EEG data classify ‘autism’ into two distinct groups

[IMAGES FROM BMC NEUROLOGY (DOI 10.1186/s12883-019-1254-1)]

The Diagnostic and Statistical Manual, 5th edition (DSM-5) established a single diagnosis of autism spectrum disorder (ASD) that includes Asperger’s syndrome, formerly considered a separate condition. The change was meant to eliminate diagnostic ambiguities, but it has encouraged schools to take a “one size fits all” approach, putting all children with autistic features in the same classroom.

This concerns many parents and professionals. “Typically, such classrooms focus on the more severely impaired, often non-verbally communicative children without helping the higher functioning children, such as those with Asperger’s,” says Heidelise Als, PhD, a psychologist at Boston Children’s Hospital.

Als and her co-investigator Frank Duffy, MD, a neurologist at Boston Children’s, decided to take an unbiased look at children diagnosed with autism, using data from their EEGs. In a paper in BMC Neurology, they conclude that autism is not a single entity, but falls into two distinct clusters — ripe for additional investigation.

Mining the EEG

While EEGs are no longer routine in children with autism, clinicians often ordered them in the past to rule out the possibility of epilepsy. This gave Duffy a pool of EEGs from children with autism, ages 2 to 12, who had been referred by senior neurologists, psychiatrists or psychologists at Boston Children’s or other Harvard-affiliated institutions. Children who actually had seizure disorders were excluded, leaving 430 children for analysis.

“We did a very wide intake intentionally,” says Duffy. “The clinicians had made many diagnoses, but we said, ‘let’s look at brain activity, with absolutely no bias.”

Duffy, originally trained as an engineer, was interested in a deeper kind of EEG data, which emerge only through computational analysis.  “There’s a huge amount of information in an EEG signal that you can’t see,” he says.

Seeking coherence

One attribute, called coherence, indicates how synchronized EEG signals are between different regions of the brain, a reflection of how the brain is wired and how it processes and integrates information. 

A standard EEG study, with electrodes placed on multiple locations on a patient’s head, as viewed here from above, yields more than 4,000 coherence variables. From these, Duffy distilled 40 factors that captured much of the variability between the children’s EEGs.

He then ran the coherence data through a statistical algorithm called NbClust, instructing it to sort patients into different groups by forming up to 15 clusters. NbClust then applied 30 different statistical tests, each “voting” on which of the 15 cluster configurations was the most frequently statistically significant grouping.

“The two-cluster configuration clearly came out best in 17 out of 30 different statistical evaluations,” says Duffy. “We conclude, with no bias of any sort, that there is a natural condensation of autistic children into two clusters, and we show they are highly distinct and separate.”

What about Asperger’s?

When 554 neurotypical control subjects were added into the mix, they didn’t overlap with either cluster. Of a separate group of 26 children with Asperger’s, 19 fell within cluster 2, four within cluster 1 and one within the control population.

Duffy and Als say that researchers should take these data into account when studying ASD.  Research datasets are often “enriched” for higher-functioning patients, such as those with Asperger’s, because they’re easier to study. But it may not be appropriate to generalize the findings to all people with ASD.

An autism research agenda

Nineteen specific coherence factors — indicating patterns of brain connectivity — were found to distinguish the two clusters:

factors separating autism clusters

The findings raise many questions to explore. What do the coherence factors and clusters mean clinically? While this study was limited by a lack of neurobehavioral data, Als and Duffy hope to examine EEG data from children evaluated with standardized neuropsychological tests to see how the findings line up with the clusters. One could also compare the clusters with brain MRI findings and, eventually, genetic data.

“If you just look at all autistic children as the same, you’ll miss the fact that today’s genetic finding may be associated with cluster 1, and tomorrow’s with cluster 2,” says Duffy. “There’s probably something different in their brains, so the underlying causes of those two autisms might be different. Behavioral tests might have different results within one cluster or the other. So this is the beginning of an exploratory clinical research process.”

This work was supported by the John Leopold Weil and Geraldine Rickard Weil Memorial Charitable Foundation, the Irving Harris Foundation, the Buehler Family, and the Intellectual and Developmental Disabilities Research Center at Boston Children’s Hospital (grant HD018655).

Read more research by Duffy and Als.