Extending the methods used to screen the WHO drug safety database towards analysis of complex associations and improved accuracy for rare events.

Post-marketing drug safety data sets are often massive, and entail problems with heterogeneity and selection bias. Nevertheless, quantitative methods have proven a very useful aid to help clinical experts in screening for previously unknown associations in these data sets. The WHO international drug safety database is the world's largest data set of its kind with over three million reports on suspected adverse drug reaction incidents. Since 1998, an exploratory data analysis method has been in routine use to screen for quantitative associations in this data set. This method was originally based on large sample approximations and limited to pairwise associations, but in this article we propose more accurate credibility interval estimates and extend the method to allow for the analysis of more complex quantitative associations. The accuracy of the proposed credibility intervals is evaluated through comparison to precise Monte Carlo simulations. In addition, we propose a Mantel-Haenszel-type adjustment to control for suspected confounders.

A bayesian recurrent neural network for unsupervised pattern recognition in large incomplete data sets.

A recurrent neural network, modified to handle highly incomplete training data is described. Unsupervised pattern recognition is demonstrated in the WHO database of adverse drug reactions. Comparison is made to a well established method, AutoClass, and the performances of both methods is investigated on simulated data. The neural network method performs comparably to AutoClass in simulated data, and better than AutoClass in real world data. With its better scaling properties, the neural network is a promising tool for unsupervised pattern recognition in huge databases of incomplete observations.

A data mining approach for signal detection and analysis.

The WHO database contains over 2.5 million case reports, analysis of this data set is performed with the intention of signal detection. This paper presents an overview of the quantitative method used to highlight dependencies in this data set. The method Bayesian confidence propagation neural network (BCPNN) is used to highlight dependencies in the data set. The method uses Bayesian statistics implemented in a neural network architecture to analyse all reported drug adverse reaction combinations. This method is now in routine use for drug adverse reaction signal detection. Also this approach has been extended to highlight drug group effects and look for higher order dependencies in the WHO data. Quantitatively unexpectedly strong relationships in the data are highlighted relative to general reporting of suspected adverse effects; these associations are then clinically assessed.

A comparison of measures of disproportionality for signal detection in spontaneous reporting systems for adverse drug reactions.

PURPOSE: A continuous systematic review of all combinations of drugs and suspected adverse reactions (ADRs) reported to a spontaneous reporting system, is necessary to optimize signal detection. To focus attention of human reviewers, quantitative procedures can be used to sift data in different ways. In various centres, different measures are used to quantify the extent to which an ADR is reported disproportionally to a certain drug compared to the generality of the database. The objective of this study is to examine the level of concordance of the various estimates to the measure used by the WHO Collaborating Centre for International ADR monitoring, the information component (IC), when applied to the dataset of the Netherlands Pharmacovigilance Foundation Lareb. METHODS: The Reporting Odds Ratio--1.96 standard errors (SE), proportional reporting ratio--1.96 SE, Yule's Q--1.96 SE, the Poisson probability and Chi-square test of all 17,330 combinations were compared with the IC minus 2 standard deviations. Additionally, the concordance of the various tests, in respect to the number of reports per combination, was examined. RESULTS: In general, sensitivity was high in respect to the reference measure when a combination of point- and precision estimate was used. The concordance increased dramatically when the number of reports per combination increased. CONCLUSION: This study shows that the different measures used are broadly comparable when four or more cases per combination have been collected.