Premise

In the United States, a common mode of death in pediatric ICUs is withdrawal of life support for patients who are terminally ill with no possible recourse. This removal of life support is frequently the removal of ventilator support, also known as compassionate or terminal extubation.

One uncertainty that families face regarding the end-of-life process is the question of how long it will take for the patient to pass away after terminal extubation. Knowing this information is valuable for family care and also, if the family wishes, to identify potential candidates for organ donation. We partnered with palliative care clinicians to train machine learning models to predict time to death after terminal extubation. This multi-institutional project resulted in a web application that used the created model.

DONATE application
DONATE application scrolled down

DONATE Web Application - Made with Flask (HTML, CSS, Python)

I was the primary data scientist for developing the DONATE model and helped create the DONATE web application. This tool predicts the probability of death within 1 hour after terminal extubation. It allows a clinician to enter information about the patient and highlights high-impact variables that increase or decrease this probability.

One hour was chosen because:
  1. It best facilitates family planning. More deaths happen within 1 hour than not, and the tool allows family to know if they have time to leave the bedside.
  2. Patients who die within one hour are eligible for organ donation, wheras those who die later are typically not. One hour is the ischemic time recommended by United Network for Organ Sharing (UNOS). It is also the most common time period institutions use if they practice organ donation after circulatory death.