A Test For Earlier Diagnosis Of Autism? Not Convincing

[Este artículo está disponible en español. La traducción al español apareció por primera vez en la revista Pensar.]

Autism is diagnosed on clinical grounds by observing the child’s behavior. There is no blood test or any other objective test to diagnose it. But that hasn’t stopped people from claiming to have found one. Among other candidates, a saliva test has been proposed, and now an eye scan.

In summer 2019, a press release from Flinders University reported an “inspiring achievement.” It proclaimed, “A ground-breaking new eye scan could help identify autism in children years earlier than currently possible.” This quick, non-intrusive test would permit early interventions for autism by diagnosing it at an earlier age. Dr. Paul Constable has been searching for a biomarker for autism and thinks he has found one. He optimistically speculates that potential biomarkers will also allow for the early detection of attention deficit hyperactivity disorder (ADHD) and other neurologic disorders.

Finally, an objective test that can discriminate between those who have autism and those who don’t. That’s good news, right? Maybe, but don’t celebrate just yet. I had a lot of questions. First, I wanted to read the study itself to see if the press release portrayed it accurately. Where was it published? It wasn’t. It hadn’t been published or peer-reviewed at the time of the press release but had only been presented as “preliminary findings” at a conference of the International Society for Autism Research in Canada. 

It was eventually published online in February 2020 as “Light-Adapted Electroretinogram Differences in Autism Spectrum Disorder” in the Journal of Autism and Developmental Disordersallowing me to read it and seek answers to the many other questions I had. 

The study compared light-adapted electroretinograms (ERGs) from ninety subjects with autism spectrum disorder (ASD) to eighty-seven control subjects. The mean age of subjects with ASD was 13 (range 6.0 to 25.8); the mean age of controls was 13.8 (range 5.4 to 26.6). So it couldn’t be expected to find evidence for “early” diagnosis.

From the abstract, “LA-ERGs were produced by a random series of nine different Troland based, full-field flash strengths and the ISCEV standard flash at 2/s on a 30 cd m−2 white background. A random effects mixed model analysis showed the ASD group had smaller b- and a-wave amplitudes at high flash strengths (p < .001) and slower b-wave peak times (p < .001). Photopic hill models showed the peaks of the component Gaussian (p = .035) and logistic functions (p = .014) differed significantly between groups.”

The text goes on to explain more details: “The modelling of the photopic hill found a group difference in the contributions of both the ON and OFF pathways that summate to produce the b-wave. Although the ON component V(bmax) showed a more significant difference that the OFF-component G(b), both were affected, implicating a difference in the way cone and bipolar cells connect and/or communicate in ASD. The ERG waveform is produced by the summation of negative and positive voltage changes in the retina with the maturation of the b-wave following the a-wave …,” and it goes on and on with details about synaptic triads, neurotransmitters, amacrine cell inhibitory circuits, the delicate ribbon synapses that form between the cone pedicles with the triad of bipolar and horizontal cells, sex differences, the heterogeneity of their population, the similarity of findings in schizophrenia, and more.

What does all this jargon mean? I didn’t understand it, and studying the detailed explanations and references in the article didn’t help. I’ve learned from experience to respect the “huh” factor. If my reaction to what I read is a bemused “Huh?” there’s a good likelihood that my lack of understanding is not my fault but is due to faulty content. It’s true that the retina is a window into the central nervous system, and it’s not implausible that changes due to ASD could be detected there, but this whole rigmarole strikes me as too complicated. They measured multiple flash strengths and wave amplitudes and relied on models and complicated analyses. Could results have been influenced by other factors not related to ASD or might they simply be due to noise in a complicated data set? I don’t know. I’m not qualified to judge. I’ll have to give them the benefit of the doubt. But there is enough information to decide that this is not a promising screening test.

If these ERGs are to be used on younger children as a screening test, we need to know the predictive value, which depends on the sensitivity and specificity. These are reported as 70 percent and 65 percent respectively, and these are not good numbers. Sensitivity is the ability of the test to correctly identify those who have the disease and specificity is the ability of the test to correctly identify those who don’t. And it varies with the prevalence of the disease in the population. The actual prevalence varies by country, but for the purposes of our calculations let’s call it 16 percent. What happens if we screen a population of 1,000 individuals with an ASD prevalence of 16 percent? 160 have ASD and 840 don’t. (Numbers are rounded).

Of every 1,000 individualsTest results positiveTest results negative
160 have ASD70 percent of 160 = 11248
840 don’t have ASD29465 percent of 840 = 546

So only 112 out of every 406 (112 + 294) individuals who test positive will actually have ASD, while 48 of those who have ASD will be deceived by a false negative result and will be mislabeled as not having ASD. So this is not very helpful. Of those who test negative, 546 of the total of 594 negative results (546 + 48) will be true negatives. So this test wouldn’t give us very useful information. It would only correctly predict 28 percent of those who actually have ASD. It would be a bit better at ruling out autism than at ruling it in: it would correctly predict 92 percent of those who don’t have ASD.

Total results that are:Number that are correctPercentage that are correct
Positive112 out of 406 (112 + 294)28 percent 
Negative546 out of 594 (546 + 48)92 percent

So this test wouldn’t improve the accuracy of diagnosis we can already achieve on clinical grounds alone. Accurate clinical diagnosis is possible by age two, although in practice it is often delayed to a much older age.

It is premature to suggest a biomarker for autism has been found, and it is pure speculation to think similar biomarkers might be predictive of other neurologic conditions.

I would be delighted if researchers could identify a biomarker for autism and even more delighted if they found a way to diagnose it at an early age. But I’ll need to see better evidence than this. 

This serves as a great example of why you can’t believe press releases, especially those that report preliminary findings. But as skeptics and critical thinkers, you already know better. You don’t believe everything you read, even if it is an impressive-looking study by respected scientists with a lot of complicated data, numbers, and statistical analyses. Hope and hype can impress, but they can’t compete with sober reality. I’m quite willing to follow the evidence wherever it may lead, and I may be proven wrong as further studies are done. But so far, I doubt that this is a ground-breaking discovery or even a meaningful one.

This article was originally published as a SkepDoc’s Corner column on the  CSI website

Dr. Hall is a contributing editor to both Skeptic magazine and the Skeptical Inquirer. She is a weekly contributor to the Science-Based Medicine Blog and is one of its editors. She has also contributed to Quackwatch and to a number of other respected journals and publications. She is the author of Women Aren’t Supposed to Fly: The Memoirs of a Female Flight Surgeon and co-author of the textbook, Consumer Health: A Guide to Intelligent Decisions.

Scroll to top