DICErClust: Deep Significance Clustering for Clinical Risk Stratification

We provide an R implementation of Deep Significance Clustering (DICE), a self-supervised learning framework designed to identify clinically meaningful and risk-stratified patient subgroups from electronic health record (EHR) data. DICE jointly optimizes deep representation learning, clustering, and outcome prediction while enforcing statistical significance between predicted outcomes and cluster membership. This integrated optimization produces subgroups that are both clinically coherent and predictive, addressing a gap where traditional unsupervised clustering methods and supervised risk prediction models alone may fail to generate actionable clinical groupings. See Huang et al. (2021) <doi:10.1093/jamia/ocab203>.

Version:	0.1.2
Imports:	argparser, ggplot2, torch, pROC
Suggests:	knitr, rmarkdown, scales
Published:	2026-05-28
DOI:	10.32614/CRAN.package.DICErClust
Author:	Sarah Ayton [aut, cre], Yiye Zhang [aut]
Maintainer:	Sarah Ayton <saa7050 at med.cornell.edu>
License:	MIT + file LICENSE
NeedsCompilation:	no
Materials:	README
CRAN checks:	DICErClust results

Documentation:

Reference manual:	DICErClust.html , DICErClust.pdf
Vignettes:	Introduction to DICErClust (source, R code) DICErClust Illustrated Example: Heart Failure Risk Stratification (source, R code)

Downloads:

Package source:	DICErClust_0.1.2.tar.gz
Windows binaries:	r-devel: DICErClust_0.1.2.zip, r-release: DICErClust_0.1.2.zip, r-oldrel: DICErClust_0.1.2.zip
macOS binaries:	r-release (arm64): DICErClust_0.1.2.tgz, r-oldrel (arm64): DICErClust_0.1.2.tgz, r-release (x86_64): DICErClust_0.1.2.tgz, r-oldrel (x86_64): DICErClust_0.1.2.tgz

Linking:

Please use the canonical form https://CRAN.R-project.org/package=DICErClust to link to this page.