cadec-query

Query the CADEC (CSIRO Adverse Drug Event Corpus). Use whenever the user asks about adverse drug event mentions in consumer health text, entity annotations from patient forum posts, MedDRA/SNOMED-CT n

SKILL.md

GitHub

CADEC Query Skill

Search CADEC annotated patient forum posts by any entity. Auto-detects type:

Input Pattern Detected As Match Logic
10019211 (8 digits) MedDRA code exact on normalizations.code
22253000 (6-18 digits) SNOMED CT code exact on normalizations.code
anything else free text substring on entity text, type, or doc_id

API

Function Input Returns
load_cadec(path) JSON path list[dict] (cached)
search(entity) single entity string list[dict]
search_batch(entities) list of entity strings dict[str, list[dict]]
summarize(hits, entity) hits + label compact text
to_json(hits) list[dict] same list (JSON-ready)

Entity Types in Corpus

Type Description
ADR Adverse Drug Reaction
Drug Medication mention
Disease Pre-existing condition
Symptom Patient-reported symptom
Finding Clinical finding

Usage

See if __name__ == "__main__" block in 34_CADEC.py for runnable examples: drug name search, ADR text search, MedDRA code lookup, batch search, JSON output.

Data

  • Source: CADEC v2, CSIRO (AskaPatient.com forum posts)
  • Local file: cadec_combined.json built from BRAT .txt + .ann files
  • Annotations: Original spans + MedDRA + SNOMED CT normalisations
  • Stats: ~1,250 documents, ~7,600 entity annotations
  • Path: DATA_PATH variable in 34_CADEC.py
  • Citation: Karimi et al., 2015. J Biomed Inform 55:73–81