Using AI To Smarter Triage During Viral Outbreaks

Scientists at Yale have developed a clever computer system that uses artificial intelligence to predict how sick a patient might get and how long they’ll need to stay in the hospital during a viral outbreak.

The system, which is described in a recent paper, consists of two things: a smart learning program and information about small molecules in cells. The goal is to help doctors care for patients better and decide where to use medical resources when there’s a big viral outbreak. This is important because such outbreaks can quickly overwhelm local hospitals. The small molecules they’re looking at are linked to how cells work.

Predicting outcomes

“Being able to predict which patients can be sent home and those possibly needing intensive care unit admission is critical for health officials seeking to optimize patient health outcomes and use hospital resources most efficiently during an outbreak,” the researchers explain.

They gathered information from 111 COVID-19 patients at Yale New Haven Hospital over two months in 2020. They also studied 342 healthy people (who were healthcare workers) for comparison. The sick patients were grouped based on how severe their illness was, from not needing much help with breathing to needing more serious treatments like using a machine to help them breathe.

The study found certain substances in the blood of COVID-19 patients that were linked to how serious their illness was. These substances included allantoin, 5-hydroxy tryptophan, and glucuronic acid.

Smarter diagnoses

One interesting discovery was that patients with higher levels of a certain type of white blood cell, called eosinophils, tended to have a more serious illness. This could be a new sign of how bad COVID-19 might get. Another surprise was that patients who needed help breathing using a machine had lower levels of a substance called serotonin in their blood. This is something researchers want to look into more.

The platform has three key components:

  1. Clinical Decision Tree: This tool uses important signs from the body to guess how sick a patient might get and how long they might need to stay in the hospital. It worked really well when they tested it.
  2. Hospitalization Estimation: The tool could guess pretty accurately how long a patient might stay in the hospital, usually within five days of the actual time. They found that a high breathing rate and certain chemicals in the blood were linked to longer hospital stays.
  3. Disease Severity Prediction: The tool was good at guessing how severe the disease was and whether a patient might need to go to the intensive care unit. This helps doctors know which patients could get very sick and start treatment sooner.

The researchers also made a simple computer program called “COVID Severity by Metabolomic and Clinical Study” (CSMC) software. This program combines machine learning with medical information to help doctors manage patients before they get to the hospital and understand how sick they are when they arrive at the emergency room.

“Our model platform provides a personalized approach for managing COVID-19 patients, but it also lays the groundwork for future viral outbreaks,” they explain. “As the world continues to grapple with COVID-19 and we remain vigilant against potential future outbreaks, our AI-powered platform represents a promising step towards a more effective and data-driven public health response.”

Facebooktwitterredditpinterestlinkedinmail