Privacy-preserving federated machine learning on FAIR health data: A real-world application

Objective

This paper introduces a privacy-preserving federated machine learning (ML) architecture built upon Findable, Accessible, Interoperable, and Reusable (FAIR) health data. It aims to devise an architecture for executing classification algorithms in a federated manner, enabling collaborative model-building among health data owners without sharing their datasets.

Materials and methods

Utilizing an agent-based architecture, a privacy-preserving federated ML algorithm was developed to create a global predictive model from various local models. This involved formally defining the algorithm in two steps: data preparation and federated model training on FAIR health data and constructing the architecture with multiple components facilitating algorithm execution. The solution was validated by five healthcare organizations using their specific health datasets.

Results

Five organizations transformed their datasets into Health Level 7 Fast Healthcare Interoperability Resources via a common FAIRification workflow and software set, thereby generating FAIR datasets. Each organization deployed a Federated ML Agent within its secure network, connected to a cloud-based Federated ML Manager. System testing was conducted on a use case aiming to predict 30-day readmission risk for chronic obstructive pulmonary disease patients and the federated model achieved an accuracy rate of 87%.

Discussion

The paper demonstrated a practical application of privacy-preserving federated ML among five distinct healthcare entities, highlighting the value of FAIR health data in machine learning when utilized in a federated manner that ensures privacy protection without sharing data.

Conclusion

This solution effectively leverages FAIR datasets from multiple healthcare organizations for federated ML while safeguarding sensitive health datasets, meeting legislative privacy and security requirements.

A Nationwide Chronic Disease Management Solution via Clinical Decision Support Services: Software Development and Real-Life Implementation Report

BACKGROUND

The increasing population of older adults has led to a rise in the demand for health care services, with chronic diseases being a major burden. Person-centered integrated care is required to address these challenges; hence, the Turkish Ministry of Health has initiated strategies to implement an integrated health care model for chronic disease management. We aim to present the design, development, nationwide implementation, and initial performance results of the national Disease Management Platform (DMP).

OBJECTIVE

This paper's objective is to present the design decisions taken and technical solutions provided to ensure successful nationwide implementation by addressing several challenges, including interoperability with existing IT systems, integration with clinical workflow, enabling transition of care, ease of use by health care professionals, scalability, high performance, and adaptability.

METHODS

The DMP is implemented as an integrated care solution that heavily uses clinical decision support services to coordinate effective screening and management of chronic diseases in adherence to evidence-based clinical guidelines and, hence, to increase the quality of health care delivery. The DMP is designed and implemented to be easily integrated with the existing regional and national health IT systems via conformance to international health IT standards, such as Health Level Seven Fast Healthcare Interoperability Resources. A repeatable cocreation strategy has been used to design and develop new disease modules to ensure extensibility while ensuring ease of use and seamless integration into the regular clinical workflow during patient encounters. The DMP is horizontally scalable in case of high load to ensure high performance.

RESULTS

As of September 2023, the DMP has been used by 25,568 health professionals to perform 73,715,269 encounters for 16,058,904 unique citizens. It has been used to screen and monitor chronic diseases such as obesity, cardiovascular risk, diabetes, and hypertension, resulting in the diagnosis of 3,545,573 patients with obesity, 534,423 patients with high cardiovascular risk, 490,346 patients with diabetes, and 144,768 patients with hypertension.

CONCLUSIONS

It has been demonstrated that the platform can scale horizontally and efficiently provides services to thousands of family medicine practitioners without performance problems. The system seamlessly interoperates with existing health IT solutions and runs as a part of the clinical workflow of physicians at the point of care. By automatically accessing and processing patient data from various sources to provide personalized care plan guidance, it maximizes the effect of evidence-based decision support services by seamless integration with point-of-care electronic health record systems. As the system is built on international code systems and standards, adaptation and deployment to additional regional and national settings become easily possible. The nationwide DMP as an integrated care solution has been operational since January 2020, coordinating effective screening and management of chronic diseases in adherence to evidence-based clinical guidelines.

A Data Transformation Methodology to Create Findable, Accessible, Interoperable, and Reusable Health Data: Software Design, Development, and Evaluation Study

BACKGROUND

Sharing health data is challenging because of several technical, ethical, and regulatory issues. The Findable, Accessible, Interoperable, and Reusable (FAIR) guiding principles have been conceptualized to enable data interoperability. Many studies provide implementation guidelines, assessment metrics, and software to achieve FAIR-compliant data, especially for health data sets. Health Level 7 (HL7) Fast Healthcare Interoperability Resources (FHIR) is a health data content modeling and exchange standard.

OBJECTIVE

Our goal was to devise a new methodology to extract, transform, and load existing health data sets into HL7 FHIR repositories in line with FAIR principles, develop a Data Curation Tool to implement the methodology, and evaluate it on health data sets from 2 different but complementary institutions. We aimed to increase the level of compliance with FAIR principles of existing health data sets through standardization and facilitate health data sharing by eliminating the associated technical barriers.

METHODS

Our approach automatically processes the capabilities of a given FHIR end point and directs the user while configuring mappings according to the rules enforced by FHIR profile definitions. Code system mappings can be configured for terminology translations through automatic use of FHIR resources. The validity of the created FHIR resources can be automatically checked, and the software does not allow invalid resources to be persisted. At each stage of our data transformation methodology, we used particular FHIR-based techniques so that the resulting data set could be evaluated as FAIR. We performed a data-centric evaluation of our methodology on health data sets from 2 different institutions.

RESULTS

Through an intuitive graphical user interface, users are prompted to configure the mappings into FHIR resource types with respect to the restrictions of selected profiles. Once the mappings are developed, our approach can syntactically and semantically transform existing health data sets into HL7 FHIR without loss of data utility according to our privacy-concerned criteria. In addition to the mapped resource types, behind the scenes, we create additional FHIR resources to satisfy several FAIR criteria. According to the data maturity indicators and evaluation methods of the FAIR Data Maturity Model, we achieved the maximum level (level 5) for being Findable, Accessible, and Interoperable and level 3 for being Reusable.

CONCLUSIONS

We developed and extensively evaluated our data transformation approach to unlock the value of existing health data residing in disparate data silos to make them available for sharing according to the FAIR principles. We showed that our method can successfully transform existing health data sets into HL7 FHIR without loss of data utility, and the result is FAIR in terms of the FAIR Data Maturity Model. We support institutional migration to HL7 FHIR, which not only leads to FAIR data sharing but also eases the integration with different research networks.

A Formal Model for the FAIR4Health Information Architecture

In the EU project FAIR4Health, a ETL pipeline for the FAIRification of structured health data as well as an agent-based, distributed query platform for the analysis of research hypotheses and the training of machine learning models were developed. The system has been successfully tested in two clinical use cases with patient data from five university hospitals. Currently, the solution is also being considered for use in other hospitals. However, configuring the system and deploying it in the local IT architecture is non-trivial and meets with understandable concerns about security. This paper presents a model for describing the information architecture based on a formal approach, the 3LGM metamodel. The model was evaluated by the developers. As a result, the clear separation of tasks and the software components that implement them as well as the rich description of interactions via interfaces were positively emphasized.

FAIR4Health: Findable, Accessible, Interoperable and Reusable data to foster Health Research

Due to the nature of health data, its sharing and reuse for research are limited by ethical, legal and technical barriers. The FAIR4Health project facilitated and promoted the application of FAIR principles in health research data, derived from the publicly funded health research initiatives to make them Findable, Accessible, Interoperable, and Reusable (FAIR). To confirm the feasibility of the FAIR4Health solution, we performed two pathfinder case studies to carry out federated machine learning algorithms on FAIRified datasets from five health research organizations. The case studies demonstrated the potential impact of the developed FAIR4Health solution on health outcomes and social care research. Finally, we promoted the FAIRified data to share and reuse in the European Union Health Research community, defining an effective EU-wide strategy for the use of FAIR principles in health research and preparing the ground for a roadmap for health research institutions. This scientific report presents a general overview of the FAIR4Health solution: from the FAIRification workflow design to translate raw data/metadata to FAIR data/metadata in the health research domain to the FAIR4Health demonstrators' performance.

Predicting 30-days Readmission Risk for COPD Patients Care through a Federated Machine Learning Architecture on FAIR Data: Development and Validation Study (Preprint)

Background

Owing to the nature of health data, their sharing and reuse for research are limited by legal, technical, and ethical implications. In this sense, to address that challenge and facilitate and promote the discovery of scientific knowledge, the Findable, Accessible, Interoperable, and Reusable (FAIR) principles help organizations to share research data in a secure, appropriate, and useful way for other researchers.

Objective

The objective of this study was the FAIRification of existing health research data sets and applying a federated machine learning architecture on top of the FAIRified data sets of different health research performing organizations. The entire FAIR4Health solution was validated through the assessment of a federated model for real-time prediction of 30-day readmission risk in patients with chronic obstructive pulmonary disease (COPD).

Methods

The application of the FAIR principles on health research data sets in 3 different health care settings enabled a retrospective multicenter study for the development of specific federated machine learning models for the early prediction of 30-day readmission risk in patients with COPD. This predictive model was generated upon the FAIR4Health platform. Finally, an observational prospective study with 30 days follow-up was conducted in 2 health care centers from different countries. The same inclusion and exclusion criteria were used in both retrospective and prospective studies.

Results

Clinical validation was demonstrated through the implementation of federated machine learning models on top of the FAIRified data sets from different health research performing organizations. The federated model for predicting the 30-day hospital readmission risk was trained using retrospective data from 4.944 patients with COPD. The assessment of the predictive model was performed using the data of 100 recruited (22 from Spain and 78 from Serbia) out of 2070 observed (records viewed) patients during the observational prospective study, which was executed from April 2021 to September 2021. Significant accuracy (0.98) and precision (0.25) of the predictive model generated upon the FAIR4Health platform were observed. Therefore, the generated prediction of 30-day readmission risk was confirmed in 87% (87/100) of cases.

Conclusions

Implementing a FAIR data policy in health research performing organizations to facilitate data sharing and reuse is relevant and needed, following the discovery, access, integration, and analysis of health research data. The FAIR4Health project proposes a technological solution in the health domain to facilitate alignment with the FAIR principles.

From Raw Data to FAIR Data: The FAIRification Workflow for Health Research

Background FAIR (findability, accessibility, interoperability, and reusability) guiding principles seek the reuse of data and other digital research input, output, and objects (algorithms, tools, and workflows that led to that data) making them findable, accessible, interoperable, and reusable. GO FAIR - a bottom-up, stakeholder driven and self-governed initiative - defined a seven-step FAIRification process focusing on data, but also indicating the required work for metadata. This FAIRification process aims at addressing the translation of raw datasets into FAIR datasets in a general way, without considering specific requirements and challenges that may arise when dealing with some particular types of data.

Objectives This scientific contribution addresses the architecture design of an open technological solution built upon the FAIRification process proposed by “GO FAIR” which addresses the identified gaps that such process has when dealing with health datasets.

Methods A common FAIRification workflow was developed by applying restrictions on existing steps and introducing new steps for specific requirements of health data. These requirements have been elicited after analyzing the FAIRification workflow from different perspectives: technical barriers, ethical implications, and legal framework. This analysis identified gaps when applying the FAIRification process proposed by GO FAIR to health research data management in terms of data curation, validation, deidentification, versioning, and indexing.

Results A technological architecture based on the use of Health Level Seven International (HL7) FHIR (fast health care interoperability resources) resources is proposed to support the revised FAIRification workflow.

Discussion Research funding agencies all over the world increasingly demand the application of the FAIR guiding principles to health research output. Existing tools do not fully address the identified needs for health data management. Therefore, researchers may benefit in the coming years from a common framework that supports the proposed FAIRification workflow applied to health datasets.

Conclusion Routine health care datasets or data resulting from health research can be FAIRified, shared and reused within the health research community following the proposed FAIRification workflow and implementing technical architecture.

A Semantic Transformation Methodology for the Secondary Use of Observational Healthcare Data in Postmarketing Safety Studies

Background: Utilization of the available observational healthcare datasets is key to complement and strengthen the postmarketing safety studies. Use of common data models (CDM) is the predominant approach in order to enable large scale systematic analyses on disparate data models and vocabularies. Current CDM transformation practices depend on proprietarily developed Extract-Transform-Load (ETL) procedures, which require knowledge both on the semantics and technical characteristics of the source datasets and target CDM. Purpose: In this study, our aim is to develop a modular but coordinated transformation approach in order to separate semantic and technical steps of transformation processes, which do not have a strict separation in traditional ETL approaches. Such an approach would discretize the operations to extract data from source electronic health record systems, alignment of the source, and target models on the semantic level and the operations to populate target common data repositories. Approach: In order to separate the activities that are required to transform heterogeneous data sources to a target CDM, we introduce a semantic transformation approach composed of three steps: (1) transformation of source datasets to Resource Description Framework (RDF) format, (2) application of semantic conversion rules to get the data as instances of ontological model of the target CDM, and (3) population of repositories, which comply with the specifications of the CDM, by processing the RDF instances from step 2. The proposed approach has been implemented on real healthcare settings where Observational Medical Outcomes Partnership (OMOP) CDM has been chosen as the common data model and a comprehensive comparative analysis between the native and transformed data has been conducted. Results: Health records of ~1 million patients have been successfully transformed to an OMOP CDM based database from the source database. Descriptive statistics obtained from the source and target databases present analogous and consistent results. Discussion and Conclusion: Our method goes beyond the traditional ETL approaches by being more declarative and rigorous. Declarative because the use of RDF based mapping rules makes each mapping more transparent and understandable to humans while retaining logic-based computability. Rigorous because the mappings would be based on computer readable semantics which are amenable to validation through logic-based inference methods.

Adverse drug event notification on a semantic interoperability framework

Adverse drug events (ADEs) are common, costly and one of the most important issues in contemporary pharmacotherapy. Current drug safety surveillance methods are largely based on spontaneous reports. However, this is known to be rather ineffective. There is a lack of automated systems checking potential ADEs on routine data captured in electronic health records (EHRs); present systems are usually built directly on top of specific clinical information systems through proprietary interfaces. In the context of the European project "SALUS", we aim to provide an infrastructure as well as a tool-set for accessing and analyzing clinical patient data of heterogeneous clinical information systems utilizing standard methods. This paper focuses on two components of the SALUS architecture: The "Semantic Interoperability Layer" (SIL) enables an access to disparate EHR sources in order to provide the patient data in a common data model for ADE detection within the "ADE Detection and Notification Tool" (ANT). The SIL in combination with the ANT can be used in different clinical environments to increase ADE detection and reporting rates. Thus, our approach promises a profound impact in the domain of pharmacovigilance.

A federated semantic metadata registry framework for enabling interoperability across clinical research and care domains

In order to enable secondary use of Electronic Health Records (EHRs) by bridging the interoperability gap between clinical care and research domains, in this paper, a unified methodology and the supporting framework is introduced which brings together the power of metadata registries (MDR) and semantic web technologies. We introduce a federated semantic metadata registry framework by extending the ISO/IEC 11179 standard, and enable integration of data element registries through Linked Open Data (LOD) principles where each Common Data Element (CDE) can be uniquely referenced, queried and processed to enable the syntactic and semantic interoperability. Each CDE and their components are maintained as LOD resources enabling semantic links with other CDEs, terminology systems and with implementation dependent content models; hence facilitating semantic search, much effective reuse and semantic interoperability across different application domains. There are several important efforts addressing the semantic interoperability in healthcare domain such as IHE DEX profile proposal, CDISC SHARE and CDISC2RDF. Our architecture complements these by providing a framework to interlink existing data element registries and repositories for multiplying their potential for semantic interoperability to a greater extent. Open source implementation of the federated semantic MDR framework presented in this paper is the core of the semantic interoperability layer of the SALUS project which enables the execution of the post marketing safety analysis studies on top of existing EHR systems.