A reproducible OMOP cohort, shared as a method, not as data.
Multi-center studies and open science stall on one problem: the data cannot travel. Polnor flips it. Each center keeps its own data in its own cloud, on an HDS-certified host, and what you share is a standardized OMOP cohort definition and analysis that runs anywhere. The heavy data preparation runs itself, so your researchers stay on the scientific question.
One diabetes cohort definition, run at every center.
You write a cohort once, against the OMOP CDM. The same definition runs at each participating center on that center's own data. You compare counts across sites, never rows. Here is the path, and how little of it is manual.
| site | patients | index events | def. version |
|---|---|---|---|
| center A | 18,540 | 21,003 | v1.2 |
| center B | 12,117 | 13,880 | v1.2 |
| center C | 9,402 | 10,644 | v1.2 |
Illustrative interface and figures, shown to explain the workflow. Not results from a Polnor customer.
Everything up to the cohort is automated.
By the time you write that definition, the hard part is done: the platform has ingested, de-identified, standardized and quality-checked the data at each center, to the same schema. You start where the science starts.
Center data in, at the FHIR standard
EHR extracts and clinical feeds from each center are pulled and normalized to FHIR R4 automatically, one typed table per resource. No bespoke ETL per site, so every center reaches the same starting point.
| resource | rows | status |
|---|---|---|
| ConditionOccurrence | 3,880,214 | ✓ auto |
| Observation | 9,410,776 | ✓ auto |
| DrugExposure | 5,201,338 | ✓ auto |
GDPR de-identification and OMOP mapping, standardized for reproducibility
Direct identifiers are masked and dates shifted by preset, then the data is converted to the OMOP CDM with OHDSI concept mapping and clinical quality control. Because every center lands on the same standard concepts, a cohort written once behaves the same everywhere.
| source | → OMOP concept | std |
|---|---|---|
| icd10:E11 | Type 2 diabetes mellitus | S |
| loinc:4548-4 | Hemoglobin A1c | S |
| atc:A10BA02 | Metformin | S |
Cohorts, notebooks and models, where the science is
This is the part your researchers own: define the cohort against the OMOP tables, characterize it in a notebook, and, if the study calls for it, train and track a model with MLflow. Everything runs on compute next to each center's data, and nothing is exported. What you circulate to collaborators is the versioned definition.
| run | metric | version |
|---|---|---|
| t2dm-characterization | logged | v1.2 |
| risk-baseline | 0.81 AUC | v1.2 |
The platform does the plumbing. Your team does the science.
The line is deliberate: the repetitive, error-prone, compliance-heavy work runs itself at every center, so your researchers spend their time on judgment, not preparation.
Science others can rerun, with a trail behind it.
Cohort definitions are versioned and their lineage recorded, every access to patient data is logged, PHI is classified, and export follows EHDS-aligned manifests. Because the data stays in each center's HDS-certified cloud and we never train our models on it, collaborators reproduce your method without ever touching your patients' records.
From raw records to a portable cohort, in every center's cloud.
Book a 30-minute demo. We run the automated pipeline on a case that looks like yours, and show you exactly where the automation ends and the science begins.