Better Clinical Decision Support for COVID-19: Identifying Patients at Highest-Risk

The clinical spectrum of COVID-19 ranges from asymptomatic to critical illness, making identifying patients at risk for respiratory deterioration a vital part of disease management. However, with so many aspects of the SARS-CoV-2 infection unprecedented on clinical and social levels, it’s not surprising that healthcare organizations don’t have an established, reliable tool to predict clinical deterioration in patients with the virus.

With in-hospital mortality for patients with COVID-19 estimated between 15.3 and 24.5 percent in 2021, a clear opportunity exists to improve outcomes for individuals at a high risk of deterioration with better clinical decision support tools. Identifying these patients early, however, requires risk scoring tools that specifically identify the onset of respiratory failure, the primary indicator of COVID-19 severity, rather than commonly used scoring tools for other acute conditions.

An Opportunity for Better Clinical Decision Support Tools: Existing Risk Scoring May Be Misleading for Patient with COVID-19

Risk-scoring tools exist for common acute conditions, such as septic shock, hypovolemic shock, or cardiogenic shock. But these methods, including the Modified Early Warning System (MEWS), Sequential Organ Failure Assessment (SOFA) scores, and National Early Warning Score 2 (NEWS2), developed in the U.K. by the National Health Service (NHS), may be misleading for patients with COVID-19. COVID-19 primarily manifests with hypoxemia (a below-normal blood oxygen level) without associated systemic disturbance. As such, using these risk scores, a patient with the virus may not appear critically ill—according to vital signs, such as a heart rate and blood pressure—while still in significant respiratory compromise.

For instance, Dr. Guy Glover, a critical care physician at Guy's and St Thomas' NHS Foundation Trust (GSTT) in the U.K., shared a case typical of how COVID-19’s unique pattern of physiology has confounded clinicians and conventional escalation pathways. In Dr. Glover’s example, a 36-year-old male presented to the emergency department with respiratory symptoms. The patient had tested positive for SARS-CoV-2 four days earlier. The hospital admitted him on conventional low flow oxygen and monitored, per the hospital protocol, with four-to-six-hour physiological observations and NEWS2 assessments.

Overnight the patient’s condition deteriorated but didn’t trigger the conventional hospital escalation protocol of oxygen saturations (SpO₂) 92 percent, respiratory rate 20bpm, HR 90bpm, SBP 120mHg, alert, temperature 37.3 degrees Celsius (99.14 Fahrenheit). The care team had escalated his oxygen to a high flow non-rebreather mask (a device to deliver oxygen in emergency situations) to maintain his SpO₂.

Despite the patient’s deterioration, with a lack of systemic physiologic disturbance and because the NEWS2 score only grades supplemental oxygen in a binary (yes or no) way, without increasing weighting for rising fraction of inspired oxygen (FiO₂), his NEWS2 value remained at 3—well below the recommended threshold for escalation for critical care review. However, he was clearly in imminent danger.

A Better Way: Data Points to Respiratory Scoring to More Accurately Predict COVID-19 Severity

In a study conducted at GSTT and led by Dr. Glover in the U.K., researchers assessed the widely used NEWS2 score versus a simple respiratory assessment, the Respiratory rate—Oxygenation (ROX) index, as a more sensitive predictor of deterioration among hospitalized patients with COVID-19. The U.K. study leveraged the from a healthcare-specific data platform (the Health Catalyst Data Operating System (DOS^TM)) and used an acute deterioration analytic accelerator dataset, developed for systemwide quality improvement work.

Employing a nimble and iterative approach, the U.K. clinicians, analysts, and data architects quickly adapted the existing population dataset to flag patients who had tested positive for COVID-19 and identify key outcomes—such as the need for advanced respiratory support alongside existing measures including cardiac arrest, critical care admission, or death. The U.K. team then extracted additional data from the data platform, including co-morbidity (Deyo-Charlson and Elixhauser-van Walraven Comorbidity Indexes) and a novel frailty assessment score to rapidly characterize the patients they were studying.

Reporting on 708 cases, the research demonstrated the earliest singular marker of clinical deterioration was a rising FiO₂. Additionally, tachypnea (abnormally rapid breathing) was a late sign of deterioration, while the cardiovascular parameters generally didn’t indicate an impending adverse event. Scoring supplemental oxygen as a binary variable had poor predictive power for deterioration, therefore demonstrating mechanistically why the NEWS2 score was relatively insensitive to deterioration in COVID-19.

The study confirmed that the predictive validity of the ROX index for COVID-19 deterioration (a composite outcome of cardiac arrest, critical care admission, or death) was significantly greater than the NEWS2. This understanding of the ROX index’s predictive capability confirms the index’s potential to translate into an earlier detection of an adverse event by up to four hours—necessary time to escalate the patient for assessment by the critical care outreach team promptly and safely before it’s too late.

While the ROX index appears to be important, the NEWS2 system retains many advantages for the U.K. NHS due to its widespread adoption and standardized approach. The research group next aims to incorporate its initial findings into a modified NEWS-FiO2 model. The scope of the data platform and the scalable nature of the datasets will allow researchers to test the model in larger patient cohorts (e.g., other respiratory viral infections) to explore generalizability of the findings.

What Health Systems Need to Create COVID-19 Deterioration Scoring

To improve risk prediction among patients with COVID-19 and adopt a respiratory deterioration scoring process along the lines of the above U.K. efforts, health systems need key features and practices in place:

1. The first step in understanding and optimizing the deterioration pathway is to aggregate historical patient data on baseline characteristics and diagnosis as well as longitudinal laboratory and physiological data, then link these values to predictor variables to patient outcomes. By characterizing typical patterns of physiological deterioration, these clinicians can explore these signals can then be explored and iteratively optimize the warning scores.
2. Subsequently, care teams must optimize response in the deterioration pathway—for example, by measuring and improving the critical care outreach team’s response times and escalation to critical care. In the U.K. study described above, the research group has further used healthcare vendor (Health Catalyst) data to report for the first time that onset of physiological deterioration to critical care admission associates with mortality. Furthermore, metrics in the vendor’s acute deterioration analytic accelerator allow the health system to track and improve their responsiveness to deterioration, with the potential to improve both clinical and operational outcomes.
3. Work of this nature requires collaboration between clinicians and data scientists, which provides the ability to identify clinical problems, pose key questions and hypotheses, and develop data-driven answers and solutions.

Harnessing the Potential for Improved COVID-19 Outcomes Globally

More robust insight into COVID-19 deterioration scoring can contribute to improving patient care and the overall pandemic by saving lives and decreasing intensive care time for patients. Applied globally, health systems would recognize potentially very sick patients earlier and intervene sooner, improving the chance for positive outcomes.

Additional Reading

Would you like to learn more about this topic? Here are some articles we suggest:

• Data Science Reveals Patients at Risk for Adverse Outcomes Due to COVID-19 Care Disruptions
• Why Data-Driven Healthcare Is the Best Defense Against COVID-19
• The Right Way to Build Predictive Models for the Most Vulnerable Patient Populations
• Healthcare Data Quality: Five Lessons Learned from COVID-19
• A Sustainable Healthcare Emergency Management Framework: COVID-19 and Beyond

PowerPoint Slides

Would you like to use or share these concepts? Download presentation highlighting the key main points.

Click Here to Download the Slides