Photo above: A multidisciplinary team developed a proprietary predictive analytics tool to determine which COVID-19 patients could be safely cared for in UMass Memorial Health’s field hospital.
A cross-functional team developed a predictive analytics tool to help emergency room physicians predict whether COVID-19 patients would continue to have mild or moderate symptoms that could be safely addressed in UMass Memorial Health’s field hospital. After being successfully used in the field hospital environment, the same predictive analytics tool has been deployed to assess patient risk elsewhere in the hospital system.
With all the COVID-19 information amassed since the start of the pandemic, it can take a moment of reflection to remember just how little was known about the virus in early 2020. As people flooded hospitals, staff members scrambled to figure out ways to treat patients while avoiding getting sick themselves. There seemed no rhyme or reason as to who would overcome the illness and who would succumb.
When David McManus, MD, talks about those early days of the pandemic, he emphasizes the lack of knowledge and extreme conditions hospitals were confronting all day, every day.
“We were making decisions without data,” said McManus, Chair, Department of Medicine, UMass Memorial Medical Center, and Professor, UMass Chan Medical School.
That was an untenable idea for him, because he knew that at least some data could be extracted from the Medical Center’s electronic medical records (EMR) system. And he wanted to figure out how to tap into it to benefit patients.
To start, McManus and Apurv Soni, MD, PhD, Physician Champion for Digital Health at UMass Memorial Health, and Assistant Professor, UMass Chan Medical School, brought together a multidisciplinary group — which included physicians, hospital administrators, medical students, and experts in digital medicine and clinical informatics. The team abstracted the EMR information of 1,000 COVID-19 patients who had been seen at the Medical Center.
“Then we asked 50 of our UMass Chan Medical School students to validate data on each patient’s symptoms, history and outcomes,” he said. “The next step was teaching a computer to look at the charts, go back into the information that would be available to an emergency room doctor, and predict whether individual patients would be okay or whether they would decompensate.”
Predicting risk for COVID-19 patients
Co-developed by McManus and Soni, the resulting proprietary predictive analytics tool — Decompensation Electronic COVID Observational Monitoring Platform Triage (DECOMP-Triage) — calculated a risk score based on factors such as vital signs, lab results, age, medical history and diagnosis of type 2 diabetes.
The DECOMP-Triage tool identified patients who were the best match for UMass Memorial’s COVID-19 field hospital, which had capacity to care for patients with mild or moderate symptoms. The UMass Memorial DCU Center Field Hospital opened in a Worcester arena and convention center on April 9, 2020, when the state of Massachusetts was dealing with its initial surge of COVID-19 cases. It was the first field hospital in the state. A total of 161 patients had been treated at the hospital by the time it closed in May 2020, although the facility would need to be reopened later in the year.
Since the field hospital did not have an intensive care unit, it was especially important to make sure only the patients least likely to decompensate were admitted for care.
“We programmed the tool directly into our medical records system to make it available for all hospital teams to use,” McManus said. A patient’s risk score was instantly accessible to the entire care team, without adding extra steps to an already-busy workflow.
Using risk scores in the field hospital
To ensure accuracy and usefulness of risk scores, the predictive modeling tool calculated them every time a patient’s vital signs or lab data were entered in the EMR and refreshed them every 15 minutes. Categorizing the scores with a DECOMP-Alert of green, yellow or red made it easy for staff to use them. When UMass Memorial reopened the field hospital with capacity for 240 patients in December 2020, case managers had already created guides to instruct team members how to use risk scores as part of existing clinical workflows. (The field hospital stayed open until mid-March 2021.)
Recommendations were based on probability of respiratory failure requiring mechanical ventilation remaining below 5% throughout the hospitalization. A patient with a DECOMP-Triage score of:
- 0–3 could be considered for the field hospital on day 0
- 4–5 could be considered for the field hospital after 24 hours, if DECOMP-Alert was green or yellow
- 6–7 could be considered for the field hospital after 72 hours, if DECOMP-Alert was green or yellow
- 8 and higher were not considered for the field hospital because their probability of respiratory failure was greater than 50%
“A big reason why DECOMP scores were so helpful with the field hospital was because there was a common language that was associated with it,” said John Broach, MD, Director, Division of Emergency Medical Services and Disaster Management, UMass Memorial Medical Center, and Director, UMass Memorial DCU Center Field Hospital. Broach also serves as an Associate Professor at UMass Chan Medical School.
“Everyone started referring to them during our noon calls when we would discuss eligible patients: ER docs, nurses, the field hospital team and case managers. Even now, ER docs refer to DECOMP scores when making decisions.”
Building on DECOMP-Triage’s success
The team continues to explore new ways to learn from and apply the predictive analytics algorithm in other care environments in the hospital system.
“The success of using the DECOMP-Triage tool in our field hospital taught us a lot about how data could be a driver of innovation in our health care system,” McManus said. “We’re building on our predictive modeling algorithms.”
The UMass Memorial team has grown rapidly and expanded the algorithms developed for the field hospital and applied them in radiology, gastroenterology, home-based acute care and chronic disease management.
These algorithms will be used throughout UMass Memorial to predict outcomes in areas such as:
- Inpatient mortality
- COPD exacerbation that could lead to emergency department visits
- Prodrome of COVID-19, before the onset of specific signs and symptoms
In UMass Memorial Health’s Hospital at Home program, which provides hospital-level care to eligible patients in their own home environments, the predictive analytics tool has become especially important. It takes data from wearable devices, radiology and other imaging, and other EHR information to provide clinical decision support.
“Everything about the patient is uploaded to a portal. We’ve saved 1,000 patient days through Hospital at Home,” McManus said. “The field hospital broke the model of how we’ve been practicing medicine. It forced us to innovate — to build a hospital where there wasn’t one before and to use data and statistical algorithms in ways we hadn’t used them before. Now we’re keeping that innovation going.”
McManus founded two new entities to help facilitate and accelerate digital innovation. The Medical School’s Program in Digital Medicine, directed by Soni, focuses on research and innovation, while the Center for Digital Health Solutions at UMass Memorial Health focuses on implementing and following through on the innovation.
“The partnership between these entities is critical,” said Soni, who serves on the leadership board of the Center for Digital Health Solutions. “By leveraging the Program in Digital Medicine’s ability to develop innovative tools like DECOMP-Triage and the Center for Digital Health System’s ability to systematically implement them, we can be at the forefront of medicine.”
View more stories of healing, advancing medicine and innovating from UMass Memorial Health.