ggsave() now works directly on a ggsurvplot object: a grid.draw() method for the ggsurvplot class is registered, so ggsave(filename, plot = p) saves the assembled plot (survival curve + risk table) without needing the print(p, newpage = FALSE) workaround, and no leading blank page is produced (#152). Same approach independently contributed by @mjakubczak (#522).ggcoxzph() now accepts a coxph model directly (running survival::cox.zph() on it automatically), instead of only a pre-computed cox.zph object — ggcoxzph(coxph_fit) previously errored with "Can't handle an object of class coxph". Passing a cox.zph object is unchanged. Contributed by @DanChaltiel (#410).
ggforest() gains a ref.display argument. Set ref.display = FALSE to omit the rows that have no hazard ratio — factor baselines and any non-estimable/aliased or strata() terms (the rows labelled "reference") — from both the plot and the table, keeping only the estimated levels; useful for compact panels. Default is TRUE (every row shown, unchanged). Contributed by @trentleslie (#563).
pairwise_survdiff() now uses anyNA() for its internal missing-grouping-value check instead of a row-wise NA %in% .row, which is faster and clearer. Results are unchanged for the usual factor/character grouping variables; the only difference is a purely numeric grouping column containing NaN, whose row is now dropped as missing (previously it formed a spurious "NaN" group). Contributed by @MichaelChirico (#635).
ggsurvplot_facet() gains a labeller argument (forwarded to ggplot2::facet_wrap()/facet_grid()) to control how the panel strip labels are formatted — e.g. labeller = ggplot2::label_both or labeller = ggplot2::as_labeller(c("1" = "Obs", "2" = "Lev", "3" = "Lev+5FU")). Default NULL keeps the current labels (unchanged); when supplied it takes precedence over short.panel.labs and composes with panel.labs. Idea contributed by @B0ydT (#667, #668).
surv_adjustedcurves() gains a fun argument (matching ggadjustedcurves()) to transform the returned survival column — e.g. fun = "event" for cumulative events, "cumhaz", or "pct". Default NULL leaves the survival probabilities unchanged (#630).
arrange_ggsurvplots() gains a layout_matrix argument (forwarded to gridExtra::marrangeGrob()) to control the position/order of the plots — e.g. layout_matrix = matrix(seq_along(splots), nrow = nrow, byrow = TRUE) orders them by row. Default is unchanged (plots fill column-wise) (#300).
ggforest() gains a global.stats argument. Set global.stats = FALSE to omit the bottom caption reporting the global statistics (number of events, global log-rank p-value, AIC, concordance index) — useful when arranging several forest plots in a panel. Default is TRUE (unchanged) (#392).
surv_pvalue() gains a pval.digits argument controlling the number of significant digits used to format the p-value in pval.txt (default 2, unchanged; e.g. pval.digits = 3 gives "p = 0.00131"). Requested for journals that report p-values to 3 digits (#343).
ggsurvtable() gains an hjust argument to control the horizontal justification of the table text (passed to geom_text). Default is 0.5 (centered, unchanged); use e.g. hjust = 0 for left-aligned counts (#629).
ggsurvplot() gains a linejoin argument controlling the line join of the survival curve. Default is "round" (unchanged); use linejoin = "mitre" for sharp corners that mark event times precisely (requires a ggplot2 version that passes linejoin through geom_step()). Previously linejoin passed via ... was silently dropped (#653).
Fix the number-at-risk table's first (t = 0) column not lining up with the survival curve's start when axes.offset = FALSE. The table panel starts exactly at x = 0 like the curve, but the t = 0 numbers were nudged right to max(xlim)/30, so they sat to the right of the curve's origin. They are now kept at x = 0 and only the t = 0 column is left-aligned (a centred number at x = 0 would be clipped by the panel edge); interior columns are unchanged. The default axes.offset = TRUE output, the log-axis path (which keeps the historical relocation, since x = 0 can't be shown on a log scale), and the inset risk.table.pos = "in" layout are all unchanged. cumevents/cumcensor tables get the same t = 0 alignment. Contributed by @nolankm (#645, #448). (The separate risk.table.pos = "in" inset misalignment in #302 is not addressed here and remains open.)
Improve the error message when ggsurvplot(fit) is called without data = for a model fitted in a non-global environment (e.g. inside a function or a {targets} pipeline). survminer re-derives the data from the fit's stored call, but the referenced object is then out of scope and a survfit stores no reference to its creation environment, so the data cannot be recovered. Instead of a cryptic object '<name>' not found, survminer now raises an actionable message telling the user to pass data = explicitly (e.g. ggsurvplot(fit, data = mydata)), while preserving the original error as context. Working inputs (data supplied, or a fit created in the global scope) are unchanged (#521).
Fix the "large dash" size not being changeable when tables.y.text = FALSE: p$table <- p$table + theme(axis.text.y = ggtext::element_markdown(size = ...)) was silently reset to the default (50) because the dash styling is re-applied when the plot is printed. The re-application now keeps a user-set size; the default is unchanged (#642).
Fix ggsurvplot(..., facet.by = , pval = TRUE) erroring with "variable lengths differ" when the model was built from a Surv object created outside the data and used as the formula left-hand side (e.g. Survival <- Surv(time, status); survfit(Survival ~ x, data = D), rather than Surv(time, status) ~ x). The per-panel p-value refit row-subset the data while the global response kept its full length; the response is now materialised once on the full data so it stays aligned under subsetting. In-formula fits are unchanged (#467).
Improve ggforest() handling of a non-converged Cox model (complete / quasi-complete separation), whose coefficient has a hazard ratio or confidence limit that overflows to Inf (or underflows to 0). Such a row can't be placed on the log axis; it previously produced a misleading full-width interval and a cryptic "log-10 transformation introduced infinite values" warning. ggforest() now emits a clear message naming the affected term(s), omits those rows from the drawn point/interval (they still appear in the table with their numeric labels), and sizes the axis from the finite rows. Converged models are unchanged (#406). The infinite/near-zero CI case was also reported by @rschauner (#582).
Fix customizing the risk table failing under ggplot2 >= 4.0 with "Can't merge the axis.text.y theme element" (e.g. p$table + theme(axis.text.y = element_text(...)), or ggsave()ing a customized table). In ggplot2 4.x theme elements are S7 objects and ggtext::element_markdown() (used to colour the risk-table y labels per strata) can no longer be merged with a user element_text(). On ggplot2 >= 4.0 the per-strata label colours are now applied with element_text(colour = <vector>) (native and mergeable); on older ggplot2 element_markdown() is kept (it merges fine there). The coloured labels are unchanged (#557).
Fix a factor level with a genuine trailing or leading space (e.g. "Obs, ") being parsed as NA in the grouping/facet columns. survfit right-pads level names in strata labels, so survminer must trim them; the trimmed value is now matched against the (also trimmed) factor levels while keeping the original levels, so such labels are recovered instead of dropped. Completes the strata special-character fixes (#616).
Fix ggsurvplot(..., risk.table = TRUE) erroring with "gridtext has encountered a tag that isn't supported yet: <blockquote>" when a strata label / legend.labs entry contains <, > or & (e.g. "> 1 Risk factor"). The risk-table y-axis labels are drawn with ggtext::element_markdown() (to colour them per strata), which parsed those characters as HTML tags; they are now HTML-escaped so they render literally. Ordinary labels are unchanged, and the plain-text path (risk.table.y.text.col = FALSE) is untouched (#532).
Fix strata parsing when a factor level contains special characters (=, >, $, etc.). survminer previously split each survfit stratum name on every =/, and mis-detected $, so such levels broke faceting, colouring, median lines, and legends. The parser now scopes the split to the known formula variable names (obtained from the fit), so levels may safely contain these characters. This fixes: ggsurvplot(..., facet.by=) erroring with "Unknown colour name" when a level has =/>= (#291, #430); a >=/< level used as facet.by erroring with "missing faceting variable" (#616); missing surv.median.line when a level has = (#599); and, for a level containing $, mis-parsed legend labels with the other variable's groups silently reordered (#680). Ordinary data (no special characters) is unchanged.
Fix ggcompetingrisks() (multi-state / survfitms competing-risks fits) sorting its facet panels alphabetically by strata name instead of following the model's strata order. The internal strata column is now a factor with levels in model order, so panels keep names(fit$strata) order. Fits whose strata order already is alphabetical are unchanged. Contributed by @muschellij2 (#470).
Fix ggforest() mislabeling the global p-value in the bottom caption as "Log-Rank": the value shown (broom::glance()$p.value.log) is the p-value of the likelihood ratio test (survival stores it in logtest), not the score/log-rank test (p.value.sc/sctest). The caption now reads "Global p-value (Likelihood ratio test)". The value itself is unchanged — only the label is corrected (#640).
Fix ggsurvplot() mis-rendering negative survival times: the x-axis was clipped at 0 (hiding the negative-time region) and a spurious vertical drop was drawn at x = 0. The plot now behaves like base plot.survfit() — the panel spans the true time range and each curve starts at survival = 1 from its first observed time, with no injected (0, 1) point. For all non-negative data (the common case) the output is byte-identical (the x-origin still resolves to exactly 0). The #523 warning that negative survival times are likely invalid is retained (#389).
ggadjustedcurves()/surv_adjustedcurves() now warn when the default method = "conditional" is used with a grouping variable that is not in the Cox model: in that case every group's curve is identical (a single visible line), which silently confused users. The warning points to method = "average"/"marginal" (which are designed for a variable absent from the model). The computed curves are unchanged (message only), and no warning fires for the other methods or when the variable is in the model (#623).
Fix ggforest() clipping the bottom global-statistics caption (number of events, global p-value, AIC, concordance index) in short plots (small figure heights): too little space was reserved below the last row. Space is now reserved so the caption always renders; it is only added when the caption is drawn, so global.stats = FALSE is unchanged (#696).
Fix ggforest() reporting a sample size that includes subjects with missing values: coxph() drops rows with a missing value in any model variable (na.action = na.omit), but ggforest() counted all rows of data, overstating the per-term/level N. The reported N now reflects the complete cases the model actually used (model$n). Models fit on data without missing values are unaffected (#597).
Fix ggsurvplot(fit, facet.by = "X") erroring with "subscript out of bounds" when every variable of the survival formula is also a facet.by variable, e.g. a null model Surv(...) ~ 1 faceted by X, or Surv(...) ~ X faceted by X. Each panel then shows a single curve with no within-panel grouping, so there is no extra strata to build; this case no longer calls the strata builder with zero variables (#304).
Fix the percentage at risk (pct.risk, used in risk.table = "percentage"/"abs_pct") exceeding 100% for a weighted survfit(): it was computed as n.risk * 100 / fit$n, but fit$n is the unweighted subject count while n.risk is weighted. When the weighting makes n.risk exceed fit$n, the denominator now falls back to the weighted number at risk at the origin. Unweighted fits (for which n.risk never exceeds fit$n) keep fit$n, so their output is unchanged (#561).
Fix break.y.by producing wrong axis breaks for transformed survival curves (fun = "cloglog", "event", "cumhaz") or a custom ylim outside [0, 1]: the y breaks were computed as seq(0, 1, by = break.y.by), so values outside [0, 1] had no breaks. They are now derived from the displayed y-range (the default survival plot is unchanged) (#378, #442).
Fix ggcoxzph() with cox.zph(..., transform = "log") drawing the fitted line on a different x-scale than the residual points (the line was squeezed to the far left): the fit line was drawn at log(pred.x) while the points and confidence bands were on the original time scale. The fit line now uses the same scale and the x-axis is log-scaled, matching survival::plot.cox.zph(log = "x") (#454, #588).
Fix pairwise_survdiff() erroring with "undefined columns selected" when the formula contains a strata() term (e.g. ~ rx + strata(sex)): strata(sex) is not a data column, so it could not be used to group/subset. strata() terms are now separated from the grouping variable and kept in the survdiff formula, giving a stratified pairwise test (#648).
Fix ggsurvplot(..., add.all = TRUE, pval = TRUE, pval.method = TRUE) drawing the p-value method (test name) as an empty string: the p-value is computed on the original fit and forwarded as text, so ggsurvplot_core() re-derived the method from the "all"-augmented fit and got "". The method is now drawn by ggsurvplot_add_all() (#673).
Fix ggsurvplot_facet(..., pval = "<string>") erroring with "argument is not interpretable as logical", and clarify the documentation: ggsurvplot_facet() computes a p-value for each panel, so (unlike ggsurvplot()) a numeric or character pval cannot be substituted. Such a value is now ignored with a warning instead of crashing (#636).
Fix ggsurvplot_facet(..., panel.labs = ...) failing with "cannot xtfrm data frames" when the data is a tibble: the panel-label code did as.factor(data[, var]), and tibble[, var] returns a one-column tibble rather than a vector. The data is now coerced to a plain data.frame at entry (#591).
Fix ggadjustedcurves(..., fun = ...) not transforming the curve: fun (e.g. "event", "cumhaz", "pct") only changed the y-axis limits, while the plotted curve stayed the raw survival probability. The transformation is now applied to the curve (a no-op when fun = NULL, the default) (#287, #498, #660).
Fix ggforest() crashing (axisTicks(): '_LARGE_ range') for a Cox model with complete/quasi-complete separation, where a coefficient is near-infinite: the x-axis range is now clamped to a finite window (with a warning) so the plot still renders (#570, #590).
Fix ggforest() producing duplicated/crossed rows for prefix-colliding non-factor term names (e.g. logical covariates add11 and add17): coefficients were matched to terms with a regex ("^var*.") that treated trailing digits as a quantifier, so each name matched the other's coefficient. Coefficients are now mapped to their term via model$assign (#689).
Fix ggforest() reference level inheriting the statistics of a similarly-named non-reference level (e.g. a reference level Bar showing the hazard ratio of Barb): term rows were matched by character row indexing, which partial-matches. They are now matched exactly, so the reference level is correctly shown as the reference (#312).
Fix ggforest() erroring with "undefined columns selected" when the Cox formula contains an in-formula factor transformation such as as.factor(x): the term is now evaluated rather than looked up as a column name (a plain column name is unaffected) (#240).
Fix the cumulative number-of-events and number-censored columns showing decimals for a weighted survfit() (in the risk/cumevents/cumcensor tables): the cumulative sums of the fractional weighted counts were not rounded, unlike the per-interval columns. They are now rounded to the same precision (#560, #554).
Fix ggsurvplot_combine() ignoring the risk-table type: risk.table = "nrisk_cumcensor" (and other types) is now honoured instead of always showing the absolute number at risk (#641).
Fix ggsurvplot_combine() ignoring risk.table.fontsize: the risk-table text size can now be set with risk.table.fontsize (as in ggsurvplot()), not only with fontsize (#514).
Fix surv_categorize() returning the raw numeric values (instead of "high"/"low") for variables whose names contain characters that make.names() alters, such as a hyphen (e.g. gene names like "A1BG-AS1"): summary.surv_cutpoint() now builds its row names with check.names = FALSE so the name still matches (#609). The hyphenated-gene-name case was also reported by @hmkim (#502).
Remove an unused, undefined alpha argument passed by surv_cutpoint() to maxstat::maxstat.test() (it only worked by lazy evaluation); no change to computed cut points (#598).
Fix ggsurvplot() clipping events that occur after the last x-axis break: the default upper x-limit was the largest "nice" axis break, which can fall below the largest event/censoring time, so those events were invisible unless xlim was set manually. The upper x-limit now extends to cover the maximum time; the axis tick breaks (and risk-table columns) are unchanged, and a user-supplied xlim is still honoured (#655).
ggsurvplot() (and the related builders such as ggsurvplot_combine()) now warn when the survival data contain negative times, which make the Kaplan-Meier curve appear to increase (not meaningful for a survival estimate). Previously such times were plotted silently. The plot itself is unchanged (#523).
Fix ggcoxfunctional() clipping most of its own points: the default y-axis limits were set to the range of the lowess smoother, which is much narrower than the martingale residuals plotted as points, so points with large residuals fell outside the visible panel. The y-axis now auto-scales to include all points by default; a user-supplied ylim is still honoured (#465).
Fix ggcoxfunctional() erroring with "'x' and 'y' lengths differ" when the Cox formula contains a covariate with missing values: model.matrix() drops rows with missing terms, but the null-model martingale residuals were computed from the full data. The data is now restricted to the complete-case (model-matrix) rows so the lengths match (#248).
Fix ggcoxfunctional() erroring with a cryptic "'x' and 'y' lengths differ" when the Cox formula contains a factor/character covariate (or a strata() term): such terms are renamed/expanded in the model matrix and cannot be checked for functional form. They are now dropped with a warning, and only continuous covariates are plotted (#357). Dropping non-continuous terms was also proposed by @DanChaltiel (#410).
Fix ggsurvplot_combine(..., surv.median.line = "hv") (or "h"/"v") not drawing the median survival lines: surv.median.line was forwarded to ggsurvplot_df() (which does not handle it) and silently ignored. The median lines are now computed from the combined fits and drawn on the plot (#316).
Fix ggsurvplot_facet(..., pval = TRUE, pval.size = <n>) ignoring pval.size: the per-facet p-value (and pval.method) text was drawn with ggplot2's default size and pval.size was silently routed into .... pval.size is now an explicit argument applied to the text; its default is 5, consistent with ggsurvplot() (previously the faceted p-value text rendered at ggplot2's default size ~3.88) (#338).
Fix ggcompetingrisks(..., multiple_panels = FALSE, conf.int = TRUE) drawing incorrect confidence bands that jumped between groups of the same event: the ribbon was grouped only by event; it is now grouped by interaction(event, group) (#490).
Fix ggsurvplot(..., add.all = TRUE) (and ggsurvplot_add_all()) erroring with "argument matches multiple formal arguments" when a legend position was supplied: legend partial-matched legend.title/legend.labs; it is now an explicit forwarded argument (#566).
Fix ggflexsurvplot() collapsing a grouped Kaplan-Meier curve to a single "All" stratum when the grouping covariate is a factor: is_factor_or_character() called ggplot2's is.facet() (a Facet-object test, always FALSE for a data column) instead of is.factor() (#408).
Fix surv_group_by() (and downstream ggsurvplot_facet() / grouped surv_pvalue()) failing with "cannot xtfrm data frame" when the input is a tibble: extract the grouping column with data[[var]] (a vector) instead of data[, var] (a one-column tibble) (#548, #670). The same one-line fix was earlier contributed by @yonicd (#549).
Fix ggflexsurvplot() erroring with "object 'data = supplied): the already-resolved data is now forwarded to the internal .extract.survfit() instead of being re-derived from fit$call$data (#436).
Fix ggsurvplot(..., ncensor.plot = TRUE) erroring at draw time with "Unknown colour name: strata" for a single-group fit (~ 1): the default color = "strata" was passed to the censoring bar plot as a literal colour when the data has no strata column. The bars now use the survival curve's own colour for that case only (falling back to black if it cannot be resolved), leaving grouped fits and explicit colours unchanged (#298).
Fix ggadjustedcurves() / surv_adjustedcurves() failing with "cannot xtfrm data frame" when a tibble is passed as data (or as reference for the marginal method): both are coerced to a plain data.frame so the internal data[, variable] extractions return a vector (#501, #628).
Fix ggsurvplot() erroring with "'names' attribute [n] must be the same length as the vector [m]" when the palette contains a duplicated colour (or a default palette rendered a repeated hue) together with risk.table / ncensor.plot: the internal colour extractor used unique(), collapsing duplicate colours; it now returns one colour per group (#397, #519, #595, #691).
Fix ggsurvplot() and surv_pvalue() erroring with "object of type 'symbol' is not subsettable" for a survfit built from a formula stored in a variable (e.g. frm <- Surv(time, status) ~ sex; survfit(frm, data)): the stored formula (a symbol) is now resolved with stats::formula(fit) instead of as.formula(fit$call$formula) (#324, #341, #602).
survMisc dependency: weighted log-rank tests (Gehan-Breslow, Tarone-Ware, Peto-Peto, modified Peto-Peto, Fleming-Harrington, test-for-trend) are now computed internally using base R, avoiding the at-risk survMisc package. All existing method arguments in surv_pvalue() and log.rank.weights in ggsurvplot() continue to work identically.size with linewidth for line geoms, and is.ggplot() with is_ggplot() (#692, #693)GeomConfint incompatibility with ggplot2 4.0.x: normalize linewidth and linetype after stairstep transformation to prevent "Aesthetics can not vary along a ribbon" error (#694)surv_fit() error when using list of formulas with list of data sets (match.fd = FALSE): replace defunct dplyr::combine() with unlist(recursive = FALSE) (#697, #699).add_surv_median()ggcoxdiagnostics() x-axis scaling when using ox.scale = "time" with Schoenfeld residuals (#608)surv.median.line = "hv" or "h" with multiple survival curves (#643)theme_survminer() to ensure proper theme object construction$layers) and development (@layers) versions%++% operator instead of + for adding themes to ggsurv objects with ggplot2 v >= 3.5.2%++% operator in theme_survminer() help\link{} targets missing package added in the following Rd files:
ggcoxzph() (#534 and #535)element_text() issues a warning when vectorized arguments are provided, as in colour = c("red", "green", "blue"). This is a breaking change affecting the function ggsurvtable(). To fix this, the function ggtext::element_markdown() is now used in place of element_text() to handle vectorized colors (issue #455 fixed by pull #503).log.rank.weights = "n" is specified in the function ggsurvplot() (#453)ggsurvplot() examples, the function gridExtra::rbind.gtable() is now replaced by gridExtra::gtable_rbind() (@jan-imbi, pull #493).ggforest() (pull 485).survfit(res.cox) returned an object of class survfit.cox. The class has been changed to survfitcox in the current survival package version. The survminer package has been now updated to take this change into account (@edvbb, #441).Fixes to adapt to dplyr 1.0.0 (@romainfrancois, #460):
conf.int in the ggsurvplot() function. To fix this issue, Now, NAs are removed by default when drawing the confidence interval (#443 and #315).surv_adjustedcurves is extracted from ggadjustedcurves. This function calculates adjusted survival curves but do not plot them. Its results may be useful for calculation of median survival or some other statistics. (@pbiecek, #423).cmprsk is no longer needed for survminer installation. The package has been moved from Imports to Suggests. It's only used in documentations (@massimofagg, #394.ggflexsurvplot(), the grouping variable can be factor or character vector (@andersbergren , #393anova() as requested (@pbiecek, #391When a factor variable name is the same as one of its level, ggsurvplot() failed (@KohSoonho, #387). Fixed now.
ggsurvplot() can now create correctly faceted survival curves (@uraniborg, #254, @BingxinS, #363)
A typo fixed in the formula for weightened log-rank test (@MarcinKosinski, #336.
surv_summary() can now handle the output of survfit(cox.model, newdata) when the option conf.type = "none" is specified by users (@HeidiSeibold, #335.
ggadjustedcurves() has now flipped labels for conditional/marginal to mach names from ’Adjusted Survival Curves’ by Terry Therneau, Cynthia Crowson, Elizabeth Atkinson (2015) (@pbiecek, #335.
ggsurvplot() can be used to plot survreg model (@HeidiSeibold, #276, #325 ).ggforest() simply returns a ggplot instead of drawing automatically the plot (@grvsinghal, #267).axes.offset argument is also applied to risk table (@dmartinffm, #243).ggsurvplot to powerpoint document using ReporteRs even if there is no risk table (@DrRZ, #314).size added in ggadjustedcurves() to change the curve size (@MaximilianTscharre, #267).ggtheme is supported when combining a list of survfit objects in ggsurvplot() (@PhonePong, #278).New function ggflexsurvplot() to create ggplot2-based graphs for flexible survival models.
The function ggadjustedcurves() handles now argument method that defines how adjusted curves shall be calculated. With method='conditional'|'marginal' subpopulations are balanced with respect to variables present in the model formula. With method='single'|'average' the curve represents just the expected survival curves.
ggcoxadjustedcurves() is replaced by ggadjustedcurves() (#229).The grouping variable to the ggadjustedcurves() function is now passed as a name (character) of grouping variable not as a vector with values of grouping variable.
New argument font.family in ggsurvtable() to change the font family in the survival tables - such as risk, cummulative events and censoring tables. For example font.family = "Courier New" (@Swechhya, #245).
Now, in ggsurvplot() the data argument should be strictly provided (@dnzmarcio, #235)
ggforest() no longer tries to bolt a table full of text to the coefficient plot (@mmoisse, #241), instead the annotations are done via ggplot2::annotate, see example at: @fabian-s, #264New argument test.for.trend added in ggsurvplot() to perform a log-rank test for trend. logical value. Default is FALSE. If TRUE, returns the test for trend p-values. Tests for trend are designed to detect ordered differences in survival curves. That is, for at least one group. The test for trend can be only performed when the number of groups is > 2 (#188).
New argument add.all added now in ggsurvplot() to add he survival curves of (all) pooled patients onto the main survival plot stratified by grouping variables. Alias of the ggsurvplot_add_all() function (#194).
New argument combine = TRUE is now available in the ggsurvplot() function to combine a list of survfit objects on the same plot. Alias of the ggsurvplot_combine() function (#195).
The standard convention of ggplot2 is to have the axes offset from the origin. This can be annoying with Kaplan-Meier plots. New argument axes.offset added non in ggsurvplot(). logical value. Default is TRUE. If FALSE, set the plot axes to start at the origin (c(0,0)) (#196).
The function ggsurvplot() can take a list of survfit objects and produces a list of ggsurvplots (#204).
New argument facet.by added now in ggsurvplot() to draw multi-panel survival curves of a data set grouped by one or two variables. Alias of the ggsurvplot_facet() function (#205).
New argument group.by added now in ggsurvplot() to create survival curves of grouped data sets. Alias of the ggsurvplot_group_by() function.
In ggsurvplot(), one can specify pval = TRUE/FALSE as a logical value. Now, it's also possible to specify the argument pval as a numeric value (e.g.: pval = 0.002), that will be passed to the plot, so that user can pass any custom p-value to the final plot (@MarcinKosinski, #189) or one can specify it as a character string (e.g.: pval = "p < 0001") (@MarcinKosinski, #193).
New argument xscale in ggsurvplot(): numeric or character value specifying x-axis scale.
New arguments censor.shape and censor.size to change the shape and the shape of censors (#186 & #187).
New argument conf.int.alpha added in ggsurvplot(). Numeric value specifying fill color transparency. Value should be in [0, 1], where 0 is full transparency and 1 is no transparency.
New function surv_group_by() added to create a grouped data set for survival analysis.
New function ggsurvplot_df() added. An extension to ggsurvplot() to plot survival curves from any data frame containing the summary of survival curves as returned the surv_summary() function. Might be useful for a user who wants to use ggsurvplot for visualizing survival curves computed by another method than the standard survfit.formula function. In this case, the user has just to provide the data frame containing the summary of the survival analysis.
New function surv_median() added to easily extract median survivals from one or a list of survfit objects (#207).
New function surv_pvalue() added to compute p-value from survfit objects or parse it when provided by the user. Survival curves are compared using the log-rank test (default). Other methods can be specified using the argument method.
New function surv_fit() added to handle complex situation when computing survival curves (Read more in the doc: ?surv_fit). Wrapper arround the standard survfit() [survival] function to create survival curves. Compared to the standard survfit() function, it supports also:
ggforest() function has changed a lot. Now presents much more statistics for each level of each variable (extracted with broom::tidy) and also some statistics for the coxph model, like AIC, p.value, concordance (extracted with broom::glance) (#178)Now, ggcompetingrisks() supports the conf.int argument. If conf.int=TRUE and fit is an object of class cuminc then confidence intervals are plotted with geom_ribbon.
Now, ggsurvplot() supports the survfit() outputs when used with the argument start.time.
Now, the default behaviour of ggsurvplot() is to round the number at risk using the option digits = 0 (#214).
pairwise_survdiff() has been improved to handle a formula with multiple variables (#213).
The argument color are updated allowing to assign the same color for same groups accross facets (#99 & #185).
For example, in the following script, survival curves are colored by the grouping variable sex in all facets:
library(survminer)
library(survival)
fit <- survfit( Surv(time, status) ~ sex + rx + adhere,
data = colon )
ggsurv <- ggsurvplot(fit, data = colon,
color = "sex",
legend.title = "Sex",
palette = "jco")
ggsurv$plot + facet_grid(rx ~ adhere)
Now, the function pairwise_survdiff() checks whether the grouping variable is a factor. If this is not the case, the grouping variable is automatically converted into a factor.
ggsurvplot(): Now, log scale is used for x-axis when plotting the complementary log−log function (argument `fun = "cloglog") (#171).
Now, the argument palette in ggsurvplot() ccan be also a numeric vector of length(strata); in this case a basic color palette is created using the function grDevices::palette().
The %+% function in survminer has been replaced by %++% to avoid breaking the ggplot2::%+% function behavior when using survminer (#199 and #200).
New argument fun added in ggcoxadjustedcurves() (@meganli, #202).
The function theme_classic2() removed.
Columns/Rows are now correctly labeled in pairwise_survdiff() display (@mriffle, #212).
Now, the pairwise_survdiff() function works when the data contain NAs (@emilelatour , #184).
Now, ggsurvplot() fully supports different methods, in the survMisc package, for comparing survival curves (#191).
ggcoxdiagnostics() function and the vignette file Informative_Survival_Plots.Rmd have been updated so that survminer can pass CRAN check under R-oldrelease.BMT added for competing risk analysis.BRCAOV.survInfo added, used in vignette filespalette argument works in `ggcoxadjustedcurves() (#174)ggsurvplot() works when the fun argument is an arbitrary function (#176).Additional data argument added to the ggsurvplot() function (@kassambara, #142). Now, it's recommended to pass to the function, the data used to fit survival curves. This will avoid the error generated when trying to use the ggsurvplot() function inside another functions (@zzawadz, #125).
New argument risk.table.pos, for placing risk table inside survival curves (#69). Allowed options are one of c("out", "in") indicating 'outside' or 'inside' the main plot, respectively. Default value is "out".
New arguments tables.height, tables.y.text, tables.theme, tables.col: for customizing tables under the main survival plot: (#156).
New arguments cumevents and cumcensor: logical value for displaying the cumulative number of events table (#117) and the cumulative number of censored subject (#155), respectively.
Now, ggsurvplot() can display both the number at risk and the cumulative number of censored in the same table using the option risk.table = 'nrisk_cumcenor' (#96). It's also possible to display the number at risk and the cumulative number of events using the option risk.table = 'nrisk_cumevents'.
New arguments pval.method and log.rank.weights: New possibilities to compare survival curves. Functionality based on survMisc::comp.
New arguments break.x.by and break.y.by, numeric value controlling x and y axis breaks, respectively.
Now, ggsurvplot() returns an object of class ggsurvplot which is list containing the following components (#158):
New function theme_survminer() to change easily the graphical parameters of plots generated with survminer (#151). A theme similar to theme_classic() with large font size. Used as default theme in survminer functions.
New function theme_cleantable() to draw a clean risk table and cumulative number of events table. Remove axis lines, x axis ticks and title (#117 & #156).
# Fit survival curves
require("survival")
fit<- survfit(Surv(time, status) ~ sex, data = lung)
# Survival curves
require("survminer")
ggsurvplot(fit, data = lung, risk.table = TRUE,
tables.theme = theme_cleantable()
)
+.ggsurv() to add ggplot components - theme(), labs() - to an object of class ggsurv, which is a list of ggplots. (#151). For example:# Fit survival curves
require("survival")
fit<- survfit(Surv(time, status) ~ sex, data = lung)
# Basic survival curves
require("survminer")
p <- ggsurvplot(fit, data = lung, risk.table = TRUE)
p
# Customizing the plots
p %+% theme_survminer(
font.main = c(16, "bold", "darkblue"),
font.submain = c(15, "bold.italic", "purple"),
font.caption = c(14, "plain", "orange"),
font.x = c(14, "bold.italic", "red"),
font.y = c(14, "bold.italic", "darkred"),
font.tickslab = c(12, "plain", "darkgreen")
)
New function arrange_ggsurvplots() to arrange multiple ggsurvplots on the same page (#66).
New function ggsurvevents() to calculate and plot the distribution for events (both status = 0 and status = 1); with type parameter one can plot cumulative distribution of locally smooth density; with normalised, distributions are normalised. This function helps to notice when censorings are more common (@pbiecek, #116).
New function ggcoxadjustedcurves() to plot adjusted survival curves for Cox proportional hazards model (@pbiecek, #133 & @markdanese, #67).
New function ggforest() for drawing forest plot for the Cox model.
New function pairwise_survdiff() for multiple comparisons of survival Curves (#97).
New function ggcompetingrisks() to plot the cumulative incidence curves for competing risks (@pbiecek, #168.
New heper functions ggrisktable(), ggcumevents(), ggcumcensor(). Normally, users don't need to use these function directly. Internally used by the function ggsurvplot().
ggrisktable() for plotting number of subjects at risk by time. (#154).ggcumevents() for plotting the cumulative number of events table (#117).ggcumcensor() for plotting the cumulative number of censored subjects table (#155).New argument sline in the ggcoxdiagnostics() function for adding loess smoothed trend on the residual plots. This will make it easier to spot some problems with residuals (like quadratic relation). (@pbiecek, #119).
The design of ggcoxfunctional() has been changed to be consistent with the other functions in the survminer package. Now, ggcoxfunctional() works with coxph objects not formulas. The arguments formula is now deprecated (@pbiecek, #115).
In the ggcoxdiagnostics() function, it's now possible to plot Time in the OX axis (@pbiecek, #124). This is convenient for some residuals like Schoenfeld. The linear.predictions parameter has been replaced with ox.scale = c("linear.predictions", "time", "observation.id").
New argument tables.height in ggsurvplot() to apply the same height to all the tables under the main survival plots (#157).
It is possible to specify title and caption for ggcoxfunctional (@MarcinKosinski, #138) (font.main was removed as it was unused.)
It is possible to specify title, subtitle and caption for ggcoxdiagnostics (@MarcinKosinski, #139) and fonts for them.
It is possible to specify global caption for ggcoxzph (@MarcinKosinski, #140).
In ggsurvplot(), more information, about color palettes, have been added in the details section of the documentation (#100).
The R package maxstat doesn't support very well an object of class tbl_df. To fix this issue, now, in the surv_cutpoint() function, the input data is systematically transformed into a standard data.frame format (@MarcinKosinski, #104).
It's now possible to print the output of the survminer packages in a powerpoint created with the ReporteRs package. You should use the argument newpage = FALSE in the print() function when printing the output in the powerpoint. Thanks to (@abossenbroek, #110) and (@zzawadz, #111). For instance:
require(survival)
require(ReporteRs)
require(survminer)
fit <- survfit(Surv(time, status) ~ rx + adhere, data =colon)
survplot <- ggsurvplot(fit, pval = TRUE,
break.time.by = 400,
risk.table = TRUE,
risk.table.col = "strata",
risk.table.height = 0.5, # Useful when you have multiple groups
palette = "Dark2")
require(ReporteRs)
doc = pptx(title = "Survival plots")
doc = addSlide(doc, slide.layout = "Title and Content")
doc = addTitle(doc, "First try")
doc = addPlot(doc, function() print(survplot, newpage = FALSE), vector.graphic = TRUE)
writeDoc(doc, "test.pptx")
ggcoxdiagnostics(), the option ncol = 1 is removed from the function facet_wrap(). By default, ncol = NULL. In this case, the number of columns and rows in the plot panels is defined automatically based on the number of covariates included in the cox model.Now, risk table align with survival plots when legend = "right" (@jonlehrer, #102).
Now, ggcoxzph() works for univariate Cox analysis (#103).
Now, ggcoxdiagnostics() works properly for schoenfeld residuals (@pbiecek, #119).
Now, ggsurvplot() works properly in the situation where strata() is included in the cox formula (#109).
A new vignette and a ggsurvplot example was added to present new functionalities of possible texts and fonts customizations.
A new vignette and a ggsurvplot example was added to present new functionalities of possible weights specification in a Log-rank test.
surv_summary() (v0.2.3) generated an error when the name of the variable used in survfit() can be found multiple times in the levels of the same variable. For example, variable = therapy; levels(therapy) --> "therapy" and "hormone therapy" (#86). This has been now fixed.
To extract variable names used in survival::survfit(), the R code strsplit(strata, "=|,\\s+", perl=TRUE) was used in the surv_summary() function [survminer v0.2.3]. The splitting was done at any "=" symbol in the string, causing an error when special characters (=, <=, >=) are used for the levels of a categorical variable (#91). This has been now fixed.
Now, ggsurvplot() draws correctly the risk.table (#93).
surv_summary() for creating data frame containing a nice summary of a survival curve (#64).ggsurvplot() by one or more factors (#64):# Fit complexe survival curves
require("survival")
fit3 <- survfit( Surv(time, status) ~ sex + rx + adhere,
data = colon )
# Visualize by faceting
# Plots are survival curves by sex faceted by rx and adhere factors.
require("survminer")
ggsurv$plot +theme_bw() + facet_grid(rx ~ adhere)
ggsurvplot() can be used to plot cox model (#67).surv_cutpoint(): Determine the optimal cutpoint for each variable using 'maxstat'. Methods defined for surv_cutpoint object are summary(), print() and plot().surv_categorize(): Divide each variable values based on the cutpoint returned by surv_cutpoint() (#41).ggsurvplot(). A logical value. If TRUE, the number of censored subjects at time t is plotted. Default is FALSE (#18).ggsurvplot() for changing the style of confidence interval bands.ggsurvplot() plots a stepped confidence interval when conf.int = TRUE (#65).ggsurvplot() updated for compatibility with the future version of ggplot2 (v2.2.0) (#68)fun. For example, if fun = "event", then ylab will be "Cumulative event".ggsurvplot(), linetypes can now be adjusted by variables used to fit survival curves (#46)ggsurvplot(), the argument risk.table can be either a logical value (TRUE|FALSE) or a string ("absolute", "percentage"). If risk.table = "absolute", ggsurvplot() displays the absolute number of subjects at risk. If risk.table = "percentage", the percentage at risk is displayed. Use "abs_pct" to show both the absolute number and the percentage of subjects at risk (#70).ggsurvplot(): character vector for drawing a horizontal/vertical line at median (50%) survival. Allowed values include one of c("none", "hv", "h", "v"). v: vertical, h:horizontal (#61).ggcoxdiagnostics() can now handle a multivariate Cox model (#62)ggcoxfunctional() now displays graphs of continuous variable against martingale residuals of null cox proportional hazards model (#63).ggsurvplot() to report the right p-value on the subset of the data and not on the whole data sets (@jseoane, #71).ggcoxzph() can now produce plots only for specified subset of varibles (@MarcinKosinski, #75)ggcoxdiagnostics function that plots diagnostic graphs for Cox Proportional Hazards model (@MarcinKosinski, #16).Survival plots have never been so informative (@MarcinKosinski, #39)ggsurvplot() documentation. (@ViniciusBRodrigues, #43)New ggcoxzph function that displays a graph of the scaled Schoenfeld residuals, along with a smooth curve using 'ggplot2'. Wrapper around \link{plot.cox.zph}. (@MarcinKosinski, #13)
New ggcoxfunctional function that displays graphs of continuous explanatory variable against martingale residuals of null
cox proportional hazards model, for each term in of the right side of input formula. This might help to properly choose the functional form of continuous variable in cox model, since fitted lines with lowess function should be linear to satisfy cox proportional hazards model assumptions. (@MarcinKosinski, #14)
New function theme_classic2: ggplot2 classic theme with axis line. This function replaces ggplot2::theme_classic, which does no longer display axis lines (since ggplot2 v2.1.0)
risk.table.y.text.col is now TRUE.ggsurvplot. logical argument. Default is TRUE. If FALSE, risk table y axis tick labels will be hidden (@MarcinKosinski, #28).New arguments in ggsurvplot for changing font style, size and color of main title, axis labels, axis tick labels and legend labels: font.main, font.x, font.y, font.tickslab, font.legend.
New arguments risk.table.title, risk.table.fontsize in ggsurvplot
New argument risk.table.y.text.col: logical value. Default value is FALSE. If TRUE, risk table tick labels will be colored by strata (@MarcinKosinski, #8).
print.ggsurvplot() function added: S3 method for class 'ggsurvplot'.
ggsurvplot returns an object of class ggsurvplot which is list containing two ggplot objects:
It's now possible to customize the output survival plot and the risk table returned by ggsurvplot, and to print again the final plot. (@MarcinKosinski, #2):
# Fit survival curves
require("survival")
fit<- survfit(Surv(time, status) ~ sex, data = lung)
# visualize
require(survminer)
ggsurvplot(fit, pval = TRUE, conf.int = TRUE,
risk.table = TRUE)
# Customize the output and then print
res <- ggsurvplot(fit, pval = TRUE, conf.int = TRUE,
risk.table = TRUE)
res$table <- res$table + theme(axis.line = element_blank())
res$plot <- res$plot + labs(title = "Survival Curves")
print(res)
ggtheme now affects risk.table (@MarcinKosinski, #1)
xlim changed to cartesian coordinates mode (@MarcinKosinski, #4). The Cartesian coordinate system is the most common type of coordinate system. It will zoom the plot (like you’re looking at it with a magnifying glass), without clipping the data.
Risk table and survival curves have now the same color and the same order
Plot width is no longer too small when legend position = "left" (@MarcinKosinski, #7).