Optimizing Sepsis Time Zero Definitions Based on Associations Between Time-to-Antibiotics and Mortality

This repository contains R code for the inverse probability weighting (IPW) analysis within the above manuscript.

Prerequisites

R version ≥4.0.2 and the following packages: data.table, progressr, and tidyverse.

File descriptions

Scripts

1. ipw_run.R

   Description: The primary script that runs the IPW analysis, including loading of the input person-time dataset, data cleanup, generation of point estimates, bootstrapping, generating summary tables, and saving output files. This script can be invoked directly and expects 2 to 4 command line arguments:

   $ ./ipw_run.R MODEL_NAME INPUT_FILE [NUM_WORKERS] [NUM_BOOTSTRAPS]
   

   MODEL_NAME specifies the variables included in the pooled over time logistic regression model. Possible values include none, baseline_HCA, and max_HCA (the fully adjusted model including time-varying covariates).

   Input files:

   • $INPUT_FILE <- the input person-time dataset, in either CSV or RDS format

   Output files:

   • $INPUT_FILE.after_setup.rds <- a cache of the dataset after cleanup/setup procedures
   • $INPUT_FILE.adj-$MODEL_NAME.est_debug.txt <- contains debug info for the point estimates
   • $INPUT_FILE.adj-$MODEL_NAME.est_summary.csv <- a summary of point estimates
   • $INPUT_FILE.adj-$MODEL_NAME.boot-$NUM_BOOTSTRAPS.rds <- estimates for the bootstrap replicates
   • $INPUT_FILE.adj-$MODEL_NAME.boot_summary-$NUM_BOOTSTRAPS.csv <- a summary of the bootstraps

Library files

1. ipw_setup.R

   Description: Setup and cleanup functions for the input person-time datasets for ipw_run.R. The most important exported function is ipw_setup(), which runs the entire setup process.

2. ipw_boot.R

   Description: Contains the primary steps of the IPW analysis that are called by ipw_run.R.

   The most important exported functions are:

   • ipw_est() <- creates point estimates
   • ipw_boot() <- creates bootstrapped estimates that can be used to construct a 95% CI
   • summarize_ipw_est() <- creates summary statistics for the point estimates
   • summarize_ipw_boot() <- creates summary statistics for the bootrapped estimates

License

MIT. See LICENSE.txt

Contact

For questions, clarifications, or further support, please contact:

Theodore Pak, MD, PhD — theodore (dot) pak (at) uci (dot) edu