In a groundbreaking study published in the prestigious journal JAMA Network Open, researchers at Columbia University School of Nursing have unearthed troubling disparities rooted within the very language clinicians use to document labor and delivery experiences. Leveraging sophisticated artificial intelligence techniques, the study reveals that Black patients admitted for childbirth are disproportionately subjected to stigmatizing language in their clinical notes compared to their White counterparts. This research not only exposes the subtleties of racial bias embedded in medical documentation but also raises profound questions about the perpetuation of healthcare inequities through seemingly routine clinical practices.
At the heart of this investigation is Dr. Veronica Barcelona, PhD, an assistant professor at Columbia Nursing, whose team harnessed the power of natural language processing (NLP)—a cutting-edge branch of AI—to sift through the clinical records of 18,646 patients admitted to two major hospitals between 2017 and 2019. The goal: to identify and categorize language within electronic health records (EHRs) that either stigmatizes or positively characterizes patients, revealing patterns tied to race and ethnicity. This large-scale textual analysis offers unprecedented insight into the complex dynamics of clinician-patient interactions as documented in medical charts, and how those narratives might influence care outcomes.
The study defined four distinct categories of stigmatizing language: bias related to a patient’s marginalized language or identity; descriptions portraying patients as “difficult”; unilateral or authoritarian clinical decision-making language; and language questioning the credibility of the patient. These categories encapsulate subtle lexical cues that can embed judgement, undermine patient autonomy, and perpetuate negative stereotypes. Additionally, the researchers analyzed two subtypes of positive language: one emphasizing patient preference and autonomy, portraying the birthing patient as an active participant in decision-making; and another reflecting power and privilege, noting markers of higher social or psychological status within the clinical narrative.
Findings from this rigorous analysis revealed that stigmatizing language was prevalent across almost half of all patients examined, appearing in 49.3% of the clinical notes overall. However, this linguistic bias was even more pronounced for Black patients, with 54.9% of their charts containing stigmatizing descriptors. The most frequently encountered stigmatizing language pertained to labeling patients as “difficult,” a trope long-recognized for its deleterious impact on patient care. Among Black patients, this “difficult” designation appeared in one-third of notes, compared to 28.6% overall.
Statistical models further quantified these disparities: Black patients were found to be 22% more likely than White patients to have any stigmatizing language in their clinical notes. Paradoxically, Black patients were also 19% more likely than Whites to have positive language documented in their charts, suggesting a complex narrative regarding how race influences documentation patterns. Meanwhile, Hispanic patients were 9% less likely to be labeled as “difficult” and 15% less likely to be described with positive language overall, whereas Asian/Pacific Islander (API) patients were significantly less represented in certain language categories, notably 28% less in marginalized identity language and 31% less in power/privilege language.
The application of natural language processing in this study exemplifies a transformative methodological advance in assessing implicit bias within healthcare systems. By algorithmically parsing thousands of clinical notes, the research team could systematically uncover linguistic patterns invisible to conventional analysis. This approach provides a scalable framework to detect and potentially mitigate bias embedded in clinician documentation, a critical step toward fostering equity in pediatric and maternal healthcare.
Crucially, the implications extend beyond linguistic analysis. These data suggest that the manner in which healthcare providers record their impressions and decisions may amplify existing racial and ethnic disparities in health outcomes. Stigmatizing language can influence contemporaneous clinical judgment, impact the continuity of care, and adversely shape subsequent providers’ perceptions, thereby perpetuating a cycle of discrimination. Furthermore, documentation that undermines patient agency or questions credibility can erode trust, a fundamental component of effective patient-provider relationships, especially during sensitive perinatal periods.
The study’s authors call for targeted interventions aimed at reshaping documentation practices, urging healthcare institutions to develop culturally sensitive guidelines and provider training programs. Such interventions could incorporate feedback mechanisms aided by AI-driven monitoring tools, enabling clinicians to identify and correct biased language patterns in real time. By fostering an environment that emphasizes patient-centered narratives and respects cultural diversity, these measures could substantially contribute to reducing health disparities during childbirth.
This research emerges amid mounting awareness of systemic racism within healthcare and aligns with broader efforts to integrate equity-focused initiatives across medical education and practice. The nuanced understanding of documentation bias complements existing evidence on differential treatment and outcomes in labor and delivery, reinforcing the need for multifaceted strategies that address structural and interpersonal dimensions of healthcare inequity.
Funding for this pivotal study was provided by the Columbia University Data Science Institute Seed Funds Program and the Gordon and Betty Moore Foundation. The interdisciplinary team, including data manager Ismael Ibrahim Hulchafo, MD, doctoral student Sarah Harkins, BS, and Associate Professor Maxim Topaz, PhD, underscores the collaborative effort bridging nursing science, data analytics, and clinical research. Their work exemplifies how leveraging data science innovations can illuminate entrenched biases and promote health justice.
Columbia University School of Nursing, renowned for its commitment to excellence in education, research, and clinical practice, spearheads this endeavor amid its mission to confront health disparities and reshape equitable healthcare policies. As part of the Columbia University Irving Medical Center, its cutting-edge research community integrates perspectives from various health disciplines, striving to advance scientific knowledge that informs real-world improvements for marginalized populations.
In sum, this study presents compelling evidence that the dynamics of language in clinical documentation are far from neutral—they reflect and reproduce societal inequities with tangible consequences for maternal health. Addressing these biases through innovative technological tools and systemic reforms offers a promising pathway to more just, respectful, and effective childbirth care for all patients, irrespective of race or ethnicity.
Subject of Research: Stigmatizing and positive language use in clinical notes related to labor and delivery, analyzed through natural language processing to assess racial and ethnic disparities.
Article Title: Stigmatizing and Positive Language in Birth Clinical Notes Associated With Race and Ethnicity
News Publication Date: May 13, 2025
Web References:
- https://jamanetwork.com/journals/jamanetworkopen/fullarticle/10.1001/jamanetworkopen.2025.9599
- https://dx.doi.org/10.1001/jamanetworkopen.2025.9599
References: Not specified within the source content.
Image Credits: Not specified within the source content.
Keywords: Nursing; Artificial intelligence; Health disparity; Health care; Health and medicine
