- Large language models may produce significantly different clinical recommendations based on patients' sociodemographic characteristics, potentially contributing to health disparities.
- Disparities were observed in triage, testing, treatment, and mental health assessment recommendations for different sociodemographic groups.
- The study revealed that the magnitude and consistency of these disparities suggest that language model outputs may be influenced more by demographic attributes than by clinical need.
- The authors emphasized the importance of developing robust bias evaluation and mitigation strategies to ensure equitable and patient-centered care.
- Sociodemographic information in clinical support tools should be evaluated and used with proper safeguards to prevent unwarranted influences.
Source: Nature Medicine
