Toward Standardized Evaluation of Synthetic Health Data Across ​Modalities for the European Health Data Space

Research and development in healthcare, including early-stage research, testing, and demonstrational data catalogues, is significantly hindered by strict data use restrictions imposed by regulations such as GDPR and HIPAA. Acquiring high-quality, real-world health data is often expensive and laborintensive for researchers, data owners, and companies. Synthetic data (SD) has emerged as a complementary approach, aiming to improve data accessibility while addressing privacy and regulatory constraints. Unlike traditional anonymization
methods for real-world health data, SD generation enables controlled trade-offs between data quality and privacy, depending on the data type and intended use. However, despite rapid methodological advances, there is no consensus on how SD quality should be evaluated across modalities, purposes of use, and intended use cases. Data privacy is often compromised by the high fidelity or resemblance
of the SD, and vice versa; therefore, the selection of metrics and evaluation thresholds should be carefully considered and tailored to the specific purpose of use and use case [1]. Current literature lacks standardized, use-case–aware quality evaluation and reporting, limiting comparability and regulatory interpretability [2]. Typically, health data authorities have no standardized criteria for reporting synthetic data quality within their evaluation frameworks, as their primary focus here is on data anonymization.