On this page
A Kaplan-Meier survival curve plots time on the x-axis and the probability of remaining event-free on the y-axis. Read it by identifying timepoints on the x-axis, tracing horizontally to the curve, then vertically to the y-axis probability. The staircase pattern shows when events occurred. Small marks indicate censored participants. Wider confidence bands later in follow-up signal reduced reliability.
- The y-axis shows cumulative probability of being event-free; the x-axis shows elapsed time from study start in the specified units.
- Each downward step in the curve represents one or more events; a flatter slope indicates fewer events and better survival prognosis.
- Censoring marks (tick marks or crosses) show participants lost to follow-up or still event-free at study end; heavy late censoring reduces reliability of later estimates.
- Confidence bands widen over time as the number of participants still at risk declines, reflecting growing uncertainty in survival estimates.
- The log-rank test compares entire survival curves between groups, not just median survival; a p-value indicates whether curves differ significantly.
Kaplan-Meier survival curves appear in nearly every clinical trial report, device market analysis, and medical literature review. They visualize whether study participants remain event-free over time, where the event might be death, disease progression, relapse, or another outcome. The curve's staircase shape, confidence bands, and censoring marks carry specific meanings that unlock the data, but only if you know what to look for.
This guide teaches you to read the graph itself: what each visual element represents, how to extract a survival probability at a specific timepoint, how to compare two treatment groups, and where readers most often misinterpret the data. You will not need to calculate anything. You will need to recognize patterns and understand what they signal.
What you'll be able to do
After working through this guide, you will be able to identify survival rates at key timepoints by reading the curve directly. You will recognize what censoring marks mean and why they matter for later estimates. You will compare two or more survival curves and understand what a p-value tells you about whether the difference is real. You will spot the most common pitfalls, small sample sizes, heavy late censoring, and misaligned timepoints, that can lead to wrong conclusions. These skills let you read a published Kaplan-Meier curve with confidence and extract the specific information you need for your analysis.
Identify the axes and what they measure
Every Kaplan-Meier curve has two axes. The x-axis represents time; the y-axis represents probability. Understanding what each axis measures is the foundation for reading anything the curve shows.
The y-axis: probability of remaining event-free
The y-axis shows the cumulative proportion of study participants who have not yet experienced the event of interest. As time passes and events occur, the curve steps downward, and the y-axis value decreases. The lower the curve, the higher the cumulative event rate.
The x-axis: time since enrollment
The x-axis represents elapsed time from the start of the study or from randomization. The units, days, weeks, months, or years, are always labeled on the axis. A point at x = 6 months means six months have passed since participants entered the study. The x-axis extends as far as the study followed participants. Some trials follow patients for years; others for weeks. The endpoint of the x-axis reflects the length of follow-up in that particular study.
Read the step pattern and extract survival rates
The characteristic staircase or step pattern of a Kaplan-Meier curve is not smooth. It descends in discrete jumps. Learning to read these steps and extract survival probabilities at specific times is the core skill.
Why the curve steps downward
Each downward step represents the moment when one or more events occurred in the study population. The size of the step depends on how many events happened at that timepoint relative to how many participants were still at risk. A large step signals that many events occurred at once, or that few participants remained when the event(s) happened. A small step means few events occurred, or many participants were still at risk. Between steps, the curve is flat because no new events happened during that interval.
Extract a survival rate at a specific time
To read a survival probability at a specific timepoint, follow these steps:
- Locate the timepoint on the x-axis. For example, find 12 months.
- Draw an imaginary vertical line upward from that timepoint until you reach the curve.
- From the point where your line meets the curve, draw an imaginary horizontal line leftward to the y-axis.
- Read the value on the y-axis.
Suppose you are reading a trial that followed patients for 24 months. At the 12-month mark, your vertical line reaches the curve at a height of 0.70 on the y-axis. This means 70% of participants remained event-free at 12 months. In a published example, the probability of a patient relapsing after 12 or more months is 85%, meaning 15% of patients relapsed within a year or less. Always read horizontally from the curve to the y-axis, not from the x-axis upward, the curve may step between your timepoint and the y-axis, and reading vertically first would give you the wrong value.
Interpret confidence intervals and uncertainty bands
Most published Kaplan-Meier curves include a shaded region or error bars around the line. This is the confidence band, and it represents the range of uncertainty around the survival estimate. A 95% confidence band means that if the study were repeated many times, the true survival curve would fall within the shaded region 95% of the time.
Confidence bands are typically narrow early in follow-up, when many participants are still at risk and the estimate is precise. As time passes and more participants experience events or are censored, the bands widen. Wide bands late in follow-up signal that the survival estimate at those later timepoints is less reliable. If the bands are very wide, be cautious about drawing strong conclusions from survival rates at those late times. When comparing two treatment groups, if their confidence bands overlap substantially, the curves may not differ meaningfully even if they appear visually separated.
Recognize censoring and its visual markers
Censoring is the core reason Kaplan-Meier analysis exists. It handles the reality that not every participant experiences the event during the study. Understanding what censoring is and how it appears on the curve is essential to reading the graph correctly.
What censoring marks mean on the curve
Censored observations are marked on the curve with small tick marks, crosses, or plus signs. Each mark indicates a participant who did not experience the event during the study period. Censoring occurs when a participant leaves the study before an event occurs, or when the study ends before the event has occurred. Right-censoring, the most common type, happens when a participant is still event-free at the end of the study period. On the curve, censoring marks appear along the flat portions between steps, showing exactly when participants were censored. A censoring mark does not cause the curve to step downward; it simply notes that a participant is no longer being followed.
Why censoring affects curve interpretation
Censoring matters because it reduces the number of participants at risk as time goes on. Early in the study, many participants are still being followed. As time passes, some experience events (causing steps) and others are censored (marked with ticks). By late follow-up, very few participants remain at risk. When few participants are at risk, the survival estimate becomes less stable. A single event late in follow-up can cause a large step because it represents a large proportion of the remaining population. Heavy censoring late in the study means the survival curve at later timepoints is based on very few people and should be interpreted with caution. Censoring should be non-informative, meaning participants who drop out should do so for reasons unrelated to the study. If people drop out because they are doing poorly (or very well), the survival curve becomes biased.
Compare survival between groups
Many trials compare two or more treatment groups, each shown as a separate curve on the same plot. Visual comparison is straightforward; statistical comparison requires the log-rank test.
Visually, the curve that stays higher on the y-axis represents better survival. If one curve is consistently above another, that group has a lower cumulative event rate at every timepoint. Look for where the curves separate: do they diverge early or late? Do they cross? Crossing curves suggest that one treatment is better early and another is better later, or that the difference is small and curves are overlapping due to chance.
The log-rank test answers whether the difference between curves is statistically significant. It tests the null hypothesis that there is no difference between the population survival curves. The p-value is based on comparing entire survival curves, not on comparing only the median survival for each group. A p-value below 0.05 typically indicates a significant difference. A p-value above 0.05 suggests the curves do not differ significantly, even if they appear visually separated. Always check whether the paper reports a p-value and what it is; visual separation alone does not prove a real difference.
Spot common misreadings and artifacts
Kaplan-Meier analysis is often misused, and researchers sometimes misunderstand it and use it incorrectly. Knowing the most common pitfalls helps you avoid them.
Small sample sizes create jagged, unstable curves. A trial with only a few dozen participants will show large steps when events occur, because each event represents a big fraction of the population. The confidence bands will be wide. Be skeptical of precise-looking survival estimates from small trials.
Heavy late censoring is a major red flag. If most participants are censored by the end of follow-up, the tail of the curve is based on very few people. Survival estimates at late timepoints become unreliable. Check the number at risk, usually printed below the x-axis or in a table. If it drops to single digits, treat late-timepoint estimates cautiously.
Crossing curves can indicate that one treatment is better early and another is better later. They can also signal that the difference is small and random variation is causing apparent crossings. Crossing curves complicate interpretation and may suggest that a single summary measure (like median survival) is misleading.
Comparing survival rates from two different studies can present challenges when measurements are taken at different timepoints. Different studies may report survival outcomes at different timepoints.
A flatter slope indicates a lower event rate, which means a better survival prognosis. Do not confuse a flat section (no events happening) with a good outcome. A flat section simply means no events occurred during that interval; it does not tell you whether the overall survival is good or bad. Look at the overall height of the curve and the cumulative step pattern, not just one flat section.
Do this automatically
Extracting survival rates manually from printed or displayed Kaplan-Meier curves requires identifying each step, reading its horizontal position against the time axis, and cross-referencing the vertical position against the probability axis, a process that becomes tedious across multiple timepoints, curves, or studies. Software can automate this extraction, converting image-based curves into structured survival data and timepoint probabilities. Healthcare research runs this kind of analysis end to end and cites every figure back to its original source, across as many subjects as you need, with Mimir attaching a citation to every figure. It will not predict a trial's outcome or advise on anyone's treatment. Those decisions stay with clinicians and regulators, and deliberately so. Every plan includes it, including the free one.
Start your first report and run it on something you already know well.
This article is general information about healthcare and life-sciences topics. It is not medical advice, diagnosis, or treatment guidance, and it is not a substitute for a qualified professional. Any treatment decision should be discussed with a clinician.
