ABSTRACT
The chapter assesses the extent to which EU law promotes gathering and use of representative health datasets to prevent and remediate statistical biases in the development of AI systems for healthcare purposes. EU regulations and directives are surveyed to assess how they address seven aspects: gathering health data, data quality, non-discrimination, sex/gender, age, disability, and ethnicity/race. The findings include that the GDPR imposes burdens on researchers seeking to gather representative datasets. The European Health Data Space and Data Governance seek to mitigate these barriers by increasing the availability of quality research data. Other regulations, including the AI Act, Clinical Trials Regulation, and Medical Device Regulation impose positive obligations to collect representative, quality data. Out of these instruments, the AIA is the most prescriptive. The CTR and MDR also contain protective elements, which can limit sponsors’ ability to include certain groups. While most of the regulations studied are silent on the topic of non-discrimination, the EU Charter imposes a cross-cutting duty to avoid discrimination of protected groups.
