Development of an ensemble machine learning prognostic model for predicting 60-day risk of major adverse cardiac events in adults with chest pain

Authors: Chris J. Kennedy, Dustin G. Mark, Jie Huang, Mark J. van der Laan, Alan E. Hubbard, Mary E. Reed

Software requirements

This code is intended to run on R 4.0.4. Exact version numbers for all software packages are listed in the renv.lock file.

Installation guide

Open the repository as an RStudio project, which will activate renv. Then run:

renv::restore()

This will install all necessary packages with the correct versions. It will take about 30 minutes to run.

Reproducing manuscript results

The following data files (EHR exports) need to be placed in the data-raw directory:

data_grace3.sas7bdat - main data file
_outrace.sas7bdat - patient race
_outgfr.sas7bdat - eGFR
_outbmi.sas7bdat - BMI
_lab_vdw.sas7bdat - A1C, HDL, LDL, triglycerides
chestpaindata.sas7bdat - MACE+ supplemental outcome

Then knit the scripts in the following order:

import-data.Rmd
estimator-superlearner.Rmd - takes 3+ days to run, depending on CPU cores.
variable-importance.Rmd
interpretation.Rmd
decision-analysis.Rmd

Note: in import-data.Rmd the GLRM grid search is disabled by default due to its CPU-intensive nature. To re-enable it, go to the section glrm_grid_search and set eval = TRUE the RMarkdown header (current value is FALSE).

ck37r package

Examples and demonstration results for the accompanying ck37r package are provided in its github repository.

License