Understanding the mechanisms driving severe disease and causing mortality or recovery among Coronavirus disease 2019 (COVID-19) patients is crucial for developing future clinical decision support systems, planning disease control strategies, and developing host-directed therapies. Older age, obesity, diabetes, hypertension, lymphopenia, and neutrophilia have been consistently identified in many studies as predictors of mortality (1, 2). HIV has also been associated with poor outcomes from COVID-19 in multi-center prospective studies, especially in people with a low CD4 cell count (3, 4). Many immunological biomarkers of mortality have been reported, such as interleukin-1α (IL-1α), IL-1β, IL-6, IL-10, IL-18, and tumor necrosis factor (TNF) (5, 6). Along with ferritin, D-dimer, C-reactive protein (CRP), and procalcitonin (PCT), these immunologic biomarkers are found in the group of people infected with Severe Acute Respiratory Syndrome coronavirus-2 (SARS-CoV-2) who develop hyperinflammation, known as a ‘cytokine storm’ (2). Excessive cytokine production from inflammatory cell death is associated with acute organ damage which is life-threatening. This is mediated in part by synergism between TNF and interferon gamma (IFNγ), which triggers the PANoptosis (pyroptosis, apoptosis, and necroptosis), a mechanistic compendium of programmed cell death pathways (7). In addition to hyperinflammation, endothelial markers such as intercellular adhesion molecule-1 (ICAM-1), vascular cell adhesion molecule-1 (VCAM-1), and E-selectin have been associated with disease severity and death. This highlights the important role of vascular endothelial cells as targets of SARS-CoV-2, and of endothelial activation with dysfunction in the pathogenesis of severe COVID-19 (8, 9).

Despite the extraordinary number of publications on this topic during the pandemic, few data from African countries are available in the public domain. Furthermore, Sub-Saharan African populations have a low vaccination coverage to mitigate the effects of future waves or variants which escape immunity acquired from previous infection (10). Therefore, data on the predictors of mortality in Sub-Saharan African populations are greatly needed. The second wave of COVID-19 in South Africa, which occurred between October 2020 and February 2021, was dominated by the B.1.351 Beta variant of SARS-CoV-2 (10). The rapid spread of infection in the population resulted in a high rate of admission of critically ill patients that overwhelmed the health care services. The B.1.351 Beta variant was associated with a high mortality rate among patients admitted in the Intensive Care Unit (ICU) and other healthcare environments (10).

In this study, our team examined the serum of critical COVID-19 patients in Cape Town, South Africa, for key immunological and endothelial cell biomarkers of mortality. The aim was to gain insight into the immune mechanisms underlying fatal COVID-19 and identify the biomarkers with the best predictive potential.

This investigation was a sub-study of a large prospective cohort spanning the whole COVID-19 pandemic, performed in the ICU of Tygerberg Hospital, Cape Town, South Africa. Participants for this sub-study were admitted during the second wave of the COVID-19 pandemic from 9 October 2020 to 10 February 2021. All adult patients admitted to the ICU during this timeframe with laboratory-confirmed SARS-CoV-2 infection on quantitative real-time reverse transcription polymerase chain reaction (RT-qPCR) from nasopharyngeal swab testing, and COVID-19 acute respiratory distress syndrome (ARDS) according to the Berlin definition (11), were recruited on the day of admission to ICU. They were followed until the primary endpoint of death or survival to hospital discharge. Eligibility for ICU admission was predetermined by the provincial Department of Health guidelines, based on the severity of illness, likely prognosis, and ICU bed availability (12).

Baseline demographic characteristics, comorbidities, medication history, and indicators of the severity of illness (i.e., type and intensity of respiratory support, arterial blood gas values, and evidence of other organ dysfunction) were collected and transferred by authorized study staff to an access-controlled Research Electronic Data Capture (REDCap®) database hosted by Stellenbosch University (13). Baseline laboratory measurements were retrieved from the National Health Laboratory Service (NHLS) Laboratory Information System (TrakCare® Lab Enterprise). Only results from blood taken on the day of ICU admission were used. In cases where a full panel blood test was not performed on the day of ICU admission, the results from blood taken within 48 hours of the date and time of ICU admission were used. The first arterial blood gas performed after admission to ICU was selected for the analysis, irrespective of the type of respiratory support at that time. Further data on the patient’s clinical progress during admission were collected, including the primary outcome of death or discharge from the hospital, progression to mechanical ventilation, and initiation of new medications. Data were verified remotely using electronic hospital records and the TrakCare® system. Serum samples were collected from all patients on day 1 (the day of ICU admission), and a second sample was collected from those who survived to day 7, to capture the trajectory of key analytes at a time point when they were likely to differentiate between survivors and non-survivors (5).

Table 2 presents the baseline characteristics of the 87 critically ill COVID-19 patients included in the analysis, of whom 55 (63.2%) died in ICU, and 32 (36.8%) survived to hospital discharge (median 3 days after ICU discharge, range 1–48 days). All patients were receiving 8mg of intravenous dexamethasone daily, and therapeutic dose subcutaneous enoxaparin sodium from admission to ICU. None of the patients received any antiviral treatment or immunomodulatory drugs for COVID-19.

In this study, we present the immunologic and vascular biomarkers of mortality in 87 patients with critical COVID-19 who were admitted during the second wave of the pandemic in South Africa. Our results show that hyperinflammation was associated with death from COVID-19, as shown by the increased levels of Th1 and Th2 cytokines, Th1/Th2-related chemokines, and acute phase inflammatory proteins in those who died.

Cytokines related to hyperinflammation have been strongly associated with severe COVID-19 in other studies (5, 6). Similar to our findings, Lucas et al. (5) reported that patients with higher incidences of coagulopathy and mortality had a baseline immune signature driven by a combination of Th1 and Th2 responses, however, our study, unlike Lucas et al., did not identify Th17 responses as part of this signature. Moreover, Lucas et al. showed that patients with severe COVID-19 maintained elevated Th1 and Th17 cytokine levels throughout the course of their disease compared to patients with moderate disease and that severe disease was also associated with a rise in Th2 responses. In line with this, elevations in Th1 cytokines were associated with mortality at days 1 and 7 in our cohort of patients with severe COVID-19, and at the later time point, the Th2 cytokines also became significant. In Abers et al., elevated levels of IL-15, soluble ST2, and MCP-1, amongst others, were independently associated with mortality (6). Similar to our findings, the longitudinal trajectory of both IL-15 and MCP-1 in Abers et al. were significantly associated with the outcome (6). IL-1Ra is a competitive antagonist of the potent Th1 cytokine IL-1β. It rises in response to IL-1 and modulates its inductive effects on IL-6 and Th17 responses. In this study, IL-1Ra was strongly associated with mortality, providing further evidence of the presence of hyperinflammation and the body’s attempt at immune regulation (17). Many of the cytokines and chemokines reported in this study are also known to be elevated and predict mortality in other forms of ARDS and sepsis, particularly IL-6, IL-8, IL-18, TNFα, and IFNγ (18–20). However, this may not represent the cytokine milieu in the lung. In one report from Saris et al., IL-6, IP-10, MCP-1, IL-10, and IFN-α were all elevated in the plasma of critically ill COVID-19 patients (21). In the bronchoalveolar lavage fluid (BALF), however, IL-6, MCP-1, and IL-10 were significantly higher than plasma, but IP-10 was not and IFN-α was undetectable. This demonstrates that while circulating biomarkers are useful for predicting severity, they do not reveal the full picture.

IL-15 is a pro-inflammatory cytokine with important effects on the activation and cytolytic activity of cytotoxic CD8 T lymphocytes and NK cells, particularly in short-term hypoxia (22). IL-15 was significantly associated with mortality in our panel overall. IL-18 is also pro-inflammatory, activated by the NLRP3 (nucleotide-binding domain, leucine-rich–containing family, pyrin domain–containing-3) inflammasome along with IL-1β and, in combination with IL-12, acts on CD4 T cells, CD8 T cells, and NK cells to induce IFN-γ production. Higher levels of IL-18 have been associated with increased severity of COVID-19 (23). In ARDS from avian influenza virus (H5N1 and H7N9), the prolonged activation of the NLRP3 inflammasome and consequently Caspase 1, results in excess IL-18 and IL-1β production. This in turn causes an IFN-γ–biased cytokine storm, pyroptosis, and lung damage (24). In the absence of IL-15, IL-18 does not induce IFN-γ production but rather plays an important role in the differentiation of naive T cells into Th2 cells and stimulates the production of IL-4 and IL-13. In this study, we found that the survivors of COVID-19 ARDS had reduced levels of IL-15 and IL-18 over the period of ICU admission to day 7 and that the levels of these cytokines were consistently higher in those who died than those who survived. Continuous stimulation with IL-15 has been shown to cause exhaustion in certain immune cells, including NK cells (25). Together, this suggests that these pathways are involved in excessive inflammatory cell death and lung damage, and reflects the development of reduced inflammatory processes in those who survived. This may have been perpetuated by the failure of the regulatory responses which was identified at day 7 in those who died.

The chemokines as a functional group, as well as the levels and trajectory of individual chemokines such as IP-10, MCP-1, and MCP-3, featured prominently in predicting the poor outcome in our study and also shown in other studies. This highlights the role of leukocyte chemoattractants in severe disease progression (26, 27). IP-10 and MCP-1 have also been associated with thrombosis, and it is hypothesized that this may be the underlying mechanism that precipitates an increased risk of mortality (27). Both IL-15 and IP-10 (along with IL-6 and IL-10) were shown to be raised in the serum of severe COVID-19 patients as compared to severe non-COVID-19 acute respiratory illness from another cohort in the Sub-Saharan Africa region (28).

Bacterial superinfection and venous thromboembolism are known complications of critical COVID-19 (29). Our patients who died had significantly higher routine PCT values on day 1, and higher Luminex®-derived PCT, ferritin, and D-dimer values on day 7, than those who survived. We also found that receiving antibiotics during admission was associated with mortality. In addition, there was considerable overlap between the markers identified in our study and those known to be associated with mortality from bacterial sepsis (30). In combination, these findings suggest that bacterial superinfection and venous thromboembolism played a role in mortality, although neither condition was verifiable with imaging or autopsy because of the COVID-19-related restrictions and resource constraints. Theoretically, a dysregulated immune response combined with the corticosteroids that were used to treat the hyperinflammation could have resulted in an individual patient’s susceptibility to secondary infection. Bacterial superinfections occur in up to 50% of critical COVID-19 patients (29, 31). They prolong ventilation, and along with fungi are common causes of mortality in critical COVID-19 (31, 32). They generally occur later in the ICU admission than our samples were taken, but it may be that some of our patients presented late in the evolution of their disease (29, 31). An elevated D-dimer, however, may be indicative of more than a complication of critical COVID-19. Endotheliitis and a progressive endothelial thrombo-inflammatory syndrome have been suggested as the main pathological mechanism of organ injury in severe disease, rather than hyperinflammation (30, 33). This is partly because cytokine levels in COVID-19 hyperinflammation are profoundly lower than in non-COVID-19 ARDS and associated bacterial sepsis (30).

In addition to the hypothesized role of IP-10 and MCP-1 in thrombosis discussed earlier, the prominence of vascular/endothelial markers in all our analyses supports the endotheliitis theory. ET-1 is a potent vasoconstrictor released from endothelial and smooth muscle cells, which stimulates interleukin and TNFα expression in monocytes, leukocyte adherence, platelet aggregation, expression of adhesion molecules including ICAM-1 and VCAM-1, production and action of growth factors, DNA and protein synthesis, and cell cycle progression (34). It is a marker of endothelial dysfunction and is higher in patients with severe COVID-19, those with respiratory failure, and those who died in hospital (35, 36). The endothelium-derived vascular adhesion molecules ICAM-1 and VCAM-1 are markers of endothelial activation, have been reported as elevated in severe COVID-19, and are associated with death among patients admitted to the ICU (9, 26, 37). In addition to this, they are critical for the recruitment of inflammatory cells from circulation into the lungs, and in this way may be mediators of lung injury in ARDS (38). Indeed, VCAM-1 has been shown to be elevated in the BALF of patients with COVID-19 ARDS (39). In our study, ET-1, ICAM-1, and VCAM-1 were significantly associated with the outcome, particularly in the day 7 analysis and the trajectory analysis, where a reduction in the level of VCAM-1 from day 1 to 7 predicted survival. ET-1 and the adhesion molecules play a critical role in the pro-atherosclerotic pathway, and it is not clear in this study whether the derangements we observed were because of pre-existing vascular disease, a new onset COVID-19 endotheliitis, or some combination of the two. GDF-15 is a stress-responsive member of the TGFβ cytokine superfamily which is produced by many cell types including activated macrophages, cardiomyocytes, adipocytes, endothelial cells, and vascular smooth muscle cells. It increases during states of tissue injury and inflammation and has a tissue-protective role in sepsis, including the regulation of injury-mediated responses in the lungs (40). GDF-15 levels have been associated with cardiovascular risk and disease, including endothelial dysfunction and atherosclerosis (41). In this particular study, a high GDF-15 was strongly associated with COVID-19 mortality. A study by Ahmed et al. found that GDF-15 was a significant marker of disease severity that correlated with IL-6 as a predictor of ICU and hospital outcomes (42). Taken together these data suggest a prominent role for endothelial dysfunction and inflammation in the pathogenesis of fatal COVID-19, possibly in the context of pre-existing vascular disease.

Our models created a clinically relevant biosignature for predicting mortality on day 1 of admission to the ICU. It only included results which should be made available within a few hours of admission: lymphocyte percentage on the differential count, pH, and PCT. With sensitivities of 82.8% and 84.8%, this signature might be worth further investigation for use as a triaging tool in the overburdened ICU settings during pandemic times. The combination of clinical and Luminex®-derived biomarkers from day 1 performed even better, and our Luminex®-alone signature from day 7 did better on performance metrics than most other reported scoring systems. However, this finding should be regarded with caution. Our sample was from a highly selected population of critically ill COVID-19 ARDS patients from a specific geographic region. Other scoring systems may suffer the same limitations as documented in a critical appraisal of 39 prognostic models for mortality risk in COVID-19, only one model did not show a high risk of bias (1). This model, the 4C mortality score, predicted in-hospital mortality using patient age, sex, number of comorbidities, respiratory rate, peripheral oxygen saturation, level of consciousness, urea level, and CRP. The AUC was 0.79 for the derivation cohort and 0.77 for the validation cohort (43). Our clinical model for use on day 1 in the ICU has fewer variables than the 4C score and our results suggest that if there was a point-of-care assay for some of the Luminex® analytes we could provide critical care clinicians with a highly sensitive and specific prediction score which is easily available at the bedside. The biosignature we found may also be valid in ARDS of other causes considering the similarity between the biomarkers that we have shown and ARDS of other causes, and the significant association between ARDS severity (the PaO₂/FIO₂) and mortality.

Two of the strongest clinical markers of prognosis in this study were the pH and PaCO₂ on arterial blood gas. COVID-19 ARDS usually presents as a Type 1 respiratory failure with severe hypoxemia, rather than Type 2 where hypercapnia is the dominant feature. In our patients who all had severe hypoxemia, a rising PaCO₂ and drop in pH likely indicated a severe ventilation-perfusion mismatch (from either lung or vascular pathology), where both the transfer of O₂ into the alveolar capillaries and the clearance of CO₂ were impaired. Consistent with this, a rising PaCO₂ trajectory has been associated with mortality from COVID-19 in mechanically ventilated patients in a large population-based cohort study (44). The complex relationship between pH and PaCO₂ was further explored in the larger cohort, which included the patients in this sub-study of biomarkers of mortality (14).

PLWH made up 15.5% of the study population. Even though the proportion of PLWH was higher among those who died, the data was insufficient to determine an exposure-outcome relationship. In addition, there could have been inherent selection bias due to the fact that the PLWH who were selected for admission to the ICU, based on the local eligibility criteria, were virologically suppressed on ART. Epidemiologic evidence from South Africa has shown a significant association between HIV infection and in-hospital mortality among critical COVID-19 patients (3). In this study patients who were not on ART or virologically suppressed were more likely to die in hospital than their counterparts (3). The mechanism underlying this increased risk remains unknown. A histopathology study found no difference in the lungs, liver, heart, or rate of bacterial co-infection (other than Mycobacterium tuberculosis) of PLWH who died of COVID-19 and the HIV-uninfected COVID-19 deceased (45). Another study done within the African continent found no significant difference in serum or nasal lining fluid cytokine responses to moderate-severe COVID-19 between PLWH and HIV-uninfected counterparts (28). However, the study was not sufficiently powered to detect statistically significant differences.

Our study was limited by the small sample size which might have underpowered its ability to detect significant differences. Other key variables such as markers of cardiac dysfunction (Troponin T and NT-proBNP) and hypercoagulability (D-dimer) were seen to be higher in patients who died, but the observed differences were not statistically significant. Other data reported from South Africa have, however, shown these biomarkers of critical illness in COVID-19 in our population to be relevant (46). Selection bias also seems to have played a role in the study as there was no significant difference between groups in age or comorbidity, despite the fact that these have been reported to be among the most common factors associated with mortality. This may be because all our patients had already developed critical illness at the time of admission such that our study sample overall was older with a high rate of comorbidity and therefore a higher risk of death. Lastly, our analysis was limited by the inclusion of only two time points, which may not provide a true reflection of the trajectory over time. Biomarker levels may have fluctuated between the two measured time points or changed significantly after day 7.

In summary, this study has added much-needed data to the pool of biomarkers of severe COVID-19 ARDS in sub-Saharan African populations. We have shown that hyperinflammation, or a severely dysregulated cytokine response, is associated with mortality in the ICU. Our results also suggest that fatal COVID-19 ARDS involves excessive activation of cytotoxic cells and the NLRP3 inflammasome. Bacterial superinfection from immune dysregulation or treatment-induced suppression, and thrombosis from underlying endothelial dysfunction, likely contributed to death in these patients. Our models have made a biosignature of fatal COVID-19 on admission to the ICU which warrants further testing.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://figshare.com/projects/COVID19_ICU_immune_and_vascular_biomarkers_of_mortality/163708.

The studies involving human participants were reviewed and approved by Stellenbosch University Health Research Ethics Committee. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

AEZ, PN, GW, HS, BA, RE, ZC, TM, AZ and NC conceived and designed the study. CS performed the assays. CK, UL, EI, NB, LS, JS, and VN collected or contributed to the data. MM and GT performed the analysis. GW, NdP, HS, AZ, BA, FE, JS, and SM informed the interpretation of the data. JS wrote the paper, which was critically reviewed and approved in the final draft by all authors. All authors contributed to the article and approved the submitted version.