Methods for Generating and Evaluating Synthetic Longitudinal Patient Data: A Systematic Review.

Area of Science:

• Health Informatics
• Data Science
• Medical Research

Background:

• Healthcare data utilization is hindered by privacy and security concerns.
• Synthetic data generation offers a privacy-preserving alternative for sensitive patient information.
• Longitudinal patient data aspects are underrepresented in existing synthetic data reviews.

Purpose of the Study:

• To systematically review methods for generating and evaluating synthetic longitudinal patient data.
• To identify challenges and gaps in current synthetic longitudinal patient data generation techniques.

Main Methods:

• Systematic literature review following PRISMA guidelines.
• Searched five databases up to May 2024.
• Identified and analyzed 39 methods for synthetic longitudinal patient data generation.

Main Results:

• Four methods addressed key challenges: temporal structure, variable types, missing values, and data imbalance.
• Most studies evaluated data resemblance and utility; fewer assessed privacy.
• No identified methods integrated privacy-preserving mechanisms, and effectiveness with small datasets is unclear.

Conclusions:

• Current synthetic longitudinal patient data methods require enhancement for privacy and robust evaluation.
• Standardized evaluation criteria and privacy-preserving mechanisms are essential for real-world applicability.
• Future research should focus on privacy, evaluation frameworks, accessible code, and regulatory guidance.