FDA Adverse Event Reporting System (FAERS)

FAERS collects individual case safety reports submitted to the FDA. Manufacturers must forward reports they receive; clinicians and patients submit voluntarily through MedWatch. Each case is coded in MedDRA and released through a public dashboard and quarterly data files. Two structural features govern how it can be used: a report is an allegation of temporal association rather than a verified diagnosis, and the database records numerators only, with no record of how many people took the drug without incident.

The database holds tens of millions of cumulative reports and grows with both prescriptions and publicity. Signal detection relies on disproportionality statistics comparing how often a drug and an event appear together against the rest of the database, using measures such as the proportional reporting ratio, the reporting odds ratio and empirical Bayes shrinkage. All are relative measures within the database and all inherit its biases. Underreporting is severe and variable, while reporting rises after media coverage, litigation advertising and product launches, an effect visible across the incretin class.

Used properly, FAERS generates hypotheses that other designs then test. It is powerful for rare, distinctive, quick-onset events no trial would be sized to see, such as anaphylaxis, and close to useless for common outcomes or anything requiring a rate.

The standard misuse is a headline counting raw reports and presenting them as cases the drug caused, often comparing a newly launched product against one marketed for thirty years. A disproportionality score is not an incidence and cannot be converted into one. The system also has a blind spot: nobody files a MedWatch report about a vial labelled for research use only, so the compounds with the least evidence generate the least surveillance data.