Performance Disparities in Healthcare AI Across Demographic Groups

Artificial intelligence (AI) is increasingly integrated into healthcare to support diagnosis, prognosis, and clinical decision-making, promising improved efficiency and outcomes. However, these systems often rely on datasets that disproportionately represent privileged populations, leading to dataset inequity. This underrepresentation of marginalized groups—due to historical research biases, socioeconomic disparities, and limited infrastructure in low-resource settings—results in algorithmic biases that perpetuate existing health inequities. Such biases manifest as disparities in diagnostic accuracy, treatment recommendations, and resource allocation, raising critical ethical, clinical, and legal concerns. This research synthesizes evidence from systematic literature reviews and experimental analyses across multiple healthcare domains, including medical imaging, oncology, and mental health. It explores the multifaceted origins of dataset inequity and the lifecycle stages where bias emerges—from data collection to deployment. The study also evaluates current mitigation strategies, such as fairness-aware machine learning techniques, bias audits, and inclusive data collection practices, while underscoring their limitations when implemented without addressing broader social and structural determinants. To advance equitable healthcare AI, this poster advocates for comprehensive approaches that integrate ethical frameworks, interdisciplinary collaboration, and stakeholder engagement. Key recommendations include developing consensus standards for dataset diversity and fairness reporting, expanding ethical auditing across healthcare systems, and investing in digital infrastructure in underserved regions. Ultimately, the vision is to reimagine healthcare AI as reparative technology that actively promotes health equity rather than merely minimizing harm.