Data quality, provenance and transparency in real-world data : Aligning quality standards with data governance legal frameworks

There have been numerous papers discussing data quality and data protection independently, but there has been little discussion on how data quality relates to data protection and other data governance regulatory frameworks. This paper is a step towards addressing that gap and makes the case for why data quality is relevant for data protection and legal compliance professionals. Real-world data in the context of healthcare refers to data that is routinely collected in the course of delivering healthcare. From a data protection regulatory perspective, Article 5 of the General Data Protection Regulation (GDPR) lists data accuracy as one of the principles for data processing. The recently adopted European Union Artificial Intelligence Act (EU AI Act) Article 10 outlines requirements for data and data governance, specifically quality criteria for datasets used to train, test and validate high-risk AI models to address concerns around algorithmic bias due to biases in the training data. The Standards for Data Diversity, Inclusivity and Generalisability (STANDING) Together consensus recommendations for dataset curators on transparency in dataset documentation enable an informed assessment of the suitability of data and examination of biases, for development of AI health technologies. This includes information on data provenance, modifications, sociodemographic composition and bias assessment findings. The Clinical Practice Research Datalink (CPRD) database is used to illustrate how these recommendations can be implemented in a practical way using unique identifiers such as digital object identifiers (DOIs), metadata, published data resource profiles with sociodemographic information and data quality assessments using validation and comparability studies. There is considerable alignment between established scientific standards, medical product regulatory and data governance legal requirements on data quality, as well as emerging international consensus which will reduce the compliance burden on curators and users of real-world data.