What Is Censoring in Clinical Trials?

Censoring is a survival analysis concept. It happens when the exact event time is unknown for a patient. In oncology trials, the event may be death, progression or relapse.

A patient may be censored if they have not had the event by the analysis date, are lost to follow-up, withdraw consent or start another therapy depending on the trial rules.

Simple example

Imagine a survival trial. A patient is alive at the last follow-up visit after 18 months. The patient has not died, so we do not know their final survival time. That patient is censored at 18 months.

Censoring does not mean the patient did poorly. It means the event had not occurred or was not observed by the analysis cutoff.

Why censoring matters

Censoring is normal. Every survival analysis has some censoring. But heavy or imbalanced censoring can make results harder to interpret.

If many patients are censored early, the survival estimate becomes less reliable. If one arm has much more censoring than the other, investors should ask why.

Censoring and median not reached

If many patients are alive at the data cutoff, median survival may not be reached. That can be positive, but it can also mean the data are immature.

Censoring on Kaplan-Meier curves

Kaplan-Meier curves often show censoring marks. These may appear as small ticks on the curve. They show when a patient left the risk set without having the event.

Informative vs non-informative censoring

Survival analysis often assumes censoring is non-informative. That means censored patients are not systematically different from uncensored patients. But in real trials, censoring can sometimes be informative. If sicker patients drop out more often, censoring may bias the results.

Censoring and event-driven trials

In event-driven trials, patients who have not had an event by analysis are censored at last follow-up. If enrollment is slow or staggered, later-enrolled patients may have shorter follow-up and more censoring.

Common investor mistake

Investors may see a long survival tail and assume it represents many long-term survivors. Sometimes it does. But if very few patients remain at risk and many are censored, the tail may be less reliable.

What investors should ask

Ask how many patients were censored, whether censoring was balanced, when censoring occurred, whether follow-up was long enough, how many patients remain at risk late, whether censoring related to treatment discontinuation and whether sensitivity analyses are available.

Bottom line

Censoring is normal in survival analysis, but it can change how reliable the data look. A survival curve is only as trustworthy as its follow-up, censoring balance and number at risk.