Synthetic Generation of Patient Service Utilization Data: A Scalability Study.

Area of Science:

• Health Informatics
• Machine Learning
• Data Privacy

Background:

• Increasing use of patient data for machine learning raises privacy and ethical concerns.
• Scalable synthetic data generation is crucial for responsible health data utilization.

Purpose of the Study:

• To evaluate the scalability and performance of advanced synthetic data generation methods for patient service utilization data.
• To compare Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), copulaGAN, and transformer models.

Main Methods:

• Five distinct synthetic data generation models were assessed.
• Evaluation focused on training/generation efficiency, data resemblance to original data, and practical utility.
• Data from a Canadian health authority was used for the study.

Main Results:

• Statistical models demonstrated superior efficiency in training and data generation.
• Most evaluated models successfully generated synthetic data that closely mirrored the characteristics of the real patient data.
• The generated synthetic data proved to be practically useful for real-world applications.

Conclusions:

• Advanced synthetic data generation methods are scalable and effective for patient service utilization data.
• A balance between model efficiency and data fidelity is achievable.
• Synthetic patient data holds significant promise for ethical and privacy-preserving health data analysis.