Category: Subgroup analysis.
Instructions
Google Chrome is required to download the ZIP files posted on this page. Copy and paste the page address in a Google Chrome window, click on the link for the ZIP file, select File in the horizontal menu and choose Download (Ctrl+S).
Software for subgroup identification
SIDES method
R package RSIDES implementing the regular SIDES method (Subgroup Identification Based on Differential Effect Search) based on Lipkovich et al. (2011) and SIDEScreen methods [last update: June 10, 2024]. The package is maintained by Alex Dmitrienko (alex.dmitrienko@gmail.com).
R package SIDES implementing the regular SIDES method [last update: October 04, 2016]. The package is maintained by Marie-Karelle Riviere (eldamjh@gmail.com).
Download the SIDESxl package (an Excel add-in) which implements the regular SIDES and SIDEScreen methods [last update: March 25, 2016]. The package is maintained by Ilya Lipkovich (ilya.lipkovich@gmail.com).
Interaction Trees method
Download the R functions and examples for the Interaction Trees method. The functions and examples are provided by Xiaogang Su (Xiaogang Su’s site). Download the R code for the Interaction Trees method [last update: Nov 15, 2024].
Virtual Twins method
Download the R code for the Virtual Twins method [last update: Dec 30, 2014]. The code is provided by Jared Foster (jaredcf@umich.edu).
R package aVirtualTwins that implements an adaptation of the Virtual Twins method by Foster et al. (2011)
GUIDE method
GUIDE package for classification and regression trees now includes methods for subgroup identification. The GUIDE package is maintained by Wei-Yin Loh (Wei-Yin Loh’s site). For more information on the subgroup identification features, see Section 5.10 of the GUIDE User Manual [last update: September 25, 2018] and paper by Wei-Yin Loh, Xu He and Michael Man.
In addition, MrSGUIDE package implements the GUIDE method for randomized trials and observational studies.
QUINT method
Quint package for QUalitative INteraction Trees. The package is maintained by Elise Dusseldorp (Elise Dusseldorp’s site) and colleagues. Reference: Dusseldorp and Mechelen (2014).
FindIt method
FindIt package for finding heterogeneous treatment effects [last update: Nov 11, 2019]. Reference: Imai and Ratkovic (2013).
Blasso method
Download the R functions for the Bayesian two-stage Lasso strategy for biomarker selection for time-to-event endpoints [last update: December 16, 2014]. The code is provided by Xuemin Gu (xuemin.gu@bms.com). Reference: Gu, Yin and Lee (2013).
ROWSi method
Download the R code for the ROWSi method (Regularized Outcome Weighted Subgroup identification). Reference: Yu et al. (2015).
Model-based Recursive Partitioning
R package partykit (A Toolkit for Recursive Partytioning) performing subgroup analyses using the functions lmtree(), glmtree() (or more generally, mob()) and ctree()).
R package model4you supporting stratified and personalized treatment effect estimation. The package is maintained by Heidi Seibold (heidi@seibold.co). See examples of subgroup analysis in Seibold et al. (2015) and Seibold et al. (2016)
Other packages
R package personalized (maintained by Jared Huling) for subgroup identification and estimation of heterogeneous treatment effects. It is a general framework that encompasses a wide range of methods including ROWSi, outcome weighted learning, and many others. See documentation and article explaining the underlying methodology.
R package SubgrID implements several algorithms for developing threshold-based multivariate (prognostic/predictive) biomarker signatures via bootstrapping and aggregating of thresholds from trees (BATTing), Monte-Carlo variations of the Adaptive Indexing Method (AIM) by Huang X. et al. (2017) and and adaptation of Patient Rule Induction Method (PRIM) for subgroup identification by Chen G. et al. (2015).
R package evalITR (Evaluating Individualized Treatment Rules). Provides various statistical methods for evaluating Individualized Treatment Rules under randomized data based on paper “Experimental Evaluation of Individualized Treatment Rules” by Imai Kosuke and Michael Li. It also provides procedures for testing hypothesis about the presence of heterogeneous treatment effect in the data generate by randomized experiments based on paper “Statistical Inference for Heterogeneous Treatment Effects Discovered by Generic Machine Learning in Randomized Experiments” by Imai Kosuke and Michael Li.
R package (grf) implements forest-based statistical estimation and inference (Generalized Random Forests by Athey, Tibshirani and Wager, 2019), including causal forests for heterogeneous treatment effects (see paper by Stefan Wager and Susan Athey).
Fu, Zhou and Faries (2016) developed a search approach that provides simple and interpretable rules defining subgroup of patients with maximizes average patients’ benefit for different treatments within a general framework of outcome weighted learning (OWL). Here you can find the C++ implementation.
R package DynTxRegime implements methods to estimate dynamic treatment regimes using Interactive Q-Learning, Q- Learning, weighted learning, and value-search methods based on Augmented Inverse Probability Weighted Estimators and Inverse Probability Weighted Estimators.
R package listdtr constructs list-based rules (lists of if-then clauses) to estimate the optimal dynamic treatment regime based on the approach by Zhang et al. (2016).
The subtee R package implements method for bootstrap-corrected estimation after subgroup selection described in Rosenkranz (2016) and a model averaging approach from Bornkamp et al. (2016).
TSDT: Treatment-Specific Subgroup Detection Tool by Chakib Battioui, Brian Denton and Lei Shen (2018).
StratifiedMedicine by Thomas Jemielita is a broad toolkit for subgroup identification and stratified/precision medicine. The package also includes a novel algorithm PRISM (Patient Response Identifiers for Stratified Medicine) by Jemielita and Mehrotra (to appear).
Policy learning via doubly robust empirical welfare maximization over trees (policytree) supports optimal policies via doubly robust empirical welfare maximization over trees. This package implements the multi-action doubly robust approach of Zhou, Athey and Wager (2018).
R package (debiased.subgroup) implements bootstrap-assisted desparsified Lasso and bootstrap-assisted R-split estimators on selected subgroup’s treatment effect estimation. The implemented estimators remove the subgroup selection bias and the regularization bias induced by high-dimensional covariates. For more information, see Guo, Wei, Wu and Wang (2021).
R package (rlearner) supports quasi-oracle estimation of heterogeneous treatment effects based on Nie and Wager (2021).
R package (causalToolBox) is available to enable metalearners for estimating heterogeneous treatment effects using machine learning based on Künzel, Sekhona, Bickel and Yu (2019).
R code (CAPITAL) for the implementation of optimal subgroup identification via constrained policy tree search based on Cai, Lu, West, Mehrotra and Huang (2021).
R package (bcf) supports causal inference for a binary treatment and continuous outcome using Bayesian causal forests based on Hahn, Murray and Carvalho (2019).
R package credsubs: Credible Subsets Functions for constructing simultaneous credible bands and identifying subsets via the “credible subsets” method based on paper by Schnell, Fiecas, and Carlin (2020).
R package knockofftools containing a suite of knockoffs functions and methods from existing R-packages and the knockoffs literature including evaluating predictive biomarkers based on paper by Sechidis, Kormaksson, and Ohlssen (2021).