Your clinical data, ready for the cohort and for regulatory evidence.
Test cohort feasibility before opening a trial, produce trial-grade Real-World Evidence, and cross-link lab, clinical and omics data. Column-level GDPR de-identification, open FHIR/OMOP standards, audit-ready traceability, and patient data that never leaves your HDS-certified cloud.
What's holding you back today.
Building a cohort takes months
Testing cohort feasibility against a real-world base before opening sites removes a large share of the recruitment risk, provided your data is unified and queryable in place.
Siloed, non-interoperable data
CTMS, EDC, clinical records, lab results, real-world data: manual reconciliation slows every decision. You need a single clinical, lab and omics foundation, without copying the data outside your perimeter.
Trial-grade RWE is now expected
The EMA, FDA and PMDA increasingly require Real-World Evidence to confirm efficacy and safety in real-world conditions. That means a pre-specified protocol, an audit trail and traceable provenance: a governed pipeline, not a one-off export.
De-identify at scale, and prove it
GDPR's data minimisation and purpose limitation require reproducible, auditable de-identification, with the guarantee that patient data never leaves the HDS-certified perimeter.
Cross-link lab, clinical and omics data
Patient stratification and biomarker discovery require integrating genomics, proteomics and clinical data. Multimodal integration closes the blind spots of the "one mutation, one test" approach, but stumbles on data harmonisation.
What's shifting in your world.
De-identified RWE enters regulatory decision-making
The FDA finalised its Real-World Evidence guidance for devices in late 2025 and accepts RWE without requiring individually identifiable data. De-identified real-world data becomes admissible in a submission, exactly what a FHIR-to-OMOP pipeline produces.
EHDS opens up secondary use of data
The EHDS Regulation (EU) 2025/327 entered into force in March 2025; secondary use (research, AI training) will apply in phases, under permit and pseudonymisation. Structuring your data now in de-identified OMOP is a regulatory advantage.
Federated learning: collaborate without centralising
The MELLODDY pharma consortium (10 competing labs) demonstrated drug discovery through federated learning, with each partner keeping its data behind its own firewall. This is the health-blind logic: training on more patients without ever moving the data.
Foundation models in digital pathology
Whole-slide foundation models (e.g. Virchow, Paige/MSKCC, published in Nature Medicine) provide a reusable backbone to fine-tune for a biomarker or a rare cancer, where annotated data is scarce. But it still requires cross-linking slides, omics and clinical data at scale.
Concretely, with Polnor.
Test a cohort before opening a trial
Ingest FHIR from your partner sites, flatten it, then query "how many patients with type 2 diabetes, HbA1c > 8 %, no insulin, over 24 months" to scope your inclusion criteria, without exporting the data.
Train and serve a risk model
From cohort to feature store, all the way to a complication-risk model (tracked in MLflow) deployed as a scoring endpoint, within your HDS perimeter and without the control plane ever seeing the data.
Produce trial-grade RWE
A reproducible pipeline: column-level de-identification, LOINC quality checks, OMOP conversion, access log and audit trail. The traceability dossier the EMA and FDA expect, replayable at the next safety signal.
Cross-link lab, clinical and omics data
Unify laboratory results (LOINC), clinical data and omics tables in a single lakehouse to identify sub-populations and biomarker candidates, without manual reconciliation.
Build an external control arm
Use a curated, versioned real-world cohort as an external comparator for a single-arm trial, then generate an EHDS-compliant export manifest, de-identified and traced end to end.
Reconcile a patient across hospitals (EMPI/INS)
Match the records of the same patient seen at multiple sites via EMPI/INS to deduplicate a multi-site cohort, without ever exposing the identity to the control plane.
From a safety signal to a regulatory RWE study.
A lab markets an antidiabetic; the authority asks it to document cardiovascular risk in real-world conditions. Here's how it plays out, without any patient data leaving its cloud.
Connection & ingestion
The FHIR servers of the partner hospitals are connected; a Bulk $export pull feeds the lab's S3 bucket, at the HDS-certified host. Nothing leaves.
Quality & structuring
FHIR is flattened into typed tables; the quality check verifies that HbA1c, LDL and creatinine are in consistent LOINC units and flags outliers before analysis.
De-identification & identity
Column-level de-identification (identifiers masked, dates shifted). EMPI/INS reconciles the same patient seen at multiple sites without exposing their identity to the control plane.
Cohort & OMOP
The cohort is built (exposed patients and external comparator), then converted to the OMOP CDM model, directly readable by regulatory epidemiologists.
Analysis, model & evidence
A notebook computes the incidence of cardiovascular events; a risk model is tracked in MLflow. Every PHI access is logged, the pipeline is versioned in Git, and an EHDS export manifest is generated.
Result. A reproducible, auditable RWE study: the audit trail the EMA and FDA require exists by construction, and the EHDS manifest is ready. Patient data never left the lab's sovereign cloud; Polnor only orchestrated SQL and models.