Who Lives and Who Dies: Two Major Healthcare Systems Providing COVID-19 Mortality Calculators | 2020-09-23

Two studies published online today seek to answer the most pressing questions for physicians examining patients with COVID-19. What is the risk of death for this person? Jabal Mousa researchers announce clinical predictions model To Lancet Digital Health The Johns Hopkins team has announced their risks calculator In Annual report of internal medicine.. Both use a small amount of readily available patient information to accurately determine the risk of rapid progression and death.

“It is very difficult to predict the mortality rate of COVID-19 patients with various complications, which hinders the prognosis and management of the disease,” said Gaurav, assistant professor of genetics and genomics at Mount Sinai Medical School. Pandey says. New York. The trajectory of the illness is very different, with some patients showing acute shortness of breath and recovering quickly, and some with “happy hypoxia” requiring immediate ventilation. Other patients are first diagnosed after experiencing a heart attack or stroke. Some patients do not develop symptoms. Some people die within hours of diagnosis. Many take weeks to recover.

“Especially, accurate prediction of clinical outcomes for patients across this range of clinical manifestations can be difficult. This problem is a prognosis for COVID-19 patients, especially in the epicenter of diseases that require triage of large numbers of patients. It presents a major challenge to management, “the authors said.

The Mount Sinai model is surprisingly simple. Researchers initially evaluated 17 factors, but the final model included three clinical functions: patient age, minimum oxygen saturation during the medical encounter process, and type of encounter (inpatient, outpatient, and distance care). And use a systematic machine learning framework to predict mortality.

The three “variables made a significant distinction between the classes of deceased and living patients,” Pandy said. BioWorld.. “I was thrilled that the rigorous machine learning framework and large datasets made it objectively determined that the combination of these three variables could distinguish these classes more accurately.”

The team suggested that the model could act as an additional “vital sign” that could be assessed on a regular basis for patients with SARS-CoV-2 infection. Researchers were keenly aware of the need for a fast and reliable model that witnessed the surge in patients in New York this spring and the extraordinary mortality rate of COVID-19. By quickly identifying patients at high risk of death, clinicians can focus appropriately on treatment and resources and minimize mortality.

“In particular, to (i) complement missing values ​​with available data, (ii) automatically select the most predictable subset of features, and (iii) develop predictive models from the selected features. We used some established data analysis and machine learning techniques, ”Pandey explained. “The resulting model produces the probability that a patient will die from COVID-19. Using this rigorous machine learning framework, this result is directly from clinical data collected on a daily basis. We were able to objectively identify the exact forecast model that was saved. “

To develop the XGBoost algorithm, the team teamed up with patient-level data from 5,051 patients treated with the Mount Sinai system in eight hospitals and 400 outpatient clinics from March 9 to April 6, 2020. Was analyzed. Researchers used machine learning-based mortality prediction models using 3,841 records from the development dataset. The test dataset contained records from 961 patients, and the algorithm was positively evaluated with records from 249 patients. This model showed high accuracy with AUC 0.91.

This study has the advantage of using the largest dataset and starting with a wide range of clinical features, with the authors saying, “What we have identified as being most strongly associated with mortality is better than previous models. Is also objective and accurate. “

Johns Hopkins model

The Johns Hopkins team sought to provide a larger step from a model that predicts how much a patient’s illness can get worse and when it is likely to progress. They found that all the more factors available at admission better tell the likely story of the patient’s trajectory.

The Johns Hopkins model includes age, home for the elderly, comorbidities, obesity, respiratory symptoms, respiratory rate, fever, absolute lymphocyte count, hypoalbuminemia, troponin levels, C-reactive protein levels, and Use the interaction between these factors. Together, these factors predict an in-hospital disease progression of AUC 0.85 on day 2 and 0.79 on days 4 and 7.

Johns Hopkins researchers use data from COVID-19’s 827 consecutive admissions to five regional hospitals in Maryland and Washington, DC, from March 4 to April 24, 2020. Developed.

Like the researchers at Mount Sinai, the Johns Hopkins team was keenly aware that time was paramount when dealing with patients with COVID-19. The authors stated that 45 patients had severe COVID-19 at admission, but 120 developed serious illness or died within 12 hours of admission. Of the 302 patients who died in the study, the median time to progression was 1.1 days.

“Rapid progression of disease after hospitalization [to the hospital] Lead author Brian Garibaldi and colleagues provide a narrow window for intervention. “Various combinations of risk factors seem to predict serious illness or death, with probabilities ranging from 90% to 5%.”

“This is part of what we’ve learned in the months since we started seeing patients with COVID-19 in our hospital,” says Garibaldi. A model called “It’s important to share knowledge with colleagues in other hospitals as we continue to work on a large number of COVID-19 infections throughout the United States.” COVID Hospitalization Risk Calculator (CIRC), Available online.