OMOP Sepsis Phenotyping Pipeline

PostgreSQL implementation of Sepsis-3 (SOFA) and CDC Adult Sepsis Event (ASE) phenotypes for OMOP CDM v5.4.

The canonical SQL path is sql/RUN_ALL_enhanced.sql. Older runners are retained only as deprecated compatibility stubs.

Run Order

export PGPASSWORD='your_password'
export DB='postgresql://user@host/omop?sslmode=require'

psql "$DB" \
  -v ON_ERROR_STOP=1 \
  -v results_schema=results \
  -v cdm_schema=omopcdm \
  -v vocab_schema=vocabulary \
  -f sql/RUN_ALL_enhanced.sql

The runner builds:

Object	Purpose
concept_set_members	Canonical ATHENA/local concept-set registry
concept_set_validation_failures	Fail-fast vocabulary validation output
view_infection_onset	Antibiotic/culture candidate infection events
sofa_hourly	Event-scoped hourly SOFA component and total scores
sepsis3_cohort	Sepsis-3 cases with sofa_delta >= 2
cdc_ase_cohort_final	CDC ASE cohort with outcomes/support flags
sepsis_cohort_comparison	Matched Sepsis-3 vs CDC ASE comparison

Concept And Vocabulary Policy

Concepts are managed in sql/03_create_concept_sets.sql.

• ATHENA vocabulary concepts are validated for existence, invalid status, expected domain when applicable, and standard concept status.
• Site-local concept IDs are allowed only when explicitly marked local_allowed = true.
• Antibiotics are expanded from the antibacterial ancestor 21602796 into standard Drug-domain descendants.
• Vasopressors, ventilation, urine output, renal replacement therapy, PaO2, FiO2, cultures, labs, vitals, and GCS use named concept sets.

If validation fails, inspect:

SELECT *
FROM results.concept_set_validation_failures
ORDER BY concept_set_name, concept_id;

Scale Notes

• sofa_hourly is built from infection-onset windows, not from each patient's full first-to-last measurement span.
• Hot joins use (concept_id, person_id, datetime) and (person_id, hr) indexes.
• RUN_ALL_enhanced.sql does not mutate SQL files at runtime.
• The SOFA table exposes a stable contract: hr, component scores, total_sofa, and components_observed.

Validation

Run static SQL checks before database execution:

bash validate/static_sql_checks.sh

Then run the pipeline with ON_ERROR_STOP=1 and review final row counts:

SELECT 'sepsis3' AS cohort, COUNT(*) FROM results.sepsis3_cohort
UNION ALL
SELECT 'cdc_ase', COUNT(*) FROM results.cdc_ase_cohort_final
UNION ALL
SELECT cohort_type, COUNT(*) FROM results.sepsis_cohort_comparison GROUP BY cohort_type;

For live scale validation, collect EXPLAIN (ANALYZE, BUFFERS) on 31_create_sofa_hourly.sql, 40_create_sepsis3_enhanced.sql, and 61_create_sepsis_cohort_comparison.sql using representative production data.