Drug and vaccine development

Development is often described as a regulatory obstacle course. It is better understood as a sequence of experiments designed to answer questions in an order that limits how many people are exposed to a wrong answer, and its attrition rate is a statement about biology, not about paperwork. Published analyses of large candidate cohorts consistently find that only around one in ten molecules entering Phase I reaches approval, and that the largest single drop occurs at Phase II — the first point at which efficacy in patients is actually tested.

Why Phase II is where things die

Phase I asks what the body does to the drug and what dose is tolerable; it is answered in tens of subjects and it is usually answered. Phase II asks whether the mechanism changes the disease, and that question was previously answered only in models. The predictive validity of animal models is the weak link: an induced condition in an inbred rodent shares a phenotype with the human disease without necessarily sharing its causal structure, and a target validated in that setting can be pharmacologically engaged in humans with no clinical benefit. Modern practice tries to separate the two failure modes by demanding evidence of target engagement — receptor occupancy imaging, a pharmacodynamic biomarker — so that a negative trial can distinguish “wrong target” from “insufficient exposure”.

Exposure is the second recurring cause. What matters is not the administered dose but the concentration-time profile at the site of action, and much of Phase I and II exists to establish the exposure-response relationship on which the Phase III dose is chosen. A dose selected for maximum tolerability rather than from a characterised exposure-response curve is a well-documented source of avoidable toxicity, and regulators have pushed explicitly toward randomised dose-comparison before confirmatory trials.

What randomisation buys

Confirmatory trials are large not because more data is better in general, but because the effects being detected are small relative to the variability of the outcome, and the number of participants follows from the expected event rate and the effect size worth detecting. Randomisation and blinding exist to make the treated and untreated groups exchangeable in every respect except the intervention; when blinding fails — as it does with drugs that produce obvious subjective effects — that exchangeability is compromised in a direction that inflates the apparent effect.

Endpoint choice is where the intellectual honesty of a programme is visible. A surrogate endpoint is only valid if intervening on it reliably moves the clinical outcome, and there are well-known cases where a treatment improved the surrogate and worsened survival. Accelerated pathways in the major jurisdictions grant approval on surrogates precisely because they are faster, and they carry confirmatory obligations for exactly this reason.

Vaccines are a different inference

A vaccine efficacy trial is event-driven: the endpoint is infection or disease in a population, so the read-out arrives only as fast as the pathogen circulates, which is why epidemiology, not enrolment, sets the timeline. Where a correlate of protection has been established — an antibody titre that reliably predicts protection — later products can be licensed by immunobridging against that correlate rather than by a new efficacy trial. Where no correlate exists, the shortcut is unavailable, and that is the single largest determinant of how long a new vaccine takes.

See also the technology article at /technology/drug-vaccine-development/.