VERTICOX: Vertically Distributed Cox Proportional Hazards Model Using the Alternating Direction Method of Multipliers

Area of Science:

• * Computational statistics
• * Machine learning
• * Health informatics

Background:

• * Survival analysis is crucial for modeling time-to-event data.
• * Cox proportional hazards models are widely used but require centralized data.
• * Federated learning offers a privacy-preserving alternative for distributed data.

Purpose of the Study:

• * To develop a federated algorithm for Cox proportional hazards models with vertically partitioned data.
• * To enable distributed model training without centralizing individual patient data.
• * To ensure accuracy and efficiency in federated survival analysis.

Main Methods:

• * Proposed VERTICOX, a novel federated algorithm based on the Alternating Direction Method of Multipliers (ADMM).
• * Computed and exchanged intermediary statistics for global model parameter estimation.
• * Validated the algorithm's performance on distributed datasets.

Main Results:

• * VERTICOX achieved accuracy equivalent to centralized Cox models.
• * The algorithm demonstrated linear convergence under the ADMM framework.
• * Showcased reduced communication costs and preserved patient data privacy.

Conclusions:

• * VERTICOX effectively supports federated survival analysis over vertically distributed data.
• * The algorithm offers a privacy-preserving and bandwidth-efficient solution.
• * Enables collaborative model building across institutions without data sharing.