Immunologic and vascular biomarkers of mortality in critical COVID-19 in a South African cohort

Abstract

Introduction: Biomarkers predicting mortality among critical Coronavirus disease 2019 (COVID-19) patients provide insight into the underlying pathophysiology of fatal disease and assist with triaging of cases in overburdened settings. However, data describing these biomarkers in Sub-Saharan African populations are sparse.

Methods: We collected serum samples and corresponding clinical data from 87 patients with critical COVID-19 on day 1 of admission to the intensive care unit (ICU) of a tertiary hospital in Cape Town, South Africa, during the second wave of the COVID-19 pandemic. A second sample from the same patients was collected on day 7 of ICU admission. Patients were followed up until in-hospital death or hospital discharge. A custom-designed 52 biomarker panel was performed on the Luminex® platform. Data were analyzed for any association between biomarkers and mortality based on pre-determined functional groups, and individual analytes.

Results: Of 87 patients, 55 (63.2%) died and 32 (36.8%) survived. We found a dysregulated cytokine response in patients who died, with elevated levels of type-1 and type-2 cytokines, chemokines, and acute phase reactants, as well as reduced levels of regulatory T cell cytokines. Interleukin (IL)-15 and IL-18 were elevated in those who died, and levels reduced over time in those who survived. Procalcitonin (PCT), C-reactive protein, Endothelin-1 and vascular cell adhesion molecule-1 were elevated in those who died.

Discussion: These results show the pattern of dysregulation in critical COVID-19 in a Sub-Saharan African cohort. They suggest that fatal COVID-19 involved excessive activation of cytotoxic cells and the NLRP3 (nucleotide-binding domain, leucine-rich-containing family, pyrin domain-containing-3) inflammasome. Furthermore, superinfection and endothelial dysfunction with thrombosis might have contributed to mortality. HIV infection did not affect the outcome. A clinically relevant biosignature including PCT, pH and lymphocyte percentage on differential count, had an 84.8% sensitivity for mortality, and outperformed the Luminex-derived biosignature.

Keywords: COVID-19; SARS-CoV-2; biomarkers; cytokines; mortality; prognostic.

Figure 1

Correlation between biomarkers identified as important in predicting COVID-19 mortality in the study sample. The figure shows the correlation (Pearson) between biomarkers identified by the Boruta algorithm as important in determining the outcome (mortality) for day 1 (A), day 7 (B), and the longitudinal trajectory between days 1 and 7 (C). The scale bar on the bottom of the figure shows the strength of the correlation (closer to 1 or -1 are strongly positive or negative respectively) with a corresponding color scale. Within each cell is a central dotted line representing 0, and the green or purple annotation represents the correlation coefficient and confidence intervals for the two biomarkers interacting in that cell, as well as the direction of the interaction. The biomarker names are shown on the labels of the rows and columns. In cells with red crosses, the confidence interval crosses the 0 line and these interactions are non-significant. Those without crosses represent the significant correlations between the biomarkers, the strength of association may be judged by their color.

Figure 2

Performance metrics for models of clinical and immunologic biomarkers in predicting COVID-19 mortality on admission to the Intensive Care Unit (Day 1). (A) shows the combined performance of both clinical and immunologic biomarkers identified by the Boruta algorithm. (B) shows the performance of a clinical biomarker panel including pH, procalcitonin (PCT), and lymphocyte percentage on the differential count. (C) shows the performance of an immunologic biomarker panel including IL-15, MPO, GDF-15, ST-2, and IL-1Ra. Each dot represents the mean, with the line extending from it representing the standard error. A solid line is the logistic regression model, and a dashed line is the random forest model, both tuned to balanced accuracy. ROC, receiver operating curve; AUC, area under the curve.

Figure 3

Performance metrics for models of immunologic biomarkers in predicting COVID-19 mortality on day 7 of admission to the Intensive Care Unit. (A) shows the performance of all day 7 analytes identified by the Boruta algorithm, without filtering. (B) shows the performance of the combination of analytes selected after correlation-based filtering, including IL-15, VCAM-1, and PCT. Each dot represents the mean, with the line extending from it representing the standard error. A solid line is the logistic regression model, and a dashed line is the random forest model. Models are tuned to balanced accuracy. ROC receiver operating curve, AUC area under the curve.

Figure 4

Biomarkers associated with mortality in critical COVID-19 which remained significant at the p<0.005 level. Red dots indicate the analyte levels in patients who died and blue dots are those who survived. (A) functional group 3, representing Th2 responses [except for Interleukin (IL)-4 and IL-5] on day 7 post-admission. (B) functional group 12, representing intercellular adhesion molecule-1 (ICAM-1) and vascular cell adhesion molecule-1 (VCAM-1) on day 7 post-admission. (C) IL-1Ra or functional group 9, representing anti-inflammatory myeloid cells on day 1. (D) Endothelin-1 (ET-1) or functional group 19, representing vascular tone and endothelial dysfunction on day 7 post-admission. Analyte levels have been scaled and transformed to Z-scores for comparability, and the means were compared with a robust t-test. ST2, growth stimulation gene-2.

Figure 5

Biomarkers with a longitudinal trajectory associated with survival or death. (A) Levels of IL-18 between day 1 and day 7 in survivors. (B) Levels of VCAM-1 from day 1 to day 7 in survivors. (C) Levels of IL-15 from day 1 to day 7 in those who died. Analyte levels have been scaled and transformed to Z-scores for comparability and the means were compared with a robust t-test.