Beyond protected characteristics: rethinking AI fairness through the lens of structural and social determinants of health

This is the authors' abstract. We don't add key points for this paper.

Fairness refers to the equitable distribution of healthcare resources, opportunities and outcomes, such that no group is systematically disadvantaged.1 While artificial intelligence (AI) systems could support fairer care, growing evidence shows that they often amplify existing inequalities by embedding them within digital systems. These risks arise across the AI lifecycle, from who is represented in data and how conditions are recorded to design choices, evaluation practices and uneven real-world implementation.2 Recent reviews have mapped these issues, synthesising definitions of fairness, sources of bias and both technical and governance-based mitigation approaches.3 However, despite this progress, we argue prevailing approaches to AI fairness remain too narrow to meaningfully address healthcare inequalities.