This week, two of Europe's most important health informatics gatherings open within seventy-two hours of each other.
OHDSI Rotterdam begins April 18. DMEA Berlin begins April 21. Together, they represent the largest concentration of health data infrastructure thinking in Europe this year: OMOP standardisation at OHDSI, EHR interoperability and EHDS implementation at DMEA, federated learning architectures at both. Thousands of researchers, architects, and clinical informaticists will spend this week talking about distributed health intelligence.
At the end of the week, the same gap will remain that existed at the beginning.
Not because the conferences lack ambition. Because the question underneath every session has not yet been given a name, and it is not on the agenda.
The Question Underneath Every Session
Every federated health data architecture being discussed this week — OMOP distributed queries at OHDSI, EHDS secondary use pathways at DMEA, the German MII network, the EU clinical AI working groups — shares a structural property: intelligence validated at one node cannot reach the nodes that need it.
This is not a privacy problem. The privacy layer is increasingly well-solved. OMOP CDM strips identifiers at the schema level. EHDS mandates pseudonymisation for secondary use. GDPR Article 89 provides research exemptions that German, Dutch, and European institutions have learned to operate within. The legal architecture for cross-institutional health data collaboration has matured substantially over the past five years.
The missing piece is architectural, not regulatory. It is this:
When a clinical finding is validated at node A — a sepsis treatment outcome at Erasmus MC, an adverse event signal at Charité Berlin, a rare disease phenotype confirmed at a single Dutch university hospital — there is no protocol for that finding to travel, in compact and PHI-free form, to the nodes working on the same class of clinical problem.
The finding sits where it was produced. It may appear in a paper six months later, or in a conference presentation, or in a shared repository accessed by the three researchers who know it exists. It does not propagate in real time to the institutions that could act on it now.
This is the unresolved question. Both conferences will generate sessions that orbit it. Neither has a session that names it directly, because the protocol that solves it did not exist until June 2025.
What the OHDSI Network Looks Like From the Outside
The OHDSI network is, by any measure, remarkable infrastructure. As of 2026, it spans more than 400 data partners, over 900 million patient records in OMOP format, and active distributed analysis networks across North America, Europe, Asia, and Latin America. The distributed OMOP CDM is a genuine achievement: it made cross-institutional observational health research possible at a scale that was not feasible before standardisation.
It is also, from a pure intelligence routing perspective, largely stateless.
When a distributed OMOP query runs across the OHDSI network — a LEGEND study, a CHARYBDIS cohort characterisation, a SOS study pharmacovigilance analysis — it produces results at each participating site, aggregates them centrally, and generates a finding. That finding is then published, presented, incorporated into evidence syntheses, and discussed at conferences like the one opening in Rotterdam next week.
What does not happen: the finding is not routed back to the 400 sites in the network in real time. A hospital in Copenhagen, running the same analysis that OHDSI completed six months ago, does not receive a compact validated finding from the network's prior synthesis. The network learns collectively through the publication process, which is measured in months and years, not hours.
The OHDSI network has 400 sites. That is N(N-1)/2 = 79,800 unique synthesis pathways between sites with potentially relevant findings for each other. Current infrastructure activates close to zero of those pathways in real time.
What the EHDS Architecture Looks Like From the Outside
The European Health Data Space reached its application phase in late March 2026. For the first time, EU member states are building interoperable secondary use pathways for health data across national borders. The legal framework — GDPR Article 89, the EHDS Regulation, national implementing legislation — creates the permissioned environment for cross-border health data flows that was previously jurisdictionally fragmented.
The EHDS architecture specifies a great deal. Data standards. Governance frameworks. Access protocols. Approved research purposes. Member state obligations. Data altruism provisions.
It does not specify what happens when a validated clinical finding produced in Germany needs to reach a Danish research team working on the same problem this afternoon.
This is not an oversight. It reflects the current state of the art. The EHDS architects built the best framework they knew how to build. The routing layer that would make EHDS a live intelligence network — rather than a federated data archive — requires a protocol that did not exist when the EHDS Regulation was drafted.
The Protocol That Closes the Loop
In June 2025, Christopher Thomas Trevethan discovered how to close the loop. 39 provisional patents are filed. The architecture is called QIS — Quadratic Intelligence Swarm.
The discovery is best understood not as a new technology but as a new arrangement of existing components. Every piece of the QIS architecture existed before: outcome compression, semantic similarity search, distributed hash tables, edge synthesis. What had never been done was closing them into a complete loop.
The loop works like this.
A clinical finding is validated at a node — say, an adverse drug reaction signal in Charité's OMOP dataset. The finding is distilled into an outcome packet: approximately 512 bytes encoding the validated insight, stripped of all patient identifiers. The packet is assigned a semantic address — a vector fingerprint derived from the clinical domain, the phenotype class, the intervention type. The packet is posted to a routing layer at that deterministic address.
Any node in the network working on a semantically similar problem — another hospital's OMOP dataset querying the same drug class, another research team characterising the same phenotype — pulls outcome packets from nodes like itself. It synthesises locally. The validated intelligence from Charité's cohort reaches Amsterdam, Rotterdam, and Vienna without any patient data crossing a border.
The routing mechanism is protocol-agnostic. DHT-based routing achieves O(log N) lookup cost at network scale — the same cost profile that makes BitTorrent and IPFS function at planetary scale. Database indices achieve O(1). The OMOP CDM already provides the standardised clinical vocabulary for defining semantic similarity. The FHIR resource layer already used across European hospital systems provides the clinical data schema. QIS does not require new infrastructure. It requires a routing layer plugged into existing infrastructure.
This is what is missing from the OHDSI network. This is what is missing from the EHDS architecture. This is the gap that will be discussed in Rotterdam and Berlin this week without being named.
The Mathematics of the Gap
The number is not abstract. It is calculable.
OHDSI Rotterdam 2026 draws from a network of 400+ data partners. Let N = 400.
N(N-1)/2 = 79,800 synthesis pathways.
That is the number of unique site-pair combinations where validated clinical intelligence could propagate bi-directionally across the network in real time. Each pathway represents an opportunity for a validated finding at one site to inform the clinical or research decisions of another.
Current OHDSI infrastructure activates, at best, a small fraction of those pathways through the publication cycle — and activates none of them in real time.
DMEA draws participants from the German MII network (36 university hospitals) and the broader European EHDS implementation ecosystem. German MII alone: N = 36, N(N-1)/2 = 630 synthesis pathways. None currently activated in real time.
EHDS at full 27-country deployment, if every member state builds a functioning secondary use node: N approaches thousands. The synthesis pathway count approaches millions. The current architecture has no mechanism for activating any of them.
QIS closes every one of those pathways simultaneously. The compute cost for each node remains O(log N) — logarithmic in the size of the network — while the intelligence synthesis capacity grows as N(N-1)/2 — quadratic. This is the phase change that neither OHDSI's current architecture nor EHDS's current specification captures.
Why This Week Is the Inflection Point
There is a specific reason this matters now, not in three years when EHDS reaches full deployment.
The architectural decisions being made this week — in Rotterdam, in Berlin, in the working sessions that precede and follow both conferences — will determine what European health data infrastructure looks like for the next decade. OMOP standardisation is largely complete. EHDS legal frameworks are in place. The implementation question is now: what does the routing layer look like?
If the answer is federated queries with centralised aggregation — the current OHDSI model — Europe builds a very sophisticated archive with a quarterly publication cycle and no real-time intelligence propagation.
If the answer includes a distributed outcome routing protocol — QIS — Europe builds a network where every validated clinical finding reaches every institution that can act on it, within minutes of validation, without any patient data crossing any boundary.
The difference is not incremental. It is quadratic.
The Routing Layer Is the Missing Piece
Both conferences will end with the same gap they started with, unless the question gets named.
The question is: Can a validated clinical finding produced at one node in a federated health data network reach every semantically similar node in that network, in real time, without transmitting patient data?
The answer, as of June 2025, is yes. The protocol is QIS. The discovery belongs to Christopher Thomas Trevethan.
OHDSI Rotterdam has the network. DMEA has the infrastructure conversation. QIS is the routing layer that connects them.
The gap has a name now. That is how closing it begins.
QIS Protocol — Quadratic Intelligence Swarm — was discovered by Christopher Thomas Trevethan in June 2025. 39 provisional patents have been filed. The architecture, technical specifications, and implementation pathways are documented at qisprotocol.com. The protocol is free for research, non-profit, and educational use.
Previous articles in this series: Why Federated Learning Has a Ceiling · The OHDSI Network Has 900 Million Patient Records — Here Is Why None of Them Talk to Each Other · DMEA 2026: Why German Health Data Infrastructure Needs Distributed Outcome Routing · The European Health Data Space Went Live on March 26 — Here Is the Routing Architecture It Needs
Top comments (0)