How prevalent is autism, actually?

If you missed my previous post about autism, you can click here.

In 2014, the CDC reported that autism prevalence is 1 in 68. This post serves to put some of the reported numbers in perspective and to venture a guess as to what the true prevalence of autism is. Is the true prevalence greater than or less than the CDC’s numbers? It’s a tough question that involves a lot of unmeasured variables. My answer is that the true rate is something lower than what the CDC quotes, but probably not too much lower.

The meat of this post concerns how the CDC’s numbers are calculated. As it turns out, there are a number of hidden “grains of salt” in the statistics that don’t get communicated when the “1-in-68” number is widely cited. It’s important to keep these in mind when considering autism’s place in society more generally.

How did the CDC calculate its 1-in-68 number?

The CDC computed the 1-in-68 statistic using what is called the Autism and Developmental Disabilities Monitoring (ADDM) Network. Importantly, CDC’s approach accounts for individuals who are not formally diagnosed. Thus, the 1-in-68 figure is indicative of the “true” prevalence, rather than official diagnoses. Taking this approach is crucial because it corrects for children who may not be diagnosed because of family circumstances (i.e. parents who do not accept that their child is different, or parents who are not involved enough to notice that their child is different).

The ADDM is a network of community organizations that span 14 different cities/states in the US who analyze health records of all children living in their area. The Network then computes autism prevalence by analyzing the information contained in the health records. (Note: only 11 areas provided statistics in the most recent report.)

The sites are located in the following areas:

• Phoenix, AZ
• Salt Lake City, UT
• Denver, CO
• St. Louis, MO
• Southeastern Wisconsin (including Milwaukee and Madison)
• The entire state of Arkansas
• Northeastern Alabama (including Birmingham and Huntsville)
• Atlanta, GA
• Central North Carolina (including Raleigh-Durham-Chapel Hill and Greensboro-High Point-Winston-Salem)
• Baltimore, MD
• Northeastern NJ (Greater New York City)
• Pennsylvania*
• South Carolina*
• Florida*

* denotes that the site did not provide data in the most recent report

Crucially, the CDC states that “The ADDM Network sites are selected through a competitive award process and are not intended to form a sample that represents the nation.”

What are the caveats with CDC’s numbers?

The ADDM is not meant to be representative

The statistics produced by the CDC are at the mercy of the composition of the population residing in the ADDM sites. In particular, ADDM sites are more likely to be in large cities which tend to have disproportionately more educated residents. This translates into higher average salaries, and potentially into higher levels of autism awareness.

Indeed, CDC finds that there are quite different rates of autism across ADDM sites: Rates are lowest in the Northeastern Alabama site (1 in 175) and highest in New Jersey site (1 in 45).

It might be tempting to try to draw some kind of conclusion about spatial differences in autism prevalence, but it is difficult to find any meaningful predictors using just 11 data points. For example, in terms of median income, Northeastern NJ is the richest of the 11 sites, while Alabama is in the bottom half. From this we might conclude that autism rates are higher among areas with higher income, or among children who live in richer circumstances. However, the correlation is not perfect. For example, Salt Lake City is second in autism prevalence but is ninth in median income.

What are the reasons that autism in general might be different than reported by the CDC?

The answer to this lies in how “representative” of a US sample the 12 ADDM sites are. If the selected sites have disproportionately high levels of factors that are correlated with autism (e.g. more skilled workforce, different ethnic background, better socioeconomic measures, etc.), then autism rates would be overstated. This is analogous to measuring sickness levels by visiting hospitals instead of households. On the other hand, the opposite argument is true: the CDC rate would be understated if the ADDM sites happen to be places where autistic children happen not to live.

A specific way in which this might operate is if the ADDM sites are better places for families with autistic children to live, such that these families choose to move to ADDM sites.

The report is missing data from four of the five largest states in the US— California, Texas, Florida, and Illinois. There’s really no way to know how this might affect CDC’s number, because the original sample upon which the 1-in-68 number is calculated is not a random sample.

The final verdict

My best guess is that the CDC rates are actually slightly overstated. In other words, the 1-in-68 number is likely to be more like 1-in-75. The primary reason is that, for an ADDM site to be established, there must be enough autism awareness in the community to put forth sufficient resources.

A secondary reason is that there are broad differences in autism rates across gender and racial groups. My view on the ADDM sites is that they skew white (and hence skew towards higher rates of autism). It’s really difficult to say how different the racial composition of the sites is relative to the population, which is why I put this as a secondary reason.

Other interesting facts

Autism prevalence is rising, but CDC’s figures are based on 8-year-olds in 2010. This suggests that the current prevalence among younger children is much, much higher

Due to the ADDM’s methodology, the CDC numbers examine prevalence among children born in 2002. The CDC report states that autism prevalence has more than doubled over a 10-year period, increasing from 1 in 150 to 1 in 68. Over a 2-year period, the rate increased from 1 in 88 to 1 in 68. Extrapolating these numbers suggests that the rate among children born in 2010 is approximately to 1 in 29, assuming a quadratic trend. (Admittedly, the true trend in the autism rate likely looks more like an S-curve, where the rate will level off at some point; we just don’t know at what level that will be.)

Autism rates differ drastically by gender and racial or ethnic background

Girls have much lower prevalence of autism, while white children have much higher rates than children of other ethnic heritage. This likely suggests some genetic reason for autism.

Source

1. CDC’s 2014 Report: http://www.cdc.gov/ncbddd/autism/states/comm_report_autism_2014.pdf