Global Clinical Data Lakes and Research Dependency Structures

Introduction

Global clinical data lakes are large-scale digital repositories that collect, integrate, and analyse heterogeneous health and research data from hospitals, laboratories, pharmaceutical companies, clinical trials, genomic databases, electronic health records, wearable devices, biobanks, insurance systems, and patient-generated data.

Unlike a conventional database, a data lake can retain enormous volumes of structured and unstructured information in relatively raw form. When such infrastructure becomes essential to pharmaceutical research, clinical-trial design, precision medicine, artificial-intelligence development, or regulatory science, it can create research dependency structures: researchers, hospitals, universities, biotechnology companies, and even public authorities may become dependent on access to a small number of datasets or data infrastructures.

The competition-law concern is therefore not simply who owns clinical data, but who controls access to an indispensable research input and under what conditions.

1. Meaning of Clinical Data Lakes

A clinical data lake may combine:

• electronic health records;
• laboratory results;
• medical imaging;
• genomic and proteomic information;
• clinical-trial data;
• adverse-event reports;
• prescription and treatment histories;
• hospital utilisation data;
• wearable-device information;
• longitudinal patient records;
• biobank information;
• disease registries;
• real-world evidence;
• synthetic or transformed clinical datasets; and
• metadata describing provenance, quality and patient populations.

The competitive value of such infrastructure increases when datasets are:

1. large;
2. longitudinal;
3. highly granular;
4. interoperable;
5. representative of particular patient populations; and
6. difficult or impossible to replicate.

A million isolated medical records may have less strategic value than a smaller but longitudinal dataset linking diagnosis, treatment, laboratory measurements and outcomes over many years.

2. What Is a Research Dependency Structure?

A research dependency structure arises where one research participant depends materially upon another entity for access to a resource necessary to conduct meaningful research.

A simplified structure is:

Patients → Hospitals → Data Platforms → Data Lakes → Researchers → Clinical Research → Pharmaceutical Products

If access to the data lake is controlled by one or a few powerful entities, the downstream research ecosystem can become dependent on them.

Example

Suppose a dominant hospital network aggregates ten years of oncology records.

A biotechnology company wants to develop an AI diagnostic model.

The company needs:

• historical patient records;
• imaging;
• pathology;
• treatment outcomes; and
• longitudinal follow-up.

If the hospital network refuses access, imposes discriminatory conditions, or provides access only to affiliated pharmaceutical companies, the data may become a competitive bottleneck.

3. Why Clinical Data Can Become a Competition-Law Asset

Data itself is not automatically an economic monopoly.

The important question is whether control over the data gives an undertaking market power.

Several characteristics can make clinical datasets strategically significant.

A. Replication difficulty

Historical patient data cannot necessarily be recreated.

A competitor cannot simply decide to collect ten years of cancer outcomes tomorrow.

B. Network effects

More patients can generate more observations.

More observations can produce better research models.

Better models can attract more hospitals and patients.

This can create:

Data → Better research → Better products → More users → More data

C. Economies of scale

Large datasets can reduce the cost of:

• clinical-trial recruitment;
• disease identification;
• patient stratification;
• drug-development research;
• safety analysis; and
• AI model training.

D. Economies of scope

The same dataset may support:

• drug discovery;
• diagnostics;
• clinical-trial optimisation;
• medical-device development;
• epidemiology; and
• AI research.

E. Switching costs

Researchers may invest heavily in:

• APIs;
• analytical infrastructure;
• data-cleaning systems;
• software;
• compliance processes; and
• machine-learning pipelines.

Once integrated, switching to another data source can become expensive.

4. The Main Competition Concerns

4.1 Refusal to Provide Access

The central concern is whether a dominant data controller can lawfully refuse access to researchers or competitors.

A competition authority may examine:

• indispensability;
• availability of alternative datasets;
• replication possibilities;
• duration of the refusal;
• discriminatory access;
• foreclosure effects; and
• legitimate justification.

The classic essential-facilities doctrine may become relevant in exceptional circumstances.

4.2 Discriminatory Data Access

A data lake operator might provide:

• favourable access to its own research division;
• inexpensive access to affiliated pharmaceutical companies;
• expensive access to independent researchers; and
• no access to competing firms.

This can create vertical foreclosure.

The competition problem becomes particularly serious where the data controller competes downstream.

5. Data Advantage and Self-Preferencing

Consider a hospital platform that operates both:

1. a clinical-data infrastructure; and
2. an AI diagnostic business.

It could potentially use privileged access to clinical data to give its own AI service an advantage.

Possible forms include:

• preferential API access;
• faster data refresh;
• greater data granularity;
• exclusive datasets;
• superior metadata;
• preferential patient recruitment;
• discriminatory pricing; or
• delayed access for competitors.

This resembles self-preferencing concerns observed in digital-platform competition law, although the health-data environment introduces additional privacy and medical-regulation constraints.

6. Data Exclusivity and Research Foreclosure

Exclusive arrangements can have legitimate purposes.

For example, a pharmaceutical company financing a clinical study may require a period of exclusive access to commercially exploit its investment.

However, excessive exclusivity may prevent rivals from:

• conducting independent research;
• validating scientific findings;
• developing competing medicines;
• improving diagnostic systems; or
• challenging incumbent scientific conclusions.

Therefore, competition analysis must balance investment incentives against downstream foreclosure.

7. Privacy Does Not Automatically Justify Every Restriction

Clinical data is highly sensitive.

Privacy, consent, medical confidentiality and data-protection rules can legitimately restrict data sharing.

However, competition law may ask a different question:

Is the claimed privacy justification genuine, proportionate and applied consistently?

A dominant undertaking should not necessarily be able to invoke privacy selectively—for example, refusing access to competitors while extensively exploiting the same information internally.

At the same time, competition law cannot simply compel disclosure of identifiable patient information contrary to privacy law.

The appropriate remedy may instead involve:

• anonymisation;
• pseudonymisation;
• secure research environments;
• federated learning;
• controlled APIs;
• data trusts;
• independent data access committees; or
• aggregated datasets.

8. Interoperability as a Competition Issue

Clinical data dependency frequently arises from incompatible systems.

A platform may control:

• data formats;
• APIs;
• identity systems;
• interoperability standards;
• metadata;
• access credentials.

If competitors cannot technically access or migrate data, the platform can acquire infrastructural power.

Interoperability therefore becomes an important competition-law remedy.

Possible measures include:

• API access;
• common data standards;
• portability;
• interoperability obligations;
• technical documentation;
• data-export mechanisms; and
• non-discriminatory access.

9. Algorithmic Research Dependency

The dependency is not limited to raw data.

Increasingly, researchers may depend upon:

• curated datasets;
• labelled datasets;
• trained models;
• embeddings;
• clinical ontologies;
• preprocessing pipelines;
• synthetic datasets; and
• proprietary analytical tools.

Consequently, the competitive bottleneck may move from:

Data → Data infrastructure → Data processing → AI model → Research ecosystem

An undertaking controlling several stages can create a vertically integrated research dependency.

10. Data Quality as a Competitive Moat

Quantity alone does not determine competitive importance.

A smaller dataset may be more valuable because it contains:

• high-quality labels;
• rare diseases;
• diverse populations;
• complete longitudinal records;
• validated clinical outcomes;
• consistent metadata.

Thus, competition authorities should evaluate data quality, uniqueness and usability, rather than merely measuring terabytes of information.

11. Relevant Case Laws

The following cases provide important legal principles that can be applied to clinical-data-lake dependency, even where the underlying disputes did not involve modern clinical data lakes themselves.

1. Bronner v Mediaprint

Case: Oscar Bronner GmbH & Co KG v Mediaprint Zeitungs und Zeitschriftenverlag GmbH & Co KG, C-7/97.

The Court of Justice applied a demanding test to refusal-to-supply claims.

The importance of the case for clinical data is that mere usefulness or commercial desirability is insufficient to establish an obligation to provide access.

The claimant generally needs to demonstrate that the facility is indispensable and that there is no realistic substitute.

Application

A researcher seeking access to a clinical data lake would therefore need to demonstrate more than:

"This dataset would make my research easier."

The argument would need to approach:

"Without access to this dataset or a realistic substitute, meaningful participation in the relevant research market is effectively impossible."

2. IMS Health v NDC Health

Case: IMS Health GmbH & Co OHG v NDC Health GmbH & Co KG, Joined Cases C-418/01.

This is particularly important for data-driven industries.

The Court considered whether refusal to license a protected information structure could constitute an abuse of dominance.

The Court identified exceptional circumstances associated with:

• indispensability;
• elimination of effective competition;
• prevention of a new product for which there was consumer demand; and
• absence of objective justification.

Clinical-data significance

A proprietary clinical-data architecture could become competitively significant where researchers cannot reasonably recreate it and access is necessary to produce innovative downstream services.

The case nevertheless demonstrates that mandatory access is exceptional rather than automatic.

3. Microsoft Corp v Commission

Case: Microsoft Corp v Commission, T-201/04.

The case concerned interoperability information and Microsoft's control over information necessary for competing products.

The General Court upheld the Commission's approach to interoperability and refusal-to-supply concerns.

Clinical-data significance

Modern clinical platforms may control:

• APIs;
• interoperability specifications;
• data schemas;
• authentication systems; and
• technical information.

Where control of these inputs prevents competitors from interoperating effectively, the Microsoft principles provide an important analytical analogy.

4. Slovak Telekom v Commission

Cases: Slovak Telekom a.s. v European Commission, Joined Cases C-165/19 P and C-152/19 P.

The litigation concerned access to telecommunications infrastructure and abusive exclusionary conduct.

The case is relevant because competition law can distinguish between:

• ordinary commercial conduct; and
• conduct by a dominant undertaking involving infrastructure that competitors require to compete.

Clinical-data significance

A dominant healthcare-data platform could potentially occupy an infrastructure-like position where researchers cannot realistically reproduce its integrated data environment.

5. Google Shopping

Case: Google and Alphabet v Commission, Case C-48/22 P.

The case concerned Google's treatment of its own comparison-shopping service within its search ecosystem.

The broader significance is the relationship between platform dominance, self-preferencing and leveraging.

Clinical-data significance

Suppose a dominant healthcare-data platform operates its own:

• clinical-research service;
• diagnostic AI;
• pharmaceutical analytics service.

Using control over the data infrastructure to systematically favour its own downstream service could raise analogous leveraging concerns.

The precise legal analysis would, however, depend on the relevant market and conduct.

6. Commercial Solvents

Cases: Commercial Solvents Corp v Commission, Joined Cases 6/73 and 7/73.

The Court recognised that a dominant undertaking controlling an upstream input could abuse its position by restricting supply to downstream competitors.

Clinical-data significance

This is highly relevant to vertically integrated clinical-data ecosystems.

Imagine:

Data controller → Clinical analytics → Diagnostic product

If the undertaking controls an upstream research input and restricts competitors downstream, competition concerns may arise.

7. Magill

Cases: Radio Telefis Eireann (RTE) and Independent Television Publications Ltd (ITP) v Commission, Joined Cases C-241/91 P and C-242/91 P.

The case concerned refusal to license copyrighted information.

It reinforced the exceptional nature of compulsory access where intellectual-property rights intersect with competition law.

Clinical-data significance

Clinical datasets may involve:

• database rights;
• contractual rights;
• copyright;
• trade secrets;
• proprietary annotations.

The existence of an intellectual-property right does not automatically immunise conduct from Article 102 scrutiny.

8. Huawei Technologies v ZTE

Case: Huawei Technologies Co. Ltd v ZTE Corp., Case C-170/13.

Although concerning standard-essential patents rather than health data, the case is important for understanding the relationship between standardisation, access and market power.

Clinical-data significance

Clinical-data ecosystems increasingly depend upon:

• common standards;
• interoperability protocols;
• technical specifications;
• certification systems.

Where participation in a standard becomes indispensable, exclusionary control can acquire substantial competitive importance.

12. Consolidated Legal Principles from the Cases

13. Research Dependency and Market Definition

A crucial question is:

What is the relevant market?

Possible markets include:

A. Clinical-data access market

The market may involve access to datasets used for:

• drug development;
• clinical research;
• diagnostics; or
• epidemiological analysis.

B. Data-infrastructure market

The relevant market may instead concern:

• healthcare cloud infrastructure;
• clinical data aggregation;
• interoperability;
• data-management platforms.

C. Downstream research markets

The competitive harm may occur downstream in:

• pharmaceuticals;
• medical devices;
• diagnostics;
• health AI;
• precision medicine.

Therefore, competition authorities should avoid treating "health data" as one homogeneous market.

14. Research Dependency Can Create a Bottleneck

The most important economic concept is the data bottleneck.

A simplified structure is:

                 PATIENT DATA                      ↓              DATA AGGREGATOR                      ↓              CLINICAL DATA LAKE                      ↓        ┌─────────────┼─────────────┐        ↓             ↓             ↓   Drug research   Health AI    Diagnostics        ↓             ↓             ↓   Pharmaceutical  AI products   Medical     markets                     products

If the central data lake is controlled by one dominant undertaking, it may become a bottleneck facility connecting multiple downstream markets.

15. Potential Abusive Strategies

A dominant clinical-data operator could theoretically engage in:

1. Exclusive dealing

Preventing hospitals or laboratories from supplying competing data platforms.

2. Refusal to supply

Withholding an indispensable dataset from competitors.

3. Discriminatory access

Different access conditions for affiliated and independent researchers.

4. Excessive pricing

Charging competitors substantially more for equivalent access.

5. Self-preferencing

Giving proprietary research services preferential access.

6. Data degradation

Providing competitors with incomplete, delayed or lower-quality data.

7. Technical exclusion

Preventing interoperability through APIs or technical barriers.

8. Bundling

Requiring researchers to purchase unrelated software or analytical services to obtain data access.

9. Data tying

Conditioning access to participation in another platform.

10. Strategic acquisitions

Acquiring hospitals, biobanks or specialist datasets primarily to eliminate competing sources.

16. Merger-Control Dimension

Clinical-data concentration can also arise through mergers.

For example:

Hospital network + health-data platform + AI company

may create a vertically integrated ecosystem.

Traditional turnover thresholds may fail to capture the strategic value of data-rich targets.

Competition authorities may therefore examine:

• future competitive significance;
• data uniqueness;
• innovation competition;
• potential entrants;
• research pipelines;
• interoperability;
• patient switching;
• access foreclosure; and
• elimination of nascent competitors.

A data-rich hospital or biotechnology company may be competitively valuable even when its current revenues are relatively modest.

17. Cross-Border Dependency

Clinical research is inherently international.

A single research project may involve:

European patient data + U.S. cloud infrastructure + Asian pharmaceutical research + global AI model development.

This produces jurisdictional problems involving:

• data localisation;
• privacy law;
• national-security restrictions;
• research-sovereignty policies;
• cross-border transfers;
• competition law;
• intellectual-property rights.

A dominant data infrastructure can therefore possess cross-border economic power even where its physical assets are located in only one jurisdiction.

18. Public and Private Clinical Data

An important distinction exists between:

Public-sector data

Examples include:

• national health services;
• public hospitals;
• disease registries;
• government research databases.

Here, competition law may intersect with:

• public procurement;
• state aid;
• administrative law;
• public-sector information rules.

Private-sector data

Examples include:

• pharmaceutical companies;
• private hospital chains;
• commercial laboratories;
• health platforms.

Here, ordinary competition-law doctrines regarding dominance, exclusion and vertical foreclosure become more prominent.

19. The "Data Lake as Essential Facility" Problem

It would be incorrect to assume:

Large dataset = essential facility.

The legal test is considerably more demanding.

Authorities should investigate:

1. Is the undertaking dominant?
2. Is the dataset indispensable?
3. Are realistic substitutes available?
4. Can competitors reproduce the data?
5. Would access eliminate or substantially reduce competition?
6. Is there a legitimate justification for refusal?
7. Can access be provided consistently with privacy law?
8. Is a proportionate remedy technically feasible?

Only an exceptionally strong combination may justify mandatory access.

20. Possible Remedies

Competition authorities could consider several remedies.

Structural remedies

In extreme circumstances:

• divestiture;
• separation of data infrastructure from downstream services;
• prohibition of certain acquisitions.

Behavioural remedies

More commonly:

• non-discriminatory access;
• transparent licensing;
• reasonable pricing;
• interoperability;
• API access;
• data portability;
• independent auditing.

Privacy-preserving remedies

These could include:

• anonymisation;
• pseudonymisation;
• secure data enclaves;
• federated learning;
• differential privacy;
• controlled-access research environments.

The objective should be to create competitive access without compromising patient confidentiality.

21. Federated Research as an Alternative

One important technological solution is federated research.

Instead of transferring patient-level information:

Hospital A ─┐ Hospital B ─┼──→ Federated research system Hospital C ─┘                    ↓             Researcher receives             aggregated results

This can reduce the need for centralised data ownership.

It may therefore weaken the competitive power of a single global data lake while preserving privacy.

22. Competition Between Data Lakes

Competition authorities should not focus exclusively on access.

They should also ask whether multiple independent data infrastructures can coexist.

A competitive ecosystem could contain:

• public research databases;
• university repositories;
• hospital networks;
• private clinical-data platforms;
• biobanks;
• federated networks.

Competition between data infrastructures can reduce dependency and prevent one platform from becoming an unavoidable intermediary.

23. The "Research Monopoly" Risk

The most significant long-term concern is not merely commercial pricing.

It is control over the direction of scientific innovation.

If one organisation controls:

• the largest datasets;
• the most sophisticated analytics;
• the dominant AI models;
• the principal clinical-trial infrastructure; and
• the relevant research platforms,

it may influence:

• which diseases receive research attention;
• which hypotheses are tested;
• which therapies are developed;
• which researchers receive access;
• which scientific results are commercially valuable.

This transforms data concentration into a form of scientific market power.

24. Economic and Social Effects

Research dependency may produce:

Positive effects

• faster clinical research;
• better patient stratification;
• lower trial costs;
• improved drug discovery;
• improved diagnostics;
• greater predictive accuracy.

Negative effects

• exclusion of smaller biotechnology firms;
• reduced independent research;
• higher entry barriers;
• concentration of pharmaceutical innovation;
• discriminatory research access;
• reduced scientific pluralism;
• dependency upon private infrastructure.

Thus, the policy objective is not necessarily maximum data sharing, but competitive and scientifically meaningful access consistent with privacy and security.

25. Overall Legal Test

A useful analytical framework is:

Clinical Data Concentration

↓

Is the data commercially/research relevant?

↓

Is the controller dominant?

↓

Is the dataset unique or difficult to reproduce?

↓

Are realistic substitutes available?

↓

Does exclusion affect a downstream market?

↓

Is there discriminatory or exclusionary conduct?

↓

Is there an objective/privacy/security justification?

↓

Would access be technically and legally feasible?

↓

Proportionate competition remedy

Conclusion

Global clinical data lakes represent a new form of infrastructural market power. Their significance lies not simply in the quantity of medical information they contain but in their ability to become indispensable inputs for clinical research, pharmaceutical development, diagnostics and health AI.

The principal competition-law danger arises when control over data becomes control over research participation.

The traditional cases—particularly Bronner, IMS Health, Microsoft, Commercial Solvents, Slovak Telekom, Magill and Google Shopping—provide the conceptual foundations for analysing these problems. They indicate that competition law can address exclusionary control over strategically important information and infrastructure, but that compulsory access remains exceptional and must be reconciled with legitimate privacy, intellectual-property, security and investment interests.

The emerging policy challenge is therefore to prevent a situation in which:

Data concentration → infrastructure dependence → research exclusion → innovation concentration → downstream market power.

The strongest regulatory model is likely to combine competition law, interoperability, privacy-preserving data access, merger control, research-governance rules and cross-border regulatory cooperation, rather than treating clinical data solely as either private property or a freely accessible public resource.