Abstracts

Rationale: A “Disease Concept Model” (DCM) enumerates the symptoms that make up a disorder and ranks their importance to caregivers. Since 2021, DCMs have been created for Angelman syndrome, SLC6A, STXBP1, and SYNGAP1. These developmental and epileptic encephalopathies (DEEs) have strikingly similar DCMs -- all identify difficulties with communication, behavior, intellectual disability, fine motor skills, gross motor skills, sleep, feeding, and gastrointestinal dysfunction as core concepts. We hypothesize that these 8 disease concepts will be shared by all DEEs. To facilitate chart review, we developed and evaluated an automated tool to identify these domains in electronic health records. Our aims were (1) evaluate the tool, and (2) use the tool to compare the distribution of disease concepts between children with epilepsy with vs. without DEE.

Methods: First, we created a “Gold Standard” for our tool by manually reviewing Weill Cornell Medicines’s Electronic Health Record for 100 children with epilepsy to identify and characterize epilepsy comorbidities. Second, we developed a tool using Python code and the Google Gemini large language model (Alphabet, Inc.) to process providers’ notes and answer the same questions. Third, we tested the tool’s accuracy by comparing its answers to the Gold Standard answers. Fourth, we deployed our tool to perform an automated chart review of 175 children with epilepsy and evaluated the 13 disease concepts and related items in children with DEEs as compared to a control group of children with epilepsy and normal development.

Results: The chart review tool correctly reproduced the Gold Standard answers better than 88% of the time for 9 items, but less than 50% for the remaining 4. When we compared 79 manually reviewed DEE patients to 113 different DEE patients reviewed by the tool, we found no significant differences in the distribution of disability for gross motor impairment (p=0.59), communication impairment (p=0.46), difficulty eating (p=0.58), and intellectual disability (p=0.65) (Figure 1). When we applied our tool to 175 children with epilepsy, we found that children with DEEs have significantly higher rates of gross motor impairment (p< 0.0001), communication impairment (p< 0.0001), difficulty eating (p< 0.0001), and moderate to high intellectual disability (p< 0.0001) compared to a control group of children with epilepsy and normal development (Figure 2).