Parthenon

Live Demo AGPL-3.0

The unified OHDSI platform — one application replaces fifteen.

Parthenon consolidates the entire OHDSI toolchain — Atlas, WebAPI, Achilles, DataQualityDashboard, Usagi, WhiteRabbit, CohortMethod, PatientLevelPrediction, and more — into a single, modern application built on Laravel 11, React 19, and OMOP CDM v5.4. One login, one Docker stack, one cohort builder that flows naturally into characterization, incidence, treatment pathways, prediction, and effect estimation — without ever switching tools.

Live Demo GitHub The origin story

The Problem

Ten tools to answer one question

Running a real-world evidence study with the classic OHDSI stack means juggling ten-plus separate applications — each with its own install, credentials, UI idiom, and learning curve. A single question — "What is the incidence of stroke in newly treated hypertensives?" — forces a researcher to hop between Atlas for the cohort, WebAPI for the SQL, Achilles for characterization, DataQualityDashboard for validation, and R packages for estimation.

Every tool jump introduces friction: auth failures, version drift, context switching. For health systems, the sprawl multiplies — Java containers, .NET services, R runtimes, and a decade-old Knockout.js front end to keep alive. Data engineers spend more time wrangling tools than wrangling data. Parthenon eliminates that friction entirely.

Capabilities

One platform, the whole research lifecycle

Cohort Builder

A drag-and-drop editor matching the Circe expression format used in Atlas — so existing cohort definitions import without modification. Real-time SQL preview shows exactly what runs, with live person-count feedback against any registered CDM source.

Vocabulary Explorer

Search 7M+ OMOP concepts by name, code, or synonym. Navigate SNOMED / RxNorm / LOINC ancestor-descendant trees, inspect source-to-standard mappings, and build concept sets with descendant expansion and AI-powered semantic search.

Characterization & Data Quality

A full Achilles characterization engine (~200 analyses) and DataQualityDashboard (~3,500 checks) built in — no R installation required. Record counts, demographic distributions, temporal trends, and HEEL violations across every domain, interactively.

Incidence & Pathways

Compute crude and adjusted incidence rates with time-at-risk windows and gender/age/calendar stratification. Visualize real-world treatment pathways as Sankey diagrams to see the sequences patients actually follow.

Prediction & Estimation

Patient-level prediction (PatientLevelPrediction) and population-level estimation (CohortMethod) run as managed R sidecars. Submit from the UI, monitor execution, and view ROC curves, calibration, and forest plots — without touching R.

Genomics & Imaging

Beyond classic OHDSI: upload VCFs for variant annotation, browse variants interactively, and view DICOM studies in an embedded OHIF / Cornerstone3D viewer — with imaging and genomic criteria usable inside cohort expressions.

AI-Powered Ingestion

Upload CSVs or FHIR bundles and the pipeline profiles the schema, suggests OMOP mappings with confidence scores, and stages a human review queue before writing to the CDM — replacing WhiteRabbit, Usagi, and hand-rolled ETL.

Care Quality & HEOR

Pre-built care bundles with constituent gap measures evaluate any population for per-patient gap status and bundle compliance — plus health-economics analytics for value-based programs.

Abby AI Copilot

Describe a cohort in plain English and Abby generates the structured OMOP expression; ask it to explain any cohort, search concepts semantically, or interpret a result. Bring your own model — Ollama / MedGemma by default, with 8 providers supported.

Product Tour

See Parthenon in action

Every screen below is the live application running on a synthetic 1M-patient CDM — click any capture to enlarge.

Parthenon research dashboard with population pyramid and source characterization — **The research dashboard.** One home for the whole study lifecycle — population pyramids, source characterization, and recent activity render the moment you land, with no tool-switching.

Parthenon cohort definition showing Circe expression and inclusion criteria — **Cohort definition.** Open any cohort to its Circe-compatible expression — entry events, inclusion rules, and the SQL it compiles to — so Atlas definitions import and run unchanged.

Parthenon concept set resolved to standard OMOP concepts with domains and vocabularies — **Concept sets & vocabulary.** Resolve a concept set across 7M+ OMOP concepts — every included code with its domain and vocabulary, descendant expansion in place of Usagi and the Atlas browser.

Parthenon data explorer with demographic and geographic distributions — **Data explorer.** Interrogate a CDM source visually — demographic pyramids, geographic spread, and record-count trends across every clinical domain, with no SQL required.

Architecture

How it works

A single Docker Compose stack orchestrates the application, AI, analytics, and search services around one OMOP database.

React 19 SPA

Feature-modular front end (cohorts, vocabulary, analyses, studies, data explorer, imaging, genomics) with TailwindCSS, Zustand, and TanStack Query.

Laravel 11 API

A typed PHP 8.4 REST API with Sanctum auth, Spatie RBAC, Horizon job queues, and an auto-generated OpenAPI spec + TypeScript SDK.

PostgreSQL + OMOP

OMOP CDM v5.4 clinical and vocabulary tables, an application schema, and pgvector for embeddings — with Redis for cache and queues.

R / HADES Sidecars

CohortMethod, PatientLevelPrediction, and SCCS run in managed R containers via a Plumber API — results stream back to the UI.

Python AI Service

A FastAPI service backs Abby: natural-language cohort generation, concept embeddings, and semantic search via Ollama / MedGemma.

Solr + Imaging

Solr powers faceted concept and cohort search; Orthanc + OHIF handle DICOM storage and 3D image viewing.

Data flow — generating a cohort

Drag conditions and drugs into the builder; the React client posts the Circe expression to the Laravel API.
A Horizon job compiles the expression to PostgreSQL-dialect SQL and runs it against the OMOP tables.
Results land in the results schema; the UI polls for progress.
On completion, person counts and age/gender distributions render — ready to feed characterization or estimation.

Who It's For

Use cases

Academic researcher

Import an existing Atlas cohort, run Achilles to understand data quality, characterize target and comparator populations, then dispatch a CohortMethod estimation to an R sidecar — and export the cohort, plan, and forest plot for a manuscript, without the context-switching of separate tools.

Data engineer

Onboard a new EHR export: the ingestion pipeline profiles the schema, suggests OMOP mappings with confidence scores, and stages uncertain mappings for review. Approve, write to the CDM, then validate with built-in Achilles and DQD — replacing WhiteRabbit, Usagi, and custom SQL.

Health-system informatics

Evaluate a diabetes care initiative against tens of thousands of patients using pre-built bundles, drill into per-clinic gap rates, and export a population dashboard for quarterly review — no R, no Atlas, just point and click.

OHDSI network participant

Keep Atlas artifact compatibility for legacy R workflows while moving the team onto a modern UI — design protocols, run cohorts locally, and report aggregate counts without moving PHI.

Why It's Different

Parthenon vs. the legacy stack

Dimension	Legacy OHDSI	Parthenon
Tools	10+ separate apps	One application
Auth	Per-tool credentials	Single sign-on, one RBAC model
Front end	Knockout.js (2010s)	Modern React 19, responsive
Characterization	R Achilles, offline	Built in, no R required
Ingestion	WhiteRabbit + manual Usagi	AI pipeline with review queue
AI assistance	None	Abby copilot, 8 providers
Imaging / genomics	Not supported	DICOM viewer + variant browser
Install	Hours (Java, R, Node)	One Docker command

Lineage

From the dimensional model that ran a health system

Parthenon's analytics core descends directly from the work documented in the Geisinger CDS Compendium — the conformed dimensional model, the rules engine, and the 410+ evidence-based measures that powered care for 620,000 patients a day. Parthenon makes that discipline reproducible: open-source, standards-based, and runnable by any health system or research center.

Read the Compendium

Under the Hood

Tech stack

Laravel 11PHP 8.4React 19TypeScriptViteTailwindCSS v4PostgreSQLOMOP CDM v5.4pgvectorRedisSolrR / HADESPlumberFastAPIOllamaMedGemmaOrthancOHIFDocker ComposeAGPL-3.0

FAQ

Common questions

Can I run Parthenon on my own infrastructure?

Yes. The Community Edition runs on any Docker-capable host — your laptop, on-prem servers, or cloud. You control the database and the deployment; nothing phones home.

I already use Atlas. Can I migrate?

Yes. Import your Atlas cohort definitions and concept sets directly via JSON — they work identically in Parthenon's modern UI, which preserves the Circe expression format.

Does it replace R?

No — it integrates R. CohortMethod, PatientLevelPrediction, and SCCS run in managed sidecars, so you don't install or manage R yourself. You can still drop into the Plumber API for custom analyses.

Do I have to use a particular AI provider?

No. Abby defaults to local Ollama / MedGemma (no external API cost) and supports eight providers, including OpenAI, Azure, Anthropic, Gemini, Mistral, Cohere, and Bedrock.

What about large datasets?

Parthenon has been exercised on 1M+ patient datasets. Heavy analyses run asynchronously through the job queue, so the UI never blocks — throughput scales with your PostgreSQL hardware.