CDCA study published in PLOS Digital Health has found that machine learning software can be used to predict attention deficit/hyperactivity disorder (ADHD) in kindergarten students.
The study, led by Bo Cao, associate professor in the faculty of medicine and dentistry at the University of Alberta, focused on at-risk populations. Yang Liu, a research assistant at the U of A, is the study’s first author.
The researchers used anonymized health data and teacher-provided developmental assessments of 23,247 children to predict ADHD diagnoses within four years. The goal of this research is to allow for earlier interventions for children who are at risk for developing ADHD, Liu stated.
“Boys are much more likely to be diagnosed with ADHD early. Girls are less likely to be diagnosed with ADHD. Also, there’s a delay in diagnosis which is much later. If [girls] are diagnosed, they’re diagnosed much later than boys. In this scenario, the girls are kind of a vulnerable group,” Liu said.
Liu added that in the future, the goal is to improve the accuracy of the model for specific groups, like girls.
Liu highlighted that kindergarten age students are usually undiagnosed, even if they do have ADHD.
“We know that ADHD is developmental, which means if you are diagnosed with ADHD at 12 years old, you probably had ADHD when you were five years old. It doesn’t just suddenly develop,” Liu explained.
Tracking ADHD early is crucial for identifying vulnerable children and making sure they receive the appropriate treatment as soon as possible, Liu said.
”As far as I know, there’s minimum concerns of risk,” Liu says
Although the study used anonymized data, there are still many privacy concerns when it comes to artificial intelligence (AI).
“There could be privacy concerns. There could be a lot of different concerns about patients, while patients need to have informed consent and also opt-in or opt-out,” Liu explained.
However, it is worth noting that all sensitive information is stripped out of the data used by the model, Liu said. “Their health records were available and anonymized. As far as I know, there’s minimum concerns of risk.”
Liu also noted that although the study had promising results, there were some puzzling results, such as higher education levels in a neighbourhood correlating with more-likely ADHD diagnoses. The model is still undergoing research, and findings like this will continue to be explored over time, Liu said.
