A lossless one-shot distributed algorithm for addressing heterogeneity in multi-site generalized linear models.

Area of Science:

• Medical Informatics
• Biostatistics
• Distributed Computing

Background:

• Generalized Linear Models (GLMs) are fundamental in medical research for analyzing various outcome types.
• Multi-institutional studies face challenges in integrating heterogeneous data while preserving patient privacy.

Purpose of the Study:

• To introduce Heterogeneity-aware Collaborative One-shot Lossless Algorithm for Generalized Linear Model (COLA-GLM-H).
• To enable privacy-preserving, lossless integration of heterogeneous multi-institutional data for GLMs.

Main Methods:

• Developed a novel one-shot lossless distributed algorithm (COLA-GLM-H).
• Reconstructed global likelihood using only institution-level summary statistics.
• Validated the algorithm in two real-world studies: a U.S. pediatric network and an international hospitalized patient network.

Main Results:

• COLA-GLM-H achieved estimates identical to pooled analyses in a centralized network.
• Effectively integrated heterogeneous data across institutions in a decentralized setting using a single communication round.

Conclusions:

• COLA-GLM-H offers a privacy-preserving, lossless, and efficient solution for multi-institutional research.
• The algorithm accounts for between-institution heterogeneity and supports diverse outcome types.
• Enables secure, scalable, and accurate collaborative clinical research.