Nat Mental Health
AI model may flag ADHD years before diagnosis
Clinical takeaway: Electronic health record-based predictive models may help clinicians identify children at risk for ADHD earlier, potentially enabling earlier evaluation and support. Further testing is needed before clinical implementation.
Attention-deficit/hyperactivity disorder (ADHD) often goes unrecognized for years, even when early signs are present. This study tested whether patterns already captured in routine pediatric records could help identify children likely to be diagnosed later, potentially creating a wider window for evaluation and support.
Researchers pre-trained an electronic health record foundation model on data from more than 720,000 patients, then fine-tuned it to predict ADHD diagnosis and timing in a pediatric cohort of more than 140,000 children from birth through age 9.
By age 5, the model accurately identified children who would be diagnosed with ADHD within the next 4 years, and performance was similar across sex, race, ethnicity, and insurance groups.
Feature analysis showed the model was picking up early developmental, behavioral, and psychiatric patterns tied to later ADHD, which could help clinicians decide who may need closer follow-up or referral.
The tool still needs prospective testing because the study did not show that using it improves diagnosis, access to care, or child outcomes.
“This is not an AI doctor,” said Matthew Engelhard, MD, PhD, senior author and faculty member in Duke’s Department of Biostatistics & Bioinformatics. “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.”
Source: Hill ED. Nat Mental Health. 2026 Apr 27. Early attention deficit hyperactivity disorder prediction from longitudinal electronic health records