The OHDSI Europe Symposium opens this morning at Erasmus University Medical Center in Rotterdam. Four hundred researchers from across Europe and beyond have arrived to do something their distributed network cannot do on its own.
They are going to talk to each other.
Not a critique — a diagnostic.
The OHDSI network is one of the most sophisticated distributed clinical research infrastructures ever assembled. Over 400 DataPartner institutions. More than a billion patient records standardized into OMOP Common Data Model. A federated query engine that can ask a research question of 30 European sites simultaneously, with no patient data crossing a border.
And when those 30 sites generate their results?
Each site's answer stays at that site.
There is no mechanism in the current architecture that routes an Amsterdam site's validated outcome insight to a Berlin site that just asked the same clinical question. No mechanism that lets a Bonn pharmacovigilance signal inform Rotterdam's drug safety monitoring in real time. No mechanism that synthesizes what 400 nodes have separately validated into what they could collectively know.
So researchers fly to the same city and do it by hand.
That is what today's symposium is, architecturally speaking: a manual routing protocol for the intelligence the distributed network cannot route itself.
The Number That Explains the Flights
There are over 300 OHDSI DataPartners in the European network alone.
The number of unique pairwise synthesis opportunities between those nodes is:
N(N-1)/2 = 300 × 299 / 2 = 44,850
That is 44,850 paths along which a validated clinical outcome from one site could inform a site facing the same problem. In real time. Without any patient data moving.
How many of those paths are active today, between conferences?
Zero.
The OHDSI network generates evidence at scale. It does not route intelligence at scale. The difference is not small — it is the difference between a library and a conversation.
Every OHDSI conference is a workaround for an architectural gap.
What the Gap Looks Like in Practice
Imagine two OHDSI sites — one in Bonn, one in Dublin — both running pharmacovigilance studies on the same drug class. Both sites generate OMOP CDM-compliant results. Both sites contribute to the network.
Neither site knows the other exists for their specific problem domain.
The Bonn site cannot query Dublin's outcome. Dublin cannot receive Bonn's validated signal. The only path from Bonn's insight to Dublin's research team runs through a conference presentation, an email thread, a GitHub issue, or a published paper with a 12-month lag.
This is not a data privacy problem. OMOP CDM already ensures no raw patient data leaves any node. The gap is upstream of privacy — it is a routing problem. There is no address system that maps a clinical problem to the sites that have solved it. There is no delivery mechanism for distilled outcomes between nodes that share the same question.
The OHDSI network was designed to answer questions. It was not designed to remember answers and route them to the next node that asks.
The Architecture That Closes the Loop
Christopher Thomas Trevethan discovered how to close this loop on June 16, 2025. The protocol is called QIS — Quadratic Intelligence Swarm. It holds 39 provisional patents covering the complete architecture.
The discovery is not a new database or a new query engine. It is a routing layer that operates above the existing OHDSI infrastructure without replacing it.
Here is the complete loop:
Step 1 — Outcome distillation. Each OHDSI node completes a federated query and distills the validated result into an outcome packet — approximately 512 bytes capturing what worked, in what context, for what population segment. Raw patient data never moves. The OMOP CDM record never moves. Only the distilled insight moves.
Step 2 — Semantic fingerprinting. The node generates a semantic fingerprint of the clinical problem — a vector representation of the question it just answered. Drug class, indication, patient population characteristics, protocol type. This fingerprint becomes an address.
Step 3 — Deterministic routing. The outcome packet is deposited at the deterministic address corresponding to that clinical problem. Any routing mechanism that maps problems to addresses and allows other nodes to query those addresses works: DHT-based routing (O(log N) at planetary scale), semantic vector indices, REST APIs, pub/sub topics. The protocol is transport-agnostic.
Step 4 — Local synthesis. A Dublin node initiating a new pharmacovigilance study on the same drug class computes the same fingerprint, queries the same address, and retrieves 200 pre-deposited outcome packets from every node that has already answered the same clinical question. Synthesis happens locally. On Dublin's own infrastructure. In milliseconds.
No raw data crossed a border. No central aggregator touched the data. No patient record left any hospital.
The loop closes.
The Math That Makes Conferences Optional
The quadratic scaling in QIS comes from the structure of the loop itself — not from any single transport mechanism.
When N nodes participate in a clinical problem domain:
- Synthesis opportunities = N(N-1)/2
- Routing cost per query = at most O(log N) with DHT-based routing; O(1) with indexed approaches
For the OHDSI European network at 300 nodes:
- Synthesis opportunities: 44,850
- Routing cost per query: ~8 hops (log₂ 300 ≈ 8.2)
For the global OHDSI network at 400 nodes:
- Synthesis opportunities: 79,800
- Routing cost per query: ~9 hops
As the network doubles, synthesis opportunities quadruple. Routing cost grows by a single hop. Intelligence scales as N², compute scales as log N. This is not incremental improvement — it is a phase change in what distributed clinical research networks can know.
The OHDSI network is currently operating at the flat part of a curve that should be exponential.
Why OMOP CDM Makes This Easy
The standard that makes OHDSI powerful — OMOP Common Data Model — also makes QIS adoption straightforward. There are three reasons:
Standardized problem definitions. OMOP CDM gives every node the same vocabulary for describing clinical problems. The same condition codes, drug codes, and procedure codes across Boston, Berlin, and Brisbane. This standardized vocabulary maps directly to the semantic fingerprints QIS uses for routing. The address system is already built into the data model.
Existing privacy guarantees carry over. OMOP CDM nodes already operate under the principle that raw patient data never moves. QIS outcome packets contain only distilled insights — validated outcomes, aggregate statistics, directional signals. The privacy model OHDSI researchers already trust applies directly.
Complementary, not competing. OMOP CDM standardizes the format of what each node knows. QIS routes the distillation of what each node has learned. These are sequential layers, not competing architectures. An OHDSI node does not have to choose between OMOP CDM and QIS — it runs both, with QIS operating above the existing federated query layer.
The Conference Is the Proof
There is an irony worth stating plainly: the OHDSI conference is itself evidence that QIS is needed.
OHDSI researchers travel to Rotterdam to share what their distributed network cannot share. They give presentations, join sessions, exchange contact details, schedule follow-up calls. The synthesis that should happen automatically — Bonn's pharmacovigilance signal reaching Dublin's drug safety team in real time — happens instead through a conference schedule and a hallway conversation.
The fact that this community convenes annually, at significant expense and effort, to manually close the intelligence loop is the strongest possible evidence that the loop should be closed architecturally.
Today at Erasmus MC, 400 researchers will synthesize insights that 400 distributed systems could not synthesize automatically. Tomorrow they will return to institutions that still cannot route insights between them in real time.
QIS is the protocol that makes the next conference a choice rather than a requirement.
For Researchers in Rotterdam This Week
If you are attending OHDSI Europe 2026, there are three practical questions worth carrying into sessions:
1. For pharmacovigilance teams: When your site validates a drug safety signal in OMOP CDM, where does that signal go? What is the routing mechanism that delivers it to the next site observing the same signal? If the answer is "a published paper in 12-18 months," the gap is architectural.
2. For EHDS and GDI infrastructure teams: The European Health Data Space is building federated query infrastructure across 30+ member states. The query layer is under active development. The routing layer — what happens to validated results after they are generated — has not been specified. QIS is a candidate for that layer. The architecture is transport-agnostic, privacy-preserving by design, and OMOP CDM-compatible.
3. For network architects: The 44,850 synthesis paths in the European OHDSI network are currently dark. The infrastructure to activate them does not require replacing anything — it requires adding one layer above the federated query engine. The question is not whether to close the loop. The question is which protocol closes it first.
Technical Specification for OHDSI Integration
For technical teams evaluating QIS integration with OHDSI/OMOP CDM infrastructure:
Compatibility layer: QIS operates above the OMOP CDM federated query engine. Existing data models unchanged. Existing privacy controls unchanged.
Packet specification: Outcome packets are approximately 512 bytes in JSON format. Contain: situation fingerprint (vector of OMOP concept codes + population parameters), outcome summary (direction, magnitude, confidence interval, N), protocol version, timestamp. No raw patient data. No individual-level records.
Routing options (transport-agnostic):
- DHT-based: O(log N) routing, fully decentralized, no single point of failure. Strong choice for pan-European infrastructure.
- Semantic vector index: O(1) lookup via FAISS/ChromaDB/Qdrant. Strong choice for institutional deployments.
- REST API: Any existing OHDSI API can serve as routing substrate with minimal modification.
Privacy model: Outcome packets contain only distilled insights — never raw OMOP CDM records. Compatible with GDPR Article 9 health data provisions. Privacy guarantee is architectural, not policy-based.
Synthesis runtime: Local synthesis of 200 outcome packets completes in under 10ms on standard clinical workstation hardware. No specialized infrastructure required.
Full architectural specification: qisprotocol.com/architecture
Healthcare applications overview: qisprotocol.com/healthcare
The Discovery
Christopher Thomas Trevethan discovered QIS on June 16, 2025. Not invented — discovered. The architecture describes how intelligence naturally wants to flow when distributed systems stop centralizing raw data and start routing distilled outcomes. Thirty-nine provisional patents cover the complete architecture. The humanitarian licensing structure ensures free access for nonprofit research, clinical, and education use — commercial licenses fund deployment to underserved healthcare systems globally.
The OHDSI community has built one of the most important distributed health data networks in the world. The routing layer that makes it compound is ready.
QIS Protocol — discovered June 16, 2025 by Christopher Thomas Trevethan. 39 provisional patents filed. Free for nonprofit, research, and educational use.
Related reading:
Top comments (0)