Synthetic data generation methods for longitudinal and time series health data: a systematic review

Area of Science:

• Healthcare Informatics
• Data Science
• Artificial Intelligence

Background:

• Synthetic data generation (SDG) offers privacy-preserving alternatives for healthcare research.
• Temporal health data (e.g., EHRs, physiological signals) present unique SDG challenges due to complexity and sensitivity.

Purpose of the Study:

• Systematically review SDG methods for longitudinal and time-series health data.
• Propose a taxonomy for the SDG landscape.
• Characterize synthesis techniques, evaluation strategies, and privacy measures.

Main Methods:

• Systematic literature review (2017-2025) following PRISMA guidelines.
• Analysis of 115 studies using structured data extraction and thematic analysis.
• Comparative synthesis of SDG methods for temporal health data.

Main Results:

• Deep generative models (GANs, AEs, diffusion) dominate SDG, with growing use of autoregressive and hybrid methods.
• Event-based EHR data are common targets; continuous/irregular time series are underexplored.
• Utility evaluations focus on statistics/prediction; privacy assessments are sparse, with limited differential privacy (DP) implementation.

Conclusions:

• Synthetic temporal data are crucial for clinical prediction, public health, and AI.
• SDG research is fragmented in terminology, evaluation, and privacy.
• A unified framework is needed for responsible AI, addressing fairness, transparency, and clinical adoption.