Autism Rising : By the Numbers

As I talked about last time, a new paper was just released that found the the prevalence of autism in school aged children was about 2%.  This figure is (yet another) sharp upwards revision from last year's surprising 1.1%.  It seems like every year or so we are treated to yet another vastly higher estimate of autism prevalence and every year we are treated to the same reasons for the increase.

Well, the good news is that the increase shown in this years data is being blamed on something slightly different.  The bad news is that this year's explanation doesn't hold up on close examination.

The official line this years seems to be that the roughly 70% jump in the autism rate is mainly attributable to children with a milder form of autism being missed at a younger age and being "discovered" later on in life.  Which is basically a variation on the "better diagnosing" line that has been used to explain away part of the increases for more years than I can count.

The conventional wisdom before this paper is that most - but not all - children with autism will be diagnosed by the time they are eight years old and that children with a diagnosis at that age are more likely than not to keep the diagnosis.  There are a couple of reasons for this belief, not the least of which is by this age most children have been in school for two or three years and that problems that their autism cause will be apparent to their teachers and when they are compared to their typical peers.  It also doesn't hurt that that it has been demonstrated empirically by several studies over the years.

This paper is seeking to amend the conventional wisdom to allow for a large number of "milder" case to be diagnosed at a later age and for this group to account for a substantial portion of the autism population.

So, the question is then whether the data underlying this paper supports that idea or not.  To answer that question I pulled the underlying data from the 2011/2012 National Survey of Children's Health and did some analysis of the data.

For the following analysis I am using the children in the NSCH data set that have a current diagnosis of autism.  I will also be using the 1.1% (1 in 88) prevalence estimate that was published by the CDC last year and will also be making some brief usage of the data from the NSCH 2003 and 2007 survey years.  These different data sets aren't completely compatible but they are good enough for my purposes.

First up, lets consider the idea that a substantial number of children in the 2012 population came a diagnosis later in life.  If the explanation offered by the paper is true and that a large part of the 0.9% increase is due to later diagnosis then that should be apparent by looking at the age when the children in the survey were diagnosed.

The good news is that the 2012 survey has this data so we can directly look at the number of children diagnosed by age.  I am going to used the following age ranges because they (should) be representative of the different times when a child might be diagnosed.

• Ages 0 to 4 are children who are not yet in the school system and represent the group that is picked up by early screening and early intervention efforts.
• Ages 5 to 8 represent the children who are just entered the school system.  Conventional wisdom tells us that most children with autism will be detected by the end of this period.
• Ages 9 to 12 are children who have been in school for a while but have not yet experienced the higher social demands that comes with adolescence
• Ages 13 to 17 are the children who are likely to experiencing the increased social demands of adolescence and where you would expect the children with the mildest forms of autism - forms that are primarily social in nature - to be detected.

Below is a chart that shows the percentage of the autism population that was diagnosed in each age range.

As you can see the majority of children, about 82%, were diagnosed by the time they were eight years old, as is expected.  A simple back of the envelope calculation suggests that 1.6 of the 2% prevalence was diagnosed by age 8 which leaves 0.4% being diagnosed later in life, although that depends on what you mean by later in life.  If later in life means older than 8 that means the 18% of the population but if later in life means the teen-aged years (13 and up) then that means only 4% of the population.

Regardless, the idea that the majority of the 0.9 increase in autism prevalence is being cause by older children being missing simply doesn't hold water even under a generous reading of "older children".

But that is across the entire data set.  Lets do a similar analysis but break down the prevalence not only by the age the child was then they were diagnosed but also by the age that they were when they were included in the survey.  The following chart does exactly that and charts the prevalence per 10,000 using the same age brackets as last time.

Now this chart shows a couple of very interesting things.

Lets start with the most obvious thing - the age of diagnosis in the oldest children shows a different profile than the data set as a whole.  In this group, the number of children diagnosed before the age of 8 roughly corresponds to last year's 1.1% prevalence and the additional increase of prevalence is coming from children who were diagnosed after the age of 8.  Just as a point of reference, the children in this group are older than the children that the CDC's 1.1% estimate was based on.

So in the oldest children, the paper's idea that the increase in prevalence comes from older children being diagnosed does hold up.  The numbers work out pretty well too, you could attribute about 0.7 of the increase in prevalence to older children being diagnosed.

But if you look at the younger age groups you see the bad news.  Both of the other groups of school aged children have a higher prevalence than the oldest age group and both groups are made up almost entirely of children diagnosed by the age of 8.  Or, in other words, the paper's idea falls completely apart and the higher prevalence in the younger children is not being caused by missed diagnosis of milder children when they get older.

Then there is the other little disturbing trend that is visible in the data - the number diagnosed in each age range is climbing as the children get younger.  Look at how the blue section (children diagnosed between 0 and 4) is larger in the 9-12 group than it is in the 13-17 and the 5-8 group is larger than the 9-12.  The same pattern appears in the red section.

Keep in mind when you look at the chart that some of the sections include the current age are going to be incomplete.  So the green section in the 9-12 year olds is going to grow as this age group gets older as will the red section in the 5-8 year olds, as will the blue bar in the 0-4 year olds.

There seem to be two trends in this data - there are children being diagnosed later in life (who are presumably milder) AND the actual prevalence of autism is growing per birth year.

To illustrate this, lets look at the data from the 2003, 2007, and 2012 NSCH survey years.  The earlier years don't have the age of diagnosis available and the 2003 survey didn't ask the exact same questions but it is close enough.

The following is the approximate prevalence per age from all three surveys, the data from the earlier surveys has been shifted to the age that the child would have been in 2012 and I removed the 0-4 age group because, as you can see above, the detection rates are quite low in this group.

The same two trends are visible here as well - the prevalence for the specific age grows each survey year but the prevalence in children born later is higher still.

I think this data makes the overall trend quite clear - the rate of autism is growing by each birth year.