On Tuesday 23 June, IDEA4RC brought researchers, clinicians, policymakers and innovators together at the Ministry for Digitalization of the Community of Madrid for a public session on how health data can be securely reused for research and better patient care. Opening the event, Giuseppe Fico framed cancer research as a mission that depends on building an ecosystem of trust across institutions, a theme that ran through the rest of the day. It was echoed by host Miguel López Valverde in his welcome speech: collaboration across institutions and countries, he said, is now essential, not optional, for rare cancer research.
At the centre of the day was IDEA4RC’s privacy-preserving approach. Eugenio Gaeta, a software engineer at the Universidad Politécnica de Madrid and the project’s technical manager, presented the zero-trust architecture behind the platform. Coordination and data custody are kept separate, so central services orchestrate analyses without ever owning clinical data, which stay under each hospital’s control. Data are harmonised locally and authorised algorithms visit the data. Only algorithms results are shared with the researchers conducting the analysis.
Gaeta offered a concrete test of that design. When Vantage6, the platform’s federated-analysis component, was targeted by an attack, the absence of a central data repository meant no breach reached the data held inside any hospital. That is what the architecture’s resilience is meant for: containing an attack before it can touch patient data is what lets hospitals keep trusting the platform with their data.
Frank Martin, a software engineer at the Netherlands Comprehensive Cancer Organisation (IKNL) who leads the platform’s federated-learning component, showed how increasingly advanced statistical analyses can now run across hospitals without centralising patient data, a capability that has matured considerably since the project’s early design work.
The day also addressed the harder edges of using AI on clinical text. Unai Zulaika, a data scientist at the University of Deusto responsible for the NLP activities, described ongoing work to extract structured information from clinical notes. Zulaika explained that current language models still struggle with variables carrying temporal information, such as disease recurrence. He also highlighted that in evaluating model performance, one should consider the possibility of fabricated answers: instances where the model returns an output for a given variable even if it was not stated at all in the source text. “Larger models behave better than smaller ones, though the problem does not disappear”, he concluded.
Angelo Solimini, policy officer at the Directorate-General for Research and Innovation of the European Commission, placed IDEA4RC within the much larger effort of the EU Mission on Cancer. So far, the Mission has invested more than 600 million euros across 75 cancer projects. Among them, Solimini pointed to two flagship initiatives, UNCAN.eu and the European Cancer Patients Digital Centre (ECPDC).
Solimini was followed by representatives of two implementation actions for UNCAN.eu: UNCAN-Conncet and CANDLE. First, Jayden Khakurel, technical manager of UNCAN-Connect, described how they are working to develop a prototype for the central platform envisaged in UNCAN.eu. Second, Lifang Liu, project manager CANDLE, explained how they are working to build the national cancer data node for UNCAN.eu starting from mapping national capacity.
The discussion closed with the host’s own perspective. José Luis Bezares, of the Community of Madrid’s Directorate-General for Digital Health, described a region of 8 million people, 36 public hospitals and 500 primary care centres already working with OMOP and HL7 FHIR, piloting NLP with John Snow Labs, and taking part in CHAIN, a cardiovascular-data project spanning 52 partners across 19 countries. “Health data managed responsibly are among the most powerful levers to improve health outcomes”, Bezares said, adding that the next challenge is making that approach sustainable. Across the day’s sessions, collaboration, interoperability and trust kept surfacing as the conditions that make this vision workable in practice.