Using Machine Learning to Identify, Risk Stratify, and Guide Personalized Treatment of COVID-19 Patients
Abstract
Coronavirus disease 2019 (COVID-19) has spread rapidly in countries around the world, including the United States, where over 2 million cases have been reported to date. COVID-19 is highly contagious and deadly, with up to 11 percent of patients requiring intensive care unit (ICU) admission and a case-fatality rate between 1 percent and 3 percent according to a recent report by the Centers for Disease Control and Prevention. The rapid spread, high severity, limited testing availability, ICU resource limitations, and lack of evidence regarding how to best treat hospitalized patients has hampered progress on limiting the morbidity and mortality from this pandemic. The objective of this proposal is to develop a novel clinical decision support (CDS) tool to detect, risk stratify, and recommend personalized treatments of COVID-19 patients using machine learning and electronic health record (EHR) integrated clinical workflows. We will combine granular multicenter data with artificial intelligence methods and human factors approaches to address several important challenges. The first challenge is that the early identification of COVID-19 relies on clinician intuition, screening guidelines that vary across sites, and testing that can be delayed or falsely negative. This results in missed diagnoses and unnecessary resource use in COVID-19 negative cases. This is especially important considering that the pandemic will likely span many months before a vaccine becomes available, thereby making continual resource-intensive manual screening impractical. A second challenge is the lack of accurate bedside tools to risk-stratify patients with COVID-19. Earlier identification and treatment of high-risk patients, such as patients with sepsis and respiratory failure, is known to improve patient outcomes, and risk stratification tools may better allocate ICU resource needs.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2022
- Accession Number
- AD1169470
Entities
People
- Matthew M. Churpek
Organizations
- University of Wisconsin–Madison