Authors :
Presenting Author: P. David Adelson, MD – WVU Rockefeller Neuroscience Institute
Pierre D'Haese, PhD – WVU Rockefeller Neuroscience Institute
Haley Delaney, BSN – WVU Rockefeller Neuroscience Institute
Ruchi Dhyani, MD – WVU Rockefeller Neuroscience Institute
Rationale:
Multidisciplinary epilepsy care aims to facilitate timely identification of drug-resistant epilepsy (DRE) by integrating comprehensive diagnostic assessments and determining optimal treatment strategies. Unfortunately, care pathways are often fragmented with incomplete workups, inconsistent adherence to guidelines, and delayed surgical referrals. Traditional registries and electronic health record (EHR) tools often depend on structured data or manual review, limiting utility. In this pilot, we implemented the Autonomous Registry and Analytics (AURA) platform within our EPIC EHR, leveraging generative AI agents to analyze free-text clinical notes, imaging reports, and assessments. We evaluated whether GenAI-driven automation could enhance surgical triage without manual burden of traditional registries.Methods:
At WVU Medicine, AURA autonomously identifies patients with potential DRE by recognizing clinical indicators such as multiple failed antiseizure medication (ASM), seizure burden, and diagnostic results suggestive of surgical candidacy. For incomplete or unclear records (e.g., missing MRI, EEG, neuropsychological evaluation), the system queries prior encounters to extract relevant data and flags unresolved gaps. Besides identifying DRE and potential surgical candidates, AURA also detects unmet standard-of-care interventions, such as folic acid supplementation for women of reproductive age. Summarized insights are provided to providers prior to visits to enable more effective neurosurgical triage and diagnostic planning as well as to facilitate review of surgical indicators, address pending diagnostics, and prepare for comprehensive neurosurgical discussion. For this pilot, we assessed impact by tracking care gaps, pre-visit readiness, and clinician-reported utility for neurosurgical decision-making.Results:
The AURA registry currently includes 3,348 patients (1,049 pediatric; 2,049 adult), serving as a longitudinal framework for epilepsy care coordination. In this prospective 90-day evaluation, AURA analyzed documentation for 820 scheduled visits, flagging care gaps in diagnostic workups—outdated or missing MRIs (54%), absent neuropsychological evaluations (91%), and missing EEGs (35%). Additionally, 88 patients (11%) exhibited clinical patterns consistent with DRE yet lacked surgical referrals. Clinicians reported improved pre-visit readiness, reduced chart review time, and more timely identification of surgical candidates. Additionally, since deployment, folic acid deficiencies were identified in 81% of eligible patients. Post-surgical outcome documentation (Engel and ILAE scores) increased from 1% to 70.9% (n=124).Conclusions:
Generative AI–powered registries like AURA may improve epilepsy care by extracting actionable insights from unstructured documentation, reducing clinician burden, and supporting earlier surgical evaluations. By enhancing pre-visit readiness and triage, AURA demonstrates a scalable decision support model that could be extended to other complex neurological and multidisciplinary care domains.
Funding: Supported by Steve A. Antoline Chair for Children's Neurosciences and the Lewis Family Research and Education Fund