What Happens to a Treatment After Approval: The Job of Real-World Data

Approval is not the end of the evidence. It is the moment a treatment moves from a few thousand carefully chosen trial participants into the hands of millions of ordinary patients, and that shift is when real-world data starts doing work no trial could do. Its job after approval is to catch the rare harms a pre-approval program was too small to see, to show how a treatment performs once it leaves the controlled setting, and to do both fast enough that a problem can be acted on before it spreads. The catch is that this same data is messy in ways that make it easy to raise a false alarm or miss a real one. Reading it well matters as much as collecting it.

To be precise about the phase: here I mean the surveillance that begins the day a product reaches the market and never quite stops while it stays in routine use. This is educational, not medical advice about any particular product.

Why does monitoring continue after a treatment is approved?

A confirmatory trial is built to answer one question well, and that means keeping the study small and clean enough to give a clear result. That cleanliness is also its blind spot. A program that enrolls a few thousand people and follows them for a year or two cannot observe an event that happens once in twenty thousand patients, or one that takes five years to appear, or one that strikes a kind of patient the trial deliberately excluded. That is not a flaw; it is the price of the design.

Post-approval surveillance is the system that watches a treatment once it is in routine use, to detect harms, benefits, and patterns of use the approval studies were never sized to find. To be confident of seeing even one instance of a one-in-ten-thousand event, you need far more than ten thousand people, and you may need to watch for years. The real world supplies both.

What does real-world data add that a trial could not?

It adds the long tail of rare events. When a treatment reaches a population hundreds of times larger than its trial, effects that were statistically invisible before become countable. Some of the most consequential safety lessons in medicine arrived this way, when a drug in wide use turned out to carry an uncommon risk that no pre-approval study had the scale to detect. The signal was always there; there were simply never enough patients to see it.

It also adds a picture of everyday performance. Trial participants tend to be younger, healthier, and more adherent than the patients who will actually take the treatment. Real-world data sees the eighty-year-old on nine other medicines and the person whose kidneys work at half the assumed rate. It shows how a treatment is genuinely used, which is rarely how the label imagines: off-label prescribing, doses drifting from what was studied, combinations nobody trialed. A safety profile built under careful conditions can shift once real use departs from it.

How is post-approval data collected?

Several streams feed the system, and they differ in speed and quality. The fastest and oldest is spontaneous reporting, where clinicians, patients, and manufacturers send reports of suspected harms to a central database. It casts a wide net and is good at raising early hypotheses, so it is often where a brand-new kind of problem first surfaces. More structured streams trade speed for rigor. Registries follow defined groups of patients over time. Large healthcare databases built from records, claims, and pharmacy refills let analysts study millions of treatment courses against comparison groups, and regulators frequently require formal phase 4 studies as a condition of approval. Each stream is strongest where the others are weak, which is why no serious program leans on a single one.

Where does post-approval data mislead?

The signature weakness of spontaneous reporting is the missing denominator. A pile of reports tells you how many events were noticed and submitted, not how many people took the treatment, so you cannot turn the count into a rate. A rate is what tells you whether anything is wrong. Ten reports against ten thousand users mean something very different from ten against ten million.

Reporting itself is biased. Most adverse events are never reported, so the absolute numbers understate reality, while media attention or a new warning can inflate reports of an event that was always happening at the same rate. That is the notoriety effect. A signal can grow because a treatment became more dangerous or because people started looking harder, and telling those apart is the whole game.

Then there is the trap that haunts all observational work. The patients who receive a treatment after launch are not a random sample, and the sickest often receive the newest option, so the treatment can inherit blame for outcomes the underlying illness was always going to produce. A reported harm may belong to the disease, to other medicines, or to chance. A report is a question, not a verdict.

How should a careful reader treat a safety signal?

Treat the signal as the start of an investigation, not its conclusion. The honest sequence is to generate a hypothesis from the fast, noisy streams, then test it in the slower, structured ones where a denominator exists and confounding can be addressed. A finding that holds up across spontaneous reports, registries, and a designed database study deserves real weight. A frightening number from one uncontrolled source deserves caution, not action on its own.

I learned to respect this from the inside. During my time in global drug development, the discipline that impressed me most was the refusal to overreact to a raw count paired with an equal refusal to dismiss one. Later, while building clinical decision-support tools, I saw the same principle from the other side: usage data was wonderful for understanding how a system behaved in real clinics, yet a genuine claim about outcomes needed a design that could carry it. The craft is matching the strength of your conclusion to the strength of your data.

This is not a story of failure. Catching a rare harm after approval is the system working as intended, because some harms stay hidden until a treatment meets the full variety of human beings who need it. The harder truth is that the same data also generates false alarms and hides true rates, so it has to be read by people who grasp both its reach and its limits. If you are weighing a specific treatment, the conversation to have is with your own clinician.