Using Machine Learning to Identify, Risk Stratify, and Guide Personalized Treatment of COVID-19 Patients
Abstract
There are over two million confirmed coronavirus disease 2019 (COVID-19) cases in the United States, and more than 100,000 of those infected have died. While some people do not have symptoms, others develop severe disease and need hospitalization, intensive care, and life support, which can lead to death. There are many challenges related to caring for patients with severe disease requiring hospitalization. First, test results are often delayed and the results are not always correct. This causes unneeded use of personal protective masks, gowns, and other equipment, the inability of loved ones to support and advocate for the sick patient at their bedside, as well as increased risk of healthcare worker exposures to patients who test negative but actually have COVID-19. Second, some hospitalized patients get better and are able to go home after only a few days, while others get sicker and need intensive care and life support. Current tools to identify high-risk patients are limited, and more accurate tools are needed so that clinicians can better identify these patients quickly, as delays and errors in triage can be deadly. Finally, there are few high-quality research studies to guide treatments for patients with COVID-19, and those that are published only display results for the typical patient in the study. It is likely that each patient responds differently to a treatment, and therefore providing doctors with new tools that recommend personalized treatments for specific patients could be lifesaving. This would enhance their ability to provide specific therapies to patients likely to benefit while avoiding therapies that are likely to do harm. This proposal will work to address these challenges by developing statistical models to identify, risk stratify, and make personalized treatment recommendations for hospitalized patients with COVID-19. These models will then be used in a clinical decision support tool that can work in the electronic health record to improve patient care. This work addresses several Focus Areas for the FY20 PRMRP Topic Areas of Emerging Viral Diseases and/or Respiratory Health, including: 1. Novel and/or innovative detection technologies or therapeutics to reduce the incidence and/or severity of acute respiratory distress syndrome (ARDS) secondary to coronaviruses, particularly COVID-19. 2. Triage of care for COVID-19 patients requiring access to resource-intensive interventions. 3. Development of improved methods for assessing and treating lung injury due to coronaviruses, particularly COVID-19. 4. Research to understand physiological biomarkers for evaluating short- and long-term health impacts from COVID-19. 5. Research on the etiology and prevention of ARDS caused by COVID-19. 6. Pharmacological and biologic interventions for COVID-19 induced complications, including ARDS and related sequelae. All Americans are at risk for infection with COVID-19 and its deadly consequences. Therefore, the proposed research would benefit active duty Service members, Veterans, military beneficiaries, and the American public by creating a novel clinical decision support tool that can work with electronic health records across the country to improve the identification, triage, and treatment of patients with COVID-19. This would lead to earlier, personalized treatment that would improve outcomes and preventable morbidity and death from this disease.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 05, 2021
- Source ID
- W81XWH2110009
Entities
People
- Matthew Churpek
Organizations
- United States Army
- University of Wisconsin–Madison