ICU_Survival_Analysis_24hrs

Business Use Case

• In clinical practice, estimates of mortality risk can be useful in triage and resource allocation
• It helps hospital to:
  • determine appropriate levels of care
  • prepare discussions with patients and their families around expected outcomes
• Also helps payers to know how the health outcomes of their policyholders will be affected, so that payers can identify useful policies

Problem Statements

MIT's GOSSIS community initiative is seeking an efficient way to address the problems with existing severity of illness systems: 

• They often lack generalizability beyond the patients on whom the models were developed, and
• The models are often proprietary, costly to use (APACHE scoring system…), and suffer from opaque algorithms

Objectives

Create a model that uses data from the first 24 hours of intensive care to predict patient survival with:

• Better prediction probability of death (as compared to apache_4a_icu_prob, apache_4a_hospital_prob)
• Minimize apache features
• Transparent (easy to explain)
• Generalizability
• Less complexity

Data

From MIT's GOSSIS community initiative

Dataset of more than 90,000 hospital Intensive Care Unit (ICU) visits from patients, spanning a one-year timeframe. This data is part of a growing global effort and consortium spanning Argentina, Australia, New Zealand, Sri Lanka, Brazil, and more than 200 hospitals in the United States.

https://gossis.mit.edu/about/

WiDS Datathon 2020

https://www.kaggle.com/c/widsdatathon2020

Model

• Logistic Regression
• Random Forest
• Light Gradient Boosting
• CatBoost
• Neural Network with PCA