A trial population is too small and too selective to show uncommon harms

The argument

Proponents of this position begin from what a clinical trial can and cannot contain. A trial enrols a defined number of people for a defined period, and the number is set by the effect it was designed to detect, which is usually a benefit rather than a rare harm. If a serious reaction occurs in one patient in several thousand, a trial of a few hundred will probably not contain a single case, and a trial of a few thousand may contain one that is indistinguishable from chance. Proponents stress that this is not a flaw in trial conduct. It follows from the arithmetic, and no amount of rigour at that stage removes it. The argument is that trials are also selective in ways that diverge sharply from use. Protocols commonly exclude the pregnant, the very old, the very young, people with kidney or liver impairment and people taking several other medicines, because those are the patients in whom an unrelated event would confound the result. Those are also, in many cases, the patients who will take the medicine in the largest numbers. Trials run for months where treatment runs for years, so harms with a long latency fall outside the window by construction. And they are conducted under supervision that ordinary prescribing cannot reproduce, which conceals how a medicine behaves when it is taken irregularly, alongside others, or for something it was not licensed for. Holders of the view conclude that the information a regulator most needs can only be gathered after approval, which is where the system should put its weight. Once a medicine reaches a population the numbers become large enough for uncommon harms to surface, the excluded groups are taking it, and the follow up is long enough to matter. Surveillance is on this account not a safety net beneath the real work but the stage at which a medicine is actually characterised.

Premises

Counter-arguments

Critics argue that the premises support a weaker conclusion than the one drawn. That trials cannot see everything is a reason to design them better, to require larger safety databases and longer follow up, and to mandate studies in the excluded groups as a condition of licensing, all of which are instruments applied before approval rather than after it. They also question how well surveillance performs, since spontaneous reporting is voluntary, heavily under used and biased towards harms a prescriber already suspects, so it detects what is expected more reliably than what is not. On this view the position identifies a real gap and nominates the weakest tool available to close it.

Rejecting the premises

[Rejecting P1] Regulators can require safety databases large enough to detect uncommon harms as a condition of licensing, so the arithmetic is a design choice rather than a fixed limit. [Rejecting P3] Post authorisation studies and mandated follow up extend observation without relying on passive reporting, so the gap does not show that surveillance should be the organising priority. [Rejecting C] Spontaneous reporting is voluntary and biased towards harms already suspected, so the conclusion rests on an instrument poorly suited to the uncommon harms the premises describe.