Verticox+: vertically distributed Cox proportional hazards model with improved privacy guarantees.

Area of Science:

• Computational biology
• Machine learning
• Privacy-preserving techniques

Background:

• Federated learning (FL) enables decentralized machine learning without data sharing, crucial for privacy.
• Cox proportional hazards models are vital for survival analysis.
• Existing federated models like Verticox require local knowledge of survival outcomes, limiting applicability.

Purpose of the Study:

• To extend the Verticox model for federated survival analysis.
• To enable Verticox to function when survival outcomes are not locally available.
• To ensure privacy preservation throughout the federated analysis.

Main Methods:

• Developed Verticox+, an extension of the Verticox federated learning model.
• Incorporated a privacy-preserving 2-party scalar product protocol.
• Integrated the protocol to handle scenarios lacking local survival outcome data.

Main Results:

• Verticox+ achieves performance equivalent to the original Verticox model.
• Demonstrated successful application in scenarios where survival outcomes are not locally known.
• Analyzed the impact on computational complexity and communication costs.

Conclusions:

• Verticox+ successfully extends federated Cox proportional hazards modeling.
• The privacy-preserving protocol allows for broader application of federated survival analysis.
• The method maintains analytical performance while enhancing data privacy.