Appendix 12
Evaluating fidelity and utility of synthetic health data
Guidance for selecting and reporting evaluation measures that demonstrate how well synthetic data preserves source patterns and supports intended analysis.
Fidelity
Check how closely synthetic data matches key distributions, relationships, and rare patterns.
Utility
Confirm the dataset supports the analyses, models, or decisions required by the approved use case.
Reporting
Document chosen metrics, thresholds, and limitations for governance review and release decisions.
How to use this appendix
Before release
- Define what "good enough" fidelity and utility mean for the approved use case.
- Select metrics that reflect those requirements and document thresholds.
- Report results alongside synthesis parameters and data limitations.
Pair with privacy checks
Strong utility does not guarantee safety. Combine fidelity and utility results with privacy evaluation methods from Appendix 7 before sharing data.