Identifying stigmatizing and positive/preferred language in obstetric clinical notes using natural language...

Area of Science:

• Medical Informatics
• Natural Language Processing
• Healthcare Bias

Background:

• Stigmatizing language in clinical notes can perpetuate healthcare disparities.
• Identifying and mitigating such language is crucial for equitable patient care.

Purpose of the Study:

• To develop and evaluate natural language processing (NLP) models for identifying stigmatizing language in obstetric clinical notes.
• To improve the accuracy and efficiency of detecting biased language in electronic health records.

Main Methods:

• Analysis of 1771 obstetric clinical notes from US birth admissions in 2017.
• Annotation of notes for stigmatizing language categories.
• Expansion of the dataset using a semantic similarity-based search approach.
• Training and validation of traditional classifiers and transformer-based models, including ClinicalBERT.

Main Results:

• The semantic similarity approach successfully expanded the dataset, particularly for low-frequency categories.
• All NLP models showed performance improvements after dataset enhancement.
• ClinicalBERT achieved the highest average F1-score of 0.78, outperforming other models.

Conclusions:

• ClinicalBERT demonstrates high efficacy in capturing nuanced, context-dependent stigmatizing language in obstetric notes.
• The semantic similarity approach enhances model performance and reduces manual annotation effort.
• These findings support the potential for NLP in real-time monitoring to reduce healthcare bias and promote equitable perinatal care.