A Kaplan–Meier curve helps nursing students understand how the estimated probability of remaining event-free changes over time. The event might be death, readmission, relapse or another clearly defined outcome. Reading the curve correctly requires more than deciding which coloured line is higher.
You need to identify the event, follow-up period, censoring marks and number of participants still at risk. You also need to distinguish descriptive differences from evidence about an intervention. This guide takes you through those decisions using an original hypothetical example.
It focuses on interpreting a figure in a nursing research paper. For broader analysis planning, see our statistical-test selection guide. The example numbers are invented for teaching and must not be presented as findings from a real study.
What a Kaplan–Meier curve estimates
The usual survival curve estimates the probability of not yet experiencing the defined event beyond a given time. “Survival” is statistical terminology. It does not always mean being alive.
If the event is first readmission, the curve describes remaining free from that readmission. If the event is wound healing, it describes remaining not yet healed. Those outcomes give a higher line very different clinical meanings.
MedCalc’s methodological documentation explains that Kaplan–Meier estimation uses event times and accommodates incomplete follow-up through censoring (MedCalc, n.d.). GraphPad similarly identifies the method as a way to analyse elapsed time until an event while accounting for censored observations (GraphPad, n.d.).
Before writing anything about a curve, complete three statements: the event is…, time starts at…, and a higher survival probability means…. These statements prevent most basic interpretation errors.
For example: “The event was first unplanned readmission. Follow-up started at hospital discharge. A higher curve indicates a greater estimated probability of remaining readmission-free.” That definition is more useful than an opening claim that one group simply had “better survival”.
Start with the axes, legend and event definition
The horizontal axis shows elapsed time. Confirm whether the unit is hours, days, months or years. Also check the starting point: admission, discharge, randomisation and intervention completion are not interchangeable.
The vertical axis often shows survival probability from 0 to 1, or from 0% to 100%. Some figures instead display cumulative events, which rise over time. Others use transformed scales. Read the label rather than assuming every time-to-event plot is a conventional survival curve.
Next, identify the groups in the legend. A line may represent an intervention, exposure category or clinical subgroup. Note whether the groups were randomised or formed through observation.
A useful reading worksheet has four fields: event, time origin, time unit and group definition. Add the study population underneath. If any field is unclear, consult the methods before interpreting the picture.
In nursing assignments, this small step improves critical discussion. It ties the graph to the actual care question and prevents you from treating an attractive figure as sufficient evidence. For the role of analysis in your research, use our guide to statistical analysis in a nursing dissertation, where applicable to your design.
Why the line falls in steps
A conventional Kaplan–Meier curve starts at 1, because participants are event-free at the defined starting point. It falls at observed event times and stays flat between those times.
A downward step does not always mean one event. Several events at the same recorded time can produce one larger step. The size also depends on how many participants remain at risk immediately before the event.
Consider a hypothetical cohort with 20 people still at risk. One event represents one twentieth of that risk set. Later, when only five remain at risk, one event represents one fifth. The same event count can therefore have a larger effect near the end of follow-up.
The line is not a trace of one patient’s condition. It is an estimate built from the cohort’s observed event history. A flat section does not prove that participants were healthy, symptom-free or receiving effective care. It means no recorded event changed that estimate during the interval.
Always check what the study counted. If the event is first readmission, subsequent readmissions by the same person are normally outside that first-event curve. A study of recurrent admissions needs a different analytical question and potentially a different method.
Read censoring marks without calling them events
Censoring occurs when follow-up ends before the event is observed. Someone may reach the study end without an event, withdraw or become unavailable for further follow-up. The analysis retains the information known before that point.
Small tick marks or crosses often indicate censored observations. However, symbols vary. Verify the figure caption rather than guessing from their shape.
Suppose a participant remains readmission-free for 40 days and then cannot be contacted. You know the event was not observed during those 40 days. You do not know whether it occurred later. Censoring records that limited knowledge.
A censored observation does not normally cause the survival curve to drop. Instead, the participant stops contributing to later risk sets. Treating the censoring mark as a death or readmission creates a false result.
Also avoid writing that censored participants “never experienced the event”. That is only known up to their last valid observation. A clearer statement is: “The participant was event-free when follow-up ended.”
If losses to follow-up are extensive or differ between groups, record the pattern as a limitation. Our missing-data guide addresses broader missingness decisions, while censoring requires the specific time-to-event interpretation described here.
Work through a small Kaplan–Meier curve example
Imagine eight patients followed from discharge to first readmission. The following invented timeline illustrates the calculation. One patient is readmitted on day 5. Another is censored on day 8. A second readmission occurs on day 10. A third occurs on day 15.
| Time | At risk before the event | Readmissions | Conditional event-free fraction | Estimated event-free probability |
|---|---|---|---|---|
| Start | 8 | 0 | 1 | 1.000 |
| Day 5 | 8 | 1 | 7/8 | 0.875 |
| Day 8: censoring | 7 before censoring | 0 | No new event step | 0.875 |
| Day 10 | 6 | 1 | 5/6 | 0.729 |
| Day 15 | 5 | 1 | 4/5 | 0.583 |
At day 5, the estimate is 7/8. At day 10, multiply that previous estimate by 5/6. At day 15, multiply again by 4/5. The estimate is therefore approximately 58.3% after the third event.
The censored patient explains why six, rather than seven, remain at risk before day 10. This example also shows why simply dividing the final number without readmission by eight does not reproduce the Kaplan–Meier estimate when follow-up is incomplete.
Check your understanding by asking what happens at day 8. The line remains at 0.875, while the later risk set becomes smaller.
Use the number-at-risk table
A number-at-risk table usually sits below the curve. It shows how many participants remain under observation and event-free at selected times. Exact conventions at a displayed time can vary, so use the study’s software definition if a boundary matters.
The table is crucial near the right-hand end of a graph. A curve supported by five participants deserves a different level of confidence from one supported by 500, even if their heights look similar.
Do not mistake the number at risk for the number who experienced the event. It excludes people who have already had the defined event and those censored before the relevant time.
When writing your appraisal, identify where the risk set becomes small. You could state: “Later estimates were based on fewer participants, which limits confidence in the apparent separation near the end of follow-up.” Give the actual counts if the paper provides them.
A missing risk table does not automatically invalidate a study. However, it makes the figure harder to assess. Look for supplementary tables or numerical survival estimates before deciding how much weight to give the tail.
Never extend a curve beyond the observed follow-up to predict long-term outcomes. The plot does not provide information about an unobserved future period merely because the last segment looks flat.
Read a probability at a particular time
To read a time-specific probability, select the relevant time on the horizontal axis, move vertically to the curve and then read across to the probability axis. Use a numerical table when available, because visual estimates from a printed figure are approximate.
If the readmission-free probability is 0.80 at 30 days, report an estimated 80% probability of remaining free from the defined readmission through that period. Do not describe it as the percentage currently still in hospital.
When the event is undesirable and no competing-event complication changes the interpretation, 1 minus the event-free estimate describes the complementary estimated probability. In the hypothetical 0.80 example, the complement is 0.20.
However, do not automatically use 1 minus Kaplan–Meier survival as real-world cumulative incidence when competing events are present. For example, death can prevent a later readmission. The estimand and method require additional attention.
Read the confidence interval alongside the estimate. A value of 80% without an uncertainty range leaves the precision unclear. Our separate confidence intervals article explains the general reporting principle.
Avoid converting a cohort estimate into a guaranteed prediction for one person. Patient-level counselling or decisions require relevant clinical evidence and professional judgement beyond a classroom reading of the figure.
Find median survival and recognise when it is not reached
The median event-free time is the point where the estimated survival curve first reaches or falls below 0.50. Read across from 0.50 to the curve, then down to the time axis.
This is not the average follow-up time. It is also not the arithmetic median of only those participants who experienced an event. Both substitutions ignore the purpose of the survival estimate.
In the eight-patient example, the curve remains above 0.50 after day 15. A median has therefore not yet been reached within that illustrated follow-up. Do not invent one by extending the line.
A paper may write “median survival not reached”. That does not prove everyone survived indefinitely. It means the available estimated curve did not reach the median threshold during the observation period.
Always identify the outcome when interpreting the median. A longer time to readmission may be favourable, whereas a longer time to healing may be unfavourable. The statistic is the same kind of summary, but the care meaning differs.
Where a study reports restricted mean survival time, note the stated time horizon. That is a separate summary of time spent event-free within a defined period. Do not relabel it as a median simply because both use time units.
Compare groups without overstating what the graph proves
A higher conventional survival curve indicates a higher estimated event-free probability at the displayed time. It does not, by itself, establish a causal intervention effect.
First, examine the design. Random allocation, baseline differences, treatment adherence and follow-up patterns influence interpretation. An observational comparison may reflect differences between patient groups as well as the exposure being studied.
Second, distinguish visual separation from a formal comparison. The log-rank test is commonly used to compare survival functions. Its p-value does not tell you the clinical size of a difference or remove confounding.
Third, inspect whether the curves cross. A single statement that one group did better throughout follow-up would then be inaccurate. Differences may change with time, and a single summary may conceal that pattern.
Fourth, do not treat a hazard ratio as a risk ratio. They describe different quantities. If the paper provides a hazard ratio, examine the model, reference group and assumptions before interpreting it.
A balanced nursing discussion could say: “The intervention group had a higher estimated readmission-free probability during early follow-up, but the observational design limits causal interpretation.” Use that wording only if the actual design and results support it.
Assess the censoring assumption and competing events
Kaplan–Meier analysis relies on an appropriate censoring assumption. In plain language, the censored participants should not systematically have a different unobserved event outlook from comparable participants who remain under observation, given the analysis context.
If participants withdraw because their condition worsens, that may undermine the assumption. If administrative follow-up ends on a planned date, censoring may be easier to justify. Neither pattern should be accepted without checking how the study was conducted.
Competing events need separate attention. In a readmission study, death can prevent readmission. Treating death as ordinary censoring may answer a different hypothetical question from the actual probability of readmission in practice.
You do not need to solve every advanced survival model in an undergraduate appraisal. You do need to recognise when the paper’s analytical choice affects its claims. Describe the issue and explain why it matters for the nursing question.
Also check whether entry dates vary. Participants recruited late may have less potential follow-up than those recruited early. A correct timeline and consistent time origin are essential for interpreting the event history.
Bring these limitations into the discussion rather than hiding them in a final list. Our critical appraisal guide helps connect methodological limitations to the strength of a conclusion.
Write a useful paragraph for a nursing assignment
Build your paragraph around the event, estimate, uncertainty and limitation. Start with what the curve measures. Next, state the relevant time-specific result or median. Then explain how the design affects the conclusion.
For a hypothetical paper, a paragraph might read: “The figure estimated time from discharge to first unplanned readmission. The intervention group appeared more likely to remain readmission-free during the first month. However, the later portion relied on a smaller risk set, and incomplete follow-up limited the precision of the long-term pattern.”
This paragraph deliberately avoids invented p-values and effect estimates. In your assignment, add verified numerical results from the actual paper, with its citation. Do not use the example as a substitute for reading the study.
Separate “the graph shows” from “the study suggests”. The first should describe the displayed estimate. The second may interpret its relevance after considering the methods and alternative explanations.
Finally, link the result to the appropriate nursing outcome. A curve about readmission cannot directly establish improved quality of life unless the study measured that outcome too. Keep each claim within the evidence.
Frequently asked questions
Does a censoring mark mean that a patient died?
No. It marks the end of usable event-free follow-up without an observed event at that point. The reason could be study completion, withdrawal or loss to follow-up. Check the paper’s definitions.
Is the highest curve always best?
No. A higher survival curve means a higher estimated probability of not yet experiencing the event. That may be desirable for death or relapse, but undesirable when the event is healing.
Why does a Kaplan–Meier curve remain flat?
No recorded event changes the estimate during that interval. Participants may still be censored, and the underlying risk set may shrink. A flat line does not prove absence of every health problem.
Can I report the median when the curve never reaches 50%?
Do not invent a median. Report that it was not reached or not estimable within the available follow-up, using the terminology and numerical results supplied by the study.
Does a significant log-rank test prove causation?
No. It provides evidence about differences between survival functions under the test’s conditions. Causal interpretation depends on the design, follow-up, analysis and possible confounding.
Conclusion
Interpret a Kaplan–Meier curve by defining the event and timeline, reading the steps and censoring correctly, and checking numbers at risk. Report time-specific estimates and medians carefully. Then judge group differences against the study design and follow-up limitations.
Get help explaining a survival figure
Bring the paper, marking criteria and your own interpretation to request research-appraisal coaching. Focused feedback can help you explain the curve accurately and build an evidence-based discussion you understand.
References
- MedCalc (n.d.) ‘Kaplan–Meier survival analysis’, Biostatistics in Practice. Available at: MedCalc methodological documentation (Accessed: 2 October 2026).
- GraphPad (n.d.) ‘Kaplan Meier survival analysis’. Available at: GraphPad survival-analysis documentation (Accessed: 2 October 2026).
