Reading Kaplan-Meier Survival Curves
Learn to read, interpret, and critically evaluate Kaplan-Meier curves — the standard visualization for time-to-event data in clinical research.
You'll learn
How to read, present, and critique Kaplan-Meier survival curves in clinical papers.
Use this when
You encounter a survival curve in a paper or in your own analysis results.
What Is a Kaplan-Meier Curve?
A Kaplan-Meier (KM) curve is a step-function estimate of the survival probability over time. At each time point where an event occurs, the curve drops. Between events, it stays flat. The distinctive staircase shape is not an artifact — it accurately reflects when individual events happened.
📖 Median Survival
The median survival is the time at which the survival curve crosses the 50% line (dashed). In the chart above, Treatment A has a median survival of approximately 24 months; the control group crosses 50% around 12 months.
How to Read a KM Curve Step by Step
- 1.Look at the Y-axis: it shows survival probability (1.0 = 100% alive, 0 = 0% alive). Never start at a value other than 1.0 unless patients were already censored at baseline.
- 2.Follow each curve: each downward step represents one or more events. A steep early drop means high early mortality; late drops mean delayed events.
- 3.Find the median: where the curve crosses the 50% horizontal reference line. This is often more meaningful than the mean for skewed survival data.
- 4.Check the number at risk below the plot: as time goes on, fewer patients remain. Curves based on very few patients at late time points are unreliable — the steps get wider and the confidence intervals explode.
- 5.Look at curve separation: curves that diverge early and stay apart suggest early and sustained benefit. Curves that cross suggest different subgroups benefit at different times.
| Pattern | Interpretation |
|---|---|
| Early sharp drop in one group | Early high-risk period; possibly early treatment failure |
| Curves diverge immediately | Treatment effect starts immediately |
| Curves diverge late | Delayed treatment effect (common in immunotherapy) |
| Curves cross | Different patients benefit differently — check subgroups |
| Flat tail on one curve | Possible cure fraction — some patients may never have the event |
The Log-Rank Test: Is the Difference Significant?
The log-rank test compares the overall survival experience across groups. Its null hypothesis is that the two survival curves are identical at all time points. It produces a p-value but no effect size estimate.
💡 Use the hazard ratio for effect size
The log-rank p-value only tells you the comparison is statistically significant. The hazard ratio (HR) from a Cox model tells you the size of the effect: HR = 0.6 means the treatment group has 40% lower hazard at any given moment. Report both.
⚠️ Proportional hazards assumption
The log-rank test and Cox model assume that the hazard ratio is constant over time (proportional hazards). If the KM curves cross, this assumption is violated — the log-rank p-value is unreliable, and you need alternatives like the restricted mean survival time (RMST).
Always report median survival with 95% CI, the hazard ratio with 95% CI, the log-rank p-value, and the number at risk at key time points. A KM figure without a risk table is incomplete.
Practice with your own dataset
Perform a survival analysis on time-to-event data.
- 1.Prepare a dataset with a time column and an event indicator (1 = event occurred, 0 = censored)
- 2.Upload and select Survival Analysis from the analysis menu
- 3.Specify the time and event variables and a grouping variable
- 4.Review the KM curves, median survival, and log-rank p-value
- 5.Fit a Cox model to obtain the hazard ratio and 95% CI
Trusted sources behind this lesson
Further reading
Read next
Sensitivity, Specificity, PPV, and NPV
Master the 2×2 table for diagnostic tests — and understand why positive predictive value changes dramatically with disease prevalence.