Exploring synthetic data to support data mobilization in public health infectious disease control

Background

Research and surveillance are legal core tasks of the infectious disease control at the Dutch Municipal Health Service (GGD). However, data accessibility is frequently hindered by GDPR, creating bottlenecks during acute outbreaks when rapid collaboration with external experts is paramount for effective public health intervention. Synthetic Data Generation (SDG) offers a solution by creating artificial datasets that mimic real-world patterns without compromising anonymity. This project explores Machine Learning models for SDG using pertussis data from GGD IJsselland (2016–2025) to facilitate outbreak analysis and forecasting while upholding privacy.

Methods

We evaluated several ML models: Conditional Tabular GAN (CT-GAN), Variational Autoencoders (VAE), and Diffusion models, to identify the most statistically valid method. Data quality was assessed through a multi-dimensional framework: representativeness (Kolmogorov-Smirnov, Jensen-Shannon Divergence, Spearman correlation), privacy (Distance to Closest Record, Propensity Score), and functionality (predictive utility). Time-series models (ARIMA and SARIMA) were trained on synthetic datasets and benchmarked against authentic data.

Results

A hybrid VAE-GAN approach yielded superior results; the VAE captured global inter-variable dependencies, while the GAN refined local distributions. Statistical metrics confirmed that this hybrid method accurately replicates realistic outbreak patterns. Furthermore, time-series analysis revealed that SARIMA models trained on synthetic data successfully captured temporal complexity for forecasting.

Conclusions

The successful deployment of the VAE-CTGAN model proves that SDG can serve as a viable alternative to restricted pertussis datasets. By enabling rapid, privacy-compliant data mobilization, this approach removes legal friction to external collaboration. This allows for a proactive GGD surveillance framework that protects both public health and individual privacy.

Key messages

• Synthetic data preserved key outbreak patterns and supported time-series forecasting, demonstrating its value for practical infectious disease control.

• Synthetic data generation can provide a privacy-preserving alternative for sharing outbreak data, reducing GDPR-related barriers and enabling faster collaboration for public health surveillance.