If you are building federated health data infrastructure in 2026 — whether inside the European Health Data Space regulatory framework, the Genomic Data Infrastructure consortium, or the broader OHDSI network — you already know what works and what does not.
What works: OMOP CDM vocabulary standardization. SNOMED CT, RxNorm, LOINC, ICD-10 mapped across hundreds of DataPartners. ATLAS cohort definitions that execute identically across institutions. ACHILLES characterization. HADES distribution. These are genuine engineering achievements that took years of sustained effort by the OHDSI Collaborative.
What does not work — yet — is continuous intelligence accumulation across nodes between study cycles. And that is not a criticism. It is a description of what the current architecture was not designed to do.
This article describes how QIS outcome routing operates as a protocol layer below the OHDSI application layer — complementing, not replacing, the infrastructure you have already built.
The Missing Layer
Consider the current OHDSI study architecture:
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 (typically)
Queries flow from coordinating center to DataPartners. Aggregate results flow back. The coordinating center synthesizes. A paper is published. The next study starts fresh.
Now consider what happens between studies. A DataPartner in Dublin running renal outcomes analysis on ACE inhibitors generates findings. A DataPartner in Columbus running the same drug class against cardiovascular endpoints generates complementary findings. Under current OHDSI architecture, these two findings meet only if a future coordinating center designs a study that requests both. The nodes themselves do not learn from each other.
This is not a flaw. OHDSI was designed for reproducible, privacy-preserving distributed research — and it achieves that. But the European Health Data Space regulation (EHDS, Regulation 2025/327) and the Genomic Data Infrastructure (GDI) consortium are now asking for something more: continuous secondary use of health data across national boundaries, with governance staying local.
That requires a routing protocol.
QIS Outcome Routing: The Protocol Layer
Christopher Thomas Trevethan discovered the Quadratic Intelligence Swarm (QIS) protocol on June 16, 2025. QIS is not a database, not a query engine, not a federated learning framework. It is a routing protocol for intelligence — operating below the application layer.
Here is how it maps to existing OHDSI infrastructure:
Application Layer: ATLAS / HADES / ACHILLES (unchanged)
Data Layer: OMOP CDM (unchanged)
───────────────────────────────────────────────────────────
QIS Protocol Layer: Outcome routing, semantic addressing,
~512-byte insight packets
───────────────────────────────────────────────────────────
Transport Layer: DHT, REST, pub/sub, database — any
The QIS layer sits between what you already have (OMOP CDM + OHDSI tooling) and the transport mechanism. It does three things:
1. Outcome Packets, Not Raw Data
When a DataPartner completes an analysis — whether an ATLAS cohort study, an ACHILLES characterization run, or a HADES-distributed analysis — the findings can be distilled into ~512-byte outcome packets. These packets contain structured insights: what was studied, what was found, confidence level, semantic address. No patient-level data. No identifiable records. The privacy boundary is architectural, not policy-dependent.
For OHDSI DataPartners already returning aggregate statistics to coordinating centers, this is a natural extension. The aggregate findings that currently flow to one coordinating center can instead be expressed as routable outcome packets that flow to any semantically relevant node in the network.
2. Semantic Addressing via OMOP Vocabularies
QIS routes outcome packets using semantic addresses — and OHDSI has already done the hard work of building those addresses. SNOMED CT concept IDs, RxNorm drug identifiers, LOINC measurement codes — these are not just data model fields. They are natural semantic addresses for QIS routing.
An outcome packet tagged with SNOMED 73211009 (diabetes mellitus) and RxNorm 1596450 (metformin extended-release) automatically routes to every node in the network that has expressed interest in those semantic coordinates. No coordinating center required. No study design phase. The routing is continuous.
3. Quadratic Intelligence Accumulation
This is the mathematical core. In an OHDSI network of N DataPartners, the current architecture accumulates intelligence linearly — each node contributes to the coordinating center, producing O(N) synthesis per study cycle.
QIS enables N(N-1)/2 pairwise synthesis paths. Every node's outcome packets can interact with every other node's outcome packets. For a 50-node OHDSI network, that is not 50 contributions — it is 1,225 potential synthesis paths. For the full OHDSI network of 300+ DataPartners, that is over 44,850 synthesis paths.
This is not a marketing number. It is combinatorics. N(N-1)/2 is the number of unique pairs in a set of N elements. The intelligence scales as I(N) = Θ(N²) while the routing cost per packet is O(log N) or better — O(1) when using indexed databases, pub/sub systems, or REST APIs as the transport layer.
Why This Matters for GDI and EHDS
The Genomic Data Infrastructure consortium connects national genomic data repositories across 20+ European countries. The European Health Data Space regulation mandates secondary use of health data across borders while maintaining local governance. Both face the same structural challenge: how do you enable cross-border intelligence without centralizing data or authority?
Current approaches lean toward federated learning — training models locally and aggregating gradients centrally. This works for a specific class of problems (model training) but does not address the broader challenge of continuous outcome intelligence flowing across the network.
QIS outcome routing addresses this directly:
- Governance stays local. Each DataPartner decides what outcome packets to generate and share. The protocol does not require a central authority.
- Privacy is architectural. ~512-byte outcome packets carry structured findings, not patient data. There is nothing to re-identify because the granular data never leaves the node.
- Transport is agnostic. QIS routes over whatever infrastructure you already have. REST APIs? Fine. Message queues? Fine. A shared database? Fine. The protocol does not prescribe the transport — it prescribes the routing semantics.
- Existing standards are preserved. OMOP CDM vocabularies become the semantic addressing scheme. ATLAS cohort definitions can trigger outcome packet generation. ACHILLES characterization data can feed into routing decisions. Nothing is replaced.
The Three Elections Applied to OHDSI
QIS operates through what Trevethan calls Three Elections — a structural framework for how intelligence accumulates without central coordination:
The Hiring Election: In an OHDSI context, this is vocabulary alignment. SNOMED CT, RxNorm, LOINC — these controlled vocabularies define what "similar" means across the network. When two DataPartners use the same OMOP concept IDs, they are participating in the same semantic space. The OHDSI community's years of vocabulary curation work is, in QIS terms, a completed Hiring Election at global scale.
The Math Election: Outcomes accumulate as evidence. When 50 DataPartners each generate outcome packets on metformin efficacy in Type 2 diabetes, those packets do not vote. They accumulate mathematically — weighted by sample size, confidence interval, population characteristics. The network's intelligence on that topic grows quadratically with each contributing node. Byzantine fault tolerance is inherent: a single node publishing anomalous results is statistically absorbed, not voted out.
The Darwinism Election: Networks compete. An OHDSI sub-network running QIS outcome routing will demonstrably outperform one that does not, because its nodes learn from each other continuously rather than episodically. This is not a theoretical claim — it is a structural property of quadratic vs. linear intelligence accumulation.
What a 50-Node OHDSI QIS Deployment Looks Like
Consider an OHDSI sub-network: 50 DataPartners across Europe, each holding OMOP CDM data, each running standard OHDSI tooling.
Without QIS: Each study produces 50 aggregate contributions to a coordinating center. Between studies, the network is silent. Intelligence accumulates in published papers that future researchers may or may not find.
With QIS outcome routing at the protocol layer:
- Each DataPartner generates outcome packets after every analysis
- Packets route via OMOP vocabulary semantic addresses to relevant nodes
- 50 nodes produce 1,225 pairwise synthesis paths
- Intelligence accumulates continuously, not per-study-cycle
- A Dublin DataPartner's renal findings automatically reach a Columbus DataPartner studying the same drug class
- No coordinating center bottleneck
- No study design phase required for cross-node learning
The application layer — ATLAS, HADES, ACHILLES — continues to function exactly as before. QIS operates below it, enabling a class of intelligence accumulation that the current architecture structurally cannot support.
For Researchers Evaluating This
QIS is protected by 39 provisional patents under a humanitarian licensing model: free for nonprofits, research institutions, and educational use forever. Commercial deployments fund humanitarian deployment in underserved regions.
The protocol is transport-agnostic. If your GDI or OHDSI infrastructure already has REST endpoints, message queues, or shared databases, QIS routes over them. No new transport infrastructure required.
Christopher Thomas Trevethan discovered QIS on June 16, 2025. The architecture is documented extensively. The math is verifiable. N(N-1)/2 is combinatorics, not a claim.
If you are building the routing layer for European federated health data, the question is not whether continuous cross-node intelligence is valuable — EHDS regulation assumes it is. The question is what protocol enables it without centralizing data, governance, or trust.
QIS is that protocol.
Christopher Thomas Trevethan discovered QIS on June 16, 2025. 39 provisional patents protect the humanitarian licensing model: free for nonprofits, research, and education forever.
Top comments (0)