A computer tool can predict which children will later receive an ADHD diagnosis, years before many families have an answer, according to a new study.
This early warning can spur children to help while school, peer relationships, and everyday confidence are still forming.
Health records show warning signs
Routine pediatric records from birth to infancy included daily visits that documented clues that frequently appeared before ADHD was officially recognized.
By following these clues, Elliott D. Hill of Duke University School of Medicine (Duke Med) linked early patterns to later diagnoses.
These patterns were not dependent on a single warning sign because the software recognized repeated combinations across development, behavior, and care.
This makes this finding useful as a screening flag rather than as evidence that a child has the condition.
Why timing is important
Doctors call ADHD (short for attention-deficit/hyperactivity disorder) a condition in which a person’s daily life is disrupted by inattention, hyperactivity, or impulsive behavior.
A national survey found that 10.47% of children ages 4 to 17 in the United States were diagnosed with ADHD in 2021-2022.
Although many children exhibit difficulties early on, families may not reach an assessment until school problems accumulate.
Early screening allows parent training, classroom support, and coping plans to begin while children are still building basic habits.
Health records reveal behavioral patterns
Routine care left a long trail in the electronic medical record, a digital file that tracks care over time.
Repeated references to development, behavior, sleep, medications, and referrals within these records allowed us to understand each child’s needs over time.
“We have this incredibly rich source of information in our electronic medical record,” Hill said.
This enrichment is important because routine care may identify concerns before they are labeled as ADHD, and records still require privacy protection.
To build this artificial intelligence tool, researchers first trained it on the records of more than 720,000 patients.
They then applied the system to more than 140,000 children, whose records tracked their care from birth to age nine.
This two-step training allowed the model to learn common medical sequences before determining each child’s subsequent ADHD risk.
Because the data comes from past treatments, this tool must be tested in a new setting before being used in patient treatment.
The model shows strong predictive accuracy
By age 5, the model reached a ranking score of 0.92 predicting diagnosis up to 4 years later.
On a scale where 1.0 correctly separates all cases after, 0.92 indicates a strong classification signal. Still, diagnosis requires clinical judgment, so a high score does not necessarily mean a child has ADHD.
A beneficial outcome for pediatricians is to reduce the list of children who deserve close follow-up before the problem worsens.
The model maintained similar performance across Duke records across gender, race, ethnicity, and insurance status.
Balanced performance is important for screening tools, as missing one group more often than another can be harmful.
Even so, no one health system can demonstrate equity for every community, insurance network, or clinic workflow.
Future tests will need to find out whether the same alerts will help children without widening existing gaps, especially in areas where there are shortages of specialists and appointments can take months.
Key factors associated with ADHD risk
When the team looked at which record events had the most impact on predictions, developmental and behavioral concerns stood out.
These entries may reflect language delays, learning concerns, emotional symptoms, or repeated visits regarding attention or behavior.
The model also found an association with mental illness. This means that mental health patterns may run parallel to later ADHD diagnosis.
Such clues can guide follow-up investigations, but they do not prove that any event caused the symptoms by itself.
Physicians still lead the final diagnosis
Predictive tools are most effective when they provide busy clinicians with a clear reason to investigate further.
“This is not an AI doctor,” says Matthew Engelhard, MD, senior author and biostatistician at Duke University School of Medicine.
Clinicians still require family interviews, teacher reports, developmental history, and judgment before making an ADHD diagnosis or referral.
Without this human step, risk scores can become just a label rather than a catalyst for better care.
Early support improves child outcomes
Early concerns give families more time to learn strategies before repeated irritations affect a child’s school life.
The American Academy of Pediatrics supports family, school, and medical input in guidelines for planning care.
Strong support can change daily life by making expectations clearer and ensuring consistent adult responses at home and school.
When alerts lead to care, the real promise is in that hands-on support, rather than worrying alone or waiting patiently.
Regular documentation, careful modeling, and clinical follow-up can transform scattered initial concerns into timely attention for children who might otherwise be waiting.
For now, the tool aims for smarter screening, but privacy, fairness, and real-world testing remain essential safeguards.
The research will be published in a journal natural mental health.
—–
Like what you read? Subscribe to our newsletter for fascinating articles, exclusive content and the latest updates.
Check out EarthSnap, a free app from Eric Ralls and Earth.com.
—–
