Skip to main content

We’ve updated our Terms & Conditions and Privacy Policy. By using this site, you agree to these terms.

All Geisinger locations are open and providing patient care. Please arrive as scheduled for your appointment unless you hear from a member of your care team. We appreciate your patience if you experience any delays during your visit.

Geisinger AI team’s entry accurately predicted long COVID risk among patients diagnosed with COVID-19

DANVILLE, Pa. – Geisinger has placed second in the Long COVID Computational Challenge (L3C), a project developed by the National Institutes of Health to use data-driven solutions to improve understanding of the risks of developing post-acute effects of COVID-19, also known as long COVID.

The nationwide challenge included submissions from universities, medical centers and public-private partnerships. Entrants developed artificial intelligence/machine learning (AI/ML) models and algorithms to identify which patients infected with COVID-19 had a higher likelihood of developing long COVID. Entries used de-identified electronic health records (EHR) data available through the National COVID Cohort Collaborative Data Enclave, a data repository that represents EHR data from more than 70 health centers across the United States.

People with long COVID experience a variety of symptoms that last weeks or months after a COVID-19 diagnosis. These conditions can include fatigue, respiratory issues, headaches, trouble sleeping, depression or anxiety, among many others. Predicting which patients are most likely to develop long COVID allows for earlier intervention and symptom management, leading to better overall health outcomes. 

Geisinger’s AI team built a portable, efficient and accurate model using the most commonly available patient information, resulting in a long COVID prediction tool that could be easily and widely integrated into the EHR and used for population-level risk stratification. The team placed second overall in the challenge, ahead of University of California-Berkeley (3rd place), University of Wisconsin-Madison (honorable mention), and the University of Pennsylvania (honorable mention). The winning team (Convalesco) is a collaboration between the University of Chicago and Argonne National Laboratory.

“Accuracy and interpretability are the twin challenges of artificial intelligence in healthcare, and our solution was based on extensive experience deploying clinical prediction models for population health at Geisinger,” said Abdul Tariq, Ph.D., director of artificial intelligence at Geisinger.

"We're thrilled to have been a part of this national, open science project, and hope that our contribution can improve clinical identification, characterization, and quality of life for those living with long COVID," said Elliot Mitchell, Ph.D., senior data scientist and a member of Geisinger’s AI team. 
 

About Geisinger
Geisinger is among the nation’s leading providers of value-based care, serving 1.2 million people in urban and rural communities across Pennsylvania. Founded in 1915 by philanthropist Abigail Geisinger, the nonprofit system generates $10 billion in annual revenues across 126 care sites — including 10 hospital campuses — and Geisinger Health Plan, with more than half a million members in commercial and government plans. Geisinger College of Health Sciences educates more than 5,000 medical professionals annually and conducts more than 1,400 clinical research studies. With 26,000 employees, including 1,700 employed physicians, Geisinger is among Pennsylvania’s largest employers with an estimated economic impact of $15 billion to the state’s economy. On March 31, 2024, Geisinger became the first member of Risant Health, a new nonprofit charitable organization created to expand and accelerate value-based care across the country. Learn more at geisinger.org or follow on Facebook, Instagram, LinkedIn and X.

Geisinger Logo

For media inquiries:

Ashley Andyshak Hayes
Marketing Strategist
Marketing & Communications

570-271-8081
arandyshakhayes@geisinger.edu

Content from General Links with modal content