mlr3cmprsk

Package website: mlr3cmprsk

mlr3cmprsk extends the mlr3 ecosystem for machine learning with competing risks right-censored survival outcomes. It provides task, learner, prediction, and measure abstractions that support model training, resampling, benchmarking, and evaluation within the mlr3 framework.

Installation

Install the development version from GitHub:

# install.packages("pak")
# pak::pak("mlr-org/mlr3cmprsk")
install.packages("mlr3cmprsk", repos = c('https://mlr-org.r-universe.dev', 'https://cloud.r-project.org'))

Example

Compare a Fine-Gray model with the Aalen-Johansen baseline using 3-fold CV on the built-in pbc task.

library(mlr3cmprsk)
set.seed(42L)

task = tsk("pbc")
task$select(c("age", "chol", "albumin", "bili"))
# Stratify by event status so all splits keep all available causes
task$set_col_roles(cols = "status", add_to = "stratum")

learners = lrns(c("cmprsk.fg", "cmprsk.aalen"))

bm_grid = benchmark_grid(task, learners, rsmp("cv", folds = 3L))
bm = benchmark(bm_grid)

measures = list(
  # AUC(t = 100)
  msr("cmprsk.auc", time = 100, id = "auc_t100"),
  # BS(t = 100)
  msr("cmprsk.brier", time = 100, id = "bs_t100"),
  # IBS: 100 equally spaced times from 0 to each test fold's 90th follow-up percentile
  msr("cmprsk.ibs", id = "ibs")
)

# By default, each measure averages the cause-specific scores, with weights
# equal to the observed event frequencies in each test fold
bm$score(measures)[, .(task_id, learner_id, iteration, auc_t100, bs_t100, ibs)]

##    task_id   learner_id iteration  auc_t100   bs_t100       ibs
## 1:     pbc    cmprsk.fg         1 0.7813865 0.1564736 0.1088478
## 2:     pbc    cmprsk.fg         2 0.7663073 0.1750273 0.1155973
## 3:     pbc    cmprsk.fg         3 0.8390308 0.1371346 0.1028685
## 4:     pbc cmprsk.aalen         1 0.5000000 0.2182071 0.1487853
## 5:     pbc cmprsk.aalen         2 0.5000000 0.2200391 0.1585370
## 6:     pbc cmprsk.aalen         3 0.5000000 0.2181846 0.1609251

Measures

All currently available measures use cumulative incidence function (CIF) predictions and are calculated with riskRegression::Score(); the Integrated Brier Score (IBS) is the default measure for competing risks tasks.

ID	Measure	Category	Prediction Type
cmprsk.ibs	Integrated Brier score	Scoring Rule	cif
cmprsk.brier	Brier score at a specified time	Scoring Rule	cif
cmprsk.auc	Time-dependent AUC at a specified time	Discrimination	cif

Learners

The package includes two learners, both of which return CIF predictions for every cause.

ID	Learner	Underlying Package
cmprsk.aalen	Aalen-Johansen estimator	survival::survfit()
cmprsk.fg	Fine-Gray subdistribution hazards model	cmprsk::crr()

For additional models, see the available list in mlr3extralearners, including CoxBoost (cmprsk.coxboost) and random forests (cmprsk.rfsrc).

We welcome contributions and pull requests to add new competing risks learners in mlr3extralearners.

Citation

If you use mlr3cmprsk in your work, please cite it using the following BibTeX entry:

@Manual{mlr3cmprsk,
  title = {mlr3cmprsk: Competing Risks Machine Learning for 'mlr3'},
  author = {John Zobolas and Andreas Bender},
  year = {2026},
  note = {R package version 0.0.7, https://github.com/mlr-org/mlr3cmprsk},
  url = {https://mlr3cmprsk.mlr-org.com},
}

Code of Conduct

This project follows a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.