The coreOpen sourcePython · Rust

The data container for medical data.

MedRecords holds diagnoses, procedures, medications, lab values, free text and everything else as entities on the patient, each with its date. When a new type appears, a node is added: no schema to extend, no data to migrate. Every product in the portfolio works with this data container.

Data model

The structure comes from the data.

Events hang off the patient, each with its date. Where the data source carries encounters, cases or stays, they are represented as a separate level. The model prescribes no intermediate level and loses none either. Point at an entry to see all entries from the same day.

Patient
EntityPatient · born 1958Every entry hangs off this entity.
DiagnosisICD-10
I10Essential hypertension14 Mar 2025
I48.0Paroxysmal atrial fibrillation02 Nov 2025
I21.4NSTEMI21 Jun 2026
ProcedureOPS
1-275.0Coronary angiography21 Jun 2026
8-837.k0PCI with stent21 Jun 2026
MedicationATC
C09AA05Ramipril 5 mg14 Mar 2025
B01AF02Apixaban 5 mg02 Nov 2025
C10AA05Atorvastatin 80 mg21 Jun 2026
B01AC04Clopidogrel 75 mg21 Jun 2026
Lab valueLOINC
67151-1Troponin T · elevated21 Jun 2026
13457-7LDL cholesterol08 Sep 2026

Schematic view of an excerpt. Further entity types are possible.

Properties

What MedRecords does with it.

An object, not a schema

A MedRecord is a unit that can be serialised, passed on and loaded directly into a model. No database that first has to be set up and connected.

Queries along time

What lies between two events, how large the intervals are, in which order something happened: that follows from the structure itself, rather than being reconstructed through sorting and nested queries.

Attributes without columns

Each event type carries the attributes it has. A medication needs dose and active substance, a lab value needs unit and reference range. There is no common row width that everything must conform to.

Open sourceNo licence fees

Included in the repository, open source, free to use.

Analytics Engine

Treatment effects from routine data. Methods for histories with uneven intervals and changing treatment.

Cohort Selector

Selection by the order of and distance between events, not just by codes inside a time window.

Audit Trail

Every transformation is recorded, so a result can be traced back to the raw data.

Data Import

OMOP tables, CSV and Parquet are read in without having to be forced into a grid first.

EnterpriseLicensed

Extensions for running it in house, with support and ongoing development.

Licence

HL7 FHIR Connector

Reads FHIR resources and writes results back, both directions over the same MedRecord.

Licence

OMOP Mapping

Maps in-house and regional codings back to OMOP vocabularies, with a record of each assignment.

Licence

Custom Backends

MedRecords works directly on existing systems, so no second copy of the data appears.

Licence

Code Embeddings

Pre-trained vector representations for all OMOP vocabularies, ready to use in your own models.

Repository

Open to inspection.

The source code is public. What the code does can be checked by your IT and your data protection officer too, before the first installation.

MedRecords on GitHub
UnderneathGraphRecords

MedRecords is the medical application of a general graph library. GraphRecords itself knows nothing about medicine, only nodes, edges and attributes.

Go to the GraphRecords page
Next step

MedRecords applied to your data.

Whether an import is worth it depends on your data sources and catalogues. A first conversation usually settles that quickly.

Get in touch