Examining Linguistic Differences in Electronic Health Records for Diverse Patients With Diabetes: Natural Language...

Area of Science:

• Medical Informatics
• Natural Language Processing
• Health Disparities Research

Background:

• Minority racial and ethnic groups face significant health disparities, partly due to clinician bias.
• Electronic health record (EHR) notes may contain linguistic markers reflecting this bias.

Purpose of the Study:

• To investigate linguistic differences in EHR notes based on patient race and ethnicity using natural language processing (NLP).
• To validate NLP findings by assessing clinician perceptions of bias related to linguistic markers.

Main Methods:

• A cross-sectional study analyzed EHR notes from over 12,000 patients (White, Black, Hispanic/Latino) with diabetes.
• Sentiment Analysis and Social Cognition Engine (SEANCE) components and word count were analyzed.
• Linear mixed-effects models examined relationships between linguistic markers and patient race/ethnicity, controlling for age.
• Clinicians rated the extent to which linguistic variations indicated bias.

Main Results:

• EHR notes for Black patients had more negative adjectives and fear/disgust words compared to White patients.
• Notes for Hispanic/Latino patients showed fewer positive adjectives, trust verbs, and joy words than White patients.
• Clinicians perceived negative adjectives and fear/disgust words as highly indicative of bias.

Conclusions:

• NLP analysis of EHR notes can identify linguistic patterns potentially associated with clinician bias.
• This method offers a tool for physicians and researchers to detect and mitigate bias.
• Reducing bias in medical interactions is crucial for addressing health disparities.