The Powers and Perils of Post-Marketing Data Analysis: Quantification and Mitigation of Biases in FAERS

From a drug safety standpoint, post-marketing data analysis is a valuable tool for gaining insight into adverse drug reactions (ADRs). However, analyzing post-marketing data without the appropriate precautions may result in the omission of crucial drug-ADR associations or the inappropriate ascribing of an ADR based on inadequate statistical treatment. We have analyzed the FDA Adverse Event Reporting System (FAERS) database with improved parsing methods, to expose the caveats that need to be considered when using this and other post-marketing datasets. To decrease noise in drug-ADR signals and reinforce the most important associations, we first minimized drug name redundancy (FAERS reports use > 0.5 million terms used to describe ~3 thousand drug products) by mapping all drugs to the 2,729 molecular ingredients encountered across the database. The unprecedented detail of our results allowed us to observe numerous idiosyncrasies in the data, such as confusion between reported indications and ADRs, and to discover that reporting of adverse events may be highly influenced by reporting trends. For example, the statistical significance of the association between pioglitazone and cardiac failure stemmed from a period of over-reporting that ended in 2004; the precise time at which another hypoglycemic drug, rosiglitazone, was reported to be the cause of numerous myocardial infarctions. The precautions and methods of analysis of FAERS that we discuss in this article will allow for improved interpretation of data in FEARS and utilization of its full potential, thus increasing the overall usefulness and impact of post-marketing data in drug safety