In order to teach science to the public it is better to tell a story about how questions are resolved in science, rather than simply to teach authoritatively the current findings of science. The process is interesting – often more so than the facts.

One such story is the true cause of the increasing autism prevalence over the last 20 years. There is no question that the number of people being diagnosed with autism is increasing. There are various theories as to why, the best, in my opinion, is that the increasing numbers are an artifact of broadening the diagnosis and increased surveillance. Meanwhile, a true increase may also be hiding in the numbers, and that would present an interesting (and important) mystery to solve.

I have written about this before (here and here on SBM, and again here on NeuroLogica). Now there is a new study which may shed some light on the question, but to quickly summarize where the evidence is: There are several ways to address the question of whether or not there is a true increase in autism numbers. One is to assess autism prevalence at various age groups. If true autism incidence is increasing, that younger age groups should have more diagnoses of autism than older age groups – when the same diagnostic criteria are applied. When this kind of analysis is done it appears that autism incidence is stable over time, which is powerful evidence against a true increase.

Another approach is to look at the same cohort today and 20 years ago (where data is available, or by comparing to historical studies) and apply a thorough survey of consistent diagnostic criteria. When this is done again we find a consistent rate of autism diagnosis.

The third approach is to look for factors that could lead to an increase in diagnosis with a true increase in incidence. When this kind of analysis is done many sources of artifactual increase are found, but the deeper question is – are they enough to account for the entire increase. This is tricky because unless you count up all the factors you will underestimate this type of increase in diagnosis rates.

Studies looking at this question have found that there is strong evidence of expanding the diagnosis of autism, diagnostic substitution (what was diagnosed 20 years ago as something else, like a language disorder, is now diagnosed as autism), increased surveillance (partly caused by increased services provided), and culture (for example, autism is more likely among the neighbors of others diagnosed with autism, suggesting that access to information about autism increases the chances of being diagnosed). However, the few studies I have seen that have attempted to quantify these effects usually focus on a few, and have not been thorough.

A recent study is another attempt to quantify the increase in autism diagnoses. Bearman and his colleagues found that 26% of the increase in autism can be accounted for by expanded diagnostic definition. A further 16% can be explained by increased surveillance. They also concluded that 11% can be accounted for by increasing parental age. This is a likely cause of a true increase, as the risk of autism increases with parental age (more definite for paternal age), and people are having children older.

These three factors add up to 53%, which means that 47% of the increase in not accounted for by Bearman’s analysis. However, this analysis does not seem to account for all possible factors, including the fact that a diagnosis of autism was a stigma to be avoided 20-30 years ago, while more recently it has lost this stigma. This means that parents are more likely to seek out the diagnosis (when appropriate) in order to get the services their child needs. In addition diagnosis falls prey to the availability heuristic – non-experts (and even experts to a lesser degree) will tend to gravitate toward available diagnoses to explain ambiguous cases. I have seen this in action personally – patients and others will fit symptoms to the limited number of diagnoses that are available to them because they are in the news or in the popular culture. There has been a dramatic change on this score for autism, and this social factor is hard to quantify.

The weakness of Bearman’s analysis is that there is uncertainty in the numbers, and therefore it is difficult to quantify exactly how much of an effect each factor has, and there are many social factors that are difficult to quantify at all. We are therefore left without a compelling reason to conclude that there is another true increase in autism rates hiding in the numbers (although again, this can also not be ruled out).

But put into the context of the other approaches to the question of autism rates, it seems that the best conclusion we can make at this time is that most of the rise in autism numbers are caused by an increase in the process of making the diagnosis, with some increase from increasing parental age. There may or may not be another factor, but if so it is likely not large.

There is also research that indicates that the true rate of autism has always been much higher than the rate of diagnosis. Thorough surveys of cohorts have found that true autism rates are about 1% of the population. We are just now approaching that true rate. If this is correct, then rates should start to level off. We are still in the middle of this story, and it is likely to be clarified further over the next decade or so.