PRomop synthetic breast-cancer cohort (n = 1000) — OMOP CDM v5.4 export underlying the trial-eligibility benchmark

This deposit contains a fully synthetic cohort of 1,000 breast-cancer patients used as the benchmark cohort for the PRomop paper ("A flattened, decision-ready projection over a standards-based longitudinal record"). The data are generated by Synthea (https://synthetichealth.github.io/synthea/), imported into the PRomop OMOP Common Data Model (CDM) v5.4 schema with oncology extensions, and selectively enriched where the Synthea generator left decision-ready fields empty. No real patients are represented and the file contains no protected health information (PHI). The single file synthea_bc_1000.json (~249 MB, UTF-8) is a JSON export produced by the PRomop export_org_patients management command. It has the structure: - export_metadata — organization slug, name, export timestamp, and patient count.- patients[] — one object per patient, each containing: - patient_record — the flattened, decision-ready projection (the "PatientRecord" 286-column read model). - person — the OMOP person row. - omop — the underlying transactional OMOP CDM rows grouped by table. - documents, trial_enrollments, language_skills, survey_responses — related PRomop records (empty for this synthetic cohort). Contents (231,078 OMOP rows total): 217,000 measurements · 7,596 observations · 1,000 condition occurrences · 4,225 drug exposures · 1,057 procedure occurrences. The measurement set includes HER2 receptor results (LOINC 48676-1) spanning positive, negative, and 346 equivocal (IHC 2+) readings, which the benchmark's decision-readiness comparison depends on. Purpose / provenance. This cohort underlies Table 3 of the paper: a controlled benchmark comparing 20-criterion trial-eligibility retrieval from the materialized PatientRecord projection (≈0.30 ms) versus a live correlated-subquery pull from the raw OMOP tables (≈11.0 ms), a ~36.8× speedup (95% CI 36.7–37.0×) over 5,000 timed samples on local hardware. It supersedes the smaller 100-patient pilot cohort (synthea_bc.json, Zenodo record 21312915) with a larger, more production-representative measurement table (~217k rows vs ~17k). Reproducing the benchmark. Import into a PRomop deployment with python manage.py import_org_patients --input synthea_bc_1000.json, then run python manage.py benchmark_trial_eligibility --org-slugs synthea-bc. Full instructions, command options, and the OMOP-to-PatientRecord field mapping are in the PRomop repository (docs/reproducing-benchmark-results.md). Reproducing the identical 15/20 field-population parity between the two paths requires the HER2-derivation fix (GitHub PR #221 / issue #220), which preserves the equivocal HER2 results; earlier code drops them. Because query latency is hardware- and cache-dependent, the absolute millisecond figures reflect the specific measurement environment; the relative speedup between the two paths is the result intended to be reproducible.