EUVIMEDEUVIMED entdecken
Open Access

Improving Data Segmentation Accuracy Through Clinician-Informed Contextual Reasoning

Mengyi Wei, Anita Murcko, Patricia Bayless, Preston Lee, Maria Adela Grando iD

Applied Clinical Informatics · 01.05.2026 · 17 (03) · 450-458

DOI: 10.1055/a-2880-9069 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Clinical Informatics, erschienen am 01.05.2026. Verfasst von Mengyi Wei und 4 weiteren Autor:innen. Beteiligte Einrichtung: Arizona State University, College of Health Solutions, Arizona, United States, Tempe. Hauptthemen laut MEDLINE: Electronic Health Records, Physicians, Data Accuracy. Der Volltext ist frei in PubMed Central verfügbar. Gefördert von: National Institute of Drug Abuse.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

A potential limitation of current data segmentation technologies is that pertinent contextual information accessible via the patient's electronic health records (EHRs) is not used to make segmentation inferences.

Objective

Learn how physicians apply contextual reasoning when classifying potentially sensitive health data in EHRs, including information related to substance use disorder (SUD), behavioral health, and other protected categories, and how these insights can enhance the design of context-aware data segmentation tools.

Methods

We conducted semi-structured interviews with 24 board-certified physicians who reviewed constructed patient summaries and categorized selected data items into predefined sensitivity categories. Their confidence levels were inferred from language cues and validated through expert review of the rationale provided. We analyzed co-occurrences between context types and data categories to identify systematic reasoning patterns.

Results

Of 24 physicians 19 (79%) explicitly used contextual reasoning, drawing on diagnoses, medical history, and medication history to inform their classifications. The most frequently categorized items were from medication history (38.9%) and laboratory results (36.1%). Diagnoses were the most common contextual reference (50%), often used to interpret medications and laboratory results. High-confidence categorizations were associated with structured data supported by clinical diagnoses; lower confidence was linked to ambiguous items, such as social history.

Conclusion

Physicians routinely draw on contextual information when assessing the sensitivity of health data—especially for laboratories and medications—which current segmentation tools often overlook. Embedding clinician-informed reasoning patterns into segmentation logic can improve classification accuracy, reduce misclassification risk, and enhance alignment with real-world care practices and privacy regulations. Future research should explore generalizability across diverse provider types and incorporate patient perspectives to support equitable, context-aware data sharing.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Electronic Health RecordsPhysiciansData AccuracyHumans

Autor:innen und Institutionen

  • Arizona State University, College of Health Solutions, Arizona, United States, Tempe.
  • Creighton University, School of Medicine, Arizona, United States, Phoenix.
  • Valleywise Health, Arizona, United States, Phoenix.

ORCID-Profile: Maria Adela Grando

Förderung

  • National Institute of Drug Abuse

Zitieren

APA

Wei, M., Murcko, A., Bayless, P., Lee, P., & Grando, M. A. (2026). Improving Data Segmentation Accuracy Through Clinician-Informed Contextual Reasoning. Applied Clinical Informatics, 17(03), 450-458. https://doi.org/10.1055/a-2880-9069

Vancouver

Wei M, Murcko A, Bayless P, Lee P, Grando MA. Improving Data Segmentation Accuracy Through Clinician-Informed Contextual Reasoning. Applied Clinical Informatics. 2026;17(03):450-458. doi:10.1055/a-2880-9069

BibTeX
@article{Wei2026,
  author = {Wei, Mengyi and Murcko, Anita and Bayless, Patricia and Lee, Preston and Grando, Maria Adela},
  title = {Improving Data Segmentation Accuracy Through Clinician-Informed Contextual Reasoning},
  journal = {Applied Clinical Informatics},
  year = {2026},
  volume = {17},
  number = {03},
  pages = {450--458},
  doi = {10.1055/a-2880-9069}
}
Kostenlos starten →