Time-to-event data (time to death, relapse or discharge) need survival methods because some patients have not had the event by the end of follow-up: they are censored. The Kaplan-Meier method estimates the survival curve for each group, and the log-rank test compares the curves. The write-up gives the reader the medians, the test and a figure with the numbers at risk.
What you need before you write a single word
- For each group: the number of patients, the number of events and the number censored.
- The median survival (time at which 50% have had the event) with its 95% CI, or survival at fixed times (for example, 2-year survival) if the median is not reached.
- The log-rank test: χ² with its degrees of freedom (groups − 1) and the P value.
- The median follow-up time.
- Ideally, the hazard ratio from a Cox model, which says how big the difference is.
The template
During a median follow-up of [X] months, [n] patients in the [group A] group and [n] in the [group B] group [died]. Median survival was [X] months (95% CI, [X-X]) with [group A] and [X] months (95% CI, [X-X]) with [group B] (log-rank χ²(1) = X.XX, P = .XX; Figure 1).
A worked example
120 patients received a new treatment (n = 60) or standard care (n = 60).
| Group | Patients | Deaths | Censored | Median survival, mo (95% CI) |
|---|---|---|---|---|
| New treatment | 60 | 27 | 33 | 26.8 (17.5-38.2) |
| Standard care | 60 | 39 | 21 | 18.9 (14.5-23.6) |
The log-rank test gives χ²(1) = 4.85, P = .03. Written up:
Of the 120 patients, 27 of 60 in the new-treatment group and 39 of 60 in the standard-care group died during follow-up. Median survival was 26.8 months (95% CI, 17.5-38.2) with the new treatment and 18.9 months (95% CI, 14.5-23.6) with standard care (log-rank χ²(1) = 4.85, P = .03; Figure 1).
The figure
- One Kaplan-Meier curve per group, survival probability from 0 to 1 on the y-axis.
- Tick marks for censored observations.
- A numbers-at-risk table under the x-axis at regular time points.
- Stop the curves where few patients remain at risk, since their tails are unreliable.
Mistakes reviewers catch
- Mean survival or simple averages of follow-up time. Censoring makes them biased; use Kaplan-Meier medians.
- No numbers at risk under the curves.
- Log-rank P value as the only result. It says the curves differ, not by how much; add the hazard ratio with its CI.
- Crossing curves. The log-rank test loses power when hazards are not proportional; say so and consider restricted mean survival time.
- Median "not reached" left unexplained. If fewer than half the patients had the event, report survival at fixed time points instead.
Let KyroStat do the write-up for you
KyroStat computes the Kaplan-Meier estimates, medians with confidence intervals and the log-rank test, draws the curves with numbers at risk, and writes the paragraph from your data.
Frequently asked questions
Does the log-rank test have an effect size? No; report the hazard ratio from a Cox model. See how to report a Cox regression.
What if I have more than two groups? The log-rank χ² then has groups − 1 df; follow a significant result with pairwise comparisons and a correction.
Which style should P values follow? Your journal's; most medical journals use AMA style. See how to report statistics in AMA style.