Scholarlink

Pharmacovigilance in Colombia faces challenges in analyzing spontaneous adverse drug reaction (ADR) reports written in free text, where clinical terms mix with colloquialisms, synonyms, abbreviations, and spelling errors. This heterogeneity complicates mapping to Medical Dictionary for Regulatory Activities (MedDRA) and delays signal detection. This study applies a semantic text-mining framework to the Colombian National Pharmacovigilance Database, focusing on 1,199 drug-level ADR narratives through multilingual embeddings, dimensionality reduction, and Structural Topic Model (STM). Drug narratives underwent preprocessing for normalization and entity harmonization via Anatomical Therapeutic Chemical (ATC) classifications, leading to a 25-topic optimal solution. The themes identified include systemic toxicities and administration-related errors, demonstrating that Spanish narratives reveal significant clinical distinctions beyond standardized terms. This novel approach facilitates scalable analysis and enhances regulatory workflows, although challenges such as underreporting and the need for validation remain. Future work should involve temporal modeling and expert feedback to improve methodologies in Spanish-speaking regulatory contexts.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From Free Text to Safety Signals: Integrating Structural Topic Model and Embeddings for Drug–Event in Spanish Pharmacovigilance (Colombia).

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Gustavo Bruges Morales,
  • Juan Manuel Luna,
  • Francisco Palencia-Sánchez,
  • Paula Alejandra Vargas,
  • Shirly Vanesa Melo,
  • María Laura Morales,
  • Mariangeles Meza

摘要

Pharmacovigilance in Colombia faces challenges in analyzing spontaneous adverse drug reaction (ADR) reports written in free text, where clinical terms mix with colloquialisms, synonyms, abbreviations, and spelling errors. This heterogeneity complicates mapping to Medical Dictionary for Regulatory Activities (MedDRA) and delays signal detection. This study applies a semantic text-mining framework to the Colombian National Pharmacovigilance Database, focusing on 1,199 drug-level ADR narratives through multilingual embeddings, dimensionality reduction, and Structural Topic Model (STM). Drug narratives underwent preprocessing for normalization and entity harmonization via Anatomical Therapeutic Chemical (ATC) classifications, leading to a 25-topic optimal solution. The themes identified include systemic toxicities and administration-related errors, demonstrating that Spanish narratives reveal significant clinical distinctions beyond standardized terms. This novel approach facilitates scalable analysis and enhances regulatory workflows, although challenges such as underreporting and the need for validation remain. Future work should involve temporal modeling and expert feedback to improve methodologies in Spanish-speaking regulatory contexts.