Targeted Learning and Variable Importance with Optimal Individualized Categorical Treatment
Authors: Ivana Malenica, Jeremy R. Coyle and Mark J. van der Laan
Suppose one wishes to maximize (or minimize) the population mean of an outcome where for each individual one has access to measured baseline covariates. We consider estimation of the mean outcome under the optimal rule, where the candidate rules are restricted to depend only on user-supplied subset of the baseline covariates (or no covariates specified). The estimation problem is addressed in a statistical model for the data distribution that is nonparametric.
The tmle3mopttx is an adapter/extension R package in the tlverse
ecosystem, that estimates the mean outcome under the following regimes:
- Optimal Individualized Treatment for categorical treatment,
- Optimal Individualized Treatment based on possibly sub-optimal rules (including static rules),
- Optimal Individualized Treatment based on realistic rules.
- Optimal Individualized Treatment with resource constraints.
The tmle3mopttx also provides variable importance analysis in
terms of optimizing the population mean of an outcome for settings
(1)-(3). It also can return an interpretable rule based on a Highly
Adaptive Lasso fit.
In addition, tmle3mopttx supports estimating the mean outcome under
regimes under the following missigness process:
- Missing outcome.
In order to avoid nested cross-validation, tmle3mopttx relies on
split-specific estimates of the conditional expectation of the outcome
()
and conditional probability of treatment
(
)
in order to estimate the rule. The targeted maximum likelihood estimates
of the mean performance under the estimated rule are obtained using
CV-TMLE.
The description of the method implemented, with simulations and data
examples using tmle3mopttx can be found in Malenica (2021). For
additional background on Targeted Learning and previous work on optimal
individualized treatment regimes, please consider consulting van der
Laan, Polley, and Hubbard (2007), Zheng and van der Laan (2010), van der
Laan and Rose (2011), van der Laan and Luedtke (2015), Luedtke and van
der Laan (2016), Coyle (2017) and van der Laan and Rose (2018).
You can install the most recent stable release from GitHub via
devtools with:
devtools::install_github("tlverse/tmle3mopttx")After using the tmle3mopttx R package, please cite the following:
@software{malenica2022tmle3mopttx,
author = {Malenica, Ivana and Coyle, Jeremy and {van der Laan}, Mark J},
title = {{tmle3mopttx}: Targeted Learning and Variable Importance with Optimal Individualized Categorical Treatment},
year = {2022},
doi = {},
url = {https://github.com/tlverse/tmle3mopttx},
note = {R package version 1.0.0}
}If you encounter any bugs or have any specific feature requests, please file an issue.
The contents of this repository are distributed under the GPL-3 license.
See file LICENSE for details.
Coyle, J R. 2017. “Computational Considerations for Targeted Learning.” PhD thesis, U.C. Berkeley.
Luedtke, A., and M. J van der Laan. 2016. “Super-Learning of an Optimal Dynamic Treatment Rule.” International Journal of Biostatistics 12 (1): 305–32.
Malenica, Ivana. 2021. “Optimal Individualized Treatment Regimes.” In Targeted Learning in r: Causal Data Science with the Tlverse Software Ecosystem.
van der Laan, M. J, and A. Luedtke. 2015. “Targeted Learning of the Mean Outcome Under an Optimal Dynamic Treatment Rule.” Journal of Causal Inference 3 (1): 61–95.
van der Laan, M. J, E C. Polley, and A E. Hubbard. 2007. “Super Learner.” Statistical Applications in Genetics and Molecular Biology 6 (1).
van der Laan, M. J, and S. Rose. 2011. Targeted Learning: Causal Inference for Observational and Experimental Data. Springer Science & Business Media.
———. 2018. Targeted Learning in Data Science: Causal Inference for Complex Longitudinal Studies. Springer Science & Business Media.
Zheng, W., and M. J van der Laan. 2010. “Asymptotic Theory for Cross-validated Targeted Maximum Likelihood Estimation.” U.C. Berkeley Division of Biostatistics Working Paper Series.