Data mining of the Australian adverse drug reactions database: a comparison of Bayesian and other statistical indicators

The Australian adverse drug reactions database is derived from 140,000 reports over 30 years, including many instances of multiple drugs and multiple reactions. There are several thousand different drugs and reactions, and so the drug-reaction table is large and sparse. To aid rapid expert assessment of new reports, Bayesian approaches are being compared with other statistical methods for the re-evaluation of historical data and to provide early indications of emerging trends. Bayesian methods provide more balanced detection criteria than either descriptive statistics such as relative risks, which are subject to large sampling variation for rare co-occurrences, or statistical significance levels which are conversely weighted towards the most common co-occurrences. In this paper the various methods are reviewed and some indicative early results are presented.