Sharing health data for research is crucial but raises privacy concerns. Synthetic data, artificially generated to resemble real data, offers a promising solution. However, a standard approach for evaluating its privacy and utility is lacking, hindering its adoption. This scoping review systematizes current knowledge on evaluating synthetic health data's privacy and utility. Analyzing 92 studies, we identified 48 utility and 9 privacy evaluation methods. A concerning trend is the tendency to trust synthetic data's safety without proper evaluation. The lack of consensus on evaluation methods hinders synthetic data's adoption in medicine. This work aims to raise awareness and provide a comprehensive taxonomy to foster the development of uniform approaches.

Bayrem Kaabachi, Jérémie Despraz, Thierry Meurers, Karen Otte, Mehmed Halilovic, Fabian Prasser, Jean Louis Raisaro