Leveraging artificial intelligence to detect stigmatizing language in electronic health records to advance health equity

Abstract

Background: The use of stigmatizing language within electronic health records (EHRs) is a significant concern, as it can impact patient-provider relationships, exacerbate healthcare disparities, influence clinical decision-making, and effective communication, which in turn affects patient outcomes.

Purpose: To identify stigmatizing language in EHRs and its associations with patient outcomes.

Methods: A retrospective analysis was conducted on 75,654 clinical notes from 500 patients with hospital-acquired conditions at an academic medical center. Machine learning techniques were utilized to detect stigmatizing language within the EHRs.

Discussion: The model demonstrated high accuracy in identifying stigmatizing language (F1 score: 0.95), and stigmatizing language had a significant association with the length of stay. The study also revealed that older patients and those with government insurance are more likely to have stigmatizing language in their notes.

Conclusion: Using AI to model language is useful for identifying care patterns and patients at risk due to stigmatizing language.

Keywords: Electronic health records; Health inequities; Hospital-acquired conditions (HACs); Informatics; Medical documentation; Natural language processing; Stigmatizing language.