Reproducibility in MIMIC-III

This repository is evaluates the reproducibility of mortality prediction studies in the MIMIC-III database. This study was presented at the Machine Learning in Healthcare (MLHC 2017) conference, and you can find the paper here (pdf).

Using this repository

Requirements

This repository assumes you have the MIMIC-III v1.4 database in a PostgreSQL database (v9.6+).

Running the code

First clone this repository recursively: git clone --recursive https://github.com/alistairewj/reproducibility-mimic.git.

This repository depends on two others:

1. The mimic-code repository
2. The mortality-prediction repository

Cloning recursively will download these repositories at the appropriate commit. These commits will be out of date. If you want to use some of the data this code generates (e.g. hourly values), you should go to the mimic-code repository instead, since everything is there and up to date.

Next, generate all the concepts in a PostgreSQL database by launching postgres in the queries subfolder and running: \i make-all.sql. You could equally run it from command line using psql -f make-all.sql. This query will take some time.

Once that is generated, you can run the notebooks present in the notebooks subfolder. These notebooks were built using Python 2.7, but should work in 3.5+ as well.

Acknowledgement

This repository can be cited using the following persistent Digital Object Identifier (DOI):

If you do find this repository useful in your work, we would be grateful if you would also cite our MLHC paper:

Johnson, A.E.W., Pollard, T.J. & Mark, R.G.. (2017). Reproducibility in critical care: a mortality prediction case study. Proceedings of the 2nd Machine Learning for Healthcare Conference, in PMLR 68:361-376

@InProceedings{johnson17reproducibility,
  title = 	 {Reproducibility in critical care: a mortality prediction case study},
  author = 	 {Alistair E. W. Johnson and Tom J. Pollard and Roger G. Mark},
  booktitle = 	 {Proceedings of the 2nd Machine Learning for Healthcare Conference},
  pages = 	 {361--376},
  year = 	 {2017},
  editor = 	 {Finale Doshi-Velez and Jim Fackler and David Kale and Rajesh Ranganath and Byron Wallace and Jenna Wiens},
  volume = 	 {68},
  series = 	 {Proceedings of Machine Learning Research},
  address = 	 {Boston, Massachusetts},
  month = 	 {18--19 Aug},
  publisher = 	 {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v68/johnson17a/johnson17a.pdf},
  url = 	 {http://proceedings.mlr.press/v68/johnson17a.html}
}