Cardiometabolic Protein Biomarkers Help With Risk Stratification of COVID-19 Patients
- This retrospective study investigated relationships between severe COVID-19 and protein biomarkers implicated in cardiometabolic disease, with the goal of predicting which patients hospitalized with COVID-19 are most likely to suffer poor outcomes
- Models to predict death or ICU admission were developed using data on 343 patients hospitalized between March 10 and April 21, 2020, ("in-sample" group) and were tested on 194 separate patients hospitalized between April 22 and June 1, 2020
- The prediction models that performed best in the out-of-sample group included two hospital laboratory measures (procalcitonin and lactate dehydrogenase) and seven protein biomarkers (IL-6, IL-1RA, ADAMTS13, VEGFD, ACE2, KIM1 and CTSL1)
- Models with the seven protein biomarkers performed better than the best-performing models built with established clinical risk factors extracted from electronic health records, such as BMI and underlying medical conditions (AUC, 0.82 vs. 0.70, P=0.001)
- These results suggest proteomic profiling could improve the triage and treatment of patients who develop COVID-19
Antiviral therapies for COVID-19 are most effective when given before severe symptoms develop, so early recognition is vital. Previous studies have consistently found significant associations between poor COVID-19 outcomes and higher body mass index, coronary artery disease, chronic kidney disease, and type 2 diabetes.
Subscribe to the latest updates from Endocrinology Advances in Motion
Researchers at Massachusetts General Hospital have now identified a set of protein biomarkers previously implicated in cardiometabolic diseases that predict which COVID-19 patients are likely to develop severe illness. Philip H. Schroeder, of the Diabetes Unit, Josep M. Mercader, PhD, a researcher in the Diabetes Unit and Center for Genomic Medicine and group leader at the Broad Institute of MIT and Harvard, Aaron Leong, MD, MSc, an endocrinologist in the Diabetes Unit and researcher in the Center for Genomic Medicine, and colleagues report the findings in Cardiovascular Diabetology.
The researchers conducted proteomic analyses on 537 discarded blood samples from patients hospitalized at Mass General with PCR-confirmed COVID-19 between March 10 and June 1, 2020. They then:
- Developed models to predict severe illness (using the proxies of death or ICU admission within 28 days of presentation) in 343 patients hospitalized early in the surge, between March 10 and April 21, 2020 ("in-sample" group, which included 221 patients, mean age 61, who died or were admitted to the ICU)
- Used those models to predict severe COVID-19 in 194 separate patients hospitalized later in the surge, between April 22 and June 1, 2020 ("out-of-sample" group, which included 82 patients, mean age 64, who died or were admitted to the ICU)
The researchers evaluated the relative importance of 92 protein biomarkers and 24 hospital laboratory tests, adjusting for age, gender, BMI, and race/ethnicity. They observed 36 significant associations with death/ICU admission: 31 protein biomarkers and five hospital laboratory tests.
Prediction in Out-of-Sample Group
The two prediction models that performed best in the out-of-sample group included two laboratory tests and seven protein biomarkers:
- Laboratory measures—Procalcitonin and lactate dehydrogenase
- Two protein biomarkers related to inflammation—Interleukin (IL)-6 (pro-inflammatory) and IL-1RA (anti-inflammatory)
- Two protein biomarkers related to thrombosis—von Willebrand factor–cleaving protease (ADAMTS13) and vascular endothelial growth factor D
- Three protein biomarkers involved in host–virus interactions—Angiotensin-converting enzyme 2 (the cellular receptor for SARS-CoV-2), kidney injury molecule 1 (has a role in viral entry and helps regulate the host response to viral infections), and cathepsin L1 (a protease that can cleave the SARS-CoV-2 spike protein, a step necessary for cellular entry)
Models built with those nine variables were more sensitive and specific than the best-performing models built with only clinical biomarkers and risk factors:
- Logistic regression model—AUC, 0.82 with protein biomarkers vs. 0.70 without (P=0.001)
- Random forest model—AUC, 0.83 vs. 0.69 (P = 3 × 10−5)
Issues for Future Research
Today, more than two years later, the population of patients hospitalized with COVID-19 is generally younger than the samples in this study and includes both unvaccinated patients and vaccinated patients with breakthrough infections or repeat infections.
These differences, plus emerging COVID-19 therapies, could influence proteomic profiles and their association with COVID-19 severity.
Protein biomarkers for risk stratification will also have to be prospectively assessed for their practical utility (e.g., access to testing, cost of testing, and speed of results reporting). Still, it seems likely that if used early in the evaluation of patients hospitalized with COVID-19, proteomic profiling could improve decisions about care pathways and whether to administer monoclonal antibodies or novel antivirals.
view original journal article Subscription may be required
Explore research in the Diabetes Unit
Refer a patient to the Diabetes Unit