Estimating covariate-balanced survival curve in distributed data environment using data collaboration...

Area of Science:

• Biostatistics
• Health Informatics
• Medical Data Privacy

Background:

• Sharing patient-level data for survival analysis is hindered by privacy concerns.
• Existing methods often require centralized data, posing privacy risks.

Purpose of the Study:

• To propose a privacy-preserving framework for distributed survival analysis.
• To enable collaborative estimation of balanced Kaplan-Meier curves from decentralized data.

Main Methods:

• Institutions share low-dimensional representations of covariate matrices after dimensionality reduction.
• Analysts reconstruct aggregated data, perform propensity score matching, and estimate survival curves.
• The framework supports both horizontal and vertical data distribution.

Main Results:

• The proposed method consistently outperformed single-site analyses in experiments.
• Achieved reliable survival curve estimation without raw data disclosure.
• Demonstrated effectiveness on simulation and five public medical datasets.

Conclusions:

• The framework facilitates collaborative survival analysis while upholding patient privacy.
• Enables reliable, privacy-preserving estimation of survival curves from distributed observational data.
• Offers a practical solution for multi-institutional medical research.