Conversational Pharmacovigilance That Preserves the Patient’s Narrative

Structured adverse-event reporting supports coding, aggregation, and regulatory review, but it can compress the patient’s account into fields that omit chronology, uncertainty, functional consequences, perceived causality, and the language through which harm is understood. This article proposes a non-empirical conversational pharmacovigilance architecture designed to preserve patient narrative while supporting structured extraction, triage, and signal work. The architecture uses a dual-representation contract: an append-preserving narrative ledger remains the source record, while a separate structured safety object contains derived medication, event, timing, seriousness, outcome, action, and uncertainty fields linked to supporting narrative spans. A bounded clarification controller may request missing safety-relevant information without silently rewriting previous statements. The preserved narrative and structured derivative are combined in a paired triage packet and passed through explicit human review gates before case assessment or signal work. Governance spans consent, privacy, accessibility, patient correction, provenance, auditability, subgroup evaluation, incident response, change control, and withdrawal authority. Evaluation must distinguish narrative completeness and fidelity from coding accuracy, triage safety, workflow consequence, patient experience, equity, and organizational readiness. The architecture does not determine causality, replace professional assessment, provide medication advice, validate safety signals, or establish regulatory compliance. Its original contribution is a provenance-preserving relationship between patient dialogue and pharmacovigilance operations that treats structured data as a useful derivative rather than a superior replacement for the patient’s account. Empirical validation is required across languages, populations, reporting channels, and medication-safety workflows.