Generating high-fidelity synthetic time-to-event datasets to improve data transparency and accessibility

Area of Science:

• Biostatistics
• Medical Informatics
• Data Science

Background:

• Lack of shared data and code hinders scientific transparency and research reproducibility.
• Information governance restricts sharing individual-level patient data.
• High-fidelity synthetic survival data can advance survival analysis methods.

Purpose of the Study:

• To develop and present methods for creating realistic synthetic survival data.
• To ensure synthetic data emulates real-world covariate patterns and survival times.
• To maintain patient privacy while enabling open data sharing.

Main Methods:

• Modeling joint covariate distributions using sequential conditional regression.
• Generating survival times with flexible parametric survival models.
• Emulating administrative censoring using follow-up data.

Main Results:

• Successfully created a synthetic colon cancer dataset (9064 patients).
• Synthetic data shows high similarity to original covariate distributions and survival times.
• No exact patient information from the original data is included.

Conclusions:

• Developed methods for accurate synthetic data generation with minimal patient identifiability risk.
• Synthetic datasets can be published openly, adhering to privacy protocols.
• Enables open data and code sharing to enhance medical research transparency and accessibility.