What Is a Safety Signal, Really?

A safety signal, in the formal sense GVP Module IX defines it, is information arising from one or more sources — including observations and experiments — that suggests a new, potentially causal association, or a new aspect of a known association, between a drug and an event, and that is judged to warrant further investigatory action. Read that definition carefully and one thing stands out: nowhere does it say the association has been confirmed. A signal is explicitly a starting point for investigation, not an endpoint.

This distinction matters enormously in practice, because it’s the single most common point of confusion for people encountering pharmacovigilance for the first time — including, often, journalists and the public, who tend to treat "a signal was detected" as equivalent to "a new risk was proven." The reality is closer to the opposite: the overwhelming majority of statistically detected signals, once a medical reviewer actually investigates them, are not validated as representing a genuine new safety concern. They turn out to reflect reporting artifacts, confounding by indication, media attention temporarily inflating reporting rates, or simple statistical noise in a database processing enormous volumes of reports.

That’s not a flaw in the system — it’s the system working as designed. Signal detection is deliberately built to be sensitive rather than perfectly specific, because the cost of missing a genuine new risk is far higher than the cost of investigating and ultimately dismissing a false alarm. This module works through that entire pipeline in order: where signals come from, the statistical methods used to detect candidate signals at scale, the human medical review that turns a statistical flag into a validated signal, how validated signals get prioritised, and finally how a validated signal can lead all the way to a label change — closing the loop Module 3 introduced as the six-stage feedback cycle.