In a groundbreaking study, Stanford Medicine researchers have harnessed artificial intelligence (AI) to comb through thousands of doctors’ notes in electronic medical records (EMRs), revealing trends that could improve care for children with attention deficit hyperactivity disorder (ADHD). This innovative use of AI promises to relieve researchers and clinicians of tedious medical chart reviews, enabling them to focus on improving patient outcomes.
The study, published on December 19 in Pediatrics, demonstrates how large language models (LLMs) can efficiently detect gaps in ADHD management and suggest improvements. “This model enables us to identify some gaps in ADHD management,” said lead author Yair Bannett, MD, assistant professor of pediatrics at Stanford Medicine. Senior author Heidi Feldman, MD, added that the insights gained could be applied broadly across healthcare.
AI Revolutionizing ADHD Follow-Up Care
Medical records often contain critical information buried in freeform notes, making it challenging for researchers to identify patterns. For children with ADHD, proper follow-up care after starting medication is essential to monitor side effects and adjust dosages. However, manually reviewing thousands of notes is time-consuming and prone to human error.
The Stanford team trained an AI tool to analyze 15,628 notes from the medical records of 1,201 children aged 6 to 11. These children, treated across 11 pediatric practices, had been prescribed ADHD medications that can cause side effects like appetite suppression. The AI was tasked with identifying whether follow-up inquiries about side effects occurred within the first three months of medication use.
By training the model on 501 human-reviewed notes, researchers achieved 90% accuracy in classifying follow-up mentions. This AI-driven approach allowed them to analyze a task that would have otherwise taken more than seven months of full-time work.
Key Findings: Insights Beyond Human Reach
The AI model uncovered patterns that manual reviews could not. For instance, it revealed that some pediatric practices frequently conducted follow-ups via phone calls, while others did not. It also showed that doctors were less likely to ask about side effects for non-stimulant ADHD medications compared to stimulants.
“These are insights you wouldn’t detect without deploying AI across thousands of notes,” Bannett said. However, the model also highlighted its own limitations. While it identified patterns, it couldn’t explain the reasons behind them — a task that required input from pediatricians.
Challenges and Ethical Considerations
The researchers noted some limitations. The AI might have missed follow-ups recorded outside the EMRs or misclassified notes on medications unrelated to ADHD. Despite these challenges, the study highlights the importance of guiding AI tools with human expertise.
“AI is ideal for sorting through vast amounts of medical data, but ethical considerations and disparities in healthcare must remain front and center,” Bannett said. In a recent editorial, he and his colleagues emphasized the need to address biases in AI models trained on existing healthcare data.
A Vision for Personalized ADHD Care
Looking ahead, AI could help doctors make more personalized decisions for ADHD management. By analyzing large populations, AI might predict which patients are at risk of specific side effects based on age, race, genetic profile, and other factors. This capability could transform ADHD care, making it more precise and patient-centered.
“Each patient has their own experience, and with AI, we can complement that with the knowledge of large populations,” Bannett said. While the potential is immense, ensuring responsible AI deployment will be key to unlocking its full benefits for ADHD care and beyond.
This study marks a significant step toward integrating AI into routine medical care, offering a glimpse of a future where technology enhances clinicians’ ability to provide better, more tailored care.