Governing synthetic athlete-monitoring data for injury-risk assessment and responsible sharing: a governance framework for quality-utility-privacy trade-offs.

OBJECTIVES: Athlete injury-risk modelling is limited by siloed monitoring data, privacy and proprietary concerns and heightened re-identification risk among publicly recognisable elite athletes. We developed and evaluated a quality-utility-privacy governance framework for assessing synthetic-data fidelity, utility, and disclosure risk before governed sharing. DESIGN: Methodological study with an independent clinical-cohort evaluation. METHODS: We analysed 5302 athlete-monitoring session-level records and a clinical injury cohort. Synthetic risk-positive records were generated using a conditional Wasserstein generative adversarial network with gradient penalty and benchmarked against four tabular generators. Utility was assessed using logistic regression, random forest, and extreme gradient boosting under targeted augmentation and minority-class substitution. Shapley Additive exPlanations rankings assessed explainability; membership-inference and distance-based diagnostics assessed privacy. Multi-objective optimisation identified a Pareto knee point. RESULTS: Augmentation improved random-forest recall but reduced F1-score and probabilistic accuracy, while logistic regression and extreme gradient boosting showed limited benefit. In the clinical cohort, augmentation affected the area under the receiver operating characteristic curve but enhanced recall for logistic regression and random forest. Feature-ranking agreement between Real-only and conditional Wasserstein generative adversarial network with gradient penalty models was moderate, while Shapley Additive exPlanations analyses revealed physiologically meaningful non-linear effects. Under the evaluated threat model, attack performance was near chance, with privacy scores of 0.96-0.98 and no widespread near-duplicate replication. The Pareto configuration balanced quality, utility, and privacy. CONCLUSIONS: This framework identifies safe-and-useful augmentation zones for synthetic athlete data. The Pareto knee point may provide a default for cross-institutional sharing with pre-release privacy attack testing and near-neighbour duplication screening.