EHAPI

EHAPI: A standardized cohort definition for hospital-acquired pressure injury based on electronic health records.

We analyze the complexity of defining hospital acquired bedsores using diverse but inconsistent data sources, provide a definition that more closely resembles nursing guidelines, and showcase the higher accuracy of a hospital acquired bedsore prediction model based on our definition on a large dataset (MIMIC-III).

Table of Contents

1. Overview
2. Infrastructure
3. Installation and Setup
4. Code Structure
5. Contributors
6. References

Overview

This repository contains the code to extract and analyze the congruence among HAPI data in clinical notes, ICD9 diagnosis codes, and chart events from the Medical Information Mart for Intensive Care III (MIMIC-III) database. The script to define the four cohorts with different criteria for HAPI based on conflicts among data sources is included, in addition to the code to rest the performance of all cohorts for HAPI classification using tree-based and sequential neural network classifiers. For more technical information about each step, kindly look at the docstrings provided in the scripts.

Infrastructure

This code has been running and tested with the following specification:

• OS: Linux 4.4.0-1107-aws
• Python version: Python 3.6.5

Dependencies required are documented in the ''requirements.txt' file.

Installation and Setup

After cloning this repository onto your local machine, do the following:

1. Install Python 3.6.5 from the follwing link:

https://www.python.org/ftp/python/3.6.5/Python-3.6.5.tgz

2. Install virtualenv, create a virtual environment and activate it:

python3.6 -m pip install virtualenv
python3.6 -m virtualenv ehapi
source ehapi/bin/activate

3. Navigate to the repository folder

cd EHAPI

4. Install the package requirements found in 'requirements.txt':

pip install -r requirements.txt

Code Structure

We assume that the MIMIC-III CSV files have been downloaded from the following link

https://physionet.org/content/mimiciii/1.4/

analyze_consistency.py

Plot upset plots for the four cohorts and the PI sources in MIMIC III, and the Venn diagrams for the overlap of data sources.

define_cohorts.py

Creates the label for each cohort using MIMIC III CSV tables.

run_model_comparisons.py

Run Gradient boosting and sequential neural networks based on different training labels from each cohort.

Contributors

• Mani Sotoodeh
• Wenhui Zhang

References

We utilized the following repositories for this project.

https://github.com/MIT-LCP/mimic-code

https://github.com/emcramer/pressure_ulcer_prediction_mimic

https://github.com/MLforHealth/MIMIC_Extract