An insight is only as good as the data behind it. Every model, statistic, and clinical recommendation carries the shape of the population it was built on. When that population is narrow, the insight is narrow too, even if the method is flawless.

In health, this has real consequences. A risk score, biomarker, or treatment signal found in one group of patients can weaken, shift, or fail when applied to another. The question is not only whether an analysis is correct, but for whom it is correct. That makes the diversity of the underlying data one of the most important, and most overlooked, factors in health AI.

Why diverse data matters

Around 84% of the global population is underrepresented in the data that health AI learns from. We call them the Remaining84. Most reference cohorts come predominantly from patients of European ancestry treated in a small number of high-income health systems.

This is a scientific issue:

• Biology varies. Mutation frequencies, risk factors, and treatment response differ across populations. A finding from one cohort may not hold in another.
• Models inherit gaps. An AI system trained on narrow data performs best on the patients who look like its training set, and worse on everyone else.
• Evidence needs replication. If an insight only appears in one dataset, it is a hypothesis.

Diverse, real-world data is how a result earns the right to be trusted beyond the cohort that produced it.

Regulators now expect representative data

Representative data is moving from best practice to legal expectation. Three EU frameworks shape how health AI is built and validated:

• EU AI Act. Health AI is classed as high-risk. Article 10 requires training, validation, and testing data to be relevant, sufficiently representative, and examined for possible bias. A model built on a single population will struggle to meet that bar.
• GDPR. Health and genetic data are special categories under Article 9. Every use needs a clear legal basis, purpose limitation, and strong safeguards, and research exemptions are implemented differently in each member state.
• European Health Data Space (EHDS). In force since March 2025, it creates a framework for secondary use of health data through national access bodies and secure processing environments. Its direction is clear: data should be used where it lives, under controlled access, rather than copied freely.

Together, these rules ask developers to prove two things at once: that their data is representative enough to be fair, and that it was accessed lawfully.

The cross-border bottleneck

The data that would close the diversity gap sits in hospitals and labs across dozens of countries. Moving it is where most initiatives stall.

• Transfer restrictions. Under GDPR, sending personal data outside the EEA requires an adequacy decision or safeguards such as standard contractual clauses. Since the Schrems II ruling in 2020, every transfer also needs an assessment of the destination’s legal regime.
• Data localisation laws. A growing number of countries restrict or prohibit the export of health and genomic data entirely, keeping it inside national borders.
• Fragmented rules. Consent models, ethics approvals, and research exemptions differ between countries, and sometimes between regions within one country.
• Incompatible formats. Even with permission, data arrives in different coding systems, languages, and structures. Without harmonisation, it cannot be compared.

The result is a paradox. The populations most missing from health AI are often the ones whose data is hardest to reach.

How PAICON closes the gap

PAICON brings representative data and global insights together in one compliant workflow.

Harmonized representative data PAICON works with partners in more than 60 countries and has built a harmonised base of over 100,000 proprietary and public cases with clinical records, molecular data, and images. This data is being integrated into PaiX Navigator, so the same question can be tested beyond European reference cohorts.

Global insights: PaiX Navigator turns this data into insights that reflect patients worldwide. Researchers and clinicians ask questions in plain language and can test whether a finding holds across regions, ancestries, and health systems. Statistics are computed by dedicated tools, and every chart and table links to the query behind it, so each global insight can be verified.

The outcome: insights that reflect the patients they are meant to serve, built on representative data under controlled, privacy-preserving access.

Access PaiX Navigator

PaiX Navigator is now open for beta testing. Request early access to explore the platform.

Note: Text drafted with the support of AI tools and verified by our editorial team.