Autism in minors: is there a new tool for its early detection?

A new digital screening tool could help speed up the diagnosis of children with autism.

According to the latest estimates from the US Centers for Infectious Diseases (CDC), one in 36 children has autism. This prevalence, which has increased in recent decades, is thought to be due to better recognition of symptoms and improved screening procedures. Despite this, families continue to face challenges such as late diagnosis. And for girls and minority children, the delay is usually longer due to inability to access appropriate specialists and the variability of symptoms from one child to another.

There is currently no medical test to detect autism. Instead, experts make a diagnosis by assessing the child’s developmental history and behavior.

“For most children, there is no objective evidence other than observation of behavior,” says Geraldine Dawson, a psychologist at Duke University in the US and lead author of the study; “We rely only on parent reports.”

For black and Hispanic children, even though their parents begin to notice signs of autism around the same time as other parents, they are still diagnosed later than their peers, says Daniel Geschwind, a physician researcher at the University of California. in USA. , whose research focuses on autism. As Geschwind notes, these children also have to visit the doctor more often and are at greater risk of receiving inaccurate diagnoses than their peers.

In a recent paper published in the journal Natural medicineThe researchers describe a digital screening device that uses machine learning to analyze various aspects of behavior and determine whether a child is at high risk of developing autism or not. When they tested this screening tool (called the SenseToKnow app) on a sample of 475 children, they found that it had high accuracy in predicting which children would eventually be diagnosed with autism.

As Dawson points out, parents are pretty good at identifying when something is wrong with their child. But communicating these concerns to doctors comes with significant challenges, whether it’s difficulty defining context or finding the right words to describe what they observe. This is even more challenging because autism manifests itself differently in each child, and the timing of symptoms may also vary.

Given how variable these signs can be, even when parents report concerns, pediatricians often do not have the necessary knowledge and training to determine that it is autism rather than something else. “There are not enough experienced specialists, and most general pediatricians do not have the necessary knowledge,” says Geschwind.

The primary screening tool, called the Modified Checklist for Autism in Toddlers-Revised with Follow-up (M-CHAT-R/F), consists of a formal screening questionnaire that includes a series of questions about the child’s behavior and developmental stages. The pediatrician then asks more questions.

The M-CHAT-R/F performs well in formal research settings, but accuracy is compromised when administered in a busy pediatrician’s office, where appointments may be urgent. This decline in accuracy disproportionately affects girls, as well as black and Hispanic boys.

“Of those (minors) who test positive, only half are referred for early intervention,” says David Mundell, a professor of psychiatry at the University of Pennsylvania whose research focuses on racial, ethnic and socioeconomic disparities in health. access to autism resources. Mundell was not involved in the study.

The new screening tool works like this: Parents ask their child to watch a 10-minute video while a camera records various aspects of their behavior. The test predicts whether a child is likely to have autism based on several factors: what they pay attention to in the video, what facial expressions they make, how they move their head, and how they respond to their name.

“We found extremely subtle differences in facial expression,” says Dawson. In practice, communicating these subtleties can be challenging for parents. “It’s very difficult for parents to quantify or even describe them,” Dawson says.

Of the 475 children who were screened using the app during a primary care visit, 49 of them were eventually diagnosed with autism, and an additional 98 children were diagnosed with other developmental delays. This higher than average prevalence was likely due to the voluntary nature of the study, which may have encouraged parents concerned about their children’s development to take part.

A good screening tool will reliably identify autistic and non-autistic children. These aspects of test accuracy are called sensitivity and specificity.

The sensitivity of a test is its ability to correctly detect autism when present; Specificity is the ability of a test to correctly detect the absence of autism.

If a test has low sensitivity but high specificity, there is a high chance that children who test positive have autism. However, there will also be many autistic children who will erroneously be tested as not autistic.

If a test has high sensitivity but low specificity, many children will be falsely labeled as autistic (false positives), but very few children with autism will be missed.

If many autistic children are neglected, this delays the provision of necessary services and accommodations; while many children being incorrectly labeled as autistic will result in long queues to see an assessor who can carry out a full assessment.

“You need to balance sensitivity and specificity, trying to find as many true positives as possible so that these children start receiving intervention services without clogging the system with a lot of false positives,” says Diana Robins, a psychologist at the University in the US. Drexel (USA), whose research focuses on autism. Robins, one of the creators of the M-CHAT-R/F screening tool, was not involved in the study. Nature.

The SenseToKnow app has been shown to have a sensitivity of 87.8% and a specificity of 80.8%.

SenseToKnow will require further research, including testing its accuracy in different groups of children, before it is ready for use in primary care settings.

“The next step is to test this in an independent population to understand the generalizability to a broader sense,” says Geschwind, who was not involved in the study; “Can you predict beyond the sample you learn about?”

Dawson and his colleagues are leading this research by testing the SenseToKnow app on a larger, more diverse set of patients to see if it can still accurately predict autism. Although the accuracy of SenseToKnow was generally quite good, these results were not consistent across all patient groups.

“The sensitivity of the black children was very good,” Mandell says; “The specificity wasn’t great.”

This lower specificity will mean a higher chance that the child will receive a false positive result (in which the test predicts that the child has autism when he or she does not). Given the relatively small number of black children in the study, this accuracy could likely be improved with more testing.

“The next step,” says Robins; “It’s to test 5,000 or 10,000 kids in reviews and see how it works.”