Guidance on the interpretation of pharmacovigilance signal detection data and analyses
Abstract
The primary objective of pharmacovigilance is the early detection and characterization of safety signals, defined as information suggesting new potential causal association between a medicinal product and an adverse event requiring further investigation. The French pharmacovigilance system is characterized by a decentralized organization relying on Regional Pharmacovigilance Centres, which provide in-depth clinical and pharmacological assessment of adverse drug reaction reports before their transmission to health authorities. The individual assessment of spontaneous adverse drug reaction reports, combined with quantitative analyses of pharmacovigilance databases (disproportionality analysis), is a key tool for safety signal detection. However, data from pharmacovigilance databases must be interpreted with caution. Causality between reported adverse reactions and the suspected drug remains uncertain due to several limitations, including incomplete information, diagnostic uncertainty, and potential confounding factors. Furthermore, the number of reported cases cannot be used to estimate the incidence of adverse reactions in the general population, notably because of selective reporting and limited data on the actual number of exposed patients. Disproportionality analyses, which are widely used to identify drug–adverse reaction associations, are useful for hypothesis generation. However, they cannot establish causality or quantify risk and are subject to numerous biases. Therefore, safety signals should be considered hypotheses requiring a multimodal assessment integrating clinical, pharmacological, and epidemiological data. Rigorous interpretation of these data is essential to avoid misleading conclusions that could have public health implications. The purpose of this work is to develop an educational tool to support the understanding and interpretation of data and signal detection analyses performed using pharmacovigilance databases.
Related
- Integrating pharmacovigilance and machine learning: bleeding signal detection and risk prediction for oral anticoagulants in cancer patients
- Pharmacovigilance signal detection of rilonacept using FAERS with exploratory Mendelian randomization analysis
- Integrating Causal Inference into Pharmacovigilance: Target Trial Emulations for Proactive Signal Detection of Atorvastatin Initiation in Medicare Beneficiaries
- Validating Artificial Intelligence Guidance for Ultrasound Acquisition and Remote Interpretation
- Hyperbolic heterogeneous graph learning for adverse drug reaction prediction, candidate prioritization, and pathway-based interpretation in pharmacovigilance
BibTeX
@article{khouri2026guidance,
title = {{Guidance on the interpretation of pharmacovigilance signal detection data and analyses}},
author = {Charles Khouri and Virginie Bres and Marion Allouchery and Romain Barus and Delphine Bourneau-Martin and Clément Cholle and Laurent Chouchana and Judith Cottin and Sophie Fedrizzi and Alex Hlavaty and Bérenger Largeau and Nathalie Massy and Anne Lise Ruellan and Lucie-Marie Scailteux and Thomas Soeiro and Laure Thomas and Aurélie Grandvuillemin and Jean-Luc Faillie},
journal = {Therapies},
year = {2026},
month = nov,
doi = {10.1016/j.therap.2026.10.005},
url = {https://hal.science/hal-05850269v1},
note = {CC BY},
}
RIS
TY - JOUR
TI - Guidance on the interpretation of pharmacovigilance signal detection data and analyses
AU - Charles Khouri
AU - Virginie Bres
AU - Marion Allouchery
AU - Romain Barus
AU - Delphine Bourneau-Martin
AU - Clément Cholle
AU - Laurent Chouchana
AU - Judith Cottin
AU - Sophie Fedrizzi
AU - Alex Hlavaty
AU - Bérenger Largeau
AU - Nathalie Massy
AU - Anne Lise Ruellan
AU - Lucie-Marie Scailteux
AU - Thomas Soeiro
AU - Laure Thomas
AU - Aurélie Grandvuillemin
AU - Jean-Luc Faillie
PY - 2026
DA - 2026/11/03
JO - Therapies
DO - 10.1016/j.therap.2026.10.005
UR - https://hal.science/hal-05850269v1
AB - The primary objective of pharmacovigilance is the early detection and characterization of safety signals, defined as information suggesting new potential causal association between a medicinal product and an adverse event requiring further investigation. The French pharmacovigilance system is characterized by a decentralized organization relying on Regional Pharmacovigilance Centres, which provide in-depth clinical and pharmacological assessment of adverse drug reaction reports before their transmission to health authorities. The individual assessment of spontaneous adverse drug reaction reports, combined with quantitative analyses of pharmacovigilance databases (disproportionality analysis), is a key tool for safety signal detection. However, data from pharmacovigilance databases must be interpreted with caution. Causality between reported adverse reactions and the suspected drug remains uncertain due to several limitations, including incomplete information, diagnostic uncertainty, and potential confounding factors. Furthermore, the number of reported cases cannot be used to estimate the incidence of adverse reactions in the general population, notably because of selective reporting and limited data on the actual number of exposed patients. Disproportionality analyses, which are widely used to identify drug–adverse reaction associations, are useful for hypothesis generation. However, they cannot establish causality or quantify risk and are subject to numerous biases. Therefore, safety signals should be considered hypotheses requiring a multimodal assessment integrating clinical, pharmacological, and epidemiological data. Rigorous interpretation of these data is essential to avoid misleading conclusions that could have public health implications. The purpose of this work is to develop an educational tool to support the understanding and interpretation of data and signal detection analyses performed using pharmacovigilance databases.
SN - 0040-5957
SN - 1958-5578
ER -