Natural Language Processing (NLP): Identifying Linguistic Gender Bias in Electronic Medical Records (EMRs).

Area of Science:

• Medical communication
• Linguistic analysis
• Health equity

Background:

• Women report increased medical doubt and discrimination.
• Physician language can impact patient care and outcomes.
• Understanding linguistic bias is crucial for addressing healthcare disparities.

Purpose of the Study:

• To explore linguistic mechanisms physicians use to express disbelief.
• To investigate gender differences in negative medical descriptions.
• To identify specific language features associated with patient bias.

Main Methods:

• Content analysis of 285 electronic medical records.
• Identification of four linguistic bias features: judging, reporting, quoting, and fudging.
• Sentiment classification, ICD-11 knowledge graph, and logistic regression for gender difference analysis.

Main Results:

• Female patients received fewer judgmental descriptions but more fudging-related language than male patients.
• Significant differences observed in judging (OR 0.69) and fudging (OR 1.38) language use.
• No significant gender differences found in reporting or quoting language.

Conclusions:

• Physician disbelief expressed through linguistic bias, particularly judging and fudging, affects both genders differently.
• Female patients face more fudging and less judging language, suggesting a pathway for healthcare quality disparities.
• Addressing these linguistic biases is essential for improving healthcare equity.