Skip to contents

SuperSurv: A Unified Framework for Machine Learning Ensembles in Survival Analysis

R-CMD-check Lifecycle: experimental

SuperSurv is an R package for building, evaluating, and interpreting ensemble models for right-censored survival data.

At its core, the package implements a Super Learner-style ensemble framework for continuous-time survival prediction under right censoring. Using inverse probability of censoring weighting (IPCW), it combines heterogeneous base learners by minimizing cross-validated prediction risk. The framework supports learners that return full survival curves as well as learners that return only risk scores, which are calibrated to a common survival-probability scale on a shared evaluation time grid.

Beyond ensemble fitting, SuperSurv provides tools for:

  • hyperparameter tuning
  • high-dimensional screening
  • time-dependent model benchmarking
  • SHAP-based interpretability
  • covariate-adjusted restricted mean survival time (RMST) contrasts

The package also provides a more user-friendly model interface through print(), summary(), coef(), and exported accessors such as event_weights(), censor_weights(), learner_names(), training_variables(), selected_variables(), and eval_times().

🚀 Get Started

The best place to start is the installation and setup tutorial:

👉 Tutorial 0: Installation & Setup

You can also browse the full documentation site here:

👉 SuperSurv website

📦 Installation

Install the CRAN release:

install.packages("SuperSurv")

Install the development version from GitHub:

# install.packages("devtools")
devtools::install_github("yuelyu21/SuperSurv")

📦 Dependency philosophy

To keep installation lightweight, several heavier machine-learning engines are listed in Suggests rather than imported as strict dependencies. This means users can install only the modeling backends they plan to use. If a requested learner is unavailable, SuperSurv will prompt the user to install the required package.

📚 Included learners and screeners

SuperSurv currently standardizes a broad set of prediction wrappers and screening methods within a unified interface.

Prediction learners

  • Machine learning: random forests, gradient boosting, XGBoost, support vector machines, BART, ranger
  • Penalized/high-dimensional: elastic net, ridge regression, CoxBoost
  • Tree-based: RPART
  • Classical/parametric: Cox proportional hazards, Weibull, exponential, log-logistic, log-normal, generic parametric models
  • Smoothing/splines: generalized additive models
  • Baseline models: Kaplan–Meier

Screening methods

  • keep all features
  • marginal Cox screening
  • variance-based screening
  • elastic-net-based screening
  • random forest variable hunting

The framework is extensible, and users can add custom learners and screeners. See the extensibility vignette for details.

📖 Documentation

The package website includes tutorials covering:

  • 0. Installation & Setup
  • 1. The SuperSurv Ensemble
  • 2. Model Performance
  • 3. Selection vs. Ensemble
  • 4. Screening Methods
  • 5. Hyperparameter Tuning
  • 6. Random Forests
  • 7. Parametric Models
  • 8. SHAP Interpretability
  • 9. Causal Inference (RMST)
  • 10. Parallel Processing
  • 11. Extending SuperSurv

📖 Citation

To cite the package, use:

citation("SuperSurv")

If you would also like to cite the accompanying preprint:

Lyu, Y., Lin, S. H., Huang, X., & Li, Z. (2026).
SuperSurv: A Unified Framework for Machine Learning Ensembles in Survival Analysis.
bioRxiv.
https://doi.org/10.64898/2026.03.11.711010

Related methodological work:

Westling, T., Luedtke, A., Gilbert, P. B., & Carone, M. (2024).
Inference for treatment-specific survival curves using machine learning.
Journal of the American Statistical Association.
https://doi.org/10.1080/01621459.2023.2205060