LoginSign Up
← Back to all sparks
S

survminer

ANALYTICS
Velocity0.0

Survival Analysis and Visualization

github.com/kassambara/survminer ↗

survminer is back after a five-year gap, and spending it cutting dependencies loose.

rsurvival-analysisggplot2dependency-reductionbiostatistics
◆Current state
survminer draws Kaplan-Meier curves and Cox diagnostics on top of ggplot2, and it sits under a large share of published survival figures. After a long dormancy it has shipped three releases since late 2024, and all of them are about staying compatible: with ggplot2 3.5, then 4.0, and now without survMisc, whose weighted log-rank tests are computed internally in base R as of 0.5.2. The plotting API itself has not changed.
◆Where it's heading
This is a revival by maintenance, not by feature work — the same pattern as its sibling packages from the same author. The dependency surface is being reduced rather than extended, and each release absorbs a breaking change from ggplot2 that would otherwise leave existing scripts producing errors or misaligned risk tables. Compatibility fixes now arrive within months of the upstream change rather than years.
◆Prediction
Continued ggplot2 4.x tracking is the most likely next release content, since 0.5.2 already carries fixes for 4.0.x aesthetics and the GeomConfint stairstep problem.

◆Recent moves

  1. 5mo ago

    survMisc dependency removed; weighted log-rank tests moved in-house

    Drops the at-risk survMisc dependency by reimplementing the weighted log-rank tests — Gehan-Breslow, Tarone-Ware, Peto-Peto, Fleming-Harrington and the trend test — in base R, with the notes stating existing method arguments behave identically. Also fixes GeomConfint under ggplot2 4.0.x and a surv_fit() failure using lists of formulas.

    View source ↗
  2. 11mo ago

    Risk-table misalignment fixed via dynamic panel detection

    Fixes chart and risk-table misalignment under ggplot2 3.5 by detecting panels dynamically instead of using hardcoded grob indices, plus several aesthetic and diagnostic fixes. The misalignment was visible in every plot with a risk table, so this is the release that made the package usable again on current ggplot2.

    View source ↗
  3. 1y ago

    ggcoxzph() sd scaling corrected; tidyverse deprecations cleared

    Raises the ggplot2 floor to 3.4.0, replaces deprecated tidyverse underscore functions and aes_string, and corrects the standard-deviation scaling factor in ggcoxzph(). That last fix changes what proportional-hazards diagnostic plots show.

    View source ↗
  4. 5y ago

    Gehan-Breslow p-value corrected for log.rank.weights = n

    Corrects the Gehan-Breslow p-value when log.rank.weights is set to n, and works around ggplot2 3.3.0 rejecting vectorized element_text colors by switching ggsurvtable() to ggtext::element_markdown(). A result-changing fix in the same weighted-test family that 0.5.2 would later reimplement.

    View source ↗
  5. 6y ago

    broom 0.7.0 compatibility fix in ggforest()

    A single maintenance change setting conf.int explicitly in ggforest()'s call to tidy.coxph for broom 0.7.0. Upstream churn absorption.

    View source ↗
  6. 6y ago

    surv_adjustedcurves() separates computation from plotting

    Splits surv_adjustedcurves() out of ggadjustedcurves() so adjusted survival curves can be computed without plotting them, making median survival and other statistics available from the same machinery. A rare separation of calculation from drawing in this package.

    View source ↗