Researchers develop EHR model to predict ADHD likelihood and timing in pediatric patients
Updated
Updated · Nature.com · Apr 27
Researchers develop EHR model to predict ADHD likelihood and timing in pediatric patients
6 articles · Updated · Nature.com · Apr 27
The model was trained on EHR data from over 720,000 patients and fine-tuned on a pediatric cohort of more than 140,000 children, achieving an AUC of 0.92 by age 5.
It maintained strong predictive performance across diverse demographics, including sex, race, ethnicity, and insurance status, and identified developmental, behavioral, and psychiatric conditions as key predictors.
These findings suggest EHR-based predictive models could help clinicians identify children at risk for ADHD earlier, potentially reducing diagnostic delays and improving long-term outcomes.
This AI predicts ADHD risk. But can our healthcare system actually help all the children it flags?
This AI works at Duke. Will it misdiagnose children in hospitals with different patient populations?
If an algorithm flags your toddler for ADHD risk, does that label follow them for life?
Public trust in medical AI is low. How can this new tool prove it's different and earn patient confidence?
When an AI 'safety net' fails and a child is harmed, who is held accountable?