ggsurvplot in R: Publication-Ready Survival Curves with survminer

Build a journal-ready Kaplan-Meier figure step by step — risk table, confidence bands, p-value on the plot, median lines, custom colors and legend, faceting, combining curves, and exporting

Biostatistics

A practical survminer reference for ggsurvplot(): start from a basic survival curve and add, one element at a time, the number-at-risk table, 95% confidence bands, the log-rank p-value on the plot, median-survival lines, censoring marks, a colourblind-safe journal palette, custom legend, axis and title, faceting by a covariate, combining multiple curves, and exporting the figure at the right size — on the NCCTG lung-cancer data.

Published

June 25, 2026

Modified

July 7, 2026

TipKey takeaways
  • ggsurvplot() (from survminer, the package built for this) turns a survfit object into a publication-ready Kaplan-Meier figure — never settle for a bare plot().
  • Build the figure one element at a time: risk.table = TRUE (number at risk), conf.int = TRUE (confidence bands), pval = TRUE (log-rank p-value on the plot), surv.median.line = "hv" (median lines).
  • Make it journal-ready: palette = "jco"/"npg" (colourblind-safe), legend.labs/legend.title, xlab/ylab/title, break.time.by, and ggtheme = theme_minimal().
  • Compare across subsets with ggsurvplot_facet() (one panel per covariate) and overlay endpoints (e.g. OS vs PFS) with ggsurvplot_combine().
  • Export at the right size with ggsave() — a ggsurvplot is a list, so save print(p) (or arrange_ggsurvplots() for several panels) at 300 dpi for print.

Introduction

You have a Kaplan-Meier curve and need it to look like the figures in The Lancet or JCO — survival steps, a 95% confidence band, the number at risk under the plot, the log-rank p-value where a reviewer expects it, and colours that survive a grayscale printer. Base R’s plot(survfit(...)) gives you none of that without a fight.

survminer’s ggsurvplot() was written for exactly this job (it’s the package the field uses, and the one we maintain). This is a recipe reference: start from the plainest survival curve and add each publication element in turn — risk table, confidence intervals, p-value, median lines, censoring marks, palette, legend, axes, faceting, combining curves, and exporting. Each section is one paragraph, the code, and the figure it produces. Copy what you need.

If you are estimating curves for the first time, read Kaplan-Meier estimation first; this lesson assumes you already have a survfit object and focuses entirely on the plot.

The data and the fit

Every recipe below plots the same survfit object, so build it once. We use lung (NCCTG, ships with the survival package) — survival by sex in 228 advanced lung-cancer patients. The dataset auto-attaches with survival, so reference lung directly (no data() call):

library(survival)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
fit
Call: survfit(formula = Surv(time, status) ~ sex, data = lung)

        n events median 0.95LCL 0.95UCL
sex=1 138    112    270     212     310
sex=2  90     53    426     348     550

Two strata (sex=1 male, sex=2 female), with their median survival and 95% CIs. Each {r} block below re-creates fit so you can copy any one of them and run it on its own.

Step 1 — a basic survival curve

ggsurvplot(fit, data) is the minimum. Pass the survfit object and the data you fit it on (survminer needs the data to build the risk table later):

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
ggsurvplot(fit, data = lung)

A basic survminer survival curve of survival by sex on the lung-cancer data: two coloured step-function curves descending from 1.0 over time, with censoring tick marks and a top legend, but no confidence band, p-value, or risk table yet.

Already a clean ggplot2 figure: two coloured curves, censoring ticks, a legend. Everything that follows adds to this call.

Step 2 — the number-at-risk table

The single most expected element in a clinical survival figure. risk.table = TRUE draws a table beneath the curves showing how many patients are still at risk at each time — it tells the reader how much data backs the tail (where curves get unreliable). Control its height with risk.table.height:

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
ggsurvplot(
  fit, data = lung,
  risk.table = TRUE,         # number at risk under the plot
  risk.table.height = 0.25   # 25% of the figure height
)

A survminer survival curve by sex with a number-at-risk table beneath it. The table lists, for each time point on the x-axis, the count of male and female patients still at risk, which shrinks toward the right end of follow-up.

For a cleaner table, add tables.theme = theme_cleantable() and risk.table.y.text = FALSE (shows a coloured bar instead of repeating the strata names). To combine counts, risk.table = "nrisk_cumcensor" shows the number at risk and the cumulative censored in one table.

Step 3 — confidence bands

conf.int = TRUE shades the 95% confidence interval around each curve — required by most journals. By default it’s a filled ribbon; conf.int.style = "step" draws stepped bounds instead (sometimes clearer when bands overlap):

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
ggsurvplot(
  fit, data = lung,
  risk.table = TRUE,
  conf.int = TRUE,            # 95% confidence bands
  conf.int.style = "ribbon"   # or "step"
)

A survminer survival curve by sex with shaded 95% confidence bands around each step-function curve and a number-at-risk table beneath. The bands widen toward the right as fewer patients remain at risk.

Notice the bands fan out at the right end — few patients remain at risk late, so the tail is uncertain. That’s the risk table’s job to expose.

Step 4 — the log-rank p-value on the plot

pval = TRUE runs the log-rank test comparing the strata and prints the p-value on the plot — no separate test call, no copy-paste. Place it with pval.coord and name the test with pval.method = TRUE:

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
ggsurvplot(
  fit, data = lung,
  risk.table = TRUE,
  conf.int = TRUE,
  pval = TRUE,                       # log-rank p-value on the plot
  pval.method = TRUE,                # also print the test name
  pval.coord = c(750, 0.85)          # x, y position of the p-value
)

A survminer survival curve by sex showing the log-rank test name and p-value annotated on the plot, with confidence bands and a number-at-risk table. The female curve stays above the male curve and the p-value indicates a significant difference.

You can also pass pval a value or string for full control — pval = "Log-rank, p < 0.001" — or change the weighting (log.rank.weights = "sqrtN" for Tarone-Ware, which is more sensitive to early differences).

Step 5 — median-survival lines

surv.median.line = "hv" draws the dashed horizontal-then-vertical guides that drop from the 0.5 survival line to each group’s median survival on the x-axis — the headline number, read straight off the figure ("h" horizontal only, "v" vertical only):

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
ggsurvplot(
  fit, data = lung,
  risk.table = TRUE,
  conf.int = TRUE,
  pval = TRUE,
  surv.median.line = "hv"     # dashed median-survival guides
)

A survminer survival curve by sex with dashed horizontal and vertical median-survival guide lines dropping from the 0.5 survival probability to each group's median time on the x-axis, alongside confidence bands, the log-rank p-value, and a risk table.

The vertical lines meet the x-axis at 270 days (men) and 426 days (women) — the medians from fit, now visible on the plot.

Step 6 — censoring marks

Censored patients (alive at last follow-up) appear as small tick marks on the curves by default — don’t mistake them for events. Toggle them with censor, and change their look with censor.shape and censor.size:

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)

# Custom censoring marks
ggsurvplot(
  fit, data = lung,
  censor.shape = "|", censor.size = 4   # vertical-bar ticks
)

Two survminer survival curves by sex side by side: the left with vertical-bar censoring tick marks shown on the curves, the right with censoring marks hidden, illustrating the effect of the censor argument.

# Hide them entirely (useful when overlaying many curves)
ggsurvplot(fit, data = lung, censor = FALSE)

Two survminer survival curves by sex side by side: the left with vertical-bar censoring tick marks shown on the curves, the right with censoring marks hidden, illustrating the effect of the censor argument.

Keep censoring marks on for a single figure (they’re informative); hide them only when many overlapping curves make the plot busy.

Step 7 — colourblind-safe journal palettes

palette = "jco" applies the Journal of Clinical Oncology colour scheme from ggsci — colourblind-safe and instantly journal-credible. Other named palettes: "npg" (Nature), "lancet", "nejm", "aaas". You can also pass "grey", an RColorBrewer name ("Dark2"), or your own hex vector:

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)

# JCO journal palette (colourblind-safe)
ggsurvplot(fit, data = lung, conf.int = TRUE, palette = "jco")

Two survminer survival curves by sex side by side using different journal colour palettes — the JCO palette on the left and the Nature Publishing Group (npg) palette on the right — both colourblind-safe, with confidence bands.

# Nature Publishing Group palette
ggsurvplot(fit, data = lung, conf.int = TRUE, palette = "npg")

Two survminer survival curves by sex side by side using different journal colour palettes — the JCO palette on the left and the Nature Publishing Group (npg) palette on the right — both colourblind-safe, with confidence bands.

# Custom colours work too: palette = c("#3a86d4", "#E7B800")

Use a named journal palette for groups; pick one and stay consistent across a paper’s figures.

Step 8 — legend, title, and axis labels

Replace the raw strata names (sex=1, sex=2) with readable labels, move the legend, and label the axes and time scale. The arguments: legend.labs (in strata order), legend.title, legend (position), xlab/ylab/title, and break.time.by (axis tick interval):

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
ggsurvplot(
  fit, data = lung,
  risk.table = TRUE,
  conf.int = TRUE,
  pval = TRUE,
  legend.labs = c("Male", "Female"),   # in strata order
  legend.title = "Sex",
  legend = "top",                      # "top","bottom","left","right", or c(x,y)
  title = "Overall survival by sex",
  xlab = "Time (days)",
  ylab = "Survival probability",
  break.time.by = 200                  # x-axis tick every 200 days
)

A fully labelled survminer survival curve: legend titled Sex with Male and Female labels, an Overall survival main title, axis labels Time (days) and Survival probability, x-axis ticks every 200 days, confidence bands, the log-rank p-value, and a number-at-risk table.

To shorten an uninformative tail, add xlim = c(0, 800) — it crops the axis without changing the survival estimates. Use surv.scale = "percent" for a 0–100% y-axis.

Transformed views: cumulative incidence and cumulative hazard

By default ggsurvplot() plots the survival probability S(t). Flip the y-axis with fun = "event" to show cumulative incidence (1 − S(t)) — the natural view when the question is “how many have had the event by time t” rather than “how many are still surviving” — and fun = "cumhaz" for the cumulative hazard, a diagnostic view (a roughly straight line suggests a constant hazard):

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)

# Cumulative incidence: 1 - S(t), the curves now RISE from 0
ggsurvplot(fit, data = lung, fun = "event", conf.int = TRUE, palette = "jco")

A survminer plot of cumulative incidence by sex on the lung-cancer data: two step-function curves rising from 0 over time with shaded 95% confidence bands in JCO colours, showing the proportion of patients who have had the event by each time point.

# Cumulative hazard (diagnostic view): fun = "cumhaz"

Same fit, different lens: fun = "event" answers the incidence question for a clinical reader, while fun = "cumhaz" is for checking the hazard shape. Everything else (risk table, pval, palette) still applies on top of the transformed axis.

Step 9 — the publication-ready figure

Put the recipe together. This is the figure you paste into a paper — risk table, confidence bands, p-value, median lines, journal palette, clean labels, and theme_minimal():

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
ggsurvplot(
  fit, data = lung,
  risk.table = TRUE,
  risk.table.height = 0.25,
  tables.theme = theme_cleantable(),
  risk.table.y.text = FALSE,
  conf.int = TRUE,
  pval = TRUE,
  surv.median.line = "hv",
  palette = "jco",
  legend.labs = c("Male", "Female"),
  legend.title = "Sex",
  xlab = "Time (days)",
  ylab = "Survival probability",
  break.time.by = 200,
  ggtheme = theme_minimal()
)

A complete publication-ready survminer survival curve of survival by sex on the lung-cancer data: the female curve stays above the male curve throughout, each with a shaded 95% confidence band in JCO journal colours, dashed median-survival lines at 270 and 426 days, the log-rank p-value on the plot, readable Male/Female legend and axis labels, and a number-at-risk table beneath.

Every element is justified: the risk table shows data behind the tail, the band shows uncertainty, the p-value answers “do they differ?”, the median lines give the headline number, and the palette is colourblind-safe. Nothing decorative.

Faceting: one panel per subgroup

To compare survival within the levels of another variable, ggsurvplot_facet() (or facet.by = inside ggsurvplot) draws a multi-panel figure — one Kaplan-Meier per subgroup, each with its own log-rank p-value. Here, survival by sex, faceted by treatment (rx) on the colon dataset:

library(survival)
library(survminer)

fit_colon <- survfit(Surv(time, status) ~ sex, data = colon)
ggsurvplot_facet(
  fit_colon, data = colon,
  facet.by = "rx",            # one panel per treatment
  palette = "jco",
  pval = TRUE,
  legend.title = "Sex",
  legend.labs = c("Male", "Female"),
  ggtheme = theme_light()
)

A faceted survminer figure with three panels, one per colon-cancer treatment group (observation, levamisole, levamisole plus 5-FU). Each panel shows male and female survival curves in JCO colours with its own log-rank p-value.

Facet by two variables with facet.by = c("rx", "adhere") for a grid. Note: ggsurvplot_facet() returns a plain ggplot, so it does not carry a risk table (that’s the trade-off for the grid layout).

Combining curves: two endpoints on one plot

To overlay separate survfit objects — classically overall survival (OS) vs progression-free survival (PFS) — use ggsurvplot_combine() with a named list of fits. Here, two endpoints built from colon (PFS simulated for illustration):

library(survival)
library(survminer)

set.seed(123)
demo <- data.frame(
  os.time   = colon$time,
  os.status = colon$status,
  pfs.time   = sample(colon$time),
  pfs.status = colon$status
)
os  <- survfit(Surv(os.time,  os.status)  ~ 1, data = demo)
pfs <- survfit(Surv(pfs.time, pfs.status) ~ 1, data = demo)

ggsurvplot_combine(
  list(OS = os, PFS = pfs),   # named list → legend labels
  data = demo,
  risk.table = TRUE,
  tables.theme = theme_cleantable(),
  censor = FALSE,
  palette = "jco"
)

A survminer figure combining two survival curves on one panel — overall survival and progression-free survival — drawn in distinct JCO colours with a shared legend and a number-at-risk table beneath, illustrating ggsurvplot_combine.

The named list keys (OS, PFS) become the legend labels. Use this whenever two curves come from different Surv() responses rather than groups of one fit.

Exporting the figure

A ggsurvplot object is a list ($plot, $table, …), not a single ggplot — so ggsave() on the raw object can drop the risk table. Save print(p) at the size and resolution you need (300 dpi for print); to export the curve with its risk table laid out, wrap a single plot in arrange_ggsurvplots():

library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ sex, data = lung)
p <- ggsurvplot(fit, data = lung, risk.table = TRUE, conf.int = TRUE,
                pval = TRUE, palette = "jco")

# Curve + risk table, sized for a journal column, 300 dpi
ggsave(
  "survival_curve.png",
  plot = arrange_ggsurvplots(list(p), print = FALSE),
  width = 7, height = 6.5, dpi = 300
)

# PDF (vector) for submission
ggsave("survival_curve.pdf",
       plot = arrange_ggsurvplots(list(p), print = FALSE),
       width = 7, height = 6.5)

Set width/height in inches to control the aspect ratio; export to PDF (vector) for journals that prefer it, PNG/TIFF at ≥300 dpi for raster submissions.

Try it live

Build your own publication figure — change the palette, the legend labels, or what you facet by. The sandbox boots on first Run.

🟢 With an AI agent

Ask Prova “build a publication-ready ggsurvplot on my survival data — risk table, confidence bands, the log-rank p-value, median lines, a colourblind-safe journal palette, and export it at 300 dpi” — it answers with survminer code you can run on your own data. The runtime is the judge. Ask Prova →

Common issues

The risk table is misaligned or cut off. The table shares the x-axis with the curve, so its alignment breaks if you crop with xlim after the fact or set an awkward figure height. Use xlim = inside ggsurvplot() (so both panels crop together), give the table room with risk.table.height = 0.25–0.3, and tidy it with tables.theme = theme_cleantable() and risk.table.y.text = FALSE.

Your legend labels are in the wrong order, or the wrong groups get the wrong colour. legend.labs follows the order of the strata in the fit, not the order you’d write them. Print fit (or names(fit$strata)) to see the order — for ~ sex it’s sex=1 (Male) then sex=2 (Female). If labels look swapped, set the grouping variable as a factor with explicit levels before survfit(), then label in that order.

The exported figure loses the risk table or comes out blurry. ggsave() on the raw ggsurvplot object saves only $plot. Save print(p), or wrap it in arrange_ggsurvplots(list(p), print = FALSE) to keep the table, and set dpi = 300 (raster) or export to PDF (vector) — never screenshot the plot pane.

Frequently asked questions

Add risk.table = TRUE to ggsurvplot(). It draws a number-at-risk table beneath the curves. Control its size with risk.table.height = 0.25, clean it with tables.theme = theme_cleantable(), and hide the repeated strata names with risk.table.y.text = FALSE.

Use the palette argument: a journal name (palette = "jco", "npg", "lancet", "nejm"), an RColorBrewer name ("Dark2"), "grey", or your own hex vector (palette = c("#3a86d4", "#E7B800")). The named journal palettes are colourblind-safe — prefer them for publications.

Add pval = TRUE — ggsurvplot() runs the log-rank test and prints the p-value on the plot. Position it with pval.coord = c(x, y), show the test name with pval.method = TRUE, or pass your own text (pval = "Log-rank, p < 0.001").

That is the risk table: set risk.table = TRUE. To combine it with the cumulative number censored or the cumulative events in one table, use risk.table = "nrisk_cumcensor" or "nrisk_cumevents".

Use ggsurvplot_facet(fit, data, facet.by = "group") (or facet.by = inside ggsurvplot) for one panel per subgroup, each with its own log-rank p-value. To overlay curves from different Surv() responses (e.g. OS vs PFS), use ggsurvplot_combine() with a named list of fits instead.

A ggsurvplot is a list, so ggsave() on it alone drops the risk table. Save print(p), or wrap it with arrange_ggsurvplots(list(p), print = FALSE) to keep the table, then ggsave("fig.png", plot = ..., width = 7, height = 6.5, dpi = 300). Export to PDF for vector-format journal submissions.

Test your understanding

  1. Build it. In the live cell below, draw a survival curve by treatment on the colon dataset (Surv(time, status) ~ rx) with a risk table, confidence bands, the log-rank p-value, and the npg palette. Fill the blanks.
  2. Interpret. A reviewer asks you to “show how many patients are still at risk over time” and to “make the colours colourblind-safe.” Which two ggsurvplot() arguments do you change, and to what?

risk.table and pval take TRUE. The Nature palette name is "npg". The strata are the three rx levels, so the legend will show three curves.

fit <- survfit(Surv(time, status) ~ rx, data = colon)

ggsurvplot(
  fit, data = colon,
  risk.table = TRUE,
  conf.int = TRUE,
  pval = TRUE,
  palette = "npg"
)

For question 2: add risk.table = TRUE (the number-at-risk table answers “how many are still at risk”), and set palette = "jco" (or "npg"/"lancet") — the ggsci journal palettes are colourblind-safe, unlike an arbitrary colour vector.

TipQuick check

Why does saving a ggsurvplot with ggsave() directly often drop the risk table?

Because ggsurvplot() returns a list of ggplot objects ($plot, $table, $cumevents, …), not a single ggplot. ggsave() saves only the main $plot. Save print(p) to render the laid-out figure, or wrap it in arrange_ggsurvplots(list(p), print = FALSE) to export the curve with its risk table.

Conclusion

You can now build a publication-ready survival curve in R with ggsurvplot(), one element at a time: the number-at-risk table (risk.table), confidence bands (conf.int), the log-rank p-value (pval), median lines (surv.median.line), censoring marks, a colourblind-safe journal palette (palette), and a clean legend, title, and axes. For comparisons across subgroups, ggsurvplot_facet() gives one panel per level; to overlay endpoints like OS and PFS, ggsurvplot_combine() puts them on one plot. Finally, export with ggsave(print(p), ..., dpi = 300) — or arrange_ggsurvplots() to keep the risk table — for a figure that drops straight into a paper.

Reuse

Citation

BibTeX citation:
@online{2026,
  author = {},
  title = {Ggsurvplot in {R:} {Publication-Ready} {Survival} {Curves}
    with Survminer},
  date = {2026-06-25},
  url = {https://www.datanovia.com/learn/biostatistics/survival-analysis/publication-ready-survival-curves},
  langid = {en}
}
For attribution, please cite this work as:
“Ggsurvplot in R: Publication-Ready Survival Curves with Survminer.” 2026. June 25. https://www.datanovia.com/learn/biostatistics/survival-analysis/publication-ready-survival-curves.