Proteomics in combination with data analysis more accurately predicted outcomes among patients with severe COVID-19 infection compared with currently available disease progression measurements, according to findings from a cohort study published in PLOS Digital Health.
Investigators conducted a prospective cohort study at 2 health care centers among patients with severe COVID-19 infection who were treated between March and September 2020. All patients had confirmed COVID-19 infection that progressed to critical disease requiring invasive mechanical ventilation and additional organ support. Patients were treated according to current clinical guidelines. The investigators sought to determine whether proteomic measurements accurately predict outcomes (death or survival) in patients with severe COVID-19 infection. The investigators assessed plasma proteomes via analysis of patients’ blood sample specimens.
Among a total of 50 patients included in the study, 72% required kidney replacement therapy (KRT), 38% required extracorporeal membrane oxygenation (ECMO), and 32% required both KRT and ECMO. All patients were treated in an intensive care unit, and the median time from sampling to the outcome was 39 (IQR, 16-64) days.
Based on analysis of area under the receiver operating characteristic curve (AUROC) values, patients’ Charlson comorbidity index (0.63), sequential organ failure assessment (SOFA; 0.65), and acute physiology and chronic health evaluation (APACHE II; 0.68) scores did not accurately distinguish between patients who died vs those who survived.
Investigators found 78 plasma proteomes with concentrations that changed significantly during patients’ disease course, 14 of which changed differently depending on whether a patient died or survived. Patients who died were found to have inflammatory proteins that significantly increased over time. Unlike patients who died, concentrations of inflammatory proteins significantly decreased over time among those who survived. Anti-inflammatory proteins also decreased over time in patients who died. The investigators noted 2 key proteins of the coagulation system — thrombin and plasma kallikrein — both of which decreased over time in patients who died and increased among those who survived.
Because obtaining time series data is impractical in an intensive care setting, the investigators developed a machine learning model based on parenclitic networks to explore the potential of assessing plasma proteomes from blood sample specimens obtained at a single timepoint to predict outcomes. Among an independent cohort of 24 patients with severe COVID-19 infection, the machine learning model demonstrated significantly accurate predictive power (AUROC =1.0; P =.000047), correctly predicting the outcome for 18 (~95%) patients who survived and 5 (100%) of those who died.
“The majority of proteins with high relevance in the model are components of the coagulation system and complement cascade, highlighting their critical role in [predicting disease] progression and outcomes [among patients with] severe COVID-19 [infection],” the investigators concluded.
Demichev V, Tober-Lau P, Nazarenko T, et al. A proteomic survival predictor for COVID-19 patients in intensive care. PLOS Digital Health. Published online January 18, 2022. doi:10.1371/journal/pdig.0000007