Rory | QIS PROTOCOL

Posted on Apr 12

PCORnet Routes Queries. QIS Routes What PCORnet Learns.

#healthinformatics #federatedresearch #distributedcomputing #clinicalresearch

The Gap No One Talks About

You run a comparative effectiveness study across 20 PCORnet sites. Forty-five thousand patients. Drug A versus Drug B in adults over 60 with type 2 diabetes and stage 3 CKD. The federated query goes out through PopMedNet. Aggregated counts come back. You compute the effect estimate, the confidence interval, the subgroup hazard ratios. You publish.

Six months later, a site in the network — one that contributed 3,200 of those 45,000 patients — launches a slightly different study: the same drugs, but in patients with concurrent heart failure. Their analysts write the query. PopMedNet routes it. The process starts over from scratch.

The validated outcome from your completed study — the one that took 18 months and $2.4 million in PCORI funding to produce — does not flow back to that site as a synthesizable finding. It lives in a PDF. The network learned something. The network does not know it learned it.

That is the gap. PCORnet routes queries. Nothing currently routes what PCORnet learns.

What PCORnet Actually Does (And Does Well)

PCORnet — the Patient-Centered Outcomes Research Network, funded by PCORI — is a federated research infrastructure spanning approximately 160 sites, including 9 Clinical Research Networks and 13 Health Plan Research Networks, with access to data on more than 160 million patients.

The architecture is deliberate and correct: raw patient data never moves. Queries travel outward from a coordinating center to participating sites. Sites run the query against their local data, return aggregated counts and statistics, and the coordinating analysis happens centrally. The PCORnet Common Data Model (CDM v6) ensures that ICD-10 diagnoses, NDC drug codes, LOINC lab values, and demographic variables mean the same thing at every site before the query even leaves the hub.

The FDA Sentinel system uses the same federated query architecture across 13 data partners representing more than 49 million patients. PopMedNet is the distributed query tool powering both.

This is not a workaround. This is the correct solution to the privacy constraint: let the data stay where it lives, and move the questions to the data.

The problem is that moving questions to data is only half the loop.

The Missing Layer: Outcome Routing

When a PCORnet study concludes, the validated finding — the effect estimate and its uncertainty — exists only in the study's output. There is no protocol for that finding to flow back into the network as an addressable, synthesizable artifact. The next study that asks a related question inherits none of that prior learning automatically.

How many synthesis pathways does this leave on the table?

PCORnet has approximately 160 sites. The number of unique site-to-site synthesis pathways is:

N(N-1)/2 = 160 × 159 / 2 = 12,720 potential pathways

FDA Sentinel adds 13 data partners:

13 × 12 / 2 = 78 additional pathways

Combined across PCORnet and Sentinel:

173 × 172 / 2 = 14,878 potential synthesis pathways

Currently, the number of those pathways that operate with any systematic outcome routing: zero.

This is the structural gap that Quadratic Intelligence Swarm (QIS) — a distributed outcome routing protocol discovered by Christopher Thomas Trevethan on June 16, 2025 — is designed to close. Not as a replacement for PCORnet's federated query layer. As the complementary return layer that PCORnet's architecture has never had.

What QIS Adds

QIS is a routing protocol. Its breakthrough is not any single component — not the addressing scheme, not the transport mechanism, not the outcome representation. The breakthrough is the complete loop: a validated finding emits as a structured outcome packet, gets addressed semantically, routes to peer nodes that have expressed compatible research intent, and arrives as synthesizable intelligence rather than a citation in a reference list.

The architecture supports multiple routing transports. A registry-based implementation — straightforward in a closed network like PCORnet where every site is already enrolled and addressable — achieves O(1) per-node routing cost using indexed lookup against the CDM's existing vocabulary. In open or partially-known networks, distributed hash table routing provides an upper bound of O(log N). The protocol does not specify the transport; it specifies what must be preserved: the semantic address, the outcome payload, and the provenance chain.

Christopher Thomas Trevethan has filed 39 provisional patents covering the QIS architecture.

PCORnet CDM as Native Semantic Addressing

Here is where PCORnet's existing infrastructure becomes an asset rather than an integration challenge.

QIS outcome packets require a semantic fingerprint — a machine-readable description of what the finding is about, sufficient for peer nodes to evaluate relevance without transmitting raw data. PCORnet CDM v6 already provides exactly this vocabulary.

A comparative effectiveness finding from a completed PCORnet study might emit as an outcome packet with a semantic fingerprint like:

{
  "condition": "ICD-10:E11.65",
  "condition_modifier": "ICD-10:N18.3",
  "intervention_a": "NDC:00169-4060-12",
  "intervention_b": "NDC:00310-0272-90",
  "outcome_measure": "LOINC:62238-1",
  "effect_estimate": -0.41,
  "confidence_interval_95": [-0.57, -0.24],
  "population_n": 45000,
  "site_count": 20,
  "study_design": "active_comparator_new_user"
}

No new vocabulary. No mapping layer. No ontology translation. PCORnet sites already speak this language. A QIS outcome packet for a PCORnet finding is just a structured envelope around what PCORnet researchers already write in their methods sections — made addressable, routable, and synthesizable by peer sites without human intermediation.

The Rare Condition Problem, Solved Differently

PCORnet's multi-site scale is its most important feature and its most significant limitation for rare disease research simultaneously.

A site contributing 12 patients with a rare autoimmune condition — say, anti-MDA5 dermatomyositis — can return anonymized counts to a PCORnet query. What it cannot do is run independent comparative effectiveness research. N=12 does not support the statistical mass required for a federated analysis to return meaningful estimates. Under federated learning approaches, N=12 is excluded entirely: per-site gradient computation requires sufficient statistical mass, and rare condition sites simply cannot contribute.

QIS emits one outcome packet per validated finding regardless of N. A site with 12 patients that has observed a clinically validated treatment response — documented, IRB-approved, methodologically sound — emits that observation as a single structured packet. Three such sites, each with 12 patients studying the same condition, produce 36 patient-observations worth of outcome intelligence through QIS routing without running a new coordinated study.

The outcome packet carries its provenance. Receiving sites evaluate confidence based on N, study design, and site metadata — not because QIS enforces a quality threshold, but because the information is available for the receiving analyst to weight as they see fit. The routing protocol does not add a reputation layer or scoring mechanism. It moves the finding. What the receiving site does with it remains a scientific judgment.

Sequential, Not Competing

It is worth being precise about the architecture relationship.

PCORnet's federated query layer is upstream. A study is designed, a query is constructed and dispatched through PopMedNet, sites return aggregated results, and a validated finding is produced. QIS is downstream of that completion event. The validated finding — which exists because PCORnet's federated query architecture worked — is the input to QIS outcome routing.

These are two sequential layers of a complete research feedback loop:

Query routing (PCORnet + PopMedNet): moves questions to data, returns aggregated statistics
Outcome routing (QIS): moves validated findings back to peer sites as synthesizable intelligence

Neither layer substitutes for the other. PCORnet without QIS produces findings that are siloed in reports. QIS without PCORnet has no federated infrastructure to generate the validated findings in the first place. Together, they close the loop that clinical outcomes research has had open since the first multi-site federated study was completed.

The 14,878 synthesis pathways sitting at zero activation across PCORnet and FDA Sentinel represent the gap between what has been built and what is possible. The vocabulary to address those pathways — ICD-10, NDC, LOINC — is already deployed at every participating site.

The protocol to route through them has been discovered. The integration surface is already there.

DEV Community