AIHealthcare Analytics

Pro-Justice Framework

Current discourse on GenAI in health focuses predominantly on technical accuracy and bias mitigation. An emerging editorial position argues for a pro-justice approach that actively confronts injustices, addresses structural inequities, and considers intergenerational responsibilities. Drawing from feminist ethics, justice theories, and decolonial practices, four principles are proposed: intersectionality (recognizing overlapping disadvantage), epistemic justice (inclusive of diverse knowledge types), social justice (fair resource distribution), and environmental justice (intergenerational responsibilities toward humans and nature).

Demographic Bias in AI-Generated Images

DALL-E generated images of "medical students" showed 76% female and 74% White representation, overrepresenting both groups relative to actual US medical school demographics. Black and Latino/Hispanic students were more commonly depicted in scrubs versus white coats. No images represented Native American/Alaskan Native or Native Hawaiian/Pacific Islander students. Such biased portrayals risk perpetuating stereotypes and hindering diversity efforts.

Assessment Equity for Non-Native English Speakers

AI detectors falsely label NNES medical student writing as AI-generated in 50.2%–61.3% of cases, compared to less than 5% for native writers. Automated scoring showed a systematic downward bias of 0.5–1.2 SD for NNES students. These findings constitute construct-irrelevant variance that disadvantages NNES students in high-stakes assessment decisions. Equity requires explicit bias auditing and policy safeguards in AI-assisted assessment.

MIMIC-IV Fairness Analysis

Deep learning mortality prediction models trained on MIMIC-IV show disparate treatment in prescribing mechanical ventilation across ethnicity, gender, and age groups. Models rely on racial attributes unequally across subgroups. High prediction performance may result from unfair utilization of demographic features, underscoring that performance metrics alone are insufficient for healthcare AI evaluation.