Automating clinical history extraction for flow cytometry panel selection using an EHR ‐integrated large language model
Cytometry Part B: Clinical Cytometry · 2026-11 · 110 (6) · 418-426
DOI: 10.1002/cyto.b.70051 ↗Kurz gefasst
Fachartikel in der Zeitschrift Cytometry Part B: Clinical Cytometry, online erschienen am 10.07.2026. Verfasst von Rebecca Rojansky und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Pathology, Stanford University School of Medicine, Stanford, California, USA. Hauptthemen laut MEDLINE: Flow Cytometry, Electronic Health Records. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 30 Literaturangaben.
Abstract (Originalsprache)
Flow cytometry immunophenotyping is essential for diagnosing hematologic malignancies, but accurate antibody panel selection depends on clinical context that is often fragmented across the electronic health record (EHR). At our institution, clinical laboratory scientists (CLS) use a standardized decision‐tree algorithm incorporating specimen characteristics, laboratory values, and prior diagnoses to select panels. This process requires manual review and synthesis of longitudinal clinical documentation and remains time‐intensive. We evaluated whether ChatEHR—an institutionally developed, EHR‐integrated large language model (LLM) platform—could automate clinical history extraction and support flow cytometry panel selection across 100 cases. Performance was compared with the current CLS‐driven workflow using hematopathologist‐reviewed panel selections as the reference standard. ChatEHR achieved comparable accuracy to CLS in extracting and categorizing prior hematologic diagnoses (78% vs. 78%). Average processing time was 20.3 s for ChatEHR versus 41 s for manual review, corresponding to an estimated annual reduction of approximately 120 staff hours. Direct panel selection accuracy was lower for ChatEHR (47% vs. 78%), primarily due to errors in deterministic protocol execution and structured data interpretation. Major error categories included incorrect decision‐tree logic (39%), diagnostic conflation (22%), failure to retrieve or apply clinical history (22%), laboratory value misinterpretation (9%), and hallucination of nonexistent panels (7%). ChatEHR demonstrated strong performance in extracting and classifying clinical history but was less reliable for deterministic protocol execution and structured data interpretation. Integration of LLM‐based clinical history extraction with rules‐based laboratory algorithms provides a practical hybrid workflow that preserves diagnostic reliability while reducing manual chart review burden. This approach enables scalable automation while maintaining appropriate hematopathologist oversight in clinical flow cytometry workflows.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Pathology, Stanford University School of Medicine, Stanford, California, USA.
- Technology and Digital Solutions, Stanford Health Care, Palo Alto, California, USA.
ORCID-Profile: Jean Oak
Zitieren
APA
Rojansky, R., Keyes, T., & Oak, J. (2026). Automating clinical history extraction for flow cytometry panel selection using an EHR ‐integrated large language model. Cytometry Part B: Clinical Cytometry, 110(6), 418-426. https://doi.org/10.1002/cyto.b.70051
Vancouver
Rojansky R, Keyes T, Oak J. Automating clinical history extraction for flow cytometry panel selection using an EHR ‐integrated large language model. Cytometry Part B: Clinical Cytometry. 2026;110(6):418-426. doi:10.1002/cyto.b.70051
BibTeX
@article{Rojansky2026,
author = {Rojansky, Rebecca and Keyes, Timothy and Oak, Jean},
title = {Automating clinical history extraction for flow cytometry panel selection using an EHR ‐integrated large language model},
journal = {Cytometry Part B: Clinical Cytometry},
year = {2026},
volume = {110},
number = {6},
pages = {418--426},
doi = {10.1002/cyto.b.70051}
}
EUVIMED