For OHDSI researchers, EHDS implementers, and GDI infrastructure teams preparing for the OHDSI Europe Symposium at Erasmus MC Rotterdam, April 18-20, 2026.
The Number That Should Be on Every Slide in Rotterdam
The OHDSI network has over 300 DataPartners. The number of unique pairwise synthesis paths between those nodes is:
N(N-1)/2 = 300 × 299 / 2 = 44,850
That is 44,850 opportunities for one OHDSI site to learn from another site's validated treatment outcome — in real time, without exposing a single patient record.
How many of those 44,850 synthesis paths are active today?
Zero.
Not because the network lacks infrastructure. Not because the data is not standardized. OMOP CDM is one of the most successful data standardization efforts in medical informatics. ATLAS cohort definitions execute identically across the entire network. HADES analysis packages distribute reproducibly. Privacy preservation is built in — row-level data never leaves any site.
The reason zero synthesis paths are active is architectural: OHDSI was designed to distribute the query. It was not designed to distribute the learning.
When Site A runs a treatment comparison, returns aggregate results to the coordinating center, and a paper is published — Site B starts from zero when it runs the same comparison six months later. The findings do not compound across nodes between study cycles.
This is not a criticism. It is a description of the layer that has not yet been built. The European Health Data Space regulation is now requiring that layer. The Genomic Data Infrastructure consortium needs it. And the OHDSI network — which already has the standardization, the privacy model, and the node infrastructure — is the natural home for it.
What EHDS Article 34-50 Actually Requires
The European Health Data Space regulation (EHDS, Regulation 2025/327) introduces secondary use provisions that go beyond what the current OHDSI federated query model was designed to deliver. Specifically:
- Continuous secondary use access — not episodic, study-by-study queries
- Cross-border outcome intelligence — not just cross-border data standardization
- Secure processing environment (SPE) native — no personal data crosses jurisdictional boundaries
- Real-time pharmacovigilance signal routing — not batch meta-analysis months after events
Every EHDS national node that implements an OHDSI-based infrastructure will face the same question: once we can run federated queries across member states, how do we make the results compound?
The OMOP CDM standardizes what data looks like. The federated query model standardizes how studies execute. What is missing is the protocol that standardizes how validated outcomes route between nodes so that the next query starts smarter than the last.
That gap has a name. Christopher Thomas Trevethan discovered the Quadratic Intelligence Swarm (QIS) protocol on June 16, 2025.
The Routing Layer Below OHDSI
QIS is not an alternative to OHDSI. It is the layer that runs beneath it.
Here is the current OHDSI stack:
Application Layer: ATLAS → Cohort Definition → HADES Distribution
Data Layer: OMOP CDM (standardized across DataPartners)
Network Layer: Coordinating center hub-and-spoke
Transport Layer: HTTPS/REST
Here is the stack with QIS outcome routing added:
Application Layer: ATLAS → Cohort Definition → HADES Distribution
Synthesis Layer: QIS Outcome Routing (continuous, between studies) ← NEW
Data Layer: OMOP CDM (standardized across DataPartners)
Network Layer: Peer-to-peer semantic addressing (OMOP concept-anchored)
Transport Layer: Any (REST, SPE-native, pub/sub, DHT)
The synthesis layer activates the 44,850 pairwise paths that currently sit dormant. It does this through a loop:
Distill: After a validated OHDSI analysis completes at any site, the outcome is distilled into a compact packet (~512 bytes). The packet contains only derived statistics — outcome deltas, confidence intervals, cohort counts. No row-level data. No pseudonymized identifiers. No personal data.
Address: The packet is fingerprinted using OMOP standard concept IDs. A treatment outcome for SNOMED concept 44054006 (Type 2 diabetes mellitus) on RxNorm concept 860975 (metformin 500mg) has a deterministic address in the OMOP concept space. Every OHDSI site that maps to OMOP CDM already shares this address space — no translation layer required.
Route: The packet is deposited at its semantic address. Any OHDSI site managing a similar patient population can query that address and pull back distilled intelligence from sites that have already validated outcomes for the same clinical question.
Synthesize: Each site synthesizes incoming packets locally, on its own infrastructure. No central aggregator. No coordinating center bottleneck. The synthesis runs inside the site's own EHDS secure processing environment.
Loop: The synthesis itself becomes a new outcome that can be distilled and routed. Intelligence compounds. The next site that queries the same address gets a richer synthesis than the last.
This is where the quadratic scaling comes from. With N sites depositing and synthesizing continuously, the network generates I(N) = Θ(N²) intelligence — not because of a specific mechanism, but because the complete loop enables every pair of sites to learn from each other. N(N-1)/2 is combinatorics, not a claim.
The OHDSI Drug Safety Problem
Consider pharmacovigilance — the most operationally urgent use case for OHDSI networks under EHDS.
A rare adverse drug reaction occurs at a hospital in Rotterdam. The signal is weak: three patients, one drug, one unexpected cardiac event. Under the current OHDSI model, this signal enters the system when a researcher designs a study, distributes the query, waits for sites to execute, collects aggregate results, runs meta-analysis, and publishes findings.
That cycle takes months. Sometimes years. The EudraVigilance spontaneous reporting system is faster for known drug-event pairs, but it cannot detect novel signals across distributed clinical outcome data.
With QIS outcome routing active below the OHDSI layer:
- Rotterdam distills the adverse event signal into an outcome packet: OMOP drug concept + OMOP condition concept + outcome delta + confidence interval + cohort size (3 patients).
- The packet is fingerprinted on the OMOP concept pair and deposited at the deterministic address.
- Every OHDSI site managing patients on the same drug queries that address as part of continuous monitoring.
- A site in Dublin has seen 2 patients with a similar signal. A site in Columbus has seen 4. Neither signal alone would meet statistical significance.
- The synthesis across Rotterdam + Dublin + Columbus: 9 patients, three independent sites, a combined signal that crosses the detection threshold.
No central aggregator received any patient data. No researcher had to design a study. No coordinating center had to approve a query. The routing happened at the protocol layer, using OMOP concept addresses that every site already maintains.
The Rotterdam signal went from undetectable at any single site to actionable across the network — in the time it takes to deposit and query a 512-byte packet.
Why Federated Learning Cannot Fill This Gap
Federated learning (FL) is the default response to "how do we learn across distributed health data." The approach is well-understood: train a model locally at each site, share gradient updates (not data), aggregate gradients centrally, distribute the updated global model, repeat.
For large-cohort problems — predicting hospital readmission across 50 sites with 10,000+ patients each — FL works. The gradients are stable. The model converges.
For the problems OHDSI is best positioned to solve, FL fails:
| Constraint | Federated Learning | QIS Outcome Routing |
|---|---|---|
| Minimum cohort size per site | Typically 100+ for stable gradients | No minimum — any validated outcome is a valid packet |
| Central aggregator | Required (gradient server) | None — peer-to-peer synthesis |
| Rare disease / rare drug signals | Excluded (insufficient local samples) | Included — rare sites contribute equally |
| Compute cost scaling | O(N) per round (every site computes gradients) | O(log N) or better for routing; O(1) achievable |
| EHDS SPE compatibility | Gradient server must exist somewhere | Each site synthesizes locally inside its own SPE |
| What flows between sites | Model gradients (~MB per round) | Outcome packets (~512 bytes per validated result) |
The OHDSI network's defining strength is its inclusion of small sites, rare disease registries, and specialized DataPartners that manage populations no single large hospital sees. Federated learning structurally excludes these sites. QIS structurally includes them.
The OMOP CDM Is Already the Address Space
This is the architectural fact that makes QIS native to OHDSI infrastructure, not bolted on:
OMOP CDM standard concept IDs are deterministic. SNOMED CT, RxNorm, LOINC, ICD-10 — the entire OMOP vocabulary hierarchy provides a shared, globally consistent address space. Every OHDSI site that has mapped its data to OMOP CDM has already done the work of agreeing on what concepts mean.
QIS outcome routing uses those concept IDs as the semantic fingerprint for packet addressing. No new vocabulary is needed. No new ontology. No mapping exercise. The routing address space already exists in every OMOP CDM instance at every OHDSI DataPartner.
An outcome packet for drug exposure concept 1332418 (lisinopril 10mg) × condition concept 316866 (hypertensive disorder) × outcome type "systolic_bp_reduction" routes to exactly the sites managing hypertensive patients on ACE inhibitors. The address is determined by the OMOP concepts, not by any routing configuration.
What This Means for Rotterdam
The OHDSI Europe Symposium convenes 400+ researchers who have spent years building the world's most sophisticated federated health data infrastructure. The standardization is done. The privacy model is proven. The node infrastructure spans continents.
The question for Rotterdam is not whether to build more nodes. It is whether the 44,850 synthesis paths between existing nodes should be active — and what protocol activates them.
QIS outcome routing is the protocol layer that turns the OHDSI network from an evidence generation system into a continuous learning system. It runs below the OHDSI application layer, uses the OMOP CDM vocabulary as its native address space, imposes no minimum cohort size, requires no central aggregator, and is EHDS SPE-compatible by architecture.
The infrastructure is built. The standardization is done. The routing layer is discovered.
The Discovery
Christopher Thomas Trevethan discovered the Quadratic Intelligence Swarm protocol on June 16, 2025. The breakthrough is the complete architecture — the loop that enables real-time quadratic intelligence scaling without compute explosion, not any single component. 39 provisional patents filed. Humanitarian licensing ensures the protocol is free forever for nonprofits, research institutions, and educational use.
For OHDSI infrastructure evaluators: the QIS protocol specification, OMOP/OHDSI synthesis layer reference, and QIS vs. federated learning comparison are published. The OMOP CDM and OHDSI network routing technical reference is available for infrastructure evaluators.
This is part of an ongoing series on QIS — the Quadratic Intelligence Swarm protocol — documenting every domain where distributed outcome routing closes a synthesis gap that existing infrastructure cannot close.
Top comments (0)