The dataset contains the Medicare claims from End-Stage Renal Disease (ESRD) patients, organized by the United States Renal Data System (USRDS). Our analysis only concerns patients that had their first ESRD service in the year of 2007 to 2011. According to the data use agreement, the data can not be made available to the public. Researchers who want to obtain access to the data should contact the USRDS for details on the procedure for requesting access to the data.
According to the data use agreement, the data cannot be made public. Those who wish to get access to the dataset could visit USRDS SAFs and submit a request. For reproducibility, we have created datasets that mimics the real USRDS data set used in the paper. The pseudo datasets are publicly available under data folder at the time of submission. The pseudo dataset is generated from the models in the paper using coefficients estimates from the original datast and simulated covariates and errors.
See usrds_desc.txt under data folder.
Researchers could use the R scripts and Shell scripts to reproduce the results in the paper, including Figures 1-2 and Tables 2. Since Figure 3-5 and Table 3-4 are generated using the original dataset, they are not reproducible. Readers can generate a pseudo version of Figure 3-5 and Table 3-4 using the pseudo datasets under data folder following the instructions provided below (Step 3 and 4). The pseudo Figure 3-5 are also provided under pseudo_figures folder for reference.
As requested by reviewers, we also provide 3D versions of the figures that appear in the paper, which can be found under 3d_figures folder. Note Figure 3-5 in this folder are generated from the original dataset, which should not be confused with the figures under pseudo_figures folder. All the 3D figures are generated using Plotly and provided in the form of HTML files. Since they cannot be viewed directly on GitHub, we suggest downloading and opening them in a browser.
A bandwidth of 1.12 is used for simulation, which is an average of the cross-validated bandwidths on 10 simulated datasets. For replication of the 10 bandwidths readers can run the following code in the command line
sh simu_semi_cv.sh
Rscript simu_semi_cv_merge.RRun simulation with 500 replications using the undersmoothed bandwidth
sh simu_semi_fit.shThis will generate intermediate results, stored under code folder. Once this is done, run
Rscript simu_semi_plot.Rto generate Figure 1 (both 2D and 3D versions) under code folder and display Table 2.
Bandwidths of 2.1 and 1.1 are used for the first and second stages of the simulation, each of which is an average of the cross-validated bandwidths on 10 simulated datasets. For replication of the 10 bandwidths in the first stage, readers can run the following code
sh simu_mixed_cv1.sh
Rscript simu_mixed_cv1_merge.RThen run first-stage simulation with 500 replications using the selected bandwidth
sh simu_mixed_fit12.shAfter that, we can obtain the second-stage bandwidths on 10 simulated datasets by running
sh simu_mixed_cv2.sh
Rscript simu_mixed_cv2_merge.RRun second-stage simulation with 500 replications using the undersmoothed bandwidth
sh simu_mixed_fit2.shThis will generate intermediate results, stored under code folder. Once this is done, run
Rscript simu_mixed_plot.Rto generate Figure 2 (2D and 3D) under the same folder.
Replicate cross-validation
sh real_semi_cv.sh
Rscript real_semi_cv_merge.RFit the model with the selected bandwidth
sh real_semi_fit.shThis will generate the intermediate result under code folder. Run
Rscript real_semi_plot.Rto generate Figure 3 (2D and 3D) and display Table 3 and 4.
Replicate first-stage cross-validation
sh real_mixed_cv1.sh
Rscript real_mixed_cv1_merge.RCreate the response for the second stage
sh real_mixed_fit12.shReplicate second-stage cross-validation
sh real_mixed_cv2.sh
Rscript real_mixed_cv2_merge.RFit the second-stage model with the selected bandwidth
sh real_mixed_fit2.shThis will generate the intermediate result under code folder. Run
Rscript real_mixed_plot.Rto generate Figure 4 and Figure 5 (both in 2D and 3D).