Periodic Reporting for period 1 - PHEMS (Pediatric Hospitals as European drivers for multi-party computation and synthetic data generation capabilities across clinical specialties and data types)
Reporting period: 2023-10-01 to 2025-03-31
The PHEMS project is a joint EU project between European children's hospitals and technology partners. Our primary goal is to improve the use of health data between countries, particularly in the fields of children's health and rare diseases. PHEMS aims to tackle challenges in accessing relevant health data essential for international collaboration between scientists, clinicians, researchers, and the health industry—without compromising patient privacy.
The project will provide European children’s hospitals with a Pediatric Health Data Space (PHDS), consisting of technical components and governance frameworks. The objective is to facilitate access to health data, advance federated health data analysis and build services for the on-demand generation of shareable, synthetized, and anonymized datasets.
To achieve this, we leverage federated analytics, decentralized machine learning, synthetic and anonymized data, and a structured contractual and governance framework to enable long-term collaboration between data controllers (hospitals) and various use cases such as clinical and operational benchmarking, scientific research based on federated data analytics, federated machine learning for algorithm development and validation, and supporting clinical trials feasibility and execution.
The PHEMS project will validate its techniques through three clinical use cases. These are designed to demonstrate the benefits of the federated analytics network, validate the utility of algorithmically anonymized and synthetic data, and demonstrate the operability of the open ecosystem. Together, they serve as proof of principle, with the potential to scale across other clinical areas and hospital networks.
• Clinical Use Case 1: Cardiology Patients' Operation Benchmarking
Aiming to improve pediatric cardiology outcomes by creating a standardized, real-time system for data collection, monitoring, and quality measures, while promoting a culture of benchmarking across institutions.
• Clinical Use Case 2: Pediatric Intensive Care Unit (PICU) Sepsis
Developing and testing Machine Learning algorithm to predict sepsis across PICUs in four major European children’s hospitals by clustering patient trajectories based on outcomes.
• Clinical Use Case 3: Hematology; Hemophilia
Developing and testing a Machine Learning-based prediction algorithm to personalize treatment for pediatric hemophilia, aiming to reduce bleeding, improve quality of life, and lower costs.
Read more about the project:
www.PHEMS.eu
www.Linkedin.com/company/phems
www.Youtube.com/@PHEMS_EUProject
The project will provide European children’s hospitals with a Pediatric Health Data Space (PHDS), consisting of technical components and governance frameworks. The objective is to facilitate access to health data, advance federated health data analysis and build services for the on-demand generation of shareable, synthetized, and anonymized datasets.
To achieve this, we leverage federated analytics, decentralized machine learning, synthetic and anonymized data, and a structured contractual and governance framework to enable long-term collaboration between data controllers (hospitals) and various use cases such as clinical and operational benchmarking, scientific research based on federated data analytics, federated machine learning for algorithm development and validation, and supporting clinical trials feasibility and execution.
The PHEMS project will validate its techniques through three clinical use cases. These are designed to demonstrate the benefits of the federated analytics network, validate the utility of algorithmically anonymized and synthetic data, and demonstrate the operability of the open ecosystem. Together, they serve as proof of principle, with the potential to scale across other clinical areas and hospital networks.
• Clinical Use Case 1: Cardiology Patients' Operation Benchmarking
Aiming to improve pediatric cardiology outcomes by creating a standardized, real-time system for data collection, monitoring, and quality measures, while promoting a culture of benchmarking across institutions.
• Clinical Use Case 2: Pediatric Intensive Care Unit (PICU) Sepsis
Developing and testing Machine Learning algorithm to predict sepsis across PICUs in four major European children’s hospitals by clustering patient trajectories based on outcomes.
• Clinical Use Case 3: Hematology; Hemophilia
Developing and testing a Machine Learning-based prediction algorithm to personalize treatment for pediatric hemophilia, aiming to reduce bleeding, improve quality of life, and lower costs.
Read more about the project:
www.PHEMS.eu
www.Linkedin.com/company/phems
www.Youtube.com/@PHEMS_EUProject
TECHNICAL ACHIEVEMENTS
During the first half of the project, the overall technical architecture for the federation network was designed; including the role of Workspace Provider serving data users, and defining the Shared Services for distributing analysis code and analytical aggregated results from workspaces to the data controllers and back. It also features federated nodes, which serve as technical intermediaries between data users and the original data hosted by data controllers, such as hospitals.
A dedicated application was developed to manage research projects, permit applications, and cohort definitions.
A Proof-of-Concept federated network was implemented across participating hospitals, with successful initial benchmarking using real-world data.
GOVERNANCE ACHIEVEMENTS
PHEMS partners aim to establish a Pediatric Health Data Space (PHDS) by building shared infrastructure, defining governance, and drafting contracts. A tailored Business Model Canvas and initial feasibility analysis for forming the ECHO Association as a long-term governance body have been completed.
USE CASE ACHIEVEMENTS
The Use Cases validate the data federation components, ensuring selected technical and analytical solutions. The Use Cases will also provide a Proof-of-Concept that can be extended to other healthcare settings. A key milestone is the development of a Use Case management framework covering dataset selection, variable methods, data quality, OMOP alignment, and FAIR principles. In PHEMS the OMOP guidelines were established to ensure consistent data harmonization and support for ETL and variable mapping.
Highlights by Use Case:
• Clinical Use Case 1: Cardiology benchmarking
- A comprehensive list of clinical data was compiled.
- Variables were selected for benchmarking purposes.
- Local registrations and ethical approvals were secured.
- A Proof-of-Concept version for benchmarking clinical outcomes was created.
- Insights from other benchmarking methods used by the Pediatric Cardiac Critical Care Consortium (PC⁴) and the UK's national cardiac morbidity and mortality reporting were discussed.
• Clinical Use Case 2: Pediatric Sepsis
- Clinical partners agreed upon the dataset and variables. Data controllers conducted internal data exploration and mapped their data to the OMOP CDM, aligning with the federated node requirements.
- All clinical sites obtained the necessary research permits and ethical approvals.
- A Hackathon for Predictive Modelling was organized, which significantly accelerated the OMOP mapping process and generated valuable insights into machine learning modelling approaches.
- The first version of the predictive algorithm for sepsis has started development (expected to be completed by September 2025).
• Clinical Use Case 3: Hematology
- Challenges related to limited data collection were identified and addressed, and strategies for overcoming them were devised.
- A Machine Learning algorithm was developed to handle missing data and limited information in population pharmacokinetic modelling (this algorithm is in testing).
During the first half of the project, the overall technical architecture for the federation network was designed; including the role of Workspace Provider serving data users, and defining the Shared Services for distributing analysis code and analytical aggregated results from workspaces to the data controllers and back. It also features federated nodes, which serve as technical intermediaries between data users and the original data hosted by data controllers, such as hospitals.
A dedicated application was developed to manage research projects, permit applications, and cohort definitions.
A Proof-of-Concept federated network was implemented across participating hospitals, with successful initial benchmarking using real-world data.
GOVERNANCE ACHIEVEMENTS
PHEMS partners aim to establish a Pediatric Health Data Space (PHDS) by building shared infrastructure, defining governance, and drafting contracts. A tailored Business Model Canvas and initial feasibility analysis for forming the ECHO Association as a long-term governance body have been completed.
USE CASE ACHIEVEMENTS
The Use Cases validate the data federation components, ensuring selected technical and analytical solutions. The Use Cases will also provide a Proof-of-Concept that can be extended to other healthcare settings. A key milestone is the development of a Use Case management framework covering dataset selection, variable methods, data quality, OMOP alignment, and FAIR principles. In PHEMS the OMOP guidelines were established to ensure consistent data harmonization and support for ETL and variable mapping.
Highlights by Use Case:
• Clinical Use Case 1: Cardiology benchmarking
- A comprehensive list of clinical data was compiled.
- Variables were selected for benchmarking purposes.
- Local registrations and ethical approvals were secured.
- A Proof-of-Concept version for benchmarking clinical outcomes was created.
- Insights from other benchmarking methods used by the Pediatric Cardiac Critical Care Consortium (PC⁴) and the UK's national cardiac morbidity and mortality reporting were discussed.
• Clinical Use Case 2: Pediatric Sepsis
- Clinical partners agreed upon the dataset and variables. Data controllers conducted internal data exploration and mapped their data to the OMOP CDM, aligning with the federated node requirements.
- All clinical sites obtained the necessary research permits and ethical approvals.
- A Hackathon for Predictive Modelling was organized, which significantly accelerated the OMOP mapping process and generated valuable insights into machine learning modelling approaches.
- The first version of the predictive algorithm for sepsis has started development (expected to be completed by September 2025).
• Clinical Use Case 3: Hematology
- Challenges related to limited data collection were identified and addressed, and strategies for overcoming them were devised.
- A Machine Learning algorithm was developed to handle missing data and limited information in population pharmacokinetic modelling (this algorithm is in testing).
Overall, the PHEMS project is expected to have a significant positive impact on health data collaboration, innovation, and healthcare outcomes in Europe.
PEDIATRIC HEALTH DATA SPACE (PHDS)
PHEMS aims to build a sustainable and scalable ecosystem that can be adopted by other hospitals and healthcare organizations.
The project’s technological impacts arise from implementing federated data access without centralized databases. This approach allows each data controller to participate with their organizational policies and local regulations. Additionally, the use of algorithmic anonymization and synthetization technologies will enable the creation of completely new kinds of datasets, and solving of previously unsolved research problems.
The PHEMS clinical use cases will provide valuable insights into the utility of federated analytics and synthetic data in real-world healthcare settings, ultimately contributing to improved patient outcomes and more efficient healthcare processes.
PEDIATRIC HEALTH DATA SPACE (PHDS)
PHEMS aims to build a sustainable and scalable ecosystem that can be adopted by other hospitals and healthcare organizations.
The project’s technological impacts arise from implementing federated data access without centralized databases. This approach allows each data controller to participate with their organizational policies and local regulations. Additionally, the use of algorithmic anonymization and synthetization technologies will enable the creation of completely new kinds of datasets, and solving of previously unsolved research problems.
The PHEMS clinical use cases will provide valuable insights into the utility of federated analytics and synthetic data in real-world healthcare settings, ultimately contributing to improved patient outcomes and more efficient healthcare processes.