Comparing Decentralized Learning Methods for Health Data Models to Nondecentralized Alternatives: Protocol for a...

Area of Science:

• Health Informatics
• Machine Learning
• Data Privacy

Background:

• Rising healthcare costs and an aging population necessitate data-driven interventions.
• Traditional data mining requires large datasets, posing privacy challenges and complex legal compliance issues.
• Decentralized learning (DL) enables health model creation without data mobilization, addressing privacy concerns.

Purpose of the Study:

• To compare the performance of health data models developed using decentralized learning (e.g., federated learning, blockchain) versus centralized or local methods.
• To evaluate privacy compromise and resource utilization across different decentralized learning model architectures.
• To synthesize evidence on the application of privacy-preserving technologies in healthcare.

Main Methods:

• Systematic review following a registered research protocol (PROSPERO 393126).
• Comprehensive search across biomedical and computational databases.
• Data extraction and bias assessment using CHARMS and PROBAST tools, reporting all effect measures.

Main Results:

• Data extraction and analysis scheduled from February 28, 2023, to July 31, 2023.
• The review will summarize state-of-the-art DL models in healthcare, comparing them to local and centralized approaches.
• Expected results will clarify consensus and heterogeneity, guiding future research in privacy-preserving health applications.

Conclusions:

• The review will present the current status of privacy-preserving technologies in healthcare.
• Findings will inform health technology assessment and evidence-based decision-making for professionals, data scientists, and policymakers.
• The study aims to guide the development and application of new tools to enhance patient privacy and future research.