Synthetic Data Generation Methods for Longitudinal and Time Series Health Data.

Area of Science:

• Health Informatics
• Data Science
• Biostatistics

Background:

• Synthetic Data Generation (SDG) is recognized for structured health data.
• Longitudinal and time series health data present unique generation challenges.
• Effective SDG is vital for privacy-preserving health data analysis.

Purpose of the Study:

• To conduct a rapid literature review on SDG methods for longitudinal and time series health data.
• To identify and categorize prominent SDG techniques in this domain.
• To provide preliminary insights into the utility, fidelity, and privacy of these methods.

Main Methods:

• Systematic search of PubMed and swisscovery databases.
• Analysis of 338 retrieved articles.
• Categorization of 14 identified SDG methods.

Main Results:

• Identified 14 prominent SDG methods for longitudinal and time series health data.
• Methods include Generative Adversarial Networks (GANs), diffusion models, Variational Autoencoders (VAEs), transformer-based models, and Bayesian methods.
• Preliminary insights into utility, fidelity, and privacy implications were gathered.

Conclusions:

• The review provides a foundational understanding of current SDG methods for complex health data.
• It serves as a guide for researchers in selecting appropriate SDG models.
• Further research is needed to refine methods and address privacy concerns in synthetic health data generation.