Discover Brigham
Poster Session

Wednesday, November 3rd, 2021 | 1pm - 3:45pm et

Virtual Event

Luwei Liu, MBI

Research Assistant
General Internal Medicine and Primary Care
DATA-DRIVEN COVID-19 RISK-ALERT Clinical Decision Support Platform

Principal Investigator: Patricia C. Dykes

Authors: Wenyu Song; Luwei Liu; Linying Zhang; Michael Sainlaire; Mehran Karvar; Min-Jeoung Kang; Avery Pullman; Anthony Massaro; Namrata Patil; Ravi Jasuja; Patricia C. Dykes
Lay Abstract

COVID-19 brings huge challenges to the current medical system. Among all patients, COVID-19 is more dangerous for the older age group. There is a higher rate of death when COVID-19 infects older patients. Therefore, it is important to identify older adults at risk of COVID-19 as early as possible.

We developed a tool for the early detection of COVID-19. We used the existing medical records in the MGB Health system. Machine Learning and statistical methods were used to develop the tool. Our results were validated and showed good performance. During the next step, we will further validate this tool in a long-term care (LTC) facility. New data collected from the LTC will improve this tool. The long-term goal is to expand and apply it in the hospital for all COVID patients.

Scientific Abstract

COVID-19 outbreak has caused unprecedented pressure on the health care system. With the disproportionately high morbidity and mortality among elderly patients, there is an urgent need to provide active monitoring and risk assessment for high-risk older adults.

During the first phase of our study, we developed a machine learning based predictive model to identify personalized risk profile for hospitalization. 1,495 COVID-19 test-positive elderly patients in the MGB Health system were included in the study cohort: case group (n=459) and control group (n=1036). 27 potential predictive variables from demographics, vital signs, disease diagnosis, and lab values were selected. Models of logistic regression, support vector machine (SVM), random forest, and neural network were trained with these variables as features and with whether patients are hospitalized during the defined window close to their COVID-19 test date as output. Our results identified albumin as a strong predictor of hospitalization risk and achieved the best predictive performance of AUC=0.81 with the random forest model.

For the next step, we are conducting a feasibility study in a long-term care facility. We will use the EHR data and sensor data from a device we developed of the residents in the facility to refine the prediction model.

Clinical Implications
This study will enhance COVID-19 clinical decision support rules by providing a prediction system for the elderly, and potentially all COVID patients.

If the PDF viewer does not load initially, please try refreshing the page.