New Artificial Intelligence-Based Risk Score May Help Avoid Unnecessary COVID-19 Hospitalizations
Researchers at Massachusetts General Hospital (MGH) have developed a diagnostic tool that uses artificial intelligence to generate a score to assess the prognosis of individual patients diagnosed with COVID-19 at the point of care in outpatient settings. The tool may enable urgent care centers and emergency departments to make rapid and automatic determinations about which […]
Researchers at Massachusetts General Hospital (MGH) have developed a diagnostic tool that uses artificial intelligence to generate a score to assess the prognosis of individual patients diagnosed with COVID-19 at the point of care in outpatient settings. The tool may enable urgent care centers and emergency departments to make rapid and automatic determinations about which patients are most likely to develop complications and need to be hospitalized.
The Diagnostic Challenge
The primary objective of most current COVID-19 diagnostics is to determine whether a person does or does not have the virus. However, once a patient tests positive, providers must make crucial decisions about hospitalization and treatment based on the patient’s individual risks and likelihood to develop complications. In addition to promoting sound care decisions, having a tool to assess the prognosis of COVID-19 patients would also help physicians avoid overburdening already extended hospitals by hospitalizing only patients at greatest risk of negative outcomes.
When the COVID-19 pandemic first began, infectious disease physician Gregory Robbins, MD, enlisted the help of his colleagues on the MGH Biothreats to develop a sophisticated method of assessing the risk of outpatients diagnosed with COVID-19. Dr. Robbins and a team of experts in neurology, infectious disease, critical care, radiology, pathology, emergency medicine and machine learning came up with a potential solution: the so-called COVID-19 Acuity Score (CoVA) based on data input from 9,381 adult outpatients seen in MGH’s respiratory illness clinics and emergency department from March 7 to May 2, 2020. “The large sample size helped ensure that the machine learning model was able to learn which of the many different pieces of data available allow reliable predictions about the course of COVID-19 infection,” noted one of the authors of the study, which was published in The Journal of Infectious Diseases.
CoVA is based on 30 different predictors, including demographics like age and gender, COVID-19 testing status, vital signs, medical history and chest X-ray results (when available). After developing it, the team tested CoVA in another 2,205 patients seen between May 3 and 14 to ensure it would actually work on new patients “in the real world.” And it did. Thus, within the prospective validation group:
- 1 percent of patients experienced hospitalization within seven days;
- 3 percent of patients experienced critical illness within seven days; and
- 5 percent of patients died within seven days.
The key point is that CoVA demonstrated excellence in predicting which patients would fall into each of these categories. Among CoVA’s 30 predictors, the top five most reliable proved to be the patient’s:
- Diastolic blood pressure;
- Blood oxygen saturation;
- COVID-19 testing status; and
- Respiratory rate.
CoVA is not the first risk score developed for predicting COVID-19 complications. However, as noted by one of the study authors, what makes it unique is that it is based on a very large patient sample and specifically designed for use in the outpatient setting, rather than for patients who are already in the hospital. It is also the only score that has been prospectively validated. Consequently, CoVA may prove to be an extremely valuable diagnostic tool, particularly at a time when U.S. COVID-19 cases and hospitalizations are once more surging.