A New Autism Study May Not Say What You Think It Says

A new article on autism is interesting, but the meaning is not obvious.

Reviewed by Devon Frye

Key points

• A new JAMA paper reports data on rates of autism diagnostic codes among over 12 million patients.
• Words in the paper could be interpreted multiple ways but its data enables deeper insights.
• A close look at the paper's data refutes some common beliefs about autism epidemiology.
• Evidence from the paper should inform priorities for public policy and research.

All scientific studies tell a story. Sometimes the story is very clear, and sometimes it’s hard to find.

A new autism epidemiology paper in JAMA, “Autism Diagnosis Among U.S. Children and Adults, 2011-2022” has garnered significant coverage (USA Today headline: “Autism is spiking in young adults, study finds”) for its focus on adults with autism.

Behind the headlines, however, it’s tougher to glean just what this study is telling us. I argue that a closer look at the data shows scant evidence for a wave of new autism diagnoses in adults—or that there are vast numbers of autistic adults who have yet to be diagnosed.

Lead author Luke Grosvenor told USA Today that “researchers involved in the study set out to assess how many adults had been diagnosed with autism because earlier studies largely focused on school-age children.” The paper itself says, “The greatest relative increase in diagnosis rate from 2011 to 2022 occurred among 26-to-34-year-olds (450%).”

What, exactly, does that mean? You might interpret it as saying that the increase among adults means the health systems identified and diagnosed a larger portion of existing autistic adults. If so, you’d be in good company. Andy Shih, the chief science officer of Autism Speaks, wrote on LinkedIn that this study is relevant to “particularly adults who may have gone undiagnosed in childhood.”

The study doesn’t say anything about previously undiagnosed adults, but that seems to be one possible interpretation. Another interpretation is that the number of autistic adults grew naturally as increasing numbers of autistic children grew older. Those two stories have very different implications in the real world.

The “recently diagnosed adults” story means there was little or no actual increase in autistic children over time. That is, there are vast numbers of undiagnosed adults and we need to identify them and provide them with appropriate services.

In contrast, according to the “increasing number of autistic children” story, the rate of occurrence of autism has increased dramatically over many years. That implies that something must have caused the increase, and it can’t be simply inherited. It also means that there are vastly fewer undiagnosed adults than some people have claimed.

Obviously, those two stories can’t both be correct. And which one is correct has huge implications for public policy and our priorities for autism research. How can you know which one is true?

The key to understanding autism epidemiology is to examine the data by birth year. Your year of birth is one of the few things about you that can never change. Your age, on the other hand, always increases by one every year.

When you compare people who were a certain age in different study years, you are comparing different people born in different years. Many papers show data organized by study year while few explicitly show birth year. You can easily convert one to the other: study year minus age equals birth year. For example, children who were eight years old in 2020 were born in 2012. This JAMA paper’s reported data are in terms of age, not birth year. But we can easily convert it to birth year.

To understand what the data has to say, we have to know what the numbers mean. Is it about rates that people were diagnosed? Or is it about prevalence—that is, how many autistic people there are at a given time? Those are different things. The paper uses both “diagnosis rates” and “prevalence.”

While the correct interpretation is not obvious, Grosvenor confirmed my guess. In fact, the data shows the proportions of members of the participating health networks who had an autism or ASD diagnostic code in their electronic health records in each study year. A diagnostic code does not mean that someone was diagnosed recently. It doesn’t necessarily mean they were ever diagnosed. Doctors treating patients decide what diagnostic codes to put in patients’ records. An autism or ASD diagnostic code might indicate that the patient was diagnosed at some point in the past, but that’s not entirely correct. We can treat it like prevalence.

The paper includes tables showing some of its data. That data is sufficient to compare the two conflicting stories: “recently diagnosed adults” vs. “increasing number of autistic children.”

Is it possible for the paper’s report of a large increase among adults aged 26-34 to show that the “recently diagnosed adults” story is true? No. Why not? Because it compares two different groups of people, born in different ranges of years.

The 26-34 group in the 2011 data was born in 1976 through 1984, while the 26-34 group in the 2022 data was born in 1987 through 1995. (In this paper’s data, the ages refer to the year before each study year.) Greater diagnosed prevalence in the later-born group vs. the earlier one could be due simply to greater prevalence among the later group.

But does the later-born group actually display a greater prevalence? We can answer that by looking more closely at the data.

The paper’s data has birth year prevalence information, but it takes a bit of effort to dig up. This JAMA paper’s eTable 7 gives the proportions of autistic patients (those with diagnostic codes) for each study year grouped into ranges of ages. I used an Excel spreadsheet to study the diagnostic code prevalence by individual birth years. The result is not as precise as we’d like, because of the age grouping, but it is still quite informative.

The following graph shows the birth year prevalence for birth years 1976 through 1995, encompassing both the earlier and later groups mentioned above. As I said, it’s not exactly prevalence but it’s the closest we can get with this data.

As you can see, there was a dramatic exponential increase in birth year prevalence: 12.3 percent per year. Note that the graph looks lumpy because the data is grouped into ranges of years.

The later-born group, born in 1987–1995, has an average birth year prevalence that is 3.82 times that of the earlier group born in 1976–1984. So, most, or possibly all, of the increase found between those two groups of adults likely came from the increasing birth year prevalence. That shows that the “recently diagnosed adults” interpretation is not correct.

Other data sources consistently show that as people from each birth year age, more are diagnosed, and this trend is similar for all birth years. So while the health systems may have diagnosed some members as adults—we can’t tell from this data—that is not something new. It has been going on for decades.

We can also examine the trend in birth year prevalence for all birth years. Here is a plot for the birth years 1990 through 2013—the last birth year that turned 8 years old by the time of the last study year.

This shows an increase of 7 percent per year of birth, which strikes me as fairly significant. It's almost the same increase as in the set of CDC autism reports on children born 8 years before each study year. The CDC reports show an increase of 7.4 percent per year from birth years 1992 through 2012.

Two scientific papers in the last few years purport to show there are vast numbers of undiagnosed adults. Both rely on the assumption that birth year prevalence has been constant all along.

I argue that both the data in this JAMA paper and the CDC reports contradict that assumption, meaning that there are relatively few undiagnosed adults now but there will be many more autistic adults in the future as the younger autistic people age. That says a lot about where public policy and research should focus.