Racial bias in clinician assessment of patient credibility: Evidence from electronic health records

Objective: Black patients disproportionately report feeling disbelieved or having concerns dismissed in medical encounters, suggesting potential racial bias in clinicians’ assessment of patient credibility. Because this bias may be evident in the language used by clinicians when writing notes about patients, we sought to assess racial differences in use of language either undermining or supporting patient credibility within the electronic health record (EHR). Methods: We analyzed 13,065,081 notes written between 2016–2023 about 1,537,587 patients by 12,027 clinicians at a large health system with 5 hospitals and an extensive network of ambulatory practices in the mid-Atlantic region of the United States. We developed and applied natural language processing models to identify whether or not a note contained terms undermining or supporting patient credibility, and used logistic regression with generalized estimating equations to estimate the association of credibility language with patient race/ethnicity. Results: The mean patient age was 43.3 years and 55.9% were female; 57.6% were non-Hispanic White, 28.0% non-Hispanic Black, 8.3% Hispanic/Latino, and 6.1% Asian. Clinician-authors were attending physicians (44.9%), physicians-in-training (40.1%) and advanced practice providers (15.0%). Terms specifically related to patient credibility were relatively uncommon, with 106,523 (0.82%) notes containing terms undermining patient credibility, and 33,706 (0.26%) supporting credibility. In adjusted analyses, notes written about non-Hispanic Black vs. White patients had higher odds of containing terms undermining credibility (aOR 1.29, 95% CI 1.27–1.32), and lower odds of supporting credibility (aOR 0.82; 95% CI 0.79–0.85). Notes written about Hispanic/Latino vs. White patients had similar odds of language undermining (aOR 0.99, 95% CI 0.95–1.03) and supporting credibility (aOR 0.95, 95% CI 0.89–1.02). Notes written about Asian vs. White patients had lower odds of language undermining credibility (aOR 0.85, 95% CI 0.81–0.89), and higher odds of supporting credibility (aOR 1.30, 95% CI 1.23–1.38). Conclusions: Clinician documentation undermining patient credibility may disproportionately stigmatize Black individuals and favor Asian individuals. As stigmatizing language in medical records has been shown to negatively influence clinician attitudes and decision making, these racial differences in documentation may influence patient care and outcomes and exacerbate health inequities.