Authors: Vassilis Koutkias MarieChristine Jaulent
Publish Date: 2015/11/21
Volume: 40, Issue: 2, Pages: 37-
Abstract
Pharmacovigilance is the scientific discipline that copes with the continuous assessment of the safety profile of marketed drugs This assessment relies on diverse data sources which are routinely analysed to identify the socalled “signals” ie potential associations between drugs and adverse effects that are unknown or incompletely documented Various computational methods have been proposed to support domain experts in signal detection However recent comparative studies illustrated that current methods exhibit high falsepositive rates significantly variable performance across different datasets used for analysis and events of interest but also complementarity in their outcomes In this regard in order to reinforce accurate and timely signal detection we elaborated through an agentbased approach towards systematic joint exploitation of multiple heterogeneous signal detection methods data sources and other drugrelated resources under a common integrated framework The approach relies on a multiagent system operating based on a collaborative agent interaction protocol aiming to implement a comprehensive workflow that comprises of method selection and execution as well as outcomes’ aggregation filtering ranking and annotation This paper presents the design of the proposed multiagent system discusses implementation issues and demonstrates the applicability of the proposed solution in an example signal detection scenario This work constitutes a step towards largescale integrated and knowledgeintensive computational signal detection
Keywords: