Stanford Medicine Study Reveals AI Tool Improves ADHD Medication Follow-Up

Stanford Medicine researchers have developed an AI tool that efficiently analyzes doctors’ notes in electronic medical records to improve follow-up care for children with ADHD. This innovation promises to revolutionize the way medical professionals monitor and manage patient care.

Stanford Medicine researchers have developed an artificial intelligence tool that can sift through thousands of doctors’ notes in electronic medical records, identifying trends and providing crucial information that could significantly improve patient care.

The AI tool, which was detailed in a study published in Pediatrics, was specifically designed to determine from medical records whether children diagnosed with attention deficit hyperactivity disorder (ADHD) received adequate follow-up care after being prescribed new medications. 

“This model enables us to identify some gaps in ADHD management,” lead author Yair Bannett, an assistant professor of pediatrics at Stanford Medicine, said in a news release.

ADHD medications, which are often prescribed to manage symptoms such as hyperactivity and inattentiveness, can have significant side effects, including appetite suppression. These side effects make it essential for doctors to monitor and adjust patients’ treatment regimens as necessary. 

Traditionally, this monitoring has required health care professionals to manually review medical charts, a time-consuming process. The new AI tool leverages large language models to automate this task, making it far more efficient and comprehensive. 

The research team applied the AI tool to assess the medical records of 1,201 children aged 6 to 11 who had been prescribed ADHD medication across 11 pediatric primary care practices.

The AI model was initially trained on 501 human-reviewed notes, identifying whether doctors asked about side effects within the first three months of starting a new medication. Using these notes, researchers verified the AI’s accuracy with an additional 90 notes and found it correctly classified about 90 percent of the notes.

Once refined, the AI tool analyzed 15,628 notes in the patients’ charts — an endeavor that would have taken over seven months of full-time human effort.

The AI implementation allowed the team to glean insights that would have been impossible to detect through manual chart reviews. For example, the AI found that some practices frequently addressed drug side effects in phone conversations with parents, while others did not.

“That is something you would never be able to detect if you didn’t deploy this model on 16,000 notes the way we did, because no human will sit and do that,” Bannett added.

Additionally, the tool revealed that pediatricians were less likely to ask about side effects of non-stimulant medications compared to stimulants.

The study underscores both the potential and the limitations of AI in health care. While the AI can detect patterns across vast amounts of data, it still requires human interpretation to understand the underlying reasons for those patterns. 

“We really had to talk to pediatricians to understand this,” Bannett added, referring to differences in how medication side effects were monitored. The AI tool’s findings highlighted variance in experience managing side effects of different types of medications.

The researchers acknowledge that their AI tool has some limitations. Some side effect inquiries might not have been documented in the electronic medical records, and the tool occasionally misclassified notes on unrelated medications. These limitations point to an ongoing need for balanced human oversight in deploying AI systems in health care settings.

An editorial co-authored by Bannett, recently published in Hospital Pediatrics, discusses potential biases embedded in health care data and the need for careful thought to mitigate these as AI tools become more widespread.

“Each patient has their own experience, and the clinician has their knowledge base, but with AI, I can put at your fingertips the knowledge from large populations,” Bannett added. “That can help doctors make personalized decisions about medical management.”

The promising results of this study highlight AI’s potential to transform the efficiency and quality of medical care, paving the way for broader applications across various health care fields.