Safety Signal Detection
Safety signal detection is the statistical and clinical screening of adverse event databases for drug-event pairs reported more often than the rest of the data would predict.
Signal detection reduces a spontaneous reporting database to a two-by-two table for every drug and event pair: reports of this event with this drug, this event with all other drugs, other events with this drug, and everything else. Disproportionality measures compare the observed cell with what the margins predict. The proportional reporting ratio and reporting odds ratio do this directly; Bayesian methods such as the empirical Bayes geometric mean and the information component shrink unstable estimates built on very few cases.
Thresholds are conventions rather than tests. A widely used flag is a proportional reporting ratio of at least two, accompanied by a chi-squared value of at least four and a minimum of three reports. Reports of suicidal ideation with GLP-1 receptor agonists were handled this way, and after review the European regulator concluded in 2024 that available evidence did not support a causal association, a conclusion supported by large cohort analyses.
The measure earns its place by being a screen, not a verdict. It tells an assessor where to look first among millions of report pairs, and the output is a hypothesis with a priority attached. What follows is causality assessment of individual cases, a review of the pharmacological plausibility, and where warranted a study in data that has a denominator.
The comparator is what people misread. Disproportionality compares a drug against the rest of the database, not against untreated patients, so it is not a relative risk and cannot be quoted as one. Media coverage inflates reporting of exactly the pair under discussion, a well-described notoriety effect that manufactures the statistic it appears to confirm. A very large number of reports for one event can also mask smaller signals for the same drug by inflating its margin.