Health data and the public interest: the OECD proposes common criteria for secondary use

A new OECD policy brief identifies five categories of criteria for deciding when access to health data truly serves the public interest. The aim is to reduce the fragmentation and delays that currently hold back research, innovation and the development of AI in healthcare.
In October 2026 the OECD published the policy brief Determining public interest in the secondary use of health data. It addresses a central question for anyone working with health data: how can we establish, consistently and transparently, whether a project that re-uses data collected during patient care pursues a public interest?
The question is particularly relevant in Europe. The European Health Data Space will soon turn secondary use into a structured procedure based on data permits. We described that procedure in The European Health Data Space: New Rights for Patients, New Access to Data for Research and Innovation. The OECD criteria show what access bodies are likely to look for when they assess an application.
Secondary use: why it matters and what slows it down
Secondary use means using data collected in clinical care for other purposes, such as research, innovation and health system management. The data involved are broad:
- diagnoses, treatments, test results and imaging;
- mental health, lifestyle and socio-economic information;
- administrative and insurance data.
Re-use makes large-scale analysis of electronic health records and registries possible, and it underpins the development of AI tools for healthcare.
When the data relate to identifiable individuals, consent is the most common legal basis. Often, however, obtaining consent is impossible, impracticable or incompatible with the purpose of the project. In those cases the decision falls to review bodies, such as ethics committees, data protection authorities and national committees. They must assess whether a legitimate public interest justifies access, with adequate safeguards.
According to the OECD, the main obstacle is fragmentation. Review bodies interpret the public interest differently, both within the same country and across borders. Where several approvals are needed, the delays add up. In France, for example, an application goes both to an ethics and scientific committee and to the CNIL.
Under the EHDS (Regulation (EU) 2025/327), each Member State must designate one or more health data access bodies by 26 March 2027. These bodies will decide on applications for secondary use, which will apply from March 2029. In Italy, the design under development provides for four bodies: the Ministry of Health, the Istituto Superiore di Sanità, AGENAS and ISTAT. Whether these bodies apply consistent criteria will largely determine whether the EHDS reduces fragmentation or reproduces it.
Five categories of criteria
The brief proposes common criteria. They are drawn from national guidelines in the UK, Australia and France and from discussions with national experts. Review bodies typically examine five areas:
- Population health and system planning. The project must make a demonstrable contribution to prevention, diagnosis and treatment, to public health policy, to epidemiological surveillance or to the quality of services. It must benefit the population, not only individuals or organisations. Review bodies also check whether the data are of sufficient quality, completeness and representativeness.
- Research and innovation. What matters is scientific merit, whether the data are fit for purpose, the credentials of the researchers and credible plans to disseminate the results. Where commercial interests are involved, safeguards must ensure that private interests do not outweigh the public benefit.
- Ethics and risk proportionality. The benefits must outweigh the risks to individuals, and the data requested must be necessary and proportionate. The public must be informed, opt-out mechanisms must exist, and security must be appropriate, for example through trusted research environments and privacy-enhancing technologies.
- Health equity. Review bodies consider whether the project improves fair access to care and knowledge, addresses neglected diseases, or meets the needs of marginalised or disadvantaged groups.
- Public communication and transparency. The use of the data must be consistent with the reasonable expectations of the public and clearly communicated.
One point stands out: the public interest is not reserved to public actors. Private organisations can also access data if they show clear public value. In that case, additional safeguards apply on scientific independence, transparency and accountability.
Many of these criteria correspond to elements of the EHDS. The Regulation lists the permitted purposes of secondary use and prohibits others. It makes data available in secure processing environments, anonymised by default, and it gives individuals the right to opt out. The OECD brief adds a more qualitative layer. The question is not only whether a purpose is permitted, but whether a specific project actually delivers public value.
Two cases in practice
France: the IQVIA data warehouse. In 2018 the CNIL authorised IQVIA to create and operate a large data warehouse based on pharmacy dispensing data, for research, studies and evaluations in health. The CNIL found that the warehouse met the public interest threshold, but required each individual study to demonstrate its own public interest purpose. It also prohibited using the data to promote health products or to make insurance decisions. The EHDS now prohibits both purposes for secondary use across the EU. A 2026 enforcement decision found shortcomings in transparency and security, but did not call into question the public interest of the warehouse.
Finland: an industry-funded study. In 2013 the Supreme Administrative Court of Finland ruled against granting access for a pharmaco-epidemiological study funded by a pharmaceutical company. Industry funding is allowed, but it raises the bar. In this case the company could comment on publications, publication of unfavourable results was not guaranteed, and the research hypotheses were not precise. For the Court, scientific independence must be built into the structure of the project, not merely declared.
Recommendations for policymakers
The brief closes with five recommendations:
- Coordinate review bodies, aligning definitions and criteria to avoid inconsistent interpretations of risk and public benefit.
- Communicate clear criteria and give applicants practical guidance, such as the CNIL’s reference methodologies for health research.
- Ensure timely decisions, because delays use up research grant periods and slow down innovation.
- Offer clear roadmaps and well-designed application forms, such as the application portal of Germany’s Health Data Lab (Forschungsdatenzentrum Gesundheit) or the UK “Five Safes” data access request form.
- Support trusted mechanisms for data sharing. These include federated infrastructures (sphin-X, OHDSI, South Korea’s medical data space) and controlled-access platforms for clinical trial data (YODA, Vivli).
Several of these recommendations are already reflected in the EHDS. The Commission has consulted on standard templates for the data access application, the data permit and the health data request. It has also adopted rules on the minimum metadata that data holders must provide to describe their datasets. National implementation will show whether the same holds for coordination between bodies and the timeliness of decisions.
What it means for organisations working with health data
The message is clear for hospitals, research centres, pharmaceutical companies and developers of AI solutions for healthcare. The public interest must be demonstrated and documented from the design stage, not simply asserted when applying for access. In practice, this means:
- defining purposes and research hypotheses precisely;
- justifying why the data requested are necessary and proportionate;
- assessing the quality and representativeness of datasets;
- putting governance, security and transparency towards the public in place;
- for private organisations, ensuring scientific independence through structures, not only commitments.
For AI developers, these requirements connect with those of the AI Act. Article 10 requires the training, validation and testing data of high-risk systems to be relevant, sufficiently representative and, as far as possible, free of errors. We discussed these requirements in AESIA Guidelines on the AI Act – Guide No. 7: Data and Data Governance.
Demonstrating public value, proportionality and equity is also close to the logic of responsible AI due diligence, discussed in AI Ethics in Practice: 7 Questions Every Company Should Ask. It is equally close to the fundamental rights impact assessment. We examined the FRIA in Fundamental rights impact assessment and AI: a new methodology for EU-funded research and in The AI Office’s Draft FRIA Template.
In Italy, Article 8 of Law No. 132/2025 provides a specific route for research on AI in healthcare. It requires approval by an ethics committee and prior communication to the Garante. The elements identified by the OECD can help prepare both.
The brief also recalls that approval may not be needed where individuals are no longer identifiable under applicable legal and technical standards for anonymisation. This is why privacy-enhancing approaches matter, including synthetic data. We first addressed synthetic data in Synthetic data, the last frontier of data protection and data sharing (in Italian). Whether such data fall outside the scope of data protection law must still be assessed case by case, on the basis of the residual risk of re-identification. We looked at this question in Synthetic Data: When Are They Really Anonymous?. The Italian Data Protection Authority’s opinion on AI training in the Province of Trento shows how demanding that assessment can be. We discussed it in Un modello di IA può dimenticare? (in Italian).
As the European Health Data Space is implemented, these elements will become the core of applications for access to health data across the EU. Organisations that learn to document them now will be better placed when the first data permits are requested.
Data Valley Consulting supports healthcare organisations, research centres and companies in the governance of health data. Our work ranges from assessing the legal bases for secondary use to designing compliant solutions such as synthetic data, and preparing for the European Health Data Space.
Sei pronto a trasformare i dati in valore per il tuo business?

