the recommended vaccines, but just didn't get them at the recommended time, and those who hadn't gotten all the shots by the time they participated in Thompson et al.'s study), finding that children who got all their recommended shots on time performed better than the other children on 12 of the 42 outcome measures --- the Boston Naming Test, the grooved pegboard (a test of fine-motor coordination), the metacognition subscale of the Behavior Rating Inventory of Executive Function, teacher ratings for inattention and hyperactivity, parent rating for stuttering, and verbal, performance and full-scale Wechsler IQ. There was no measure on which they performed worse than the less-vaccinated group.

Because there were demographic differences between the families of children who were vaccinated on time vs. children who weren't, the researchers also did a multivariable analysis to try and control for those factors. That analysis found only two (small) between-group differences, but both of these were also in the fully-vaccinated kids' favor. (They scored 1 point higher on the speeded-naming subtest of the Developmental Neuropsychological Assessment's Language and Communication domain, and also tended to have slightly higher performance IQs).


They also did a secondary analysis, dividing the study participants into three groups, based on "timeliness" of vaccine receipt. The "most timely" group (n = 310) was created by dividing all the children who'd gotten all 10 of the vaccines recommended for the first year of life into quintiles based on how old they were when they completed the schedule. The first two quintiles, or the 40% of children who completed the schedule earliest (i.e., within the first seven months of life), constituted the most timely group, while the children who had gotten six or fewer of the ten recommended vaccines by the time they were seven months old made up the "least timely" group (n = 112).

(The rationale for doing this secondary analysis was to look for any effect of spacing all the vaccinations really close together --- the children in the "most timely" group had crammed the ten first-year vaccines into the smallest window of time. Similarly, the group they're being contrasted with represents the opposite extreme in terms of vaccine compliance --- these are the children who've gotten the fewest shots within that same seven-month window of time. If there is a spacing effect, it ought to show up in a comparison between those two groups, if it shows up anywhere.)

The results of the secondary analysis were a lot like those of the primary analysis: univariate analysis turned up fifteen outcome measures on which the most-vaccinated children outperformed the least-vaccinated ones (including most of the same measures identified as showing a difference between groups in the primary analysis), while multivariate analysis found no difference between the groups. Overall, this study failed to find evidence of any deleterious effects of multiple vaccinations on neuropsychological development, even when those vaccines are spaced very close together very early in life.

(One of the potential objections the authors deal with in their Discussion section involves the fact that the vaccination schedule has changed since the mid-1990s, but they figure that, since most of the children in their study received the version of the combined diphtheria, tetanus and pertussis vaccine that includes whole-cell pertussis, their overall antigenic load would be heavier, not lighter, than what children are exposed to today. See my note below for more explanation about changes in the pertussis vaccine.)


Sources:
Smith, M., & Woods, C. (2010). On-time Vaccine Receipt in the First Year Does Not Adversely Affect Neuropsychological Outcomes PEDIATRICS, 125 (6), 1134-1141 DOI: 10.1542/peds.2009-2489

Thompson WW, Price C, Goodson B, Shay DK, Benson P, Hinrichsen VL, Lewis E, Eriksen E, Ray P, Marcy SM, Dunn J, Jackson LA, Lieu TA, Black S, Stewart G, Weintraub ES, Davis RL, DeStefano F, & Vaccine Safety Datalink Team (2007). Early thimerosal exposure and neuropsychological outcomes at 7 to 10 years. The New England journal of medicine, 357 (13), 1281-92 PMID: 17898097

*Actually, while it looks like we're throwing more things at young children's immune systems now than before, when you just look at the number of vaccines children receive now versus thirty years ago, the composition of vaccines has changed, too, so that each vaccine now contains fewer antigens than they used to. Looking at this table in a 2002 review on vaccine safety by Paul Offitt and colleagues, the switch from using the whole bacterium that causes whooping cough, Bordetella pertussis, to just using the toxin it secretes cut the number of distinct antigens in the pertussis vaccine from around 3,000 to just two to five (depending on how many other bacterial cell components are still mixed with the active ingredient, the toxin). That whole-cell pertussis vaccine looks like the proverbial pig in the python, swelling the total number of antigens in all the recommended vaccines to 3,000+ during the time period in which it was used. Now, with the adoption of the acellular pertussis vaccine, the total number of antigens in all the early-childhood vaccines is about 125, give or take a few. So the question of antigen overload is actually a lot more complicated than just comparing the number of vaccines recommended at different points in time, and, if anything, a closer look seems to tell us that young children's developing immune systems are less overloaded by the current vaccine schedule, not more.

PLoS ONE: Children from Wealthier Families More Likely to Meet Criteria for an ASD

EXECUTIVE SUMMARY: A recent study shows that wealthier families are somewhat likelier than middle- and low-income families to have autistic children. The study compares surveillance data on a cohort of eight-year-old children in fourteen areas of the U.S. in 2002 and 2004 whom the CDC decided, based on school and medical records, met diagnostic criteria for ASDs with U.S. Census data from 2000 on eight-year-old children living in those same areas of the U.S. Statistical analyses were done on both groups' demographic data, which showed that the autistic children tended to come from somewhat better-off neighborhoods than the general population of eight-year-old children. Because of the study's use of public-school records, and its inclusion of a sizable group of never-before-diagnosed autistic children, its authors believe the association they found between higher socioeconomic status and higher rates of autism prevalence is not solely due to richer people's unfettered access to diagnostic services, but instead might also reflect some other factor that positively correlates with both autism and higher SES: the authors suggest both advanced parental age and excessive hygiene as potential candidates for this mystery factor.
___________________________________

This article in the Kansas City Star alerted me to a study published in PLoS ONE on July 12, giving the first indications I've seen that the conventional wisdom about autism and social class --- that autism is mainly a disorder of the upper classes --- might have more to it than just the simple fact that people with unfettered access to health care are a lot more likely than people without such access to get diagnosed with autism-spectrum conditions.

The way I had understood it, there was no big mystery there: people who can afford to see a doctor in non-emergency situations are going to get all sorts of things diagnosed at a higher rate than people who can't afford that kind of preventive care. In that framework, a higher rate of autism diagnoses among children from wealthy families says nothing about autism per se; it just serves as one more example of the huge inequality of access to health care here in the U.S.

The study I mentioned --- carried out by Dr. Maureen S. Durkin and her colleagues --- tries to address how much of the association between autism and higher socioeconomic status (which Durkin says shows up about half the time in the existing literature on the topic) is due to this gap in access to health care and how much may be due to other factors yet to be uncovered.

To try to include autistic children lacking formal diagnoses in their analysis, Durkin et al. relied on the CDC's Autism and Developmental Disabilities Monitoring Network, whose methodology is described here* on the CDC's website:

In 2000, CDC established the Autism and Developmental Disabilities Monitoring (ADDM) Network to track the prevalence and characteristics of ASDs in the United States. The ADDM Network is a multiple-source, active, population-based surveillance system that reviews developmental records at educational and health sources and employs a standardized case algorithm to identify ASD cases.

There is also a flowchart showing where they get their data from and how they decide which children are actually autistic; unfortunately, the flowchart is still pretty vague about which "educational and health sources" they use. It does mention that they use "multiple" of each, though.

So, in the years 2002 and 2004, the ADDM Network collected data on 407,578 (in 2002) and 172,335 (in 2004) eight-year-old children living in fourteen areas around the country: the northern half of Alabama; the central region of North Carolina; the Coastal and PeeDee regions of South Carolina; the entire states of Arkansas and West Virginia; parts of the cities of Denver, Colorado; Atlanta, Georgia; St. Louis, Missouri; and Salt Lake City, Utah; and the cities and surrounding areas of Phoenix, Arizona; Baltimore, Maryland; Newark, New Jersey; Philadelphia, Pennsylvania; and Milwaukee, Wisconsin**. Of these children, 2,685 met diagnostic criteria for an ASD in 2002; in 2004, the number was 1,376.

Durkin et al. took the data on all the children determined to be autistic --- minus the ones living in Utah and West Virginia, since those sites didn't provide enough information to determine socioeconomic status for the children living there --- and compared prevalence rates for ASDs across three socioeconomic "tertiles": high, middle and low socioeconomic status, as determined by three factors: 1) percentage of people in the area living above the poverty line, 2) percentage of adults 25 and over in the area who have at least a bachelor's degree, and 3) median household income for the area. They also compared those factors between the autistic children and the general population of eight-year-olds living in the same areas at the time of the 2000 census.

Compared with the general population of eight-year-olds living in the same areas, autistic eight-year-olds are somewhat less likely to live in "poverty areas," or areas defined in the 2000 census as having 20% or more of its families with children living on poverty-level incomes, with 16.8% of autistic children living in such areas and 25.8% of all children living in them.

Similar small-but-noticeable differences show up with respect to the other two factors, median household income and proportion of adults with college degrees. Among the general population, the median household income was $42,898; among families with autistic children, it was $50,114. Autistic children, on average, tended to live in census blocks where a larger proportion of adults 25 and older had bachelor's degrees; 30.3% had them compared with the general-population figure of 24.8%.

These differences kept showing up, even when the comparisons were set up differently: besides comparing autistic children with their (approximate) peers on measures of socioeconomic status, the researchers also compared autism prevalence rates across the three socioeconomic-status categories they'd created, and found that the richest group had the highest proportion of children meeting criteria for autism, the middle-income group had the second-highest, and the poorest group had the lowest. This pattern showed itself across all the racial categories included in the study, too.

Now, the question is, does the design of this study correct sufficiently for (well-established) class differences in access to specialized medical and educational services? Can its findings of autism prevalence increasing with socioeconomic status be taken at face value?

Part of the reasoning behind the authors' contention that at least part of the association between autism and wealth is real lies in their use of school-based, as well as clinic-based, documentation. Theoretically, all children in the U.S. going to public schools are going to be monitored by their teachers and screened for learning and developmental disabilities if they start to have problems with classwork, behavior or social interaction. But not all schools have the resources to make this sort of individualized attention a reality. It probably won't surprise you to hear that schools in districts where poor people live have less money than schools in districts where richer people live, and accordingly the poorer school districts are less likely to employ psychologists, paraprofessionals or even college-educated teachers. With less space and less staff, also, even very dedicated, insightful and observant teachers aren't able to give individual attention to every student they see struggling.

The U.S. educational system might be more equitable than its health-care system --- last I heard, public schools were still legally required to educate every student living in their districts --- but similar issues of accessibility plague both systems. Even if our educational system is technically socialized --- available for free to everyone, paid for with taxes --- there are still huge differences between the kind of education (including special education) you can get at a well-funded school in a primarily upper- or middle-class district and the kind of education you can get at a school in a poor district. The richer districts can levy additional taxes to give more money to their schools, and students going to those schools also pay student fees to cover equipment and activities.

All of this leads me to suspect that ascertainment bias can be just as big a factor in an educational setting as in a medical setting.

The study authors tackle this issue in their Discussion section:

An important limitation of this study was that the ADDM Network surveillance system relies on information for children who have access to diagnostic services for developmental disabilities. We could not rule out the possibility that the quantity and quality of evaluations and the information available for case ascertainment might have varied by SES. We looked for evidence of this by examining the number of evaluations per child with ASD recorded in the ADDM Network surveillance system, reasoning that if the higher prevalence of ASD among children of higher SES was due to increased access to diagnostic services, high SES might be associated with a higher number of diagnostic evaluations per child. However, we found no association between the number of evaluations per child and SES. We also examined the mean ages at diagnosis by SES and found that children of high SES received an ASD diagnosis at an average age of 58.0 months, 1.1 month earlier than those of middle SES (p = 0.2838) and 2.7 months earlier than those of low SES (p <>

The other important factor that somewhat mitigates the effect of differential access to the educational and medical professionals who can diagnose autism and provide autism-related services on this study's outcome is the inclusion of a subgroup within the group of autistic children they studied who had no prior diagnosis of autism; there were enough of these children (1,244) to do a separate statistical analysis of this group to determine whether the association between autism and socioeconomic status still holds up among children who had never been evaluated. The association did still stand: even among those children with no previous diagnoses of ASDs, the ratio of autism prevalence as determined by the CDC of low- to middle- to high-SES children was 0.78:1:1.09. (That is, the poorest children were somewhat less likely to be autistic than the middle-income children, who in turn were somewhat less likely to be autistic than the wealthiest children). It's worth pointing out that the ratio for children with previous ASD diagnoses is stronger in both directions, at 0.70:1:1.25.

While I'm not sure this study goes far enough to correct for huge systemic inequities in availability of services, those two findings --- that number of evaluations per child does not vary with SES, and that autism prevalence does vary with SES among never-diagnosed children --- introduces some doubt in my mind where there had been none before. Maybe there *is* more to this socioeconomic-status thing than just access to diagnostic services. I remain skeptical, but no longer absolutely convinced that ascertainment bias explains everything.

*A more complete description exists in this 2007 article in Pediatric and Perinatal Epidemiology, for which I cannot find the full text anywhere online for free.

**The list of fourteen areas applies to 2002; in 2004, only eight of those areas participated --- Alabama, Arizona, Georgia, Maryland, Missouri, North Carolina, South Carolina and Wisconsin. That's why the numbers are so much lower for 2004.

Durkin, M., Maenner, M., Meaney, F., Levy, S., DiGuiseppi, C., Nicholas, J., Kirby, R., Pinto-Martin, J., & Schieve, L. (2010). Socioeconomic Inequality in the Prevalence of Autism Spectrum Disorder: Evidence from a U.S. Cross-Sectional Study PLoS ONE, 5 (7) DOI: 10.1371/journal.pone.0011551