A high ratio is a reason to look, never an answer. Screening rules turn the statistics into signals of disproportionate reporting (SDRs):
- A widely used rule for the PRR (Evans and colleagues, 2001): PRR at least 2, chi-squared at least 4, and at least 3 cases.
- GVP Module IX Addendum I: thresholds usually combine a disproportionality statistic (often its lower confidence limit rather than the point estimate) with a minimum number of cases. Very low thresholds flood the team with false positives; very high ones delay or miss real reactions.
What the numbers cannot do (ICH E2E and GVP IX Addendum I):
- they do not quantify the risk: a PRR of 10 does not mean the risk is ten times higher;
- they do not remove confounding or the biases of spontaneous reporting (publicity, stimulated reporting, time on the market, duplicates);
- they are used with, never instead of, a review of the individual cases;
- no signal does not mean no problem;
- not every SDR needs further investigation, and some combinations that are not SDRs still do.
Disproportionality can also be masked: when a database is full of reports of an event for one class of products, a new product's ratio for that event looks smaller than it is.