Visual representation bias in artificial intelligence-generated depictions of anesthesia, pain, and intensive care professionals.

ObjectiveTo evaluate the demographic and hierarchical representation biases in artificial intelligence-generated depictions of healthcare professionals in anesthesiology, pain medicine, and intensive care, with a focus on sex, age group, skin tone, ethnicity, and professional role hierarchy.MethodsThis cross-sectional comparative study analyzed 4400 artificial intelligence-generated images produced by 4 text-to-image models (DALL·E 3, Midjourney, Leonardo AI, and Gemini). Ten professional roles in anesthesiology, pain medicine, and intensive care, ranging from trainees to department heads, were evaluated. Two independent anesthesiologists with clinical experience in anesthesiology and intensive care assessed each image using a structured digital evaluation form, recording sex, age group, skin tone, ethnicity, and professional role hierarchy. Statistical comparisons were performed using chi-square test with Bonferroni-adjusted post hoc proportion analyses.ResultsMale representation predominated across roles and models (mean: 68.5%), with leadership positions showing the highest male proportion (up to 90.0%). Light skin tones (mean: 70.6%) and Caucasian ethnicity (mean: 68.0%) were the most commonly depicted categories. Younger individuals (age: <40 years) were overrepresented (mean: 53.1%), whereas individuals aged >60 years were rarely depicted (<3.0%) and appeared mainly in leadership roles. Significant differences in sex representation were observed in 9 out of 10 professional roles across artificial intelligence models, with the exception of the anesthesia specialist role (p = 0.051).ConclusionsText-to-image artificial intelligence systems consistently reproduced results with demographic and hierarchical biases in depictions of healthcare professionals in anesthesiology, pain medicine, and intensive care. The predominance of male, light skin tone, and Caucasian ethnicity, particularly in senior roles, highlights the need for more diverse training datasets and greater transparency in the development of artificial intelligence systems.