A child fidgets through class, stares out the window instead of at the board, forgets their homework three days in a row. Teachers are likely to scold, parents worry, but nobody says the word yet.

This is what the gap before an ADHD diagnosis looks like. Now, artificial intelligence wants to close that gap. The question is whether we are ready for what that means.

What the Research Says

A new study from Duke Health, published in Nature Mental Health, has found that an AI tool can analyse routine electronic health records to predict which children are likely to develop ADHD, way before a formal diagnosis is ever made.

Researchers trained the model on data frommore than 140,000 children, both with and without ADHD, tracking patterns from birth through early childhood.

The system identified combinations of developmental, behavioural and clinical signals that consistently appeared years ahead of a diagnosis, and proved highly accurate for children aged five and older, with consistent results across sex, race, ethnicity and insurance status.

"We have this incredibly rich source of information sitting in electronic health records," said Elliot Hill, lead author and data scientist at Duke University School of Medicine.

"The idea was to see whether patterns hidden in that data could help us predict which children might later be diagnosed with ADHD, well before that diagnosis usually happens."

A Problem That Needs Solving

The scale of the issue makes this development significant.

A 2023 meta-analysis covering over 3.2 million participants found a global ADHD prevalence of8% in children and adolescents, with boys twice as likely to be diagnosed as girls.

Yet despite these numbers, the road to diagnosis remains slow. The average age of ADHD diagnosis fallsbetween five and nine years, but many children, particularly girls, spend years with unrecognised symptoms before anyone connects the dots.

Girls are more likely to display inattentive symptoms, which are less visibly disruptive, making it easier for their struggles to be dismissed as daydreaming, shyness, or simply being "quiet."

The Case for Getting There Sooner

The benefits of catching ADHD early are well-documented.Research confirms that timely diagnosis and access to intervention are associated with better social, mental health, educational, and functional outcomes for children with the disorder.

Early support, whether through behavioural therapy, classroom accommodations, or parent training, gives children a framework to understand themselves before years of academic failure and self-doubt can settle in.

"Children with ADHD can really struggle when their needs aren't understood and adequate supports are not in place," said Naomi Davis, associate professor in the Department of Psychiatry and Behavioral Sciences at Duke and co-author of the study.

For families who have watched their child fall behind, be labelled "difficult," or internalise the idea that they are broken, this tool represents something genuinely valuable: time.

The Case for Caution

There are real concerns that cannot be set aside. The first is the risk of labelling.

A prediction is not a diagnosis. If an AI flags a three-year-old as high-risk for ADHD, what happens next?

Without careful clinical follow-through, that flag could translate into stigma, premature medication, or a self-fulfilling prophecy built into a child's medical record before they have had a chance to develop.

Research onAI in ADHD assessment has consistently noted that algorithm biases and fairness gaps must be actively managed at every stage of a model's lifecycle, warning that poorly designed tools risk deepening existing inequalities rather than narrowing them.

There is also the question of privacy. Health records are among the most sensitive data that exist, especially for children.

Using them to train predictive AI models raises legitimate questions about consent, data governance, and who ultimately controls what a child's early medical history is used to conclude about their future.

Then there is the risk of over-medicalisation. A2025 analysis flagged concerns that young children are already being prescribed ADHD medication too quickly.

Introducing an AI tool that flags risk earlier could make this problem worse if not implemented carefully.

A Tool, Not a Verdict

The researchers behind the Duke study made it clear that this is not meant to replace doctors. "This is not an AI doctor," said Matthew Engelhard, senior author of the study. "It's a tool to help clinicians focus their time and resources, so kids who need help don't fall through the cracks or wait years for answers."

Used responsibly, early AI prediction could redirect clinical attention toward children who are currently overlooked, particularly girls and children in under-resourced communities where specialist access is limited.

Used carelessly, it could flood those same communities with labels that follow children into adulthood.

The technology is arriving faster than the ethical frameworks needed to govern it.

The question that needs to be asked is whether our healthcare systems, schools and policies are ready to act on that information in ways that genuinely serve children, rather than simply categorise them.