How to Read a Kaplan-Meier Curve: What the Steps, the Spread, and the Censoring Marks Mean

A Kaplan-Meier curve is a running estimate of the share of a group that has not yet had the event being tracked, plotted against time since each person entered the study. It begins at one hundred percent, steps downward each time an event occurs, and holds flat between events. The vertical gap between two curves shows how differently two groups fared. The small tick marks show people who left the analysis before any event, which is called censoring. The curve answers one question well, by this time what fraction were still event-free, yet it cannot tell you why, for whom, or what happens past the last data point. This is general education about reading evidence, not medical advice, and questions about your own care belong with a clinician who knows your history.

I read these curves as a reviewer and as an author of peer-reviewed research, and most of the careful reading lives in what the picture quietly leaves out.

What a Kaplan-Meier curve actually plots

The vertical axis is the estimated probability of being event-free. The horizontal axis is time, usually counted from randomization or study entry rather than from the calendar. The event is whatever the study defined in advance, often death, but just as often a first hospitalization, a relapse, or a device failure. Survival is a convention inherited from the method's origins, not a promise that the chart is about dying.

The estimate is built one event at a time. When an event occurs, the curve recalculates the fraction surviving among those still being followed and carries that figure forward. Between events the line holds steady. That is why it is a staircase rather than a smooth slope.

Why the curve looks like a staircase

Each downward step marks one or more events, and its height depends on how many people were still at risk when they happened. Late in the study, when few remain, a single event can produce a large step that reflects a thin sample rather than a sudden change in biology. A jagged, plunging tail looks alarming. It usually shows only how few people were left to measure. The far right of a curve carries the least information and invites the most overreading.

How to read the spread between two curves

When a trial plots two arms, the eye jumps straight to the gap, which is the visible difference in event-free fraction at a given time. A wide, stable separation that holds across the follow-up is the cleanest signal a survival curve can give. The gap at a single time point, though, is one slice of a moving story. Two curves can separate early and then run parallel, which suggests an effect that lands up front and then holds. They can also sit together for a long stretch and diverge only late, which is a more fragile claim. The pattern of the separation, rather than its widest moment, tells you when and how the difference emerged.

What median survival does and does not tell you

The median is where a curve crosses the fifty percent line, the time by which half the group has had the event. It is a useful single number and only one point on a rich curve. Two groups can share an identical median while differing enormously in their tails, where one curve flattens into a plateau of long-term survivors and the other keeps falling. Reading the median alone discards the part that often matters most.

What the censoring marks mean

Censoring is the honest accounting of people who stopped contributing information before they had the event, whether they were observed for too short a window, moved away, withdrew, or were still event-free when the study ended. The tick marks flag these exits, and the method uses each person's data right up to the moment they leave.

The quiet assumption underneath is that censoring is uninformative, meaning people who drop out were not at systematically different risk from those who stayed. That assumption is usually invisible and occasionally wrong. If sicker patients leave before their event, the remaining group looks healthier than the truth and the curve drifts optimistic. Heavy censoring, especially when it differs between arms, is a reason to read the result with both hands on the rail.

Why late separation and crossing curves need care

A difference that appears only late in follow-up is harder to trust than an early one, because by then the sample has thinned and censoring has accumulated. The late gap may be real, yet it rests on the fewest people, so it deserves a look at how many remained at risk when the lines pulled apart.

Crossing curves are their own warning. When one group does better early and the other does better late, the relationship reverses partway through, so no single summary like a hazard ratio describes the picture honestly. Crossing can signal a real trade-off, an early harm that buys a later benefit, or it can be noise in sparse data. Either way the standard tools that assume a steady proportional difference between groups no longer fit.

What a Kaplan-Meier curve cannot tell you

The curve is an estimate, not a measurement, so the line you see is a best guess surrounded by uncertainty the plain plot often hides. A confidence band, or the number-at-risk table printed below the axis, tells you how thin the evidence is at each point. A curve without either is asking for more trust than it has earned.

It also cannot tell you why the groups differed or whether the difference will hold for a person unlike the people studied. It describes a population over a defined window and stops at the last observed time. Extending a curve past its data is the most common way a hopeful reader is misled.

I have lived inside these choices from the building side. With EASY Diabetes, an AI clinical decision-support tool I co-developed, we ran a registered randomized controlled trial (NCT03258268), and the discipline a survival curve demands is the discipline a trial demands: define the event before you start, follow people honestly, account for everyone who leaves, and read no drama into a thin tail.

How I read one as a reviewer

I start at the bottom, with the number at risk, before I let myself look at the gap. A survival curve rewards a slow, suspicious eye. The question to keep asking is plain: how many people are actually holding up this part of the line?