Imagine transforming weeks of tedious chart reviews into just minutes of actionable insights. This is now possible thanks to Stanford Medicine researchers, who leveraged artificial intelligence (AI) to analyze medical records for children with ADHD.
Their work not only revealed important care gaps but also highlighted new opportunities for improving patient outcomes.
Large Language Models Revolutionize Chart Review
Traditionally, reviewing medical charts for quality care involved experts painstakingly sifting through thousands of doctors’ notes. The Stanford team, led by Drs. Yair Bannett and Heidi Feldman, developed an AI tool powered by a large language model.
This system was trained on hundreds of annotated notes to read and interpret the nuanced, often freeform text in electronic medical records. Its primary goal: to identify whether proper follow-up care, especially discussions about medication side effects, occurred after starting children on new ADHD medications.
Once validated, the AI tool evaluated more than 15,000 notes from over 1,200 pediatric patients, a task that would have consumed more than seven months of full-time human effort.
Impressively, the model correctly classified about 90% of notes, efficiently flagging cases with and without documented follow-up on side effects.
Revealing Patterns and Gaps in ADHD Treatment
With this large-scale automated approach, the AI uncovered patterns that would likely remain invisible in manual reviews.
Researchers found significant variability among pediatric practices regarding how often they checked in with families about medication side effects. For instance, some practices regularly used phone follow-ups, while others did not.
The AI also discovered that doctors were less likely to discuss side effects for non-stimulant medications compared to stimulant prescriptions, pointing to a potential blind spot in care that manual analysis would struggle to identify.
However, while AI excelled at spotting these trends, it could not explain the underlying reasons. Follow-up interviews with pediatricians revealed that clinicians felt more confident managing stimulant side effects, demonstrating the continued importance of human expertise alongside AI-driven insights.
Limitations and Ethical Issues in Medical AI
Not every side effect discussion may have been recorded in the electronic medical records, and the AI sometimes misclassified unrelated notes. The tool’s accuracy also depends heavily on the quality and completeness of the underlying data, a common challenge in healthcare AI.
Dr. Bannett emphasized that while AI is ideal for quickly sorting massive amounts of medical data, it cannot replace human oversight, especially when interpreting results or addressing ethical challenges.
Because AI models are trained on real-world healthcare data, they may inadvertently reflect and perpetuate existing biases and disparities. Responsible use of AI in clinical research requires vigilance to prevent these issues from being amplified.
Looking Ahead: Personalized and Equitable ADHD Care
The team envisions a future where AI plays a pivotal role in medicine beyond chart review. With continued development, AI could help predict individual risks, like the chance of adverse reactions based on a patient’s demographics or genetics, making care more personalized and effective. Despite ongoing challenges, these advancements promise to reduce administrative burdens and enhance outcomes for both clinicians and families.
Takeaway
Stanford’s innovative use of AI demonstrates how technology can transform the painstaking process of chart review, surfacing actionable insights and care gaps that might otherwise go unnoticed. Combining AI’s analytical power with clinical expertise offers a path toward more proactive, precise, and equitable care in pediatric ADHD and the broader field of medical research.
How AI Is Changing Pediatric ADHD Chart Reviews and Care