Hospitals are looking to new approaches that can help them manage COVID-19-related capacity challenges without adversely affecting patient care. Researchers developed an automated text messaging approach that can monitor patients who have been discharged from the ED. Other investigators have leveraged artificial intelligence to train an algorithm to help emergency clinicians better predict outcomes and manage resources.

  • COVID Watch is a text messaging tool that checks on patients twice a day, only escalating patients to a live clinician when their conditions warrant.
  • Emergency clinicians report it is easy to enroll patients in COVID Watch during discharge, a process that won broad approval from emergency staff.
  • Patients deemed at higher risk also may receive pulse oximeters upon discharge so their blood oxygen readings can be monitored remotely.
  • At Mount Sinai, investigators have trained an algorithm to enable frontline clinicians to better predict which COVID-19 patients require more aggressive monitoring or treatment, using results from routine bloodwork, blood pressure readings, and a chest X-ray to produce a severity score.

In the early days of the pandemic, some patients with COVID-19 were deteriorating a few days after their initial contact with a physician in the ED.

“For a lot of respiratory illnesses like the flu or colds, we generally expect those illnesses to get better when patients leave the ED. Because this particular viral illness was defying our typical expectations for how viral respiratory illnesses behave, we were trying to figure out if there were any risk factors [to explain] why patients might return,” explains Austin Kilaru, MD, MSHP, an emergency physician and researcher at the Perelman School of Medicine at the University of Pennsylvania.

Consequently, Kilaru and colleagues examined the outcomes of about 1,400 patients with COVID-19 who were treated and discharged from the ED between March and May 2020. Investigators found 5% returned to the ED within 72 hours of their initial visit and required admission. Another 3.5% required hospital admission within a week of their initial ED visit.

Additionally, the researchers found (perhaps not surprisingly) age was a significant risk factor. “Patients who were older than 60 compared to younger patients between the ages of 18 and 39, had three times the probability of coming to the hospital within 72 hours vs. 2.5% [in the younger patients],” Kilaru reports.

Other risk factors associated with a return visit to the ED included low blood oxygen levels (a pulse oximetry reading less than 95%), an abnormal X-ray, or a fever upon presentation to the ED. Each factor was associated with double the probability of a return visit to the ED within 72 hours.

“We also looked at patients who came back [to the ED] within seven days. For that group, we found additional risk factors [for a return visit to the ED] were obesity as well as having hypertension as a comorbid illness,” Kilaru says.

Monitor Patients

Knowing which COVID-19 patients may need closer monitoring is helpful, but carrying out such a task when staff resources are strained is tough. To address this issue, the University of Pennsylvania Center for Health Care Innovation developed COVID Watch, an automated text messaging approach that checks in with patients twice a day. The tool escalates concerning cases to a team of telemedicine clinicians who are available to respond 24/7. If a patient reports any deterioration in his or her ability to breathe or other worsening symptoms, that case will go to a clinician who will follow up with the patient quickly. The clinician can refer the patient to the ED, if necessary, or arrange for further assessments or care, as needed.

Clinicians in the ED greatly appreciate COVID Watch because they know their patients will be followed once they leave the ED. Patients seem to appreciate the monitoring, too. “The goal of the program is to make sure patients are feeling better, but also to make sure that our outpatient colleagues aren’t overwhelmed with calls or concerns so we can all be more efficient,” Kilaru says.

Further, there now is an additional program for COVID-19 patients who meet higher-risk criteria. These patients will receive both the automated text messages through COVID Watch and a pulse oximeter upon discharge from the ED. “We collect ... their pulse oximeter readings, and are able to again escalate [patients with] increasing hypoxia to a pool of nurses and a physician. If necessary, we are able to bring patients back to the hospital,” Kilaru explains.

These tools help prevent patients from becoming so sick that no one can help them. “We want patients to come back if they are starting to get sicker,” Kilaru adds. “We have better therapeutics, and we can put those patients on oxygen."

Tally the Benefits

COVID Watch is available to a broad patient pool. “Anybody who interacts with our health system, whether they are in the hospital setting or outpatient setting, can be enrolled in this program,” says Kilaru, adding there is no cost to the patients or their insurance companies. “This program significantly benefits our health system in terms of triage and capacity as well as patient satisfaction, but this is not a profit-seeking enterprise.”

Of the first 3,000 patients invited to participate in COVID Watch, researchers found 83% were managed through the tool without escalating to a clinician for follow-up.1 They also found 78% of patients who were offered the program accepted enrollment and remained engaged for a mean of about 12 days. Further, about half of participants asked to extend their involvement with the 14-day COVID Watch monitoring period to 21 days.

About 2% of participating patients were escalated to a nurse every day. Of all the patients who escalated to a nurse during the study period (396), 83 patients were advised to go to the ED. An additional 26 patients were in the ED or admitted to the hospital by the time a COVID Watch nurse responded.1

More data should be forthcoming soon. Funding from the Patient-Centered Outcomes Research Institute is enabling investigators to rigorously study outcomes from patients enrolled in the COVID Watch program. “We are essentially comparing patients enrolled in the program to patients not enrolled, and examining outcomes for 30 days after getting symptoms from COVID-19 and getting tested,” Kilaru observes. “That study is actively going on right now."

Researchers also are analyzing the benefits of providing higher-risk patients with pulse oximeters to determine if this approach enables clinicians to identify worsening symptoms faster, and whether earlier detection can help patients recover without returning to the hospital.

Prioritize Ease of Use

COVID Watch does not require a smartphone. Most patients own some kind of cellphone that allows them to participate. Kilaru says a goal of the program is to cut the number of check-in phone calls staff have to make “so that we are only responding to patients who are worsening or reporting some kind of concern rather than calling every patient every day. The use of these automated programs reduces the amount of human power needed.”

Another plus is the ease with which emergency clinicians can enroll patients, essentially with one click in the electronic medical record during discharge. “We had to have a system that was essentially very easy to implement in the ED,” Kilaru says. “We purposefully didn’t want it to take an hour to get someone into the program.”

Further, investigators note that automated text messaging backed up by clinician support carries potential beyond the monitoring of COVID-19 patients. Researchers at the University of Pennsylvania have produced a similar approach for patients with COPD, and they are in the process of developing programs for patients with other chronic conditions, such as congestive heart failure and hypertension.

Leverage AI

Another way to better manage available resources is to develop a tool that can provide frontline providers with keener insight on which patients who present to the ED with mild symptoms from COVID-19 are at the highest risk for requiring intubation or succumbing to the illness within 30 days. Researchers at the Icahn School of Medicine in the Mount Sinai Health System in New York City created an artificial intelligence-driven algorithm they say can deliver such insight based on routine test data and a chest X-ray. This is information emergency physicians generally have at hand.

Fred Kwon, PhD, a researcher in biomedical sciences at Icahn, says this algorithm differs from other predictive tools. “There are algorithms that use only imaging data or that only use the clinical data from EMRs. What our algorithm does is combine [both sets of data] together similar to would a clinician would do,” he explains.

Further, Kwon notes most other prognostic algorithms only take information from patients who have been admitted or have undergone more advanced imaging tests, such as a CT scan. “You only need to take information that is obtained within days, if not hours, of patients coming to the ED to get an idea of the potential severity so that you can triage the patient and make sure appropriate resources are allocated,” he says.

To develop the algorithm, researchers used data from 338 patients ages 21 to 50 years with COVID-19 who presented to EDs in the Mount Sinai Health System between March 10 and March 26, 2020. The data included chest X-rays, basic blood work (including a metabolic panel and complete blood count), and blood pressure readings. Researchers applied the algorithm to adult patients of all age groups.

Researchers reported the algorithm offers 82% sensitivity in predicting intubation and death within 30 days of initial arrival to the hospital, producing a severity score clinicians can use to plan care. For instance, a patient with a higher severity score likely would be placed under close observation and perhaps given more aggressive or rapid treatment.

However, Kwon acknowledges these results are based on a patient population served by EDs within the Mount Sinai Health System. While this cohort was diverse, other investigators are collaborating with 20 different hospitals around the globe to train a more generalizable algorithm with data from more than 60,000 patients.

“Our algorithm can be readily adapted to be used in patients not just positive for COVID-19, but hopefully in the future, other acute respiratory syndromes and other respiratory illnesses,” Kwon says. “Pneumonia and acute respiratory distress syndrome — those are the big ones.”


  1. Morgan A, Balachandran M, Do D, et al. Remote monitoring of patients with COVID-19: Design, implementation and outcomes of the first 3,000 patients in COVID Watch. NEJM Catalyst. July 21 2020.