EBGM (Empirical Bayes Geometric Mean)
What is it?
EBGM is one of the statistical measures used in disproportionality analysis to flag drug-event pairs that are reported together more often than chance would predict. Its distinguishing feature is the Bayesian shrinkage behind it: for drug-event pairs with only a handful of reports, EBGM pulls the estimate toward a more conservative, stabilized value instead of letting a tiny sample size produce a wildly inflated signal.
That makes it particularly useful for large safety databases with many rare drug-event combinations, where a simpler ratio measure could easily overreact to two or three coincidental reports.
Watch the shrinkage
Same drug-event pair, three different report counts. Pick a scenario and watch how far the raw ratio swings compared to EBGM.
Click a scenario to move the markersEBGM: pulled toward center on thin data
2 reports against 0.1 expected: the raw ratio spikes to roughly 20. EBGM shrinks that down to around 2.3, still worth a look, far less alarming.
8 reports against 1.2 expected: the two measures are already converging, EBGM's shrinkage matters less once there's more data behind it.
40 reports against 6 expected: with a real sample size, raw ratio and EBGM land close together. Shrinkage barely moves the needle here.
The "raw ratio" above is the Reporting Odds Ratio (ROR). Many safety databases calculate both side by side: ROR is simpler and reacts faster, EBGM is steadier on sparse data. Neither one proves causation on its own, see disproportionality analysis for how a flagged signal actually gets worked.
