A pilot deployment of a generative artificial intelligence (AI) registry identified patients with drug-resistant epilepsy (DRE) who were not referred for surgical evaluation and flagged those overdue for EEG, MRI, or other standard assessments, early results show.

Using AI to identify gaps in care could reduce the burden on treating clinicians and enhance epilepsy management, study investigator P. David Adelson, MD, Steve A. Antoline Endowed Chair for Children’s Neurosciences and professor of neurosurgery, West Virginia University Medicine in Morgantown, West Virginia, told Medscape Medical News.

“AI has the real potential to improve the quality of care of patients, in this case those with epilepsy, and to be proactive in identifying patients who would benefit from further investigation,” he said.

The findings were presented on December 6 at the American Epilepsy Society (AES) 79th Annual Meeting 2025.

Opportunity for Real-Time Decision-Making

Epilepsy care is inherently multidisciplinary and data dense, with each patient’s trajectory unfolding over years of changing medication regimens, diagnostic tests, and specialist evaluations.

Traditional registries, which depend on manual review, often become outdated or incomplete, limiting their usefulness for real-time decision-making.

Working with data scientists and bioinformatics experts, Adelson noticed that epilepsy surgery referral rates appeared “inordinately low” and set out to understand the reasons behind the gap.

The study used the Autonomous Registry and Analytics platform, which currently includes 3348 adults and children with epilepsy.

Over a 90-day evaluation period, a large language model-based AI system analyzed documentation for 820 scheduled patient visits. The system reviewed medical records — including clinical notes, imaging reports, and prior assessments — to identify where there were issues or where things came up short, said Adelson.

One concern was the identification of DRE. Adelson said the AI combed through medical records and pinpointed patients who met criteria for medical intractability — those who had tried at least two antiseizure medications yet continued to have seizures.

Guidelines recommend that such patients be referred for surgical evaluation. Yet among the 38% of patients flagged as having DRE, only 11% were referred for a surgical assessment.

Failing to receive a surgical evaluation can have serious consequences. Epilepsy is a potentially fatal condition, and an estimated 1%-2% of patients die each year from sudden unexpected death in epilepsy, Adelson noted.

Adelson emphasized that uncontrolled seizures can lead to progressive cognitive decline — “it basically burns out the brain” — making timely, appropriate treatment essential.

Filling Gaps in Care

Guidelines also recommend that patients with epilepsy get an MRI every 2-3 years. The AI scan showed recommended MRIs were either missed or outdated in 54% of cases.

An EEG is also recommended approximately every 3 years, yet the AI review showed that 35% of patients had no recent EEG on record. The system also flagged a substantial gap in neuropsychological assessments, which were missing in 91% of cases.

The system also identified gaps in folate counseling, which is especially important for women taking antiseizure medications because their offspring face a higher risk for birth defects. Yet the study showed that only 19% of clinicians documented discussing folate supplementation with their patients.

Adelson acknowledged that the proportion of patients missing recommended elements of care is high but noted that neurologists face substantial pressures — including the sheer volume and complexity of patient documentation and a shortage of specialists.

The team plans to develop reminder systems for neurologists and primary care clinicians to flag when a patient should be referred, undergo imaging, or have an EEG, Adelson said.

He emphasized that AI is not intended to replace clinicians but to “take away some of the busy work” by automating labor-intensive chart reviews. “We want to give them extra manpower, for lack of a better word, in the form of an artificial ‘fellow.’”

Next steps include using the AI platform to evaluate surgical outcomes and expanding the system to stroke and movement disorders, Adelson added.

‘Stand-Out’ Research

Commenting on the research, Aatif M. Husain, MD, professor of neurology and chief of epilepsy, sleep, and neurophysiology at Duke University Medical Center in Durham, North Carolina, said the study stood out among the many AI-focused studies presented at the AES meeting.

Husain said the work advanced key areas of interest in epilepsy care — particularly the ability of AI to review medical records and highlight missed opportunities for treatment, including surgical referral, which is an effective therapy but “highly underutilized.”

He added that a particularly important finding was how many women were not receiving information about the “critically” important issue of folate. All women who have the potential to become pregnant should be treated with folic acid, regardless of their intentions, because pregnancy is not always planned, he noted.

He added that AI is rapidly expanding in epilepsy care, with emerging tools designed to assist in interpreting brainwave data, selecting the most appropriate medications, and enhancing MRI interpretation.

Husain said AI could significantly ease the demands on physicians, but clinicians will need to adapt as these tools become part of routine care. He noted that integrating AI into practice has the potential to improve efficiency and allow physicians to focus their time where it’s most needed.

The investigators and Husain reported having no relevant disclosures.