Related Experiment Videos
A knowledge based approach for automated signal generation in pharmacovigilance
Corneliu Henegar1, Cédric Bousquet, Agnès Lillo-Le Louët
1Laboratoire de Santé Publique et d'Informatique Médicale, INSERM ERM 202, Faculté de médecine Broussais Hôtel Dieu, 15 rue de l'Ecole de Médecine, 75006 Paris, France. corneliu.henegar@spim.jussieu.fr
Studies in Health Technology and Informatics
|September 14, 2004
Summary
A new knowledge-based approach enhances automated signal detection for adverse drug reactions (ADRs), improving sensitivity without sacrificing specificity. This method leverages semantic information from medical vocabularies for more accurate drug safety monitoring.
Area of Science:
- Pharmacovigilance
- Medical Informatics
- Data Mining
Background:
- Manual review of spontaneous reporting systems is used by pharmacovigilance experts to detect adverse drug reactions (ADRs).
- Automated signal generation aims to identify potential drug-adverse event associations disproportionally present in databases.
- Current signal detection methods do not utilize the semantic information from controlled vocabularies like the Medical Dictionary for Regulatory Activities (MedDRA).
Purpose of the Study:
- To enhance the performance of existing signal detection algorithms.
- To incorporate a knowledge-based approach into automated signal generation for ADRs.
Main Methods:
- Development of a formal ontology for adverse drug reactions (ADRs).
- Creation of a data mining tool utilizing description logic representations of MedDRA terms.
- Grouping of medically related case reports based on semantic similarity.
Main Results:
- The knowledge-based approach significantly increased the sensitivity of signal detection.
- No decrease in the specificity of signal detection was observed.
- Improved identification of true drug-adverse event associations.
Conclusions:
- A knowledge-based approach demonstrably improves the performance of pharmacovigilance signal detection tools.
- The significant workload associated with knowledge engineering currently limits the scalability of this approach for machine learning applications.