A digital tool first released in 2023 uses a touchpad computer like an iPhone or iPad to effectively screen young children for autism at scale, promising to open the door for support and intervention earlier than previously possible. But doctors and researchers developing the tool wanted to know whether conditions like ADHD might be challenging detection in some cases.
Now, in an Aug. 12 article in the journal Nature Scientific Reports, researchers at Princeton and Duke universities have demonstrated that the tool, called SenseToKnow, reliably identifies autism in children who also have been diagnosed with attention-deficit/hyperactivity disorder (ADHD). Because autistic children often exhibit other conditions such as ADHD, this is a key step to ensure individuals receive needed diagnosis and intervention.
“If someone has certain conditions, one can mask the other and make diagnosis difficult,” said Guillermo Sapiro, a professor of electrical and computer engineering at Princeton and the co-lead researcher. “If someone has diabetes and a heart problem, it might be difficult to distinguish because some of the behaviors are similar.”
The researchers noted that autism diagnosis is frequently delayed because autistic children often also have ADHD, which can mask and challenge the autism diagnosis. Co-occurring ADHD has been linked to delayed autism diagnosis in 20% of children, with average delays of 12-20 months compared to children without ADHD, the researchers wrote.
Sapiro, who came to Princeton from Duke in 2024, co-led the creation of the SenseToKnow tool with a team at Duke. He has continued to work with colleagues at Duke to refine and further validate it, including a clinical study just completed with the goal of an FDA submission.
The researchers said that SenseToKnow is extremely accurate in detecting autism in children as young as 18 months. However, it’s less effective in diagnosing ADHD, as expected from a tool designed specifically for one condition, and the researchers said SenseToKnow is not recommended as a diagnostic tool for that condition. Given that 40-60% of autistic children also have ADHD, Sapiro said it was critical to confirm that the app’s designed weakness in identifying ADHD and specificity to autism was not affecting its actual diagnosis of autism.
The researchers checked SenseToKnow’s accuracy by comparing it to two standard diagnostic tools, the Social Responsiveness Scale and Vineland Adaptive Scale measurements. They found no significant differences in measurements when comparing autistic children and children with both autism and ADHD. SenseToKnow’s measurements showed a clear signal for autism in both groups of children. The researchers said they retested children 12 days after the initial test, and the results showed consistently similar results.
“Are we able to detect autism with ADHD? The answer is yes,” said Sapiro, who is the Augustine Family Professor in Electrical and Computer Engineering at Princeton.
At Princeton, Sapiro serves as associate director of Princeton Precision Health, a research and academic center that works to develop predictive models of health conditions including autism.
SenseToKnow began about 10 years ago, when Sapiro wanted to use commonly available sensors to help clinicians diagnose autism in children. With early detection, therapy and intervention can be helpful for many children, Sapiro said.
SenseToKnow uses cameras on a standard tablet, such as an iPad, along with a touch screen to gather information for an evaluation. The child is shown a series of videos and plays a game on the tablet, which records 23 biomarkers such as eye movement, facial expression, head movement and touch screen response. The test takes about 10 minutes to perform. The results of the test are uploaded to the secure SenseToKnow server, and a machine-learning system analyzes the inputs to deliver an explainable report.
The researchers developed the app in close collaboration with experts at the Duke hospital system. Sapiro said coordinating with healthcare workers has been a critical aspect of the research.
“The only way to get good work in this field is to work closely with clinicians,” in particular the project co-lead Geraldine Dawson, a professor of psychiatry at Duke, he said.
The researchers noted that this study was not designed for individual diagnoses, but rather to determine the abilities of the SenseToKnow tool. The researchers also noted that the study’s sample size, 183 children with ages ranging from 3 to 8 years, and geographic distribution were limited. They recommended future studies with a larger sample size.
Sapiro said the researchers are working with the FDA on developing SenseToKnow as a diagnostic tool. He said the proposal is to first develop a tool for clinicians to use and explain to caregivers rather than distribute it for public use. Sapiro likened it to a blood test, which is reported to a clinician who then explains the implications and possible treatments to the patient.
“The initial goal is to have a system that is ready to give to clinicians,” he said. “All they need to run the app is to create a server account.”
The article, Quantifying autism related behaviors through digital phenotyping in preschool- and school-age children diagnosed with autism and co-occurring ADHD, was published in the journal Nature Scientific Reports on Aug. 12. Besides Sapiro, authors include co-lead researcher Geraldine Dawson, of Duke University; Vikram Aikat, Kimberly L.H. Carpenter, Juan Matias Di Martino, Pradeep Raj Krishnappa Babu, Steven Espinosa, Naomi Davis, Lauren Franz, and Marina Spanos of Duke University. The work was supported in part by the National Science Foundation, the National Institute of Health, the National Institute of Mental Health, the Office of Naval Research and the Simons Foundation.
