OHDSI ATLAS

In one line

ATLAS is the visual cockpit of the OHDSI ecosystem: point-and-click design of patient cohorts and analyses that run, unchanged, against any database in the OMOP Common Data Model — anywhere in the world. Licence: Apache 2.0 (fully open).

The problem it solves

Defining a research cohort — "adults started on metformin, no prior insulin, with a follow-up visit" — as hand-written SQL is slow, error-prone, and unsharable: every analyst writes it differently. ATLAS lets you build it visually, once, against the standard OMOP CDM, and generates the correct, standardised SQL for you — so the same definition runs identically everywhere.

What you can build

Sitting on top of an OMOP database, ATLAS generates the analytic SQL for:

• Cohorts — inclusion/exclusion rules over conditions, drugs, observations.
• Characterization — who is in the cohort (demographics, comorbidities).
• Incidence rates — how often an outcome occurs.
• Pathways — treatment sequences over time.
• Patient-level prediction — models for an outcome of interest.

Why "runs anywhere" matters

Because every analysis is expressed against the standard CDM and vocabularies, the same study definition can be shipped to dozens of sites and executed locally — the data never leaves, only the study design and aggregate results travel. This is how OHDSI runs network studies across hundreds of millions of patients while respecting every site's privacy.

Where it shows up in digital health

ATLAS is the working tool behind real-world-evidence and pharmacoepidemiology programmes. It is the production destination for the skills in the OMOP Data Harmonization lab: once source data is mapped to the CDM (WhiteRabbit/Rabbit-in-a-Hat plan that), ATLAS is where cohorts and analyses are built — the reason harmonization is worth the effort.

Common pitfalls

• Garbage in, confident garbage out — ATLAS is only as good as the ETL beneath it; validate data quality first.
• Cohort definition drift — small rule changes change results; version and document definitions.
• Treating generated SQL as a black box — understand what it does, especially for publication-grade work.

Key takeaways

• ATLAS = visual, point-and-click design of OMOP cohorts and analyses → standardised SQL.
• Build once; run unchanged on any OMOP database, locally, sharing only results.
• Powers OHDSI network studies at population scale.
• The destination for OMOP harmonization work — open source (Apache 2.0).