Bayesian confidence propagation neural network

Abstract

A Bayesian confidence propagation neural network (BCPNN)-based technique has been in routine use for data mining the 3 million suspected adverse drug reactions (ADRs) in the WHO database of suspected ADRs of as part of the signal-detection process since 1998. Data mining is used to enhance the early detection of previously unknown possible drug-ADR relationships, by highlighting combinations that stand out quantitatively for clinical review. Now-established signals prospectively detected from routine data mining include topiramate associated glaucoma, and the SSRIs with neonatal withdrawal syndrome. Recent advances in the method and its use will be discussed: (i) the recurrent neural network approach used to analyse cyclo-oxygenase 2 inhibitor data, isolating patterns for both rofecoxib and celecoxib; (ii) the use of data-mining methods to improve data quality, especially the detection of duplicate reports; and (iii) the application of BCPNN to the 2 million patient-record IMS Disease Analyzer.