Code files corresponding to the manuscript Development and Validation of a Risk Score Predicting Death Without Transplant in Adult Heart Transplant Candidates by Zhang, Narang, et al, available at: https://jamanetwork.com/journals/jama/article-abstract/2814884.
All files are written and tested using R 4.3.1.
Kevin Lazenby (coauthor and collaborator) performed a reproducibility check on 10/14/2023.
Scientific Registry of Transplant Recipient (SRTR) data files are required to run these files, which are restricted under a data usage agreement. A premade dataset generated from SRTR data (status_changes_pubsaf2303.csv) is also used; details provided at https://github.com/kevinlazenby/heart_data_pipeline.
An app for calculating US-CRS scores is available at https://kevinz1194.shinyapps.io/US-CRS_Score. The corresponding code file is app.R.
Files should be run in this order:
Creates 14-day interval (blocked) dataset from SRTR data, including 6-week mortality outcome for patients not transplanted. Checks against original source dataset, as well as pre- and post- blocking, are performed.
Datasets required:
status_changes_pubsaf2303.csv (helper dataset)
cand_thor.sas7bdat (SRTR dataset)
JustFormHRStat1.sas7bdat (SRTR dataset)
JustFormHRStat2.sas7bdat (SRTR dataset)
JustFormHRStat3.sas7bdat (SRTR dataset)
JustFormHRStat4.sas7bdat (SRTR dataset)
Updated 7/9/2023: Added.
Updated 7/11/2023: Imputation procedure changed, code cleaned to fully account for edge cases.
Updated 8/6/2023: BNP type (NT-proBNP) accounted for.
Updated 8/14/2023: Added lower bound for sodium, and censoring date now uses the maximum of removal date, last active date, or last inactive date.
Updated 10/15/2023: Typo with calculation of short-term MCS is fixed, error with IV dosage changed, new loop for maximum time.
Updated 11/24/2023: Imputation of bilirubin and albumin added.
Updated 1/19/2024: One-line change to directory so that it is clear that it should be adjusted based on user's file location.
Updated 4/16/2024: Reworked definition of durable LVAD.
Sample iteration of elastic net models with parameter tuning on alpha and lambda. Requires large amount of computational resources. Optional. Estimated runtime of several hours on a high-memory node.
Updated 7/4/2023: Added.
Updated 8/11/2023: Models add NT-proBNP.
Updated 10/15/2023: Dataset version number corrected.
Updated 4/16/2024: Reworked definition of durable LVAD.
Sample iteration of gradient-boosted models with parameter tuning. Requires large amount of computational resources. Optional. Estimated runtime of several days on a high-memory node.
Updated 7/4/2023: Added.
Updated 8/11/2023: Models add NT-proBNP.
Updated 10/15/2023: Dataset version number corrected.
Updated 4/16/2024: Reworked definition of durable LVAD.
Runs logistic regression models, creates predicted results and the main figures for the blocked (14-day interval) dataset.
Updated 7/9/2023: Added.
Updated 7/10/2023: Figure colors changed.
Updated 7/11/2023: Main figure structure revised, now includes calibration and facet plots of score.
Updated 7/13/2023: Colors changed, confidence intervals added.
Updated 8/11/2023: Models add NT-proBNP, new French-CRS score added.
Updated 8/18/2023: Sensitivity analysis using MCS without balloon pump added. Model changed to drop hemodynamics.
Updated 8/22/2023: Make additional predictions on model with hemodynamics.
Updated 10/8/2023: AUC, in addition to AIC, is used for model selection. An additional set of models without BNP is added. Gender and race disparities are examined as a sensitivity analysis. Non-informative censoring is examined. An additional mixed-effects model is added. A sensitivity analysis with a different temporal split is added.
Updated 10/14/2023: Code for specific thresholds of sensitivity and specificity are added.
Updated 10/15/2023: Minor change to legend of AUC curves.
Updated 10/22/2023: Sensitivity and specificity code revised.
Updated 11/28/2023: Robust standard errors clustered by patient are now used. Time-dependent AUC curves added. Some calculations use only the first interval.
Updated 11/29/2023: Time-dependent AUC curves fixed, specific intervals added. Sensitivity analyses for 2-week is added.
Updated 1/19/2024: One-line change to directory so that it is clear that it should be adjusted based on user's file location.
Updated 4/16/2024: Reworked definition of durable LVAD.
Figures and discrimination, calibration from the full, continuous (no 14-day intervals) dataset.
Updated 7/9/2023: Added.
Updated 7/11/2023: Unblocking method changed to every justification form change.
Updated 7/13/2023: Colors changed.
Updated 8/11/2023: Accounts for NT-proBNP, new French-CRS score.
Updated 8/18/2023: Name changed, updated to work with new data. Cox PH model added.
Updated 8/22/2023: Added MCS components to Table 1, added c-index calculations for US-CRS model with hemodynamics.
Updated 11/25/2023: Section with Harrell's c removed. Integrated AUC added.
Updated 8/25/2023: IABP, durable and perc LVAD variables added.
Updated 10/14/2023: Updated notes on Harrell's c, also draws an AUC curve. Generalized c-index added.
Updated 10/27/2023: Name changed to results_continuous, bootstrapping and time-dependent AUC's added.
Updated 1/19/2024: One-line change to directory so that it is clear that it should be adjusted based on user's file location.
Updated 4/16/2024: Reworked definition of durable LVAD.