JMIR Publications

ABSTRACT

Background:

Considering the soaring health-related costs directed towards a growing, aging, and comorbid population, the health sector needs effective data-driven interventions, attempting to achieve desirable cost-utility ratios and respecting economic constraints. Progresses made in data mining processes have become prevalent in health interventions, from epidemiological surveillance to mortality prediction. To adequately address current problems, those tools demand quality Big Data. However, growing privacy concerns have hindered large scale data sharing. In parallel, newly introduced legal instruments have had a controversial impact and complex implementation, especially when it comes to medical research purposes. New privacy-preserving technologies, such as decentralized learning, make it possible to create health models without mobilizing datasets by using distributed computation principles, which can parcel the inferential processes out. Several multinational partnerships, including a recent agreement between the United States and the European Union, are adopting these techniques for the next generation of data science. While these approaches are promising, there isn't a clear and robust evidence synthesis of healthcare applications.

Objective:

The main aim is to compare the performance (e.g., AUROC curve) among health data models developed using decentralized learning approaches (e.g., federated, blockchain) to those developed using non-decentralized methods (e.g., centralized, local). Secondary aims are comparing the privacy compromise (e.g., privacy budget) and resource utilization (e.g., computation power) among model architectures.

Methods:

We will conduct a systematic review using the first-ever registered research protocol for this topic, following robust search query and methodology, including several biomedical and computational databases. This work will directly compare health data models differing in development architecture, grouping them according to their clinical applications. For reporting purposes, a PRISMA 2020 flow diagram will be presented. CHARMS-based forms will be used for data extraction and to assess the risk of bias, alongside with PROBAST. All effect measures in the original studies will be reported.

Results:

With this protocol, and subsequent study, we aim to summarize the progress and findings from state-of-the-art decentralized learning models in healthcare, in comparison to their local and centralized counterparts. Results are expected to clarify the consensuses and heterogeneities reported and help guide the research and development of new robust and sustainable applications to address the health data privacy problem, with applicability in real-world settings.

Conclusions:

It is expected that such work will have implications for this budding research field and policymaking, especially concerning health data privacy. This review will highlight the advances and shortcomings of these approaches to inform the development and application of new tools in service of patients’ privacy and guide future research. Clinical Trial: Protocol Registration: Registered to PROSPERO (#393126).