Representation Learning for Integrating Multi-domain Outcomes to Optimize Individualized Treatment
Authors: Yuan Chen, Donglin Zeng, Tianchen Xu, Yuanjia Wang
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on simulated data and a real-world clinical trial data show that the learned treatment polices compare favorably to alternative methods on heterogeneous treatment effects, and have broad utilities which lead to better patient outcomes on multiple domains. |
| Researcher Affiliation | Academia | Yuan Chen Department of Biostatistics Columbia University New York, NY 10032 yc3281@cumc.columbia.edu Donglin Zeng Department of Biostatistics University of North Carolina at Chapel Hill Chapel Hill, NC 27516 dzeng@email.unc.edu Tianchen Xu Department of Biostatistics Columbia University New York, NY 10032 tx2155@cumc.columbia.edu Yuanjia Wang Department of Biostatistics Columbia University New York, NY 10032 yw2016@cumc.columbia.edu |
| Pseudocode | No | The paper describes its computational algorithm in paragraph text under 'Computational algorithms' in Section 3.2 but does not provide a formal pseudocode block or algorithm figure. |
| Open Source Code | No | The paper states, 'We built the proposed model using Pytorch' in Section 4.1, but it does not provide an explicit statement about releasing the source code or a link to a code repository for their implementation. |
| Open Datasets | Yes | We assumed there were three latent domains, and the observed outcomes consisted of nine discrete items and five continuous items. We simulated data where the true optimal treatments are known. The detailed data generating mechanism are included in the supplementary material. (...) STAR*D [22] is a multi-site, multi-level randomized clinical trial designed to compare various treatment regimes for patients with major depressive disorder (MDD). The citation [22] refers to 'A. J. Rush, M. Fava, S. R. Wisniewski, P. W. Lavori, M. H. Trivedi, H. A. Sackeim, M. E. Thase, A. A. Nierenberg, F. M. Quitkin, T. M. Kashner, et al. Sequenced treatment alternatives to relieve depression (STAR*D): rationale and design. Controlled Clinical Trials, 25(1):119 142, 2004.' |
| Dataset Splits | Yes | For the proposed method, we tuned over the hyper-parameters including the number of hidden layers (ranges from 1 to 3) and hidden units (10 to 30), and number of iterations (1 to 10) through cross-validations on the training data. (...) We conducted 4-fold cross validations to evaluate the estimated ITRs (i.e., fit ITRs on 3 folds of data and compute the empirical value function from the remaining fold). |
| Hardware Specification | Yes | For model fitting, the computing time is 5s, 10s, 35s, and 72s under sample size of 200, 500, 1000, and 2000 with a batch size of 100, 250, 250, and 500 on a 2.7 GHz Intel Core i5 processor under the specified training procedure. |
| Software Dependencies | No | The paper mentions 'Pytorch' and 'Adam' as software used ('We built the proposed model using Pytorch'; '6 epochs of SGD was implemented with Adam'), but it does not specify any version numbers for these software dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | For the proposed method, we tuned over the hyper-parameters including the number of hidden layers (ranges from 1 to 3) and hidden units (10 to 30), and number of iterations (1 to 10) through cross-validations on the training data. Model with 2 hidden layers of 20 and 10 units was selected. 6 epochs of SGD was implemented with Adam [9] under a learning rate of 0.1 before 1 step of exact search for Zi0, and we ran in total 6 iterations of the update process. (...) with a batch size of 100, 250, 250, and 500... |