The Field Guide · No. 48

Median survival and survival curves: what "lived months longer" means

Median overall survival is the time, counted from diagnosis or the start of treatment, at which half the people in a group are still alive; it is read where a Kaplan-Meier survival curve crosses 50 percent, and it says nothing about how long the longest survivors live.

Updated

When a cancer drug is said to extend survival by four months, the number is usually a difference in median overall survival. The National Cancer Institute defines it as the length of time, from diagnosis or the start of treatment, at which half of a group of patients are still alive. If the median is 12 months, half the group lived past 12 months and half did not. It is not an average, and it is not a forecast for any one patient.

The median is read off a survival curve, usually drawn with the Kaplan-Meier method. The curve starts at 100 percent and steps down each time someone in the group dies. It is drawn as a staircase because the proportion surviving stays flat between deaths. Patients who leave the study, or are still alive when it ends, are called censored: they count for the time they were followed, and the method assumes their prospects were the same as those of the patients still being followed. The median is the time at which the curve crosses 50 percent.

Two traps follow. First, a median is one point on the curve. Some treatments, immunotherapies among them, keep a small share of patients alive for a long time, which shows up as a plateau at the right-hand end of the curve, and a 2019 review in Frontiers in Oncology says medians are unable to recognize that plateau. Second, many trials report progression-free survival, the time a patient lives without the disease getting worse, in place of overall survival. That is a stand-in, and a 2015 review of 65 published correlations between such stand-ins and survival in cancer found more than half were weak.

So when a survival gain makes news, ask which survival it is, overall or progression-free. Ask for the curves, not only the medians, and look at the right-hand end, where they flatten or keep falling. A survival rate at a fixed point, such as five years, can say more than a gap at the median. Put the months next to the hazard ratio and the absolute numbers, and the headline gain has a size.

The survival terms, defined

Cancer results use a handful of survival measures that sound alike and answer different questions. These definitions follow the National Cancer Institute.

Survival measures in cancer trials
TermWhat it measures
Overall survivalHow long patients are still alive, from diagnosis or the start of treatment
Median overall survivalThe time by which half of the group are still alive (and half are not)
Progression-free survivalHow long a patient lives with the disease without it getting worse, during and after treatment
Survival rateThe percentage of a group still alive a set time after diagnosis or the start of treatment
Five-year survival rateThe percentage still alive five years on; the disease may or may not have come back
Paraphrased from the National Cancer Institute's Dictionary of Cancer Terms.

How to read a Kaplan-Meier curve

The vertical axis is the share of the group still alive, or still free of the event being counted. The horizontal axis is time since the study started. Each step down is an event. The flat stretches between steps are periods when no one in the group had one.

To find the median, run a line across from 50 percent on the vertical axis to where it meets the curve, then drop down to the time axis. If the curve never falls to 50 percent before the study ends, the median has not been reached, and results are reported that way.

Toward the right-hand end, fewer people are still being followed, because some have died and others have been censored. Each step there rests on fewer patients, so the tail of a curve is less certain than its start. When two arms are drawn together, look at where they separate, whether the gap holds, and whether the lines cross.

When the median misses the tail

Two treatments can share a median and differ completely in how many patients are alive years later. The 2019 review describes the pattern: some treatments, such as immunotherapies, produce durable survival in a small percentage of patients, and those long-term survivors form a plateau in the right tail of the curve. The cancer societies ASCO and ESMO have added after-the-fact corrections to their value frameworks that raise a drug's score when such a plateau is present.

The review discusses two ways of measuring the tail directly. The milestone method reads the share of patients alive at a fixed long-term point in follow-up. The area-under-the-curve method measures the whole curve, optionally up to a set time, which amounts to an average survival rather than a median. Both, the authors conclude, capture a plateau that a median misses.

Progression-free survival is a stand-in

A surrogate end point stands in for the outcome patients care about, and in cancer trials progression-free survival is among the most common. The hope is that a drug that delays progression will also extend life. Vinay Prasad and colleagues tested that hope in 2015 by gathering trial-level analyses of how well surrogate end points tracked overall survival in cancer.

They found 36 articles reporting 65 correlations. Of those, 34 (52 percent) were of low strength, 16 (25 percent) medium and 15 (23 percent) high. Their conclusion was that the evidence supporting surrogate end points in oncology is limited. A gain in progression-free survival is a reason to watch for the overall-survival result, not a substitute for it.

What to remember

From the record

The length of time from either the date of diagnosis or the start of treatment for a disease, such as cancer, that half of the patients in a group of patients diagnosed with the disease are still alive.

National Cancer Institute NCI Dictionary of Cancer Terms: median overall survival, 2026

Asked often

What does median overall survival mean?

It is the length of time, from diagnosis or the start of treatment, at which half of a group of patients are still alive. A median of 12 months means half the group lived longer than 12 months and half did not. It describes the group, not any single patient.

What is the difference between overall survival and progression-free survival?

Overall survival measures how long patients stay alive. Progression-free survival measures how long they live without the disease getting worse. Progression-free survival is reported sooner, but in cancer trials it often tracks overall survival weakly, so a gain in one does not guarantee a gain in the other.

What does censored mean on a survival curve?

A censored patient is one whose outcome is not known for the full study, because they left it or were still alive when it ended. The Kaplan-Meier method counts them for the time they were followed and assumes their prospects were the same as those of the patients still being followed.

Why can median survival be misleading?

Because it is one point on the curve. When a treatment keeps a small share of patients alive for years, the curve flattens into a plateau that the median does not register. Two treatments with the same median can leave very different numbers of people alive at five years.

Read the news better, every morning.

We send you the day's real progress in science, medicine and beyond, told plainly and sourced. Free, forever.