Leveraging Artificial Intelligence GPT Models to Streamline Research Data Extraction from Clinical Neuropsychology Reports in Pediatric Epilepsy
Abstract number :
3.081
Submission category :
11. Behavior/Neuropsychology/Language / 11B. Pediatrics
Year :
2025
Submission ID :
486
Source :
www.aesnet.org
Presentation date :
12/8/2025 12:00:00 AM
Published date :
Authors :
Presenting Author: Rebecah Kaplun, BS – Boston Children's Hospital
Alyssa Ailion, PhD – Boston Children's Hospital & Harvard Medical School
Rationale: Research on pediatric epilepsy surgery often includes numerous clinical variables (i.e., seizure semiology) as well as neuropsychological assessment data. However, these data are documented in free-text clinical notes that vary widely in structure and language, which makes it time consuming and laborious to extract requisite data and conduct research in neuropsychology in general, and specifically as it relates to epilepsy surgery outcomes. This project aims to evaluate and test the feasibility of using the Boston Children’s Hospital’s HIPAA-compliant internal GPT platform to 1) extract structured clinical variables from retrospective unstructured pediatric epilepsy documentation and 2) develop a prospective structured GPT clinical template that can streamline future prospective data extraction.
Methods: Structured prompt templates were applied to de-identified neuropsychological reports and pre-surgical consult notes from 25 epilepsy surgery candidates, targeting the extraction of seizure semiology descriptors (e.g., motor onset, lateralizing signs, language involvement), cognitive scores (e.g., FSIQ, WISC subtests), and behavioral domains (e.g., attention, memory, executive function). Essential variables from data extraction were incorporated into a prospective GPT compatible template to streamline future clinical data collection.
Results: GPT outputs were qualitatively compared to curated summaries and documentation used in clinical decision-making to ensure accuracy of data extraction models. Prospective clinical template with data smart fields for automatic research extraction will be illustrated.
Conclusions: This feasibility study demonstrates the potential of large language models to structure heterogeneous clinical documentation in a way that supports and streamlines research abstraction and cross-case comparison. This work highlights the potential for AI GPT to enhance both clinician workflows and the improving the cost-effectiveness, feasibility, and overall efficiency of clinical research broadly, and specifically as it applies to epilepsy surgery outcomes.
Funding: None
Behavior