N3C Data Help Validate Two Clinical Tools to Predict COVID-19 Mortality Risk

A nurse is talking with an older patient and writing down answers. (Shutterstock)

November 20, 2023

For people hospitalized with COVID-19, two new risk-prediction tools, tested in part using the NCATS National COVID Cohort Collaborative (N3C), could help clinicians make faster, better treatment decisions. The tools use a patient’s age and five common lab tests to analyze their risk of death more rapidly and accurately.

Scientists at Rutgers University reviewed health records from 969 adults hospitalized with COVID-19 from March to May 2020. The researchers used machine learning to examine 77 patient-health variables and find optimal combinations that best predicted severe COVID-19 outcomes.

The machine-learning analysis led to the discovery of two particularly predictive models, named PLABAC and PRABLE. Each model includes four shared patient-health variables: age, platelet count, lactate and blood urea nitrogen. PLABAC adds aspartate aminotransferase and C-reactive protein, and PRABLE uses red cell distribution width and eosinophil count.

The two models topped the traditional five-variable CURB-65 pneumonia severity prediction tool at assessing mortality risk among the people hospitalized with COVID-19, which is often used to predict COVID-19 outcomes. PLABAC and PRABLE performed on par with CURB-65 at identifying patients at greater risk of death. However, they surpassed CURB-65 at identifying patients at lower risk of death. That difference could help clinicians rapidly decide which patients hospitalized with COVID-19 face the greatest risk of death and require more intensive treatment.

To show how effective their prediction tools were, the researchers then tested PLABAC using N3C’s clinical health data, which represents more than 8 million COVID-19-positive patients. PRABLE could not be proven using N3C because the N3C data set lacked two of the PRABLE tool’s variables: red cell distribution width and eosinophil count. PLABAC performed well in predicting mortality using the N3C data from the periods both before widespread vaccination and after.

The researchers explained that the two prediction tools are “a practical first step for the inclusion of machine learning into clinical decision-making for COVID-19 that can serve as a template for future prognostic models.”

The PLABAC clinical predictive tool is available online. Learn more about the study findings in the journal mBio.