A Kaplan-Meier curve is a graph that shows the probability of survival over time. It’s the most common way to visualize results in clinical trials and medical research, especially when researchers want to answer a straightforward question: how long did patients survive, and did one treatment keep people alive longer than another? You’ll see these curves in nearly every cancer study, but they’re used across medicine whenever time-to-event data matters.
How the Curve Works
The Kaplan-Meier method (also called the “product-limit estimate”) calculates survival probability at each point in time when an event occurs. Here’s the basic logic: at every moment a death or other event happens, the method recalculates the fraction of people who survived out of everyone still being tracked. It then multiplies that fraction by all the previous survival probabilities to produce a running, cumulative estimate of survival.
For example, if 100 people start a study and one dies on day 10, the survival probability at day 10 is 99/100, or 99%. If another dies on day 30 (with 99 still at risk), the probability for that interval is 98/99. The cumulative survival probability at day 30 is 99/100 × 98/99 = 98%. This multiplication chain continues throughout the study, producing a step-by-step picture of how survival changes over time.
Reading the Graph
A Kaplan-Meier curve has a distinctive staircase shape. The Y-axis shows survival probability (starting at 100% or 1.0), and the X-axis shows time. Three visual elements carry all the information:
- Horizontal lines represent periods where no events occurred. The longer the horizontal stretch, the longer that interval of stable survival lasted.
- Vertical drops represent events (usually deaths). The size of each drop reflects how much that event changed the cumulative survival probability. A big drop means the event happened when relatively few people were still at risk.
- Tick marks (small vertical lines or dots on the curve) indicate censored patients, meaning people who left the study without experiencing the event. They weren’t counted as surviving or dying from that point forward.
The staircase shape is important. These are not smooth lines. They’re step-by-step estimates, which means reading an exact survival probability at a precise time point can be tricky. The curve only changes at the moments when events actually happen.
What Censoring Means
Not every person in a study will experience the event researchers are tracking. Some people are still alive when the study ends. Others move away or become unreachable. Some experience a completely different medical event that makes continued follow-up impossible. In all these cases, researchers know the person survived at least up to a certain point, but they don’t know what happened afterward.
This is called right censoring, and it’s one of the main reasons the Kaplan-Meier method exists. Rather than throwing out incomplete data or pretending these patients survived forever, the method removes censored patients from the “at risk” count going forward. This preserves the accuracy of the survival estimate without requiring every patient to reach a definitive endpoint. When you see tick marks scattered along a curve, each one represents a censored patient dropping out of the calculation at that time point.
Comparing Two Groups
Most clinical trials plot two or more Kaplan-Meier curves on the same graph, one for each treatment group. When one curve sits consistently above another, that group had better survival. But eyeballing the gap isn’t enough to know whether the difference is real or just due to chance.
That’s where the log-rank test comes in. It’s the most widely used statistical test for comparing survival curves, and it evaluates whether there’s a meaningful difference between groups across the entire follow-up period, not just at a single time point. The log-rank test produces a p-value. A p-value below 0.05 is generally considered statistically significant, meaning the observed difference is unlikely to be explained by chance alone. One limitation: the log-rank test tells you whether a difference exists, but not how large it is.
Hazard Ratios and What They Add
While Kaplan-Meier curves show survival visually, hazard ratios put a number on the difference between groups. A hazard ratio compares the risk of the event happening at any given moment in one group versus another. A hazard ratio of 0.5, for instance, means patients in the treatment group had half the risk of dying at any point compared to the control group.
Hazard ratios come from a more advanced statistical model (Cox proportional hazards) that can also account for other variables like age or disease stage. This matters because Kaplan-Meier curves and the log-rank test treat each group as a whole, without adjusting for differences between patients. In one published example, the log-rank test found a statistically significant difference between two treatments (p = 0.032), but when age was included as a factor in the Cox model, the p-value shifted to 0.052. The conclusion was similar, but the added context from adjusting for age changed the precision of the result.
The hazard ratio does carry an important assumption: the relative difference between groups stays constant over time. If one treatment works well early but loses its advantage later, a single hazard ratio can be misleading. When Kaplan-Meier curves cross or converge, that’s a visual sign this assumption may not hold.
Median Survival Time
One of the most useful numbers you can pull from a Kaplan-Meier curve is the median survival time. This is the point on the X-axis where the survival curve drops to 50%. It tells you the time by which half the patients in that group experienced the event. In a cancer trial, for example, if the median overall survival is 14 months, that means half the patients in that group were still alive at 14 months.
Median survival is preferred over average survival because it’s less distorted by outliers. A few patients who survive far longer (or shorter) than everyone else can skew an average dramatically, but the median stays anchored to the middle of the group. If the curve never drops to 50%, typically because the study ended before enough events occurred, the median survival is reported as “not reached,” which generally signals a favorable outcome.
Key Assumptions to Know
Kaplan-Meier analysis relies on two core assumptions. First, censored patients must have the same likelihood of survival as those who remain in the study. If sicker patients are the ones dropping out, the curve will overestimate survival. This assumption is difficult to test directly. Second, patients who enter the study early must have the same survival prospects as those who enroll later. If treatment protocols or patient characteristics change over the enrollment period, this assumption can break down.
When you’re reading a study, heavy censoring (lots of tick marks, especially early in the curve) is a reason to interpret the results more cautiously. Confidence intervals around the curve also widen as fewer patients remain at risk, which typically happens toward the right side of the graph. Those late portions of the curve are inherently less reliable because they’re based on a shrinking number of people.
Where You’ll See These Curves
Kaplan-Meier curves are the standard visual in oncology trials, where they display both overall survival (time until death from any cause) and progression-free survival (time until the cancer grows or spreads). But the method works for any time-to-event question. Cardiology studies use it to track time to heart attack. Surgical research uses it to measure how long a joint replacement lasts before needing revision. Transplant medicine uses it to show graft survival rates.
The “event” doesn’t have to be death. It can be disease recurrence, hospital readmission, or any clearly defined outcome. The method is the same regardless: track when events happen, account for people who drop out, and plot the cumulative probability of remaining event-free over time.