Prediction of Clostridium Difficile Infection In ICU Patients Utilizing MIMIC-III Database

This project is my capstone project completed as part of the data science immersive program with General Assembly. For this project, I utilized the MIMIC-III ICU database (part of the physionet collaboration project between MIT and Beth Israel Medical Center). Project notebooks, presentations and a summary of findings/results below are in the process of being uploaded - check back soon for more content!

MIMIC-III data background

Dictionary and additional info can be found here: https://mimic.physionet.org/about/mimic/

Schema: https://mit-lcp.github.io/mimic-schema-spy/

GitHub Code Respository: https://github.com/MIT-LCP/mimic-code

Project Notebooks

The following 3 notebooks have been uploaded so far:

1 - Create Patient Admits Dataframe

Creates an initial dataframe of unique patient admissions and ICU stays alongside key demographics.
Makes use of 3 MIMIC-III tables: Patients, Admissions, and ICUStays.

2 - C. Diff Infection Labeling and Inclusion Criteria Finalization

C. Diff Case Definition:
  Timestamped lab diagnosis of CDI prior to cutoff of 2 days post-ICU intime

  Utilizing a standard cutoff of 2 days post-ICU admission aligns with timing of other risk assessments typically made such as
  mortality/morbidity predictions with scales like APACHE (made around 1 days after ICU admit). As such, a tool based on this model
  could be implemented into existing clinical workflows.

Resulting Patient Population:
  
  Overall CDI rate of ~1.42%
  Total 670 cases / 47,000 ICU Stays

3 - Basic Demographics

Demographic features explored:
    age
    gender
    insurance
    language
    ethnicity
    marital status

Note that underlying data files are not and will not be included given data use agreement signed to gain access to the data