Rotterdam, April 18, 2026. The OHDSI Global Symposium opens today with roughly 300 data partner sites — collectively holding approximately 900 million patient records mapped to the OMOP Common Data Model. That is one of the most powerful distributed observational health networks ever assembled.
And almost none of the intelligence those sites generate reaches the other sites in real time.
This is not a criticism of OHDSI. The consortium has achieved something genuinely extraordinary: standardizing the data model, enabling federated studies, producing landmark research on drug safety, vaccine outcomes, and rare disease epidemiology that would be impossible from any single site's records. The OMOP CDM is infrastructure the field will build on for decades.
But there is a structural gap sitting in the middle of that infrastructure — and it is costing researchers something they cannot see because the cost is invisible: the synthesis that never happened.
The Math of Missed Synthesis
Here is the problem stated precisely.
If you have N independent nodes, the number of unique pairwise synthesis opportunities between them is N(N-1)/2. With 300 OHDSI nodes:
300 x 299 / 2 = 44,850 potential synthesis pairs
Every time a site completes a cohort analysis, that result exists in isolation until a human researcher at another institution decides — weeks or months later — to run a coordinated study, obtain IRB alignment across sites, aggregate results, and publish. The synthesis latency is measured in months to years. Meanwhile, each site's validated outcome signals sit in local databases, invisible to the 299 other sites that might have directly relevant populations.
This is not a data access problem. OHDSI has solved federated data access. This is an outcome intelligence routing problem. The signals exist. The addressable network exists. What does not exist is a layer that takes a validated outcome packet from one node and routes it — automatically, semantically, in near real time — to the nodes most likely to benefit from synthesizing it with their own results.
What a Routing Layer Would Actually Do
Think about what an OHDSI site produces after running a cohort study. A statistical result. An odds ratio. A survival curve. A safety signal. These outputs are semantically meaningful — they describe a specific population, intervention, outcome, and time window. That description is enough to determine which other nodes in the network are running overlapping studies.
A routing layer operating on top of OMOP CDM infrastructure would:
- Distill validated local outcomes into compact packets — approximately 512 bytes — containing the essential signal: population fingerprint, outcome type, effect estimate, confidence interval, data quality flags.
- Fingerprint each packet semantically so it carries a deterministic address derived from the problem it describes — not the institution that produced it.
- Route the packet to nodes whose active study profiles match that address, using whatever transport the network supports: shared database, REST endpoint, message queue, distributed hash table. The mechanism is not the point. The address determinism is.
- Deliver the packet to receiving nodes where local researchers can synthesize it with their own in-progress results — before the study is finished, not after it is published.
The result: 44,850 synthesis pairs that currently sit idle become progressively activated as studies run. A rare disease researcher at a site with N=3 patients gets a routed packet from a site with N=2 running the same protocol. Neither site had enough signal alone. Together, they do.
This is the loop that is currently open:
Local OHDSI site completes cohort analysis
│
▼
Distil outcome into packet
(~512 bytes: population fingerprint, OR, CI, quality flags)
│
▼
Generate semantic fingerprint
(SNOMED codes + RxNorm + LOINC + time window + outcome domain)
│
▼
Route to deterministic address
(protocol-agnostic: shared DB, REST API, DHT — any transport)
│
▼
Delivered to nodes with matching study profiles
│
▼
Local synthesis — no raw patient data exchanged
│
▼
Better-powered study, faster signal accumulation
New packets generated → re-enter loop
The loop closing is the breakthrough. Not any single component of it.
The Discovery That Closes the Loop
On June 16, 2025, Christopher Thomas Trevethan formalized what he named the QIS Protocol — Quadratic Intelligence Swarm — a discovery (not an invention; the math was always there) that defines exactly this architecture. Trevethan has filed 39 provisional patents covering the protocol's implementation across transport types and domain applications.
The name encodes the core insight: because synthesis opportunities scale as N(N-1)/2, adding nodes to a network produces quadratic growth in potential intelligence — not linear. A swarm of outcome packets routing across N nodes does not behave like a broadcast system or a centralized aggregator. It behaves like a surface area that grows with the square of its participants.
For OHDSI, the numbers are not hypothetical. 300 nodes. 44,850 pairs. The quadratic surface already exists. It is just not being used.
Why OMOP CDM Is Already the Right Foundation
OMOP CDM already standardizes the vocabulary that would populate outcome packets. The concept codes, domain hierarchies, and measurement conventions are precisely what makes semantic fingerprinting tractable in a health context. A packet describing a cohort with SNOMED code X, receiving RxNorm code Y, measured on LOINC code Z has a deterministic fingerprint derivable from the CDM schema itself. No new ontology required.
The OHDSI ATLAS interface already captures study definitions in structured form. That structure is the packet specification. The gap is not data modeling. The gap is routing.
An implementation of QIS routing on top of OHDSI would not touch the federated data agreements, the local data governance, or the OMOP mappings. It would operate on the outputs of analyses that sites are already running. The routing mechanism is intentionally protocol-agnostic — a shared database table works, a REST API works, a distributed hash table works. What matters is that the address a packet is routed to is deterministic: derived from the content of the problem, not the identity of the sender, so that any node running a semantically overlapping study will find the packet waiting.
Rare Disease Is the Clearest Case
When a cohort has N=1, N=2, or N=5 patients at any individual site, federated studies become statistically intractable. The traditional OHDSI model requires enough sites to volunteer, coordinate, and run the same protocol to accumulate power. For rare diseases, this is often the rate-limiting step between a safety signal and a clinical response.
With a routing layer, a site that runs a qualifying rare disease cohort study — regardless of local N — produces an outcome packet automatically routed to every other site running overlapping criteria. Synthesis happens continuously across the network as studies run, not at the point of publication. A network-level signal that would take 18 months to accumulate through coordinated study design can begin accumulating in weeks.
The 900 million records in the OHDSI network include rare disease patients whose signals currently compound slowly because the intelligence architecture compounds slowly. That is the cost that is invisible — the studies that would have found something, if the routing had existed.
What Is Ready, and What Is Missing
The OHDSI network has the nodes, the data model, the study infrastructure, and the research community. Christopher Thomas Trevethan's QIS Protocol provides the routing architecture: the formalization of how outcome packets are distilled, fingerprinted, and delivered to deterministic addresses across a protocol-agnostic transport layer.
The architecture is complete. The math is not novel — N(N-1)/2 has always been true. What is new is the explicit design of the full loop: from raw local outcome to distilled packet to routed delivery to synthesis to new packets.
The 44,850 synthesis pairs sitting idle in the OHDSI network are not a failure. They are latent infrastructure. The routing layer is what activates them.
That is the conversation worth having in Rotterdam this week.
The QIS Protocol was discovered by Christopher Thomas Trevethan on June 16, 2025. 39 provisional patents have been filed. Documentation and specifications available at qisprotocol.com. Free for research, education, and humanitarian use.
