Nexus 5% Sample — real, de-identified SUS data in OMOP CDM

Records are real and de-identified — this is not a synthetic dataset. The Nexus cohort comes from the deterministic linkage of the SUS outpatient (SIA) and inpatient (SIH) systems, described in Pharmacoepidemiology and Drug Safety (2026); this sample is a random ~5% cut of its patients, standardized to OMOP CDM v5.3 with OMOP vocabularies (ICD-10 → SNOMED, SIGTAP → procedures and drugs).

• 10 clinical tables: personobservation_periodvisit_occurrencecondition_occurrenceprocedure_occurrencedrug_exposuredevice_exposuremeasurementobservationdeath
• Format: one Parquet (zstd) file per table, read directly by DuckDB, Arrow/R, Spark and pandas; manifest with row counts, period and SHA-256; README with recipes for DuckDB, R and ATLAS (Broadsea).
• De-identification: no direct identifiers; opaque IDs; date of birth standardized to January 1st of the birth year (use year_of_birth only); location, care_site, provider and note excluded; clinical event dates kept.
• Vocabularies: not included — download for free from OHDSI Athena (the source version is in the manifest).
• Limits: the Nexus cohort is a selected subpopulation (enriched for specialized care) and this is a 5% sample of it — good for learning, testing ETL and practicing longitudinal analyses, not for population inference, incidence/prevalence or clinical decisions.
• License: Data Use Terms (educational and non-commercial use; no re-identification; no redistribution; attribution). Acceptance is recorded before the links are generated. For commercial use or the full cohort, talk to us.