The Hospital's Missing Layer

In brief

Every research hospital in Europe holds more evidence about its patients than it can use. The records of their care, the registries that follow their diseases, the samples in the biobank, the images in the archive, and the files of the trials they joined all sit in separate systems, each run by its own department and each built for something other than research. A researcher who wants to ask one question across them must find the data, request it from each owner, and join it by hand. Most questions are never asked, because the cost of asking is too high.

The consequences are concrete. At Nantes University Hospital, a research warehouse that recorded about 1.97 million patients as living turned out, once linked to France's national death records, to hold 205,698 patients who had died: about three times the deaths it knew of (Guardiolle et al., 2022). At the Medical University of Vienna, a median of 926 days passed before a stored sample was first used, and four types of sample were never requested at all (Mucher et al., 2026). In the European rare kidney disease network, centres were feeding more than 60 separate registries, none of them used by every centre, before they agreed on one (Bassanese et al., 2021). Studies from other European countries, set out in section 2, show the same problem. The data exist, but the hospital cannot link them.

Where a hospital links its data, patients benefit. Lausanne University Hospital built a sepsis registry fed every day from its own clinical data warehouse. On the wards in the programme it found sepsis in 9 to 10% of stays, where diagnosis codes had recorded 2 to 4%, and on those wards in-hospital mortality among patients flagged with sepsis fell from 20.54% to 15.27% between the periods before and after the system was introduced (Despraz et al., 2026). Helsinki's hospital biobank screened genetic data it already held for high-risk variants in the BRCA1, BRCA2, and PALB2 genes; about three quarters of the carriers it found had not been identified in clinical care, and for 17 of the 19 newly identified women carriers, care changed (Pehrsson et al., 2026).

Between hospitals the gap widens. In the EU's own pilot of cross-border data access, approval and access took between one and twelve months per data node (Schneider et al., 2025). Europe's main network for observational research has 210 data sources in 30 countries, with Italy, Great Britain, and Spain alone supplying 13%, 12.5%, and 11.5% of them, and none in Poland, Slovakia, Slovenia, Latvia, or Lithuania (Blacketer et al., 2025). Meanwhile the European Economic Area's share of global commercial clinical trials fell from 22% in 2013 to 12% in 2023 (European Commission, 2025a). Industry sponsors run most of Europe's trials, and hospitals depend on them, but about 40% of EU trials are sponsored by academia (EMA, 2023), and many of the questions that matter most to patients carry little commercial reward. A hospital can pursue those questions only with data it can find and use itself.

The law has now set a timetable. From March 2029, under the European Health Data Space (EHDS) Regulation, a hospital that holds health data must describe it for a national catalogue and supply requested data to its national access body within three months, and datasets built with public money must carry a data quality label (Regulation (EU) 2025/327). A hospital that cannot say what research data it holds will meet these duties with the same manual effort its researchers face today.

Some governments have started to pay for the answer. France has put more than EUR 100 million since 2023 into hospital data warehouses across 31 university hospitals and more than 50 other establishments (Ministère de la Santé, 2026). Germany has committed more than EUR 400 million to its Medical Informatics Initiative, which every university hospital has joined. These programmes reach university hospitals first, and none yet guarantees the running costs. In much of Central and Eastern Europe we found no comparable programme at all. Most of Europe's research hospitals will have to build this capacity themselves.

We propose that every research hospital in Europe build a research data hub of its own: a governed function inside the hospital that brings its research data together under common standards, answers questions from researchers, sponsors, and networks without the data leaving its control, and prepares data for the access body when the law requires it. To help hospitals judge where they stand, we built the Hospital Research Data Readiness Framework, a six-dimension, five-level assessment with a 20-question self-assessment in Annex A.

What to do now

• In the next 90 days, bring research leadership, IT, the data protection officer, and the biobank and registry leads together around the self-assessment, and agree who in the hospital is accountable for research data.
• During 2027, reach level 2 on all six dimensions, and bring one dataset that matters to your researchers to a prepared and labelled state.
• By March 2029, if your hospital is a data holder under the EHDS, be ready to describe your datasets in the national catalogue and supply data on request, with one access committee and written rules for every source you hold.

1. A hospital holds more data about its patients than it can use

Every research hospital already collects the data that research, trials, and better care need. What most lack is a function that brings those data together and governs their use.

Consider a question any oncologist might ask: of the patients treated here for a particular cancer over the past five years, how many relapsed within two years, and what did their scans and tumour samples look like at diagnosis? Every piece of that answer is inside the hospital. The diagnoses are in the electronic record, the relapses in follow-up notes and the cancer registry, the scans in the imaging archive, the samples in the biobank, and some of the patients in the files of trials the hospital ran. In most European hospitals, answering the question means a research nurse, a data manager, or a doctoral student spending weeks requesting extracts from each department, matching patients by hand, and cleaning what comes back. When the study ends, the work is usually lost, and the next researcher starts again.

A research data hub removes that repeated work. We use the term for a governed function inside the hospital, with its own staff, rules, and infrastructure, that does four things. It brings the hospital's research data together, from records, registries, biobanks, imaging, and trials, and links them under common standards. It checks and documents their quality. It decides, through a committee with written rules, who may use which data and for what. And it answers questions from outside, from a sponsor counting eligible patients, a research network running a federated study, or a national access body, without the hospital handing over control of the data.

Such hubs already run in several countries. Germany's data integration centres, set up at university hospitals under the Medical Informatics Initiative, extract, integrate, harmonise, quality-assure, and provide clinical data for research, and the data remain decentralised at each centre (Albashiti et al., 2024). The centre at Halle runs as a joint core facility of the hospital and the medical faculty, with an operational team inside hospital IT and a scientific team beside it, offering identity management, pseudonymisation, consent management, recruitment support, and study databases (SMITH, 2025). At Regensburg, one use and access committee, with the data integration centre, the hospital board, and the biobank represented, reviews feasibility, scientific merit, and legal basis for every request, including requests for biobank samples (Universitätsklinikum Regensburg, 2025). In the Netherlands, the Princess Máxima Center links patient records, biobank data, diagnostics, and clinical trial data in a single shared registry (Health-RI, 2022). In Switzerland, every participating hospital installs common software that transforms, de-identifies, and checks its data before anything leaves (SPHN, 2026).

In each of these examples, research data are handled in a system separate from the one that runs patient care. A clinical record system is built around the individual case; research needs a subset of the data for a whole group of patients (Gehrmann et al., 2023). At Lille University Hospital, the anaesthesia department found that its vendor's system stored data in a proprietary structure that could not be changed, so it built a separate warehouse. Over a ten-year project, that warehouse came to hold 636,784 anaesthesia records on 353,152 patients and to support about 15 studies a year and ten clinical dashboards, with one full-time data scientist working alongside anaesthetists and computer scientists (Lamer et al., 2022). Research data therefore need their own system, alongside the care systems.

What a hub gives back

The case for a hub rests on what it lets a hospital do that it cannot do now. The evidence comes mostly from single hospitals, and few studies have controls, but it points one way.

A hub finds patients that routine records miss. Lausanne's sepsis registry and Helsinki's biobank screening, described above, are the clearest cases. In both, the hospital already held the data; what changed was its ability to join them and act.

A hub also makes research possible at a scale no single study could afford. Nantes University Hospital's warehouse, built from 2018 at a cost of EUR 2.6 million over six years, had handled 577 requests and supported 269 projects by the time it was reported, covering about 1.5 million patients (Karakachoff et al., 2024). The warehouse of Assistance Publique-Hôpitaux de Paris (AP-HP) holds data on 19 million patients and has supported 241 research projects (AP-HP, 2025). In 2020, Germany's university hospitals ran the same federated query at each site to answer a national question about hospital admissions during the first wave of COVID-19; the shared software reached 20 of them, and the 18 that obtained ethics approval, together accounting for about one million inpatient admissions a year, delivered results (Kapsner et al., 2021).

Sponsors notice the difference. In interviews with pharmaceutical companies and contract research organisations in the Nordic countries, sponsors said they favour sites that can support their patient count estimates with health record data (Laaksonen et al., 2021). A feasibility platform rolled out to 33 German university hospitals lets a researcher send a count query to each hospital's repository and receive only aggregate numbers back (Gruendner et al., 2022).

AI depends on the same capability. Models trained in one hospital rarely work unchanged in another. A certified intensive care model developed at Amsterdam UMC lost accuracy when moved to Leiden University Medical Center and regained it only after retraining on Leiden's own data, which first had to be mapped across two different record systems (de Hond et al., 2023). Across 572 studies of AI scoring systems in intensive care, only 84 had been tested on data from outside the place they were built (Rockenschaub et al., 2025). Clinicians want that testing done: in a survey of 303 physicians at eight German university hospitals, 91.1% wanted AI evaluated before routine use, and 83.2% wanted to use anonymised patient data for their own research (Maassen et al., 2021). A hospital that can prepare its own data is a hospital that can test, adapt, and trust the tools it buys.

2. Inside the hospital, research data are often scattered

Registries, biobanks, imaging archives, and consent records usually sit with different departments, in different systems. Researchers cannot easily find or link what the hospital holds.

Most hospitals never planned their research data; it accumulated. Each registry was started by a clinical team to answer its own questions, each biobank collection by the researchers who needed the samples, and each imaging archive by radiology for diagnosis. Nobody designed them to work together, and in most hospitals nobody is responsible for making them do so. The consequence is that any question spanning more than one source becomes a project of its own: someone has to find the data, ask each owner for it, and match the patients by hand.

Registries multiply

Registries are the oldest form of organised research data in European hospitals, and the most scattered. Germany's national catalogue lists 423 medical registries, and for more than half of them it records no legal form, so it cannot say who owns them (Niemeyer et al., 2026). Europe has more than 800 rare disease registries (Hageman et al., 2023). In a review of 37 of them, only 22% used a common data set and only 24% coded their data to a standard ontology (Hageman et al., 2023). The authors warn that a lack of focus on improving and sustaining existing registries produces data that is "not always usable nor shareable".

The rare kidney disease network shows what this means on the ground. When its centres surveyed their registries in 2017, they found more than 60 in use, mostly disease-specific; none was used by every centre, and few tracked outcomes (Bassanese et al., 2021). Data entry depended on very few people: in 27 centres, a single person was the only active user. Even Sweden, whose national quality registries are used in every hospital and funded nationally, shows the limits of the model: a validation of its renal registry found diagnoses of two kidney diseases correct in 95% of cases, while the share of kidney biopsies registered rose only from 37% to 58% over its first five years; detailed data were entered by hand through a web form (Swedish Renal Registry, 2024).

Biobank samples wait years to be used

Europe's biobank directory lists 4,617 collections. Of these, 2,573 declare that they hold medical records alongside their samples and 413 that they hold imaging (BBMRI-ERIC Directory, 2026). A declaration says the records exist somewhere. It does not say that anyone has linked them to the samples.

The most careful study of how a hospital biobank is used comes from the Medical University of Vienna. Six to seven years after deposit, 55% of the collections stored in 2019 had been used at least once, but only 25.4% of individual submissions and 3.5% of aliquots. When samples were used, the median wait for first use was 926 days, and four sample types were never requested at all (Mucher et al., 2026). The authors caution that low aliquot use is not simply waste, since biobanks store material for future questions. They also name the structural problem: the clinician who collected the samples "is also the gatekeeper of the linked clinical data." Without a hospital function that holds the link, the samples are only as useful as one person's availability.

Warehouses rarely include imaging or genomic data

France offers the best measure of what hospital warehouses contain, because it studied them. In 2022, researchers interviewed teams at all 32 French university and regional hospitals. Fourteen had a warehouse in production, five were experimenting, five were planning one, and eight had no project at all (Doutreligne et al., 2023). Among the 21 warehouses examined, only four had integrated imaging, and none had integrated genomic data. No single data model was used; nine warehouses relied mainly on the OMOP common data model. Only six offered researchers a secure analysis environment, so in the others researchers still worked on extracts. Internal documentation existed in ten and was never public.

Other countries describe the same pattern in their own terms. In Sweden, data sources exist in silos, each region chooses its own record system, and there is no national repository (TEHDAS, 2023a). In Denmark, records sit in regional warehouses (TEHDAS, 2022). Belgian authors describe their country's health data as rich but fragmented and not FAIR enough, with no harmonised route for secondary use (Saelaert et al., 2023). Estonia, often held up as Europe's most digital health system, converted a national sample of its records to the OMOP model and mapped 95% of entries on average, but only 38.9% of the local service codes hospitals used, because local coding is where national systems break down (Oja et al., 2023). Much of the most valuable information is not coded at all: French cancer centres estimate that about 80% of pertinent clinical information sits in free text (Guérin et al., 2024).

Most patients agree, but few decisions are recorded

Patients usually agree to research use when asked. At University Hospital Münster, 96% of adult cancer patients with a documented decision on broad consent had agreed. Yet only 4,835 of 38,819 patients, about one in eight, had a documented decision at all (Hoffmann et al., 2026). In interviews, German data protection officers and health IT staff described why. One described the different consent versions in use as making reuse "extremely complex"; another said that with larger numbers "you cannot handle it on paper anymore" (Wiertz and Boldt, 2024). In France, the national health authority found that a patient has no way of knowing whether their data are included in a given project (HAS, 2022).

Too few people to do the work

The work of joining data falls to small teams. French warehouse teams ranged from half a full-time post to 80 people at AP-HP, with a median of six, and usually began with one or two people (Doutreligne et al., 2023). Private companies offered implementation services in 15 of the 21 cases studied. In a European survey of data stakeholders, only 15.6% reported clearly defined roles for data quality, and 68.8% had no dedicated quality manager or team (Declerck et al., 2026). At Leiden University Medical Center, a project to make COVID-19 data reusable ran, in the authors' words, "without dedicated FAIR stewards" (Queralt-Rosinach et al., 2022).

Digital maturity scores do not capture this gap. The EU's eHealth index, which measures citizens' online access to their records, reached 87% across the EU in 2025 (European Commission, 2026). German hospitals' digital maturity averaged 55 out of 100 in the latest national assessment (DigitalRadar, 2026). Neither measures whether a hospital can find, link, and prepare its data for research. A hospital can be digitally advanced in care and still unable to answer a simple research question about its own patients.

3. What fragmentation costs

The price of disconnected data is paid in missed patients, unreliable findings, AI models that fail outside the hospital that built them, and research that never starts.

Patients are missed

When data are not joined, the hospital does not see what has happened to its patients. Nantes discovered that its warehouse missed most deaths after discharge only when it linked to the national death register (Guardiolle et al., 2022). The authors note that vital status after discharge is often missing or uncertain in hospital warehouses, though it is central to their research value, and that linkage itself worked less well for some groups, including people born outside France. A study of survival or treatment outcomes run on the unlinked warehouse would have reached the wrong answer.

The same holds for diagnoses. Lausanne's sepsis registry found several times as much sepsis as routine codes had captured (Despraz et al., 2026). Helsinki's biobank found that only about a quarter of the carriers of high-risk breast and ovarian cancer variants in its genetic data were already known to clinical care (Pehrsson et al., 2026). In the rare kidney disease network, active centres had on average enrolled an estimated quarter to a third of the patients available to them (Bassanese et al., 2021).

Findings become unreliable

Preparing data badly can be worse than not preparing it at all. At Tübingen's medical data integration centre, a rare disease dataset was converted to make it findable, accessible, interoperable, and reusable. Afterwards it contained 1,988 data quality issues, and 1,977 of them had been introduced by the conversion (Tahar et al., 2026). One French warehouse data scientist, interviewed about the work, described results that could have been published but were not, because the team was not confident in the data (Priou et al., 2023). The authors of a federated French oncology network conclude that data quality variability "must be acknowledged, quantified, and considered" (Guérin et al., 2024). A hospital that does not own the quality of its data cannot vouch for the research built on it.

AI models fail outside the hospital that built them

AI models lose accuracy when they move between hospitals because each hospital codes and structures its data differently, and a model learns those differences. The authors of the German physician survey call for research databases "applicable to the situation in Germany" (Maassen et al., 2021). An account of five European AI projects on big data infrastructures for imaging found efforts "fragmented, based on single-institution, size-limited, and annotation-limited datasets" (Kondylakis et al., 2023). Without prepared local data, a hospital can neither validate the tools it is sold nor contribute to building better ones.

Research slows, or never starts

At Lille, a study takes weeks if the variables it needs already exist in the warehouse, and months if they must be built (Lamer et al., 2022). In Finland, where secondary use runs through a national permit authority, 72% of 346 clinical researchers surveyed said the permit process had slowed their work, and 51% reported higher costs (Sanmark et al., 2025). During the pandemic, European research projects funded at speed found hospitals hard to engage for observational studies, and authorisation of trials and studies took up to six months (Tacconelli et al., 2022). The authors' conclusion is direct: federated research "requires the data owners to be equipped with the infrastructure and expertise to maintain the local nodes."

No study we found puts a price on fragmentation itself, and none links it to a specific lost trial. The cost is visible, though, in what hospitals already spend to work around it. Supplying a single prepared dataset for a data permit cost Helsinki University Hospital a median of EUR 4,700 (Findata, 2026). A French working group estimated that a hospital warehouse costs about EUR 3 million a year to run (Marchand-Arvier, 2023). Hospitals are paying for research data work either way. The question is whether they pay for it once, in a standing function, or again for every study.

4. Between hospitals, research networks exist, but many member hospitals are not ready

Europe has built large research networks. Few of them measure how ready their member hospitals are, and many hospitals are not in them at all.

The second level of the problem is sharing between hospitals. Here Europe has invested heavily, and the networks are real. The European Health Data and Evidence Network (EHDEN) has 210 harmonised data sources in 30 countries, whose partners keep their data locally and share only aggregate results (Blacketer et al., 2025). Its largest study drew on more than 50 data partners. In one exercise, drug safety signals were assessed across ten data partners in the UK, Spain, Norway, the Netherlands, and Serbia, with 15 to 60 minutes of execution per signal; patient-level data stayed at each site and only aggregates were shared (Gauffin et al., 2023). DARWIN EU, the regulator's data network, had 40 data partners across 18 countries by June 2026 (EMA, 2026). The 24 European Reference Networks for rare diseases link 1,606 expert centres in 375 hospitals across every EU member state and Norway (Graessner et al., 2026). The European Federation for Cancer Images connected 83 imaging datasets on about 107,000 subjects by September 2025 (European Commission, 2025b).

These networks show that federated research works. A shared registry for rare kidney disease enrolled 7,607 patients from 21 countries in 24 months once it existed, while each centre kept access to its own data and only aggregated or anonymised data left (Bassanese et al., 2021). French cancer centres built Consore, a federated network across 11 centres covering about 1.3 million patients, precisely to avoid a central warehouse (Guérin et al., 2024).

Every network depends on its member hospitals being ready

What the networks cannot do is make their members ready. EHDEN scored applicants before funding them (Blacketer et al., 2025). The cancer imaging federation assessed 82 datasets, including datasets from major academic hospitals: 30 met no tier at all, 48 reached the first tier, four the second, and none the third (EUCAIM, 2025). Making registry data FAIR is "not yet systematically implemented across all ERNs" (Graessner et al., 2026). In the 2020 German federated study, 20 university hospitals installed the shared software, and 18 obtained ethics approval to take part (Kapsner et al., 2021). Installing the software was quick; waiting for each hospital's own approval took longer.

Access between countries is slower still. In the EU's own pilot for cross-border access, HealthData@EU, approval and access took between one and twelve months per node, and the authors found that the federated design had not sped up approval (Schneider et al., 2025). In the pilot's COVID-19 use case, three of five nodes reached the data, and authorisation in Flanders was delayed by 11 months without explanation (Vande Catsyne et al., 2025). The pilot drew mainly on use cases that relied on infrastructure already in place (Svingel et al., 2025). An EU joint action found access procedures across member states diverse, lengthy, and sometimes nontransparent (TEHDAS, 2023b).

Who is missing

The map of who takes part is uneven. Italy, Great Britain, and Spain together supply more than a third of EHDEN's sources (Blacketer et al., 2025). Central and Eastern European hospitals do appear: Semmelweis University and Pécs in Hungary, four Croatian hospitals including University Hospital Centre Zagreb, the Clinical Centre of Serbia, and a hospital in Travnik, Bosnia and Herzegovina, are EHDEN data custodians, and so are general hospitals in Kavala, Greece, and Santarém, Portugal. But EHDEN lists no data source at all in Poland, Slovakia, Slovenia, Latvia, or Lithuania. Estonian authors note that data from Eastern Europe are "much less represented" in the common data model community (Oja et al., 2023). No network we found publishes how many of Europe's hospitals are connected, or how ready its members are.

A hospital that has not organised its own data joins these networks late and waits longest for access, whatever its research strengths. A hospital that has organised them can join any network that suits its research, on its own terms.

5. Patients will share their data, on conditions

Europeans are broadly willing to let their health data be used for research. What they ask in return is to know, to have a say, and to trust who holds the data.

Any argument for doing more with patient data has to start with the patients. The evidence is encouraging and conditional at once.

Willingness is high, especially among people who are ill. In a survey of 838 cancer patients in Baden-Württemberg, 96.7% expressed general willingness to make their clinical data available for research, and only 5.8% opposed use by researchers at universities or university hospitals (Köngeter et al., 2022). A review of 116 studies with more than 228,000 participants found people readier to share data for their own care than for secondary use, with willingness in studies of biobank data ranging from 10% to 98% (Cascini et al., 2024).

Who holds and uses the data matters a great deal. In a discrete choice study of 5,015 people across 12 European countries, being informed about data sharing and asked for consent mattered most, and a technology company collecting the data was among the least acceptable options; in every group of respondents, people preferred their own health care provider to a technology company (Biasiotto et al., 2023). A survey of 1,004 German adults found about 43% willing to share their health data with public research institutions and about 29% with private research institutions (Kühnel and Wilke, 2025). In Belgium, almost three in four people said they would share data for scientific research (King Baudouin Foundation, 2021, cited in Saelaert et al., 2023).

What people want in exchange is consistent across studies. They want to be told when their data are put to a new use, when results come back, and if confidentiality is breached (Cumyn et al., 2023). One study in that review found people worried more about who has the right to access their data than about the mechanisms used to protect it. Asked who should decide on access, the largest group of German cancer patients chose a committee of experts in which patients' views are represented (Köngeter et al., 2022).

England shows what happens when these conditions are ignored. Its national care.data programme failed, in the words of a later review, "as a result of poor communications, weak governance processes, and an unclear opt-out mechanism" (Mathers et al., 2017). When a second national scheme was announced in 2021, 1,339,862 people opted out in a single month, raising the opt-out rate from 2.77% to 4.97%, and the effect on research, the authors note, is not quantifiable (Tazare et al., 2024).

Patients already trust their hospital and its doctors more than most other users of their data. That trust is an asset a hospital can lose if its data leave its control without explanation, and one it can strengthen if it governs research use visibly: with a named committee, patients at the table, a public register of projects, and a clear route for withdrawal. University College London Hospitals has done this with a Data Trust Committee of patients and staff that oversees research access (UCLH Biomedical Research Centre, 2023). The research data hub is where those commitments can be kept.

Almost all of this evidence comes from Western and Northern Europe; we found no survey of attitudes to health data research in any Central or Eastern European country. And willingness is not uniform: in the 12-country study, three of the four groups of respondents preferred not to share by default, though for some the conditions offered could change that (Biasiotto et al., 2023).

6. Research beyond the commercial agenda

Industry-sponsored trials are essential to European hospitals, and Europe is losing them. But a hospital's research base cannot rest on commercial interest alone.

Pharmaceutical and biotechnology companies sponsor about 60% of the clinical trials registered in the EU, and academia about 40% (EMA, 2023). For hospitals, industry trials bring patients early access to new treatments, bring clinicians into international research, and bring income. Europe's position is weakening. The European Economic Area's share of global commercial trials fell from 22% in 2013 to 18% in 2018 and 12% in 2023, and about 60,000 fewer patients took part in trials involving an EEA country (EFPIA, 2024). The figures do not say why, and many causes are at work. But the one study we found that asked sponsors directly, a small interview study in the Nordic countries, suggests that sponsors favour sites that can back their patient counts with health record data (Laaksonen et al., 2021). A hospital that can answer a feasibility question in days from its own data is a hospital a sponsor can plan around.

Commercial interest, though, follows products. Of 577 randomised trials registered in 2006 by the 15 largest industry sponsors, only 18 compared products from different companies (Lathyris et al., 2010). The questions that matter most to patients and health systems are often different ones: which of two existing treatments works better, how long to treat, at what dose, for which patients. European clinical researchers note that academic treatment optimisation studies "frequently address questions for which commercial incentives are limited" (Berdel et al., 2026). There are between 6,000 and 8,000 rare diseases, a disease being rare in the EU when it affects fewer than one in 2,000 people, and research on them depends on registries to gather enough patients to study (Kölker et al., 2022).

Academic research cannot simply take up the slack when industry leaves. In Denmark between 1993 and 2006, academic and commercial drug trials declined together (Berendt et al., 2008). Both need the same things from a hospital: patients it can find in its own records, and data it can prepare and release under its own rules. A hospital that builds its research data hub serves both. It makes itself a better site for industry, and it makes possible the research no company will fund: comparing treatments already on the market, following patients for years, testing whether a finding in one hospital holds in others.

A hospital exists to improve the care of its patients, and a research base built in the patients' clinical interest belongs to that purpose as much as the trials that sponsors bring. A hospital with one can set its own research agenda; a hospital without one mostly runs studies that others design.

7. Why now: the law sets a clock, and AI raises the stakes

The European Health Data Space turns research data readiness from an aspiration into a duty, with dates. AI makes prepared hospital data more valuable every year.

The EHDS timetable

The European Health Data Space Regulation entered into force in 2025. Its chapter on secondary use, which governs research use of health data, applies in general from 26 March 2029 (Regulation (EU) 2025/327, Article 105). From then, national health data access bodies will decide applications for data and issue permits, and must do so within three months of a complete application, extendable by three more (Articles 57 and 68).

For hospitals, the duties fall on them as data holders. A hospital holding requested data must supply it to the access body within three months of the request, with a possible extension in justified cases (Article 60). It must describe its datasets for the national catalogue and check those descriptions at least once a year. Datasets built with Union or national public funding must carry a data quality and utility label, which covers documentation, technical quality, coverage, the time needed to access the data, and how the dataset has been linked or modified (Article 78). The Commission must set the label's specifications by March 2027. Some parts of the chapter apply from 2027, and some categories of data only from 2031 (Article 105).

The regulation also recognises that preparing data costs money. Fees charged to data users may cover the cost of compiling and preparing data, and the part of the fee for the holder's work is paid to the holder (Article 62). Practice today varies widely: in an EU joint action survey, Slovak respondents reported providing data free of charge while Swedish respondents recovered their full costs (TEHDAS2, 2026). Helsinki University Hospital's median cost of supplying one dataset, EUR 4,700, gives a sense of the work involved (Findata, 2026). A hospital that has organised its data will spend less to meet each request, and can recover more of what it spends.

Hospital bodies have noticed. Germany's hospital federation has warned that hospitals, as data holders, must expect considerable effort to provide high-quality, coded data for research and AI, and must be compensated for it (Deutsche Krankenhausgesellschaft, 2025). Critics writing while the regulation was being negotiated warned that it would need "a widespread, technically complex, and massively costly digital infrastructure" (Marelli et al., 2023). The access bodies themselves expect hiring enough technical and legal staff to be a central challenge (Svingel et al., 2025). The EU's own pilot drew mainly on infrastructure that already existed. The system will work first, and best, for hospitals that have prepared.

AI raises the value of prepared data

AI changes the stakes. Hospitals across Europe are signing data partnerships with AI developers; AP-HP renewed its partnership with one for five years in 2024 (AP-HP, 2024). Europe's cancer imaging federation is designed to train AI at hospital data warehouses without moving the data (EIBIR, 2023). The EU's AI Act already requires organisations deploying AI to ensure their staff are AI literate, and its obligations for high-risk systems follow over the coming years (Regulation (EU) 2024/1689).

For a hospital, the question is whether it takes part in this on its own terms. A hospital with prepared, documented data can validate the AI tools it buys, as Leiden did when it retrained an imported model on its own patients. It can contribute to federated training without handing over records, and it can negotiate partnerships knowing what its data are worth. A hospital without prepared data can do none of these. It will buy tools it cannot test and sign agreements it cannot fully assess.

8. How Europe is responding

Several countries have begun to fund hospital research data infrastructure. France has gone furthest. None has yet solved who pays to keep it running, and much of Europe has not started.

France: paying for hospital data as infrastructure

France is the clearest case of a government treating hospital research data as national infrastructure. In March 2022 it announced a EUR 50 million call to build health data warehouses in hospitals and link them into a national network, later raised to EUR 75 million (Ministère des Solidarités et de la Santé, 2022; Bpifrance, 2023). By the end of 2023, 16 projects with 54 partners were funded to build a network of 62 warehouses (Ministère de la Santé, 2023). By September 2026 the programme covered 31 university hospitals and more than 50 other establishments, public investment since 2023 had passed EUR 100 million, and the ministry described the warehouses as vehicles for the secure sharing of hospital health data and as a way for hospitals to prepare for the EHDS (Ministère de la Santé, 2026).

France's experience also shows what remains unsolved. A working group drawing on the experience of seven university hospital warehouses estimated that a warehouse costs about EUR 2 million a year to develop over five years and EUR 3 million a year to run (Marchand-Arvier, 2023). The 2026 call to widen the networks has EUR 2.325 million in total, paid once, and applicants must fund half their project themselves (Ministère de la Santé, 2026). The ministry says it is working on permanent funding from 2027. The national health authority found warehouse building very uneven and costly, and not within reach of every hospital (HAS, 2022). Marc Cuggia, professor of medical informatics at CHU de Rennes, put the question plainly in a press interview: the funding launched the projects, but "what about the landing?" (Beurnez, 2026).

Germany, Switzerland, and the Netherlands: strong at the top

Germany has committed more than EUR 400 million to the Medical Informatics Initiative since 2018, which every university hospital has joined, and about EUR 390 million over two periods to the Network University Medicine (Medical Informatics Initiative, n.d.; BMFTR, 2025). Data integration centres now operate at 38 university and three non-university locations (Albashiti et al., 2024). Beyond the university hospitals, progress is slower. Pilots extending the infrastructure to other hospitals found that data could rarely be taken from primary systems and that long-term operation could not be guaranteed on current funding; as their authors note, many treatment pathways neither start nor end in a university hospital (Krefting et al., 2024). Authors at one private hospital group, which scaled its own research platform to 77 hospitals, observe that real-world data research in Germany has been limited to university hospitals (Bockhacker et al., 2025).

Switzerland funded its Personalized Health Network with CHF 68 million for 2017 to 2020 and CHF 67 million for 2021 to 2024, building infrastructure that holds consented data on more than 700,000 patients (SIB, 2025). Cantonal hospitals began joining in 2024 (SIB, 2024). Federal funding for 2025 to 2028 has fallen to CHF 20.7 million, for coordination only, and funding for research projects has ended. A 2025 governance study compared the roles, bodies, and processes that govern data sharing at the five Swiss university hospitals one by one (Egli et al., 2025).

The Netherlands has built a national model around its eight university medical centres, each acting as a regional hub, funded by EUR 47 million from the National Growth Fund on top of EUR 22 million (Rijksoverheid, 2023). Dutch university hospitals have called for sufficient structural funding to sustain it.

The Nordic countries and Estonia: national systems, hospital gaps

The Nordic countries have the longest experience of national health data. Sweden runs around one hundred national quality registries used in every hospital, funded jointly by the regions and the state with SEK 168 million in 2024 (Nationellt system för kunskapsstyrning, 2025). Finland created a national permit authority, Findata, and a network of hospital biobanks; 185,000 people have provided a sample to Helsinki University Hospital's biobank (HUS, 2025). These systems prove that national routes can work. They also show their limits. Data still sit in regional silos in Sweden and regional warehouses in Denmark (TEHDAS, 2022; 2023a). Access in Denmark takes one to six months on average. Finnish researchers report that the permit process has slowed their work, and from May 2026 Finland's amended law lets applicants apply to public data controllers directly as an alternative to Findata (Findata, 2026). Estonia's national OMOP conversion shows that even a fully digital system struggles with hospitals' local codes (Oja et al., 2023).

Southern Europe: networks without national programmes

Italy has funded a network of its research hospitals, the IRCCS, with local platforms and a central platform, with EUR 55 million over ten years from 2019; we could not confirm its current status (Alleanza Contro il Cancro, n.d.). Spanish, Portuguese, and Greek hospitals take part in European networks, including Hospital Universitario 12 de Octubre in Madrid, which works with both the OMOP and openEHR standards, and general hospitals in Santarém and Kavala. But we found no national programme in Spain, Portugal, or Greece that funds hospital research data infrastructure, and the evidence on Southern European hospitals is thin.

Central and Eastern Europe: the widest gap

In Central and Eastern Europe the gap is widest. We searched for hospital-level research data programmes in Slovakia, Czechia, Poland, Slovenia, Romania, Hungary, Croatia, and Estonia and found none. Slovakia is building parts of its health data access body with an EU grant of EUR 972,587.20, with no stated role for hospitals (NCZI, 2025), and its audit office found seven of eleven national health registers still only partly functional at the end of 2025 (NKÚ, 2026). Individual hospitals have connected to European networks on their own initiative in Hungary, Croatia, and Serbia, and the Children's Memorial Health Institute in Warsaw is a DARWIN EU data source. The European clinical research infrastructure network is working to bring more Polish centres into multinational trials (ECRIN, 2026). Countries in the region win a growing share of EU research participation, but a smaller share of the money (European Commission, 2024).

For a hospital director in Warsaw, Bratislava, or Bucharest, no national programme is likely to build this before 2029. The EHDS duties will apply all the same. The hospitals in the region that are already in European networks got there by organising their own data.

What the comparison shows

Every national programme began with university hospitals, and reaching other hospitals is at pilot stage or has not begun. Governments fund the building and hesitate over the running; France, Germany, Switzerland, and the Netherlands all face the question of permanent funding. And national infrastructure, however good, leaves the work inside the hospital where it was. Every federated network, every national catalogue, and every access body depends on hospitals that can describe, prepare, and supply their own data. National systems work only as well as the hospitals that supply them.

9. What good looks like, and what it does not

Hospitals that have organised their research data share six features. The efforts that stalled share common failures.

Good practice

The hospitals that have made progress did not all follow one model, but they share recognisable features.

One accountable function. Nantes runs its warehouse through a validation board that meets weekly and named roles across research administration, data stewardship, IT, and data protection (Karakachoff et al., 2024). Halle runs its data integration centre as a joint core facility of hospital and faculty (SMITH, 2025). Each has someone whose job is the hospital's research data as a whole.

Data linked across domains. The Princess Máxima Center brings records, biobank data, diagnostics, and trial data together in one shared registry (Health-RI, 2022). At Helsinki, a biobank integrated with the hospital could return findings to patients' care, and 99% of donors had agreed to be contacted (Pehrsson et al., 2026).

Data that work for care as well as research. Lausanne's warehouse feeds a daily sepsis registry that changed practice on the wards (Despraz et al., 2026). Lille's anaesthesia warehouse supports ten clinical dashboards alongside its research (Lamer et al., 2022). A hub that clinicians use every day is easier to fund than one used only for research.

Federation by default. Consore, the rare kidney disease registry, and the German federated studies all kept data at each hospital and moved only queries and aggregate results (Guérin et al., 2024; Bassanese et al., 2021; Kapsner et al., 2021). Each centre kept its own data and could still take part.

Quality treated as continuous work. Estonia's team describes its conversion as "a continuous and iterative process" rather than a one-time project (Oja et al., 2023). Biobank accreditation to ISO 20387 raised metadata completeness to 99% to 100% in one study (Zucca et al., 2026).

Clear, visible governance. Regensburg runs one use and access committee for all requests, including biobank samples, under written rules (Universitätsklinikum Regensburg, 2025). UCLH goes further and puts patients on its Data Trust Committee. In both, a researcher, a sponsor, or a patient can see who decides and on what basis.

Failure modes

The efforts that stall share features too.

A warehouse nobody can see into. Only eight of 14 French production warehouses had a public portal listing their studies, and some listed a dozen while their teams cited several hundred (Doutreligne et al., 2023). A patient has no way to know whether their data are used.

Registries multiplied instead of shared. A network whose centres feed more than 60 registries, or a registry that depends on one person in each of 27 centres, still has fragmented data, now spread across more systems (Bassanese et al., 2021).

Samples without their data. A biobank whose samples wait a median of 926 days for first use, and whose linked clinical data sit with the collecting clinician, holds material that researchers find hard to use (Mucher et al., 2026).

Consent on paper. Münster's one-in-eight documentation rate and the German interview evidence show that paper consent cannot keep up once patient numbers grow.

Preparation that adds errors. In Tübingen, the conversion created almost all of the dataset's quality problems (Tahar et al., 2026). Preparation has to be checked as carefully as the source.

Dependence without control. Private companies offered implementation services in 15 of 21 French warehouse cases (Doutreligne et al., 2023), and a French warehouse data scientist described warehouses as fully dependent on hospital information systems (Priou et al., 2023). Vendors bring skills that hospitals often lack, yet a hospital that does not hold its own data model, documentation, and access rules cannot change supplier, and cannot answer the access body without asking someone else first.

Build money without running money. France, Germany, and Switzerland all show programmes funded to build and uncertain how to run. A hub funded only as a project tends to stop when the project money stops.

10. How ready is your hospital?

Readiness can be measured. We built a framework to help a hospital see where it stands and what to do next.

National digital health indices measure whether citizens can see their records online or how digital a hospital's care processes are. We found no European instrument that measures whether a single hospital's research data are ready for use. So we built one: the Hospital Research Data Readiness Framework. It has not yet been validated in hospitals, and we publish it so that others can test and improve it.

The framework scores a hospital on six dimensions, each tied to a principle of good data stewardship and to the duties of the EHDS. Each dimension has five levels, from 0, where no one is responsible for research data, to 4, where the hospital can answer outside questions without its data leaving its control. Levels are defined by what a hospital can show, not by how it describes itself.

The result is a profile, not an average, because a single weak dimension can stop research on its own: a well-integrated warehouse is of little use if nobody can say whose data may be used, or who decides. A hospital is research-ready at level 3 or higher on all six dimensions. It is connected, ready to act as a node in national and European networks, when it also reaches level 4 on custody and description. We suggest an interim target of level 2 on all six by the end of 2027.

The profile also tells a hospital what its data can support today. Research depends mainly on custody, description, and integrity. Trial feasibility and recruitment depend on those and on patients. Finding what routine records miss, and using AI safely, depend on integrity and on governance of AI. Sharing between sites needs level 4 on custody, description, and patients.

The 20-question self-assessment in Annex A lets a hospital place itself on each dimension in a single working session, with the right people in the room.

11. Objections, and the limits of our evidence

The strongest arguments against a hospital research data hub, and the places where our evidence runs out.

"National platforms will do this." They will not, and they are not designed to. A national access body decides who may use data and provides a secure environment to analyse it. It does not know how a hospital's data were produced, and it cannot fix a hospital's coding, linkage, or consent records. The EHDS places the duty to describe and supply data on the holder. Sweden's national quality registries and Finland's Findata are the strongest evidence that national routes work, and we take them seriously. But even there, data sit in regional silos, and Finnish researchers report that the national permit route has slowed them (Sanmark et al., 2025). National platforms depend on hospitals that are ready.

"Buy it from a vendor." Many hospitals will, and should, buy software and services. Private companies offered implementation services to most French warehouses. A bought-in tool still depends on the hospital's own data, people, and rules, and a hospital that does not hold its own data model, documentation, and access decisions cannot change supplier or answer an access body without its vendor. Every vendor should be asked what the hospital keeps if the contract ends.

"Hospitals cannot afford it." This is the strongest objection, and we do not dismiss it. A French working group estimated that a university hospital warehouse costs about EUR 3 million a year to run (Marchand-Arvier, 2023), beyond the reach of most hospitals. Three things change the picture. A hub does not need to start as a full warehouse: one accountable lead, one access committee, one catalogue, and one well-prepared dataset are within reach of any research hospital, and they are what levels 2 and 3 ask for. Hospitals also need not build alone: one German hospital group carried a shared research platform across 77 hospitals (Bockhacker et al., 2025), and regional or group models can spread the cost. And the EHDS lets holders recover the cost of preparing data through fees. None of this makes the money appear. Hospitals are already paying for research data work, study by study, and that a standing function costs less over time than repeated one-off effort.

"Patients will not accept it." The evidence suggests the opposite, on conditions. Most patients, asked by their own hospital, agree to research use, and they trust their hospital more than most other users of their data (Köngeter et al., 2022; Biasiotto et al., 2023). What they reject is opaque use, unclear opt-outs, and data passed to companies without explanation, as England learned. A hospital hub that governs research use visibly answers those concerns.

"There is no proof it wins trials or improves care." That is fair. We found no study showing that a hospital research data hub causes more trials to be placed or better outcomes for patients. The best evidence of benefit, from Lausanne and Helsinki, comes from single hospitals without controlled comparison, and the sponsor evidence is one small interview study. We think the direction of the evidence is clear, and that waiting for proof while the EHDS deadlines approach is the riskier course. Readers should weigh the evidence as it is.

"The author sells the remedy." ZINOS offers hospitals a research data hub, and the author leads ZINOS. Readers should know that, and we state it in full at the end of this paper. The framework in section 10 and the self-assessment in Annex A are free to use, and work whatever system a hospital runs or buys. Nothing in our argument depends on choosing ZINOS.

Limits of the evidence

The published evidence leans heavily towards university hospitals in France, Germany, Switzerland, the Netherlands, and the Nordic countries. Southern Europe appears mainly through network membership, and Central and Eastern Europe almost entirely so; we found no patient attitude survey and no hospital-level case from the region. We found no costed example of research data capacity in a mid-sized or non-university hospital, no measured time from request to delivery at any hospital, and no study that counts how many separate research data systems a typical hospital runs. Many of the commentaries on the EHDS were written about the 2022 proposal, not the adopted regulation. And our framework is untested in practice. We hope this paper prompts the studies that would fill these gaps.

12. What a hospital should do

Start with one conversation and one dataset. Build the function step by step, and be ready by March 2029.

In the next 90 days. Bring together research leadership, IT, the data protection officer, and the leads of the biobank and the main registries, and complete the self-assessment in Annex A. Agree one person who is accountable for the hospital's research data as a whole. Make a first list of every research data source the hospital holds, with its owner and legal basis. Most hospitals will find themselves at level 1 or 2 on most dimensions. That is the normal starting point, and knowing it is the first step.

During 2027. Aim for level 2 on all six dimensions: a register of sources, an internal catalogue, a log of extractions, a central consent record, a named team and budget line, and one use and access committee with written rules for every source. Choose one dataset that matters to your researchers, such as a disease registry linked to records and samples, and bring it to level 3: mapped to a common data model, checked for quality, with every link documented. Use it to answer a real question. Watch for the Commission's specifications for the data quality label, due by March 2027.

By March 2029. If your hospital is a data holder under the EHDS, be ready to describe your datasets in the national catalogue to the required standard, check those descriptions each year, and supply data within three months of a request. For datasets built with public money, be ready to label them. Aim for level 3 on all six dimensions as the floor for the research the hospital runs itself, and for the connected profile, level 4 on custody and description, if you intend to take part in national and European networks.

Throughout. Talk to the hospitals around you. Shared platforms, regional nodes, and group models can carry much of the infrastructure, as long as each hospital keeps governance of its own data. And tell your patients what you are doing, because every other step depends on their trust.

About this paper

About the author. Daren Wilson is Chief Executive Officer of ZINOS and Co-Founder and Senior Partner of HealthSeed AG, a Swiss healthcare venture studio. He leads ZINOS's work with hospitals that want to hold and govern their research data and form multi-site research coalitions. Contact: daren.wilson@zinos.com.

Competing interests. The author is Chief Executive Officer of ZINOS, which offers hospitals a commercial research data hub and developed the Hospital Research Data Readiness Framework described in section 10. No ZINOS product was assessed in the evidence for this paper.

How this paper was prepared. This is a structured evidence review, not a systematic review. It draws on peer-reviewed studies, government and regulatory documents, the published records of European research networks, and the text of Regulation (EU) 2025/327, gathered between October and November 2026 through ten structured research modules, searches of the biomedical literature, and direct reading of primary sources. Every figure is checked against its source before publication. The full methods note, evidence annex, and references follow.

Suggested citation. Wilson D. The Hospital's Missing Layer: Why Europe's hospitals need research data hubs of their own. ZINOS White Paper 02. November 2026. zinos.com

Licence. © 2026 ZINOS. Licensed under CC BY 4.0.

Sources

Sources

1. Acteurs Publics (2026). Les hôpitaux, pionniers des données de santé, encore peu reconnus (in French). News article, 27 August 2026. https://acteurspublics.fr/articles/les-hopitaux-pionniers-des-donnees-de-sante-encore-peu-reconnus/ (accessed 6 November 2026)
2. Albashiti F, Thasler R, Wendt T, et al. (2024). Data integration centers: from a concept in the Medical Informatics Initiative to its local implementation in the Network of University Medicine (in German). Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 67(6): 629 to 636. https://doi.org/10.1007/s00103-024-03879-5 (accessed 2 November 2026)
3. Alleanza Contro il Cancro (n.d.). Health Big Data. Web page, undated; project start 16 December 2019. https://www.alleanzacontroilcancro.it/en/projects/health-big-data/ (accessed 2 November 2026)
4. Assistance Publique-Hôpitaux de Paris (AP-HP) (2024). Accélérer l'innovation IA au service des patients (in French). Web page, published 30 September 2024. https://www.aphp.fr/actualites/accelerer-linnovation-ia-au-service-des-patients (accessed 6 November 2026)
5. Assistance Publique-Hôpitaux de Paris (AP-HP) (2025). L'entrepôt de données de santé, un outil innovant majeur pour l'AP-HP (in French). Web page, last updated 2 June 2025. https://www.aphp.fr/actualites/lentrepot-de-donnees-de-sante-un-outil-innovant-majeur-pour-lap-hp (accessed 2 November 2026)
6. Bassanese G, Wlodkowski T, Servais A, et al. (2021). The European Rare Kidney Disease Registry (ERKReg): objectives, design and initial results. Orphanet Journal of Rare Diseases 16(1): 251. https://doi.org/10.1186/s13023-021-01872-8 (accessed 6 November 2026)
7. BBMRI-ERIC (2026). BBMRI-ERIC Directory, live GraphQL query run on 25 October 2026. https://directory.bbmri-eric.eu/ERIC/graphql
8. Berdel WE, Dreyling M, Schrappe M, et al. (2026). Academic treatment optimization studies in oncology and internal medicine: a risk- and burden-adapted regulatory framework for the European Union. Lancet Regional Health Europe 71: 101888. https://doi.org/10.1016/j.lanepe.2026.101888 (accessed 2 November 2026)
9. Berendt L, Håkansson C, Bach KF, et al. (2008). Effect of European Clinical Trials Directive on academic drug trials in Denmark: retrospective study of applications to the Danish Medicines Agency 1993 to 2006. BMJ 336(7634): 33 to 35. https://doi.org/10.1136/bmj.39401.470648.BE (accessed 2 November 2026)
10. Biasiotto R, Viberg Johansson J, Alemu MB, et al. (2023). Public preferences for digital health data sharing: discrete choice experiment study in 12 European countries. Journal of Medical Internet Research 25: e47066. https://doi.org/10.2196/47066 (accessed 6 November 2026)
11. Blacketer C, Schuemie MJ, Moinat M, et al. (2025). Advancing real-world evidence through a federated health data network (EHDEN): descriptive study. Journal of Medical Internet Research 27: e74119. https://doi.org/10.2196/74119 (accessed 2 November 2026)
12. Bockhacker M, Martens P, von Münchow C, et al. (2025). Lessons learned from building a data platform for longitudinal, analytical use cases and scaling to 77 German hospitals: implementation report. JMIR Medical Informatics 13: e69853. https://doi.org/10.2196/69853 (accessed 2 November 2026)
13. Bpifrance (2023). France 2030 : six premiers lauréats pour constituer et consolider des entrepôts de données de santé hospitaliers (in French). Press release, 3 April 2023. https://presse.bpifrance.fr/france-2030-six-premiers-laureats-pour-constituer-et-consolider-des- entrepots-de-donnees-de-sante-hospitaliers/?lang=fra (accessed 2 November 2026)
14. Bundesministerium für Forschung, Technologie und Raumfahrt (BMFTR) (2025). Cem Özdemir: Erfolgsgeschichte des Netzwerks Universitätsmedizin kann fortgeschrieben werden (in German). Press release 28/2025, 7 April 2025. https://www.bmftr.bund.de/SharedDocs/Pressemitteilungen/DE/2025/04/070425-NUM.html (accessed 6 November 2026)
15. Cascini F, Pantovic A, Al-Ajlouni YA, et al. (2024). Health data sharing attitudes towards primary and secondary use of data: a systematic review. eClinicalMedicine 71: 102551. https://doi.org/10.1016/j.eclinm.2024.102551 (accessed 6 November 2026)
16. Cumyn A, Ménard JF, Barton A, et al. (2023). Patients' and members of the public's wishes regarding transparency in the context of secondary use of health data: scoping review. Journal of Medical Internet Research 25: e45002. https://doi.org/10.2196/45002 (accessed 6 November 2026)
17. de Hond AAH, Kant IMJ, Fornasa M, et al. (2023). Predicting readmission or death after discharge from the ICU: external validation and retraining of a machine learning model. Critical Care Medicine 51(2): 291 to 300. https://doi.org/10.1097/CCM.0000000000005758 (accessed 25 October 2026)
18. Declerck J, Eklund N, Sáez C, et al. (2026). Health data quality skill gaps and training needs among European health data stakeholders: cross-sectional survey. Journal of Medical Internet Research 28: e86878. https://doi.org/10.2196/86878 (accessed 2 November 2026)
19. Despraz J, Matusiak R, Nektarijevic S, et al. (2026). An artificial intelligence-powered learning health system to improve sepsis detection and quality of care: a before-and-after study. npj Digital Medicine 9(1): 106. https://doi.org/10.1038/s41746-025-02180-2 (accessed 25 October 2026)
20. Deutsche Krankenhausgesellschaft (DKG) (2025). Thesen und Forderungen der Deutschen Krankenhausgesellschaft: Nutzung von Künstlicher Intelligenz (KI) im Krankenhaus (in German). Position paper, September 2025. https://www.dkgev.de/fileadmin/Mediapool/1_DKG/1.3_Politik/Positionen/2025-10-23_DKG-Positionspapier_KI_im_Krankenhaus.pdf (accessed 2 November 2026)
21. DigitalRadar Krankenhaus (2026). Pressemitteilung zu den ersten Ergebnissen der dritten Erhebung (in German). Press release, 17 June 2026. https://www.digitalradar-krankenhaus.de/pressemitteilung-zu-den-ersten-ergebnissen-der-dritten-erhebung/ (accessed 25 October 2026)
22. Doutreligne M, Degremont A, Jachiet PA, et al. (2023). Good practices for clinical data warehouse implementation: a case study in France. PLOS Digital Health 2(7): e0000298. https://doi.org/10.1371/journal.pdig.0000298 (accessed 2 November 2026)
23. ECRIN, European Clinical Research Infrastructure Network (2026). ECRIN visits Agencja Badań Medycznych. 24 March 2026. https://ecrin.org/news/ecrin-visits-agencja-badan-medycznych
24. EUCAIM (European Federation for Cancer Images) (2025). D2.5: The FAIR Report: Data Maturity Model Adoption on the EUCAIM data federation. Project deliverable, Grant Agreement 101100633, delivered 23 December 2025. https://cancerimage.eu/wp- content/uploads/2026/05/D2.5_The-FAIR-Report_-Data-Maturity-Model-Adoption-on-the-EUCAIM-data-federation.pdf (accessed 25 October 2026)
25. European Commission (2024). Country participation in the EU R&I framework programmes. Publications Office of the European Union. https://op.europa.eu/en/publication-detail/-/publication/8925c85b-90f0-11ef-a130-01aa75ed71a1/language-en (accessed 6 November 2026)
26. European Commission, DG CONNECT (2025a). European Cancer Imaging Initiative. Web page, last updated 10 November 2025. https://digital-strategy.ec.europa.eu/en/policies/cancer-imaging (accessed 2 November 2026)
27. European Commission, DG SANTE (2025b). Update on policy development at the EU level. Presentation by E. Szepessy to the ACT EU Multi- stakeholder Platform Advisory Group, 18 September 2025. https://accelerating-clinical-trials.europa.eu/system/files/2025- 09/Update%20on%20policy%20development%20at%20the%20EU%20level%20-%20E.Szepessy%20%28EC%29.pdf (accessed 2 November 2026)
28. European Commission, DG CONNECT (2026). The 2026 Digital Decade eHealth Indicator Study. 17 June 2026. https://digital- strategy.ec.europa.eu/en/library/2026-digital-decade-ehealth-indicator-study (accessed 25 October 2026)
29. European Federation of Pharmaceutical Industries and Associations (EFPIA) (2024). 60,000 fewer clinical trial places for Europeans, despite global surge in research projects. Press release, 22 October 2024. https://www.efpia.eu/news-events/the-efpia- view/statements-press-releases/60-000-fewer-clinical-trial-places-for-europeans-despite-global-surge-in-research-projects/ (accessed 2 November 2026)
30. EUCAIM consortium, European Institute for Biomedical Imaging Research (EIBIR) (2023). Radiology leads flagship of Europe's Beating Cancer Plan: kick-off for federated European infrastructure for cancer images [press release]. Brussels, 24 January 2023. https://www.eibir.org/projects/eucaim/pressreleases/radiology-leads-flagship-of-europes-beating-cancer-plan-kick-off-for-federated- european-infrastructure-for-cancer-images/
31. European Medicines Agency (EMA) (2023). Accelerating Clinical Trials in the EU (ACT EU): priority areas for possible EU-level funding to support the transformation of the European clinical trials environment. EMA/75374/2023, 15 November 2023. https://accelerating- clinical-trials.europa.eu/system/files/2023-11/ACT%20EU%20priority%20areas%20for%20possible%20EU-level%20funding.pdf (accessed 2 November 2026)
32. European Medicines Agency (EMA) (2026). DARWIN EU data partners onboarded in year I, II, III, IV and V. Document, 16 June 2026. https://www.ema.europa.eu/en/documents/other/darwin-eu-data-partners-onboarded-year-i-ii-iii-iv-v_en.pdf (accessed 6 November 2026)
33. European Parliament and Council of the European Union (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L, 12 July 2024. https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng (accessed 6 November 2026)
34. European Parliament and Council of the European Union (2025). Regulation (EU) 2025/327 of the European Parliament and of the Council of 11 February 2025 on the European Health Data Space and amending Directive 2011/24/EU and Regulation (EU) 2024/2847. Official Journal of the European Union, 5 March 2025. https://eur-lex.europa.eu/eli/reg/2025/327/oj/eng (accessed 3 November 2026)
35. Findata, Finnish Social and Health Data Permit Authority (2026). Average cost estimates. Web page, updated 11 September 2026. https://findata.fi/en/pricing/average-cost-estimates/ (accessed 2 November 2026)
36. Gauffin O, Brand JS, Vidlin SH, et al. (2023). Supporting pharmacovigilance signal validation and prioritization with analyses of routinely collected health data: lessons learned from an EHDEN network study. Drug Safety 46(12): 1335 to 1352. https://doi.org/10.1007/s40264-023-01353-w (accessed 6 November 2026)
37. Gehrmann J, Herczog E, Decker S, et al. (2023). What prevents us from reusing medical real-world data in research. Scientific Data 10(1): 459. https://doi.org/10.1038/s41597-023-02361-2 (accessed 6 November 2026)
38. Government of the Netherlands (2023). Kabinet trekt definitief 397 miljoen euro uit voor vijf Nationaal Groeifondsprojecten (in Dutch). News release, Rijksoverheid, 24 February 2023. https://www.rijksoverheid.nl/actueel/nieuws/2023/02/24/kabinet-trekt-definitief-397- miljoen-euro-uit-voor-vijf-nationaal-groeifondsprojecten (accessed 2 November 2026)
39. Graessner H, Ripp S, Pereira AM, et al. (2026). European Reference Networks: a flagship activity of the EU in the field of rare and complex diseases: from 2017 to 2025. Orphanet Journal of Rare Diseases 21(1). https://doi.org/10.1186/s13023-026-04341-2 (accessed 25 October 2026)
40. Gruendner J, Deppenwiese N, Folz M, et al. (2022). The architecture of a feasibility query portal for distributed COVID-19 Fast Healthcare Interoperability Resources (FHIR) patient data repositories: design and implementation study. JMIR Medical Informatics 10(5): e36709. https://doi.org/10.2196/36709 (accessed 6 November 2026)
41. Guardiolle V, Bazoge A, Morin E, et al. (2022). Linking biomedical data warehouse records with the national mortality database in France: large-scale matching algorithm. JMIR Medical Informatics 10(11): e36711. https://doi.org/10.2196/36711 (accessed 2 November 2026)
42. Guérin J, Nahid A, Tassy L, et al. (2024). Consore: a powerful federated data mining tool driving a French research network to accelerate cancer research. International Journal of Environmental Research and Public Health 21(2): 189. https://doi.org/10.3390/ijerph21020189 (accessed 6 November 2026)
43. Hageman IC, van Rooij IALM, de Blaauw I, et al. (2023). A systematic overview of rare disease patient registries: challenges in design, quality management, and maintenance. Orphanet Journal of Rare Diseases 18(1): 106. https://doi.org/10.1186/s13023-023-02719-0 (accessed 6 November 2026)
44. Haute Autorité de santé (HAS) (2022). Entrepôts de données de santé hospitaliers : la HAS publie un panorama inédit en France (in French). News release, 17 November 2022. https://www.has-sante.fr/jcms/p_3386076/fr/entrepots-de-donnees-de-sante-hospitaliers-la-has- publie-un-panorama-inedit-en-france. Report: Entrepôts de données de santé hospitaliers en France (in French). https://www.has- sante.fr/upload/docs/application/pdf/2022-11/rapport_entrepots_donnes_sante_hospitaliers.pdf (accessed 2 November 2026)
45. Health-RI (2022). Building blocks of a health data infrastructure: together with and for data users and data holders. 8 November 2022. https://www.health-ri.nl/en/news/building-blocks-health-data-infrastructure-together-and-data-users-and-data-holders
46. Helsinki University Hospital (HUS) (2025). HUS year 2024: Networks and collaboration. Annual report web page, updated 4 April 2025. https://www.hus.fi/en/about-us/annual-reports/hus-year-2024/research-teaching-and-development/hus-year-2024-networks-and (accessed 6 November 2026)
47. Hoffmann JS, Lenz P, Duda A, et al. (2026). Evaluation of the broad consent distribution at a university hospital: is it representative or biased? Studies in Health Technology and Informatics 340: 268 to 273. https://doi.org/10.3233/SHTI261012 (accessed 2 November 2026)
48. Kapsner LA, Kampf MO, Seuchter SA, et al. (2021). Reduced rate of inpatient hospital admissions in 18 German university hospitals during the COVID-19 lockdown. Frontiers in Public Health 8: 594117. https://doi.org/10.3389/fpubh.2020.594117 (accessed 2 November 2026)
49. Karakachoff M, Goronflot T, Coudol S, et al. (2024). Implementing a biomedical data warehouse from blueprint to bedside in a regional French university hospital setting: unveiling processes, overcoming challenges, and extracting clinical insight. JMIR Medical Informatics 12: e50194. https://doi.org/10.2196/50194 (accessed 2 November 2026)
50. King Baudouin Foundation (2021). Take care of your data. Health data information brochure. Cited in Saelaert M, Mathieu L, Van Hoof W, et al. (2023), reference 15. https://media.kbs-frb.be/nl/media/8912/2022_PUB_3832_Zorg%20voor%20je%20data (accessed 6 November 2026)
51. Kondylakis H, Kalokyri V, Sfakianakis S, et al. (2023). Data infrastructures for AI in medical imaging: a report on the experiences of five EU projects. European Radiology Experimental 7(1): 20. https://doi.org/10.1186/s41747-023-00336-x (accessed 6 November 2026)
52. Kölker S, Gleich F, Mütze U, et al. (2022). Rare disease registries are key to evidence-based personalized medicine: highlighting the European experience. Frontiers in Endocrinology 13: 832063. https://doi.org/10.3389/fendo.2022.832063 (accessed 6 November 2026)
53. Köngeter A, Schickhardt C, Jungkunz M, et al. (2022). Patients' willingness to provide their clinical data for research purposes and acceptance of different consent models: findings from a representative survey of patients with cancer. Journal of Medical Internet Research 24(8): e37665. https://doi.org/10.2196/37665 (accessed 6 November 2026)
54. Krefting D, Bavendiek U, Fischer J, et al. (2024). Digital health progress hubs: data integration beyond university hospitals (in German). Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 67(6): 701 to 709. https://doi.org/10.1007/s00103-024-03883-9 (accessed 2 November 2026)
55. Kühnel E, Wilke F (2025). Health data sharing in Germany: individual preconditions, trust and motives. Frontiers in Public Health 13: 1538106. https://doi.org/10.3389/fpubh.2025.1538106 (accessed 6 November 2026)
56. Laaksonen N, Bengtström M, Axelin A, et al. (2021). Clinical trial site identification practices and the use of electronic health records in feasibility evaluations: an interview study in the Nordic countries. Clinical Trials 18(6): 724 to 731. https://doi.org/10.1177/17407745211038512 (accessed 2 November 2026)
57. Lamer A, Moussa MD, Marcilly R, et al. (2022). Development and usage of an anesthesia data warehouse: lessons learnt from a 10-year project. Journal of Clinical Monitoring and Computing 37(2): 461 to 472. https://doi.org/10.1007/s10877-022-00898-y (accessed 6 November 2026)
58. Lathyris DN, Patsopoulos NA, Salanti G, et al. (2010). Industry sponsorship and selection of comparators in randomized clinical trials. European Journal of Clinical Investigation 40(2): 172 to 182. https://doi.org/10.1111/j.1365-2362.2009.02240.x (accessed 2 November 2026)
59. Maassen O, Fritsch S, Palm J, et al. (2021). Future medical artificial intelligence application requirements and expectations of physicians in German university hospitals: web-based survey. Journal of Medical Internet Research 23(3): e26646. https://doi.org/10.2196/26646 (accessed 6 November 2026)
60. Marchand-Arvier J, Allassonnière S, Hoang A, et al. (2023). Fédérer les acteurs de l'écosystème pour libérer l'utilisation secondaire des données de santé, Tome 2 : Annexes (in French). Report, December 2023, vie-publique.fr. https://www.vie- publique.fr/files/rapports/fichiers_joints/295922-Annexes.pdf (accessed 2 November 2026)
61. Marelli L, Stevens M, Sharon T, et al. (2023). The European health data space: too big to succeed? Health Policy 135: 104861. https://doi.org/10.1016/j.healthpol.2023.104861 (accessed 6 November 2026)
62. Mathers N, Sullivan R, Dhillon A, et al. (2017). The use of NHS patient data: report by the National Data Guardian for Health and Care. British Journal of General Practice 67(655): 56 to 57. https://doi.org/10.3399/bjgp17X688933 (accessed 6 November 2026)
63. Medical Informatics Initiative (n.d.). About the initiative. Web page, undated. https://www.medizininformatik-initiative.de/en/about- initiative (accessed 2 November 2026)
64. Ministère de la Santé (2023). France 2030 : 19 projets numériques pour améliorer les prises en charge médicales (in French). Press release, Paris, 14 December 2023. https://www.info.gouv.fr/upload/media/content/0001/08/d49e3a7855924090645ea90dc158800019ac2f5c.pdf (accessed 2 November 2026)
65. Ministère de la Santé (2026). Note d'information n° DGRINES/2026/137 du 23 septembre 2026 relative au lancement d'un appel à projets concernant l'élargissement des réseaux d'entrepôts de données de santé hospitaliers (EDSH) (in French). NOR SFHI2624561N, Bulletin officiel Santé, 24 September 2026. https://bulletins-officiels.social.gouv.fr/sites/textes-officiels/files/2026-09/SFHI2624561N_0.pdf (accessed 2 November 2026)
66. Ministère des Solidarités et de la Santé and Secrétariat général pour l'investissement (France 2030) (2022). Joint press release, 17 March 2022, no printed headline: call for projects of EUR 50 million for hospital health data warehouses (in French). https://health-data- hub.fr/sites/default/files/2022-03/France2030_entrep%C3%B4t%20de%20donn%C3%A9es_VF.pdf (accessed 2 November 2026)
67. Mucher P, Bayer M, Flieder I, et al. (2026). Utilization patterns of hospital biobank samples: time to first use and material-specific demand. Journal of Clinical Medicine 15(11): 4292. https://doi.org/10.3390/jcm15114292 (accessed 6 November 2026)
68. Národné centrum zdravotníckych informácií (NCZI) (2025). Zriadenie častí orgánu pre prístup k zdravotným údajom v SR (in Slovak). Web page, last updated 17 November 2025. https://www.nczisk.sk/Medzinarodna-spolupraca/EHDS/Pages/Zriadenie-casti-organu-pre- pristup-k-zdravotnym-udajom-SR.aspx (accessed 2 November 2026)
69. Niemeyer A, Veit C, Semler SC, et al. (2026). From fragmentation to transparency: exploring and mapping the landscape of medical registries in Germany. Studies in Health Technology and Informatics 336: 1113 to 1117. Published 21 May 2026. https://doi.org/10.3233/SHTI260371 (accessed 6 November 2026)
70. Oja M, Tamm S, Mooses K, et al. (2023). Transforming Estonian health data to the Observational Medical Outcomes Partnership (OMOP) Common Data Model: lessons learned. JAMIA Open 6(4): ooad100. https://doi.org/10.1093/jamiaopen/ooad100 (accessed 25 October 2026)
71. Pehrsson M, Jakkula E, Ala-Kulju K, et al. (2026). Opportunistic screening of high-risk breast cancer variants in hospital biobank setting. Cancer Epidemiology, Biomarkers and Prevention 35(6): 883 to 893. https://doi.org/10.1158/1055-9965.EPI-25-1757 (accessed 25 October 2026)
72. Priou S, Lamé G, Jankovic M, Kempf E (2024). "In conferences, everyone goes 'health data is the future'": an interview study on challenges in re-using EHR data for research in Clinical Data Warehouses. AMIA Annual Symposium Proceedings 2023: 579 to 588. PMID 38222365; PMC10785853. https://pmc.ncbi.nlm.nih.gov/articles/PMC10785853/ (accessed 6 November 2026)
73. Queralt-Rosinach N, Kaliyaperumal R, Bernabé CH, et al. (2022). Applying the FAIR principles to data in a hospital: challenges and opportunities in a pandemic. Journal of Biomedical Semantics 13(1): 12. https://doi.org/10.1186/s13326-022-00263-7 (accessed 6 November 2026)
74. Rehnberg J, Segelmark M, Ludvigsson JF, Emilsson L (2024). Validation of IgA nephropathy diagnosis in the Swedish Renal Registry. BMC Nephrology 25(1): 78. https://doi.org/10.1186/s12882-024-03512-2 (accessed 6 November 2026)
75. Rockenschaub P, Akay EM, Carlisle BG, et al. (2025). External validation of AI-based scoring systems in the ICU: a systematic review and meta-analysis. BMC Medical Informatics and Decision Making 25(1): 5. https://doi.org/10.1186/s12911-024-02830-7 (accessed 25 October 2026)
76. Saelaert M, Mathieu L, Van Hoof W, et al. (2023). Expanding citizen engagement in the secondary use of health data: an opportunity for national health data access bodies to realise the intentions of the European Health Data Space. Archives of Public Health 81(1): 168. https://doi.org/10.1186/s13690-023-01182-4 (accessed 6 November 2026)
77. Sanmark E, Ponkilainen V, Brück O, et al. (2025). Toisiolaki tukahduttaa yhä lääketieteellistä tutkimusta (in Finnish; English version: The Secondary Use Act continues to hinder medical research). Suomen Lääkärilehti 80: e43720, 10 April 2025. https://www.laakarilehti.fi/tieteessa/alkuperaistutkimukset/toisiolaki-tukahduttaa-yha-laaketieteellista-tutkimusta/en (accessed 2 November 2026)
78. Schneider KL, Alves de Sousa L, Sava I, et al. (2025). The European Health Data Space in communicable diseases surveillance and monitoring of medicines and vaccines: achievements of a pilot project examining the user journey. European Journal of Public Health 35(Suppl 3): iii5 to iii10. https://doi.org/10.1093/eurpub/ckaf115 (accessed 2 November 2026)
79. SIB Swiss Institute of Bioinformatics (2024). Cantonal hospitals are joining the Swiss Personalized Health Network. News release, 10 October 2024. https://www.sib.swiss/news/cantonal-hospitals-are-joining-the-swiss-personalized-health-network (accessed 2 November 2026)
80. SIB Swiss Institute of Bioinformatics (2025). The Swiss Personalized Health Network moves from set-up to sustainable infrastructure. News release, 20 March 2025. https://www.sib.swiss/news/the-swiss-personalized-health-network-moves-from-set-up-to- sustainable-infrastructure (accessed 2 November 2026)
81. SMITH consortium (2025). A look into the Data Integration Centers: paving the way for research and digital innovation in medicine. The Data Integration Center Halle. News item, 30 April 2025 (date taken from the URL). https://www.smith.care/en/2025/04/30/a-look- into-the-data-integration-center-halle/ (accessed 6 November 2026)
82. Egli M, Kiss Blind J, Maurer J (ELSI Team, Swiss Personalized Health Network) (2025). Exploring data governance for data sharing at the Swiss university hospitals. Swiss Personalized Health Network (SPHN), 18 November 2025. https://sphn.ch/document/exploring-data- governance/ (PDF: https://sphn.ch/wp-content/uploads/2025/11/Pub_20251028_DGWG-external_Exploring-data-governance-for- data-sharing-at-the-swiss-university-hospitals-2.pdf)
83. SPHN, Swiss Personalized Health Network (2026). 2025 SPHN Annual Report. Report, announced 22 May 2026. https://www.samw.ch/dam/jcr:c60c86ad-e494-47a5-99fd-3ae355bfe88b/annual_report_sphn_2025.pdf (accessed 6 November 2026)
84. Supreme Audit Office of the Slovak Republic (NKU) (2026). Slovakia still developing health policy blindly as data registers remain incomplete. News release, 10 April 2026. https://www.nku.gov.sk/web/nku-en/-/slovakia-still-developing-health-policy-blindly-as- data-registers-remain-incomplete (accessed 2 November 2026)
85. Nationellt system för kunskapsstyrning Hälso- och sjukvård (Sveriges regioner i samverkan, hosted by SKR) (2025). Nationella kvalitetsregister: Årsrapport 2024 (in Swedish). 18 August 2025. https://kunskapsstyrningvard.se/download/18.5611e984198e5d7e93bc90c/1756297570576/Arsrapport-nationella-kvalitetsregister- 2024.pdf
86. Svingel LS, Jensen CE, Kjeldsen GF, et al. (2025). Shaping the future EHDS: recommendations for implementation of Health Data Access Bodies in the HealthData@EU infrastructure for secondary use of electronic health data. European Journal of Public Health 35(Suppl 3): iii32 to iii38. https://doi.org/10.1093/eurpub/ckaf033 (accessed 6 November 2026)
87. Tacconelli E, Gorska A, Carrara E, et al. (2022). Challenges of data sharing in European Covid-19 projects: a learning opportunity for advancing pandemic preparedness and response. Lancet Regional Health Europe 21: 100467. https://doi.org/10.1016/j.lanepe.2022.100467 (accessed 2 November 2026)
88. Tahar K, Verbuecheln R, Martin T, et al. (2026). Sharing EHR data of patients with rare diseases for research: the role of quality assessments in a national federated research data infrastructure. BMC Medical Informatics and Decision Making 26(1). https://doi.org/10.1186/s12911-025-03153-x (accessed 2 November 2026)
89. Tazare J, Henderson AD, Morley J, et al. (2024). NHS national data opt-outs: trends and potential consequences for health data research. BJGP Open 8(3). https://doi.org/10.3399/BJGPO.2024.0020 (accessed 25 October 2026)
90. TEHDAS joint action (2022). TEHDAS Denmark country visit factsheet. Prepared by Sciensano, October 2022. https://tehdas.eu/app/uploads/2023/03/denmark-country-visit-factsheets-10-2022.pdf (accessed 2 November 2026)
91. TEHDAS joint action (2023a). TEHDAS Sweden country visit factsheet, March 2023. https://tehdas.eu/app/uploads/2023/03/sweden- country-visit-factsheet-03-2023.pdf (accessed 2 November 2026)
92. TEHDAS joint action (2023b). Member states' readiness to benefit from the EHDS regulation varies. Web page, 28 April 2023. https://tehdas.eu/tehdas1/results/member-states-readiness-to-benefit-from-the-ehds-regulation-varies/ (accessed 2 November 2026)
93. TEHDAS2 joint action (2026). D4.1: Guideline for health data access bodies on fees and penalties for non-compliance related to the EHDS Regulation. Deliverable, version 31 March 2026. https://tehdas.eu/wp-content/uploads/2026/05/d4.1-guideline-for-health-data- access-bodies-on-fees-and-penalties-for-non-compliance-related-to-the-ehds-regulation.pdf (accessed 6 November 2026)
94. UCLH Biomedical Research Centre (2023). New UCLH service to streamline access to data for research. News item, launch 16 October 2023. https://www.uclhospitals.brc.nihr.ac.uk/news/new-uclh-service-streamline-access-data-research (accessed 6 November 2026)
95. Universitätsklinikum Regensburg (UKR), Clinical Data Integration Centre MEDIZUKR (2025). Data Use and Access Committee (DUAC). Web page; Rules of Procedure v3.0, 18 February 2025. https://www.ukr.de/en/research/research/how-you-can-support-our-research- work/clinical-data-integration-centre-at-ukr-medizukr/access-committee/data-use-and-access-committee-duac and https://www.ukr.de/fileadmin/GO__D_UAC_V3.0f_2025-02-18_.pdf
96. Vande Catsyne CA, Slot M, Buble T, et al. (2025). Feasibility of mapping cross-country population coronavirus disease 2019 metrics in a federated design: learnings from a HealthData@EU pilot use case. European Journal of Public Health 35(Suppl 3): iii11 to iii17. https://doi.org/10.1093/eurpub/ckaf017 (accessed 2 November 2026)
97. Wiertz S, Boldt J (2024). Ethical, legal, and practical concerns surrounding the implemention of new forms of consent for health data research: qualitative interview study. Journal of Medical Internet Research 26: e52180. https://doi.org/10.2196/52180 (accessed 25 October 2026)
98. Zucca A, Biagi E, Nocchi F, Tola J, Lapi S (2026). Governance stabilization under operational expansion: a longitudinal analysis of ISO 20387 accreditation in a hospital-based biobank. Biopreservation and Biobanking, article 19475535261464279. Published online 4 July 2026. https://doi.org/10.1177/19475535261464279

13. Why ZINOS exists

Health is data. The hospitals that structure and govern their research data now will be the ones that lead clinical research in Europe over the next decade: running the trials, answering the questions no sponsor will fund, and taking part in AI on their own terms.

ZINOS exists to help them do that. We offer hospitals the Scientific and Research Data Hub. It joins the research records a hospital holds, system by system, into one record per patient, inside the hospital's walls, with one permission system and one audit trail covering all of it. Researchers, sponsors, networks, and access bodies can put their questions to it without the data leaving the hospital's control. It is aligned with the European Health Data Space. The platform behind the Scientific and Research Data Hub runs a national disease register, with full image management in the record.

The place to start is a conversation. We invite any hospital to complete the self-assessment in Annex A, and we will review the result with your team. We will publish what hospitals tell us only as aggregate, anonymised findings, so that the evidence on hospital readiness in Europe grows.