Quorum-based model learning on a blockchain hierarchical clinical research network using smart contracts.

Area of Science:

• Distributed machine learning
• Blockchain technology
• Healthcare informatics

Background:

• Collaborative privacy-preserving modeling across institutions enables more generalizable predictive models while safeguarding patient data.
• Hierarchical networks in healthcare can face site availability challenges, impacting model training and continuity.

Purpose of the Study:

• To address site unavailability in hierarchical networks for collaborative modeling.
• To design an immutable, transparent, and source-verifiable quorum mechanism for robust distributed learning.

Main Methods:

• Developed a hybrid approach combining a hierarchical learning algorithm with a novel Proof-of-Quorum (PoQ) consensus protocol.
• Integrated blockchain smart contracts to create QuorumChain, a system designed to handle site unavailability during modeling.
• Evaluated QuorumChain on three healthcare/genomic datasets, simulating site unavailability during model initialization and iteration.

Main Results:

• QuorumChain demonstrated significant improvements in predictive correctness compared to a standard HierarchicalChain when sites became unavailable.
• The Area Under the receiver operating characteristic Curve (AUC) improved from 0.068 to 0.441 (p < 0.001) in QuorumChain, indicating enhanced model performance.
• Unlike HierarchicalChain, QuorumChain maintained functionality despite site disruptions.

Conclusions:

• QuorumChain effectively addresses site unavailability in collaborative modeling by forming a quorum to sustain the process.
• The system enhances learning continuity on network-of-networks and ensures data/software immutability, transparency, and provenance.
• These features are crucial for accelerating clinical research and building trustworthy predictive models in healthcare.