Real-World Safety Can't Be Captured in a Clinical Trial Bubble.

Real-World Safety Can't Be Captured in a Clinical Trial Bubble.

One Idea Worth Acting On.

A patient on a newly approved oncology drug is bitten by a dog.

He reports it as an adverse event.

The company is required to report the dog bite to the FDA.

The industry treats this as an amusing anomaly of postmarketing surveillance.

It is a symptom of something far more serious.

When patients cannot distinguish a drug toxicity from an unrelated event — and when the system has no way to help them make that distinction — the safety data the industry depends on starts to break down.

What's Actually Happening

In a clinical trial, safety is monitored as close to perfectly as the industry knows how to do. Investigators report everything. Monitors review everything. The data is dense, structured, and supervised.

After approval, that structure disappears. Patients receive a package insert listing dozens of possible adverse events in clinical vocabulary and are expected to navigate their safety experience independently.

As Dr. Gary Zuckerman, Medical Safety Assessment Physician at Bristol Myers Squibb and contributor to Voices of Oncology, describes it: patients go online, try to read everything, and become incredibly afraid. They cannot distinguish between a true drug toxicity, a symptom of their cancer, and an entirely unrelated incident. The safety information exists. The framework for understanding it does not.

The result is systematic distortion in both directions. Patients over-report unrelated events — producing noise that obscures real signals. And they under-report or dismiss actual toxicities because the adverse event list was not written in a way that would help them recognize what they are experiencing.

Why This Matters

When safety data degrades after approval, two things happen simultaneously. Real signals are harder to find because they are buried in noise. And genuine toxicities go unreported because patients do not know they are experiencing something that should be reported.

Poor safety reporting also produces poor adherence. A patient who is afraid of misunderstood side effects stops taking the drug. A patient who does not understand which symptoms require clinical attention delays seeking it. In both cases, the therapy never delivers its full clinical benefit — not because of the drug, but because of the gap between how the safety information was communicated and what the patient could actually do with it.

Where It Breaks in the Real World

The dog bite is the absurd end of a real continuum. At the serious end is the patient who experiences a grade 3 immune-related adverse event, reads the package insert, concludes it sounds like the fatigue everyone said to expect, and does not call the clinic. The clinical team does not know. The pharmacovigilance system does not know. The safety signal does not form because it never entered any reporting pathway.

The clinical trial would have caught it. The real world did not — because the real world was never designed to.

What Needs to Change

Safety communication must evolve from regulatory compliance to patient activation. The patient who understands what a meaningful symptom looks like, knows which changes require immediate clinical attention, and has a simple mechanism to report what they are experiencing — that patient is a participant in the safety system, not just its subject.

Digital tools are part of this — continuous monitoring applications, patient-reported outcome platforms, symptom diaries that convert patient experience into reportable data. But technology without plain-language understanding produces more data from confused patients, not better data from informed ones.

The Bottom Line

The safety profile the industry relies on after approval is only as good as the patients who report into it.

Until safety communication is designed to produce understanding — not just disclosure — the real-world safety picture will remain systematically incomplete.

Add Your Voice — Where are you seeing this? What's working — and what's not?