Reconciling Clinical Research with AI Fairness Requirements: a Methodological Challenge in AI based Healthcare research

The implementation of AI fairness requirements in high-risk medical applications reveals a tension between regulatorymandates and established clinical research practices. While frameworks like the EU AI Act and ongoing harmonisedstandards (e.g., CEN/CLC TR 18115:2024) demand representativeness of training data as a requirement for data qualityand fairness across social groups in healthcare AI based systems, traditional clinical study design often restrictpopulation heterogeneity (e.g. through strict inclusion criteria), or implicitly flatten social differences by considering thestudy population as a homogeneous group.