AI tool predicts diseases: ‘The risk is that we might interpret it the wrong way.’

Recently, INDICATE co-lead Michel van Genderen shared his expertise on the Dutch national news program NOS Nieuwsuur, discussing the risks associated with AI tools such as Delphi-2M, which predicts the likelihood of developing diseases.

How much risk does someone have of developing a disease like cancer? Or of having a heart attack? The Delphi-2M system can predict that. But does it really help us? Or does it also come with risks?

“When I read about it, I thought: yet another prediction model. Research shows that only two percent of all AI innovations are actually used in healthcare. Innovations often don’t fit well with real-world practice,” says Michel van Genderen, internist-intensivist and co-founder of the AI ethics lab REAiHL, a collaboration between Erasmus MC and TU Delft.

Van Genderen and his colleagues advise the World Health Organization (WHO) on the use of AI in healthcare: “Embracing AI in healthcare will go wrong if there are no moral ground rules. It must be safe and ethical.”

Read the full article here (in Dutch) or watch the news item here (in Dutch). The item starts at 30:30 minutes.

Source: NOS Nieuwsuur

We are pleased to welcome the European Society of Intensive Care Medicine (ESICM) to the External Expert Advisory Board of the INDICATE Project!

This collaboration strengthens our shared European reach and accelerates progress toward secure, interoperable ICU data sharing, fully aligned with the ambition of the European Health Data Space (EHDS).

Together, we aim to support Intensive Care Units across Europe in responsibly connecting data, advancing research, improving clinical decision-making, and ultimately enhancing patient care. A key element will be the development of a federated data infrastructure, enabling hospitals to collaborate without moving data outside the hospital, while maintaining privacy, trust, and local control.

ESICM will be represented by Maurizio Cecconi in the External Expert Advisory Board of INDICATE.

The high-level event on Virtual Human Twins (VHTs), held on the 21st of October in Brussels, Belgium, brought together leading voices from the European Commission, academia, industry, and patient organisations, all united by a shared goal: accelerating innovation in health through AI, data, and trustworthy digital ecosystems.

Key takeaways:

• Europe has the technology, talent, and ambition to lead, but success depends on access to high-quality, interoperable data. This highlights the importance of our INDICATE project, as well as other European health data initiatives such as Genome Of Europe, EUCAIM, and PHEMS, all of which strengthen the foundation for innovation.
• Trust remains essential. It is one of the key underlying principles of the European Health Data Space (EHDS) Regulation, and aims to give individuals more control over their health data while enabling safe and secure use for research and innovation.
• VHTs offer a unique testbed for AI in healthcare, helping translate innovation from lab to clinic.
• Collaboration across disciplines, sectors, and borders is vital to build an inclusive, patient-centred ecosystem.
• Robust regulatory frameworks Medical Device Regulation (MDR), In Vitro Diagnostics Regulation (IVDR), AI Act and European health data spaces will be key to scaling impact safely and responsibly.

Our technical lead of INDICATE, Jan van den Brand, had the honour to join the panel discussion on Harnessing data for advancing integrated VHTs with a focus on EU data spaces and infrastructures. He shared how our federated approach to data availability will enhance access to critical ICU population data, ensuring that valuable insights can be generated without compromising privacy or data sovereignty.

Discussions during the technical and regulatory sessions further highlighted the importance of secure data environments, data harmonisation, and clear guidance for AI-enabled research. Experts agreed that VHTs can accelerate clinical evidence generation, potentially reducing animal testing and enabling personalised care. The event concluded with a shared message: VHTs sit at the crossroads of digital, data, and health policy.

That intersection brings enormous potential, but also challenges, as it means VHTs touch upon regulatory frameworks across multiple domains. Building trust, ensuring quality, and fostering collaboration will determine how quickly Europe can turn this vision into real impact for patients.

At the II National Congress on Innovation for the Advancement of Health Data Management and Evaluation (INNODATA 2025), held on 24–25 September 2025 in Seville (Spain), the Computational Health Informatics Group from the Virgen del Rocio University Hospital, as part of the Andalusian Health Service (Servicio Andaluz de Salud—SAS), presented their work on advancing federated integration of ICU data in Europe.

The audience—composed of healthcare professionals, clinical researchers, biomedical informatics experts, data engineers, governance managers, and institutional stakeholders—gathered to hear the presentation titled “ETL Guide for the Federated Integration of ICU Data in Europe”, delivered by Maria Parra Rodriguez-Armijo, Celia Alvarez Romero, Maria Gonzalez Lopez, Alberto Garcia-Agundez Blanco, Silvia Rodriguez Mejias, and Carlos Luis Parra-Calderón.

The talk began by setting the stage with the motivation behind the work: the pressing need for federated infrastructures that comply with FAIR principles in order to enable the secondary use of ICU data and facilitate the deployment of Artificial Intelligence in critical care. The talk also outlined key regulatory hurdles, including GDPR compliance, governance responsibilities, and the challenge of building trust in data sharing, alongside persistent technical barriers such as data fragmentation, limited semantic interoperability, and the absence of common ETL (Extract, Transform, Load) pipelines.

The session then introduced the ETL pipeline structure of INDICATE, which operates within a federated architecture, ensuring that data remain under local governance. The pipeline is organised into three essential layers:

1. Extraction through HL7 FHIR resources.
2. Transformation and storage in OMOP-CDM for analytics and AI model development.
3. FAIR metadata to support data discovery, interoperability, and reuse.

At the heart of the presentation was the ETL Guide (INDICATE Data Provider Handbook), conceived as a living document designed to support data providers in navigating the complexities of ICU data integration. The Handbook encompasses:

• A dual approach using both OMOP-CDM and HL7 FHIR.
• Adoption of semantic and syntactic standards, including SNOMED CT, LOINC, ATC, ICD-10, and DICOM.
• Practical guidance on ETL processes, de-identification, and GDPR compliance.
• Data quality assessment frameworks such as OHDSI DQD (Data Quality Dashboard from the Observational Health Data Sciences and Informatics (OHDSI) community) and validation metrics.
• A publication and maintenance strategy to ensure the document evolves continuously with emerging needs.

The closing remarks emphasized the Handbook’s pivotal role as a bridge between local ICU data and the INDICATE federated infrastructure, guaranteeing standardisation, GDPR compliance, interoperability, and scalability. By underpinning the six INDICATE clinical use cases, the Handbook contributes directly to the European Health Data Space (EHDS), aligns with the FAIR principles, and lays the groundwork for AI-readiness in critical care.

Abbreviations 

• ETL (Extract, Transform, Load)
• Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM)
• OHDSI DQD = the Data Quality Dashboard from the Observational Health Data Sciences and Informatics (OHDSI) community.
• HL7 FHIR (Fast Healthcare Interoperability Resources)
• SNOMED CT: Systematized Nomenclature of Medicine – Clinical Terms
• LOINC: Logical Observation Identifiers Names and Codes
• ATC: Anatomical Therapeutic Chemical Classification System
• ICD-10: International Classification of Diseases (Tenth Revision)
• DICOM: Digital Imaging and Communications in Medicine

End of September: INDICATE Design Workshop in Rotterdam at KPMG – lots of bright minds in the room thinking about how we can advance patient-centered care and promote the use of Intensive Care data by developing and implementing responsible and trustworthy AI models.

One of the AI use-cases discussed was the model to predict sepsis using ai on the neonatal intensive care department. The main goal is to develop and validate a continuous prediction model which detects late onset neonatal sepsis before symptoms occur.

Other AI use-cases discussed are the atlas of anonymous acute care cases covering infectious diseases, cancer, heart diseases, and brain injury, the early detection of organ failure, and a digital twin for mechanical ventilation.

As you can imagine, when developing these use-cases, a fine balance is needed between innovation and dealing with sensitive data of patients.

Read the full programme here.

Indicate Plateau 2 Workshop – Rotterdam

Program Wednesday, September 24^th  

Evening program

Program Thursday, September 25^th

Program Friday, September 26^th