Inflammatory biomarkers to predict the prognosis of acute bacterial and viral infections


 Mortality in acute infections is mostly associated with sepsis, defined as ‘life-threatening organ dysfunction caused by a dysregulated host response to infection’. It remains challenging to identify the patients with increased mortality risk due to the high heterogeneity in the dysregulated host immune response and disease progression. Biomarkers reflecting different pathways involved in the inflammatory response might improve prediction of mortality risk (prognostic enrichment) among patients with acute infections by reducing heterogeneity of the host response, as well as suggest novel strategies for patient stratification and treatment (predictive enrichment) through precision medicine approaches. The predictive value of inflammatory biomarkers has been extensively investigated in bacterial infections and the recent COVID-19 pandemic caused an increased interest in inflammatory biomarkers in this viral infection. However, limited research investigated whether the prognostic potential of these biomarkers differs between bacterial and viral infections. In this narrative review, we provide an overview of the value of various inflammatory biomarkers for the prediction of mortality in bacterial and viral infections.



Introduction
Infectious diseases have been a threat to mankind since prehistoric times [1].While infection-related mortality has been greatly reduced by the discovery of antibiotics and development of supportive treatments in intensive care units, infectious diseases still account for about 15% of deaths worldwide [2].The great majority of these deaths are due to sepsis, defined as 'life-threatening organ dysfunction caused by a dysregulated host response to infection' [3].Upon invasion by a pathogen, including bacteria and viruses, the immune system is activated and orchestrates the release of multiple pro-inflammatory cytokines and acute phase proteins, including interleukin (IL)-1β, IL-6, IL-8, ferritin, and Creactive protein (CRP).Although this pro-inflammatory response is necessary for elimination of pathogens, overzealous 'hyperinflammation' can result in adverse effects for the host, including shock, coagulation disorders, and collateral tissue damage, resulting in organ dysfunction and death.Next to the pro-inflammatory response, activation of the immune system upon infection also encompasses an antiinflammatory response aimed to restore homeostasis and mitigate detrimental effects due to collateral damage.This response includes the release of anti-inflammatory cytokines such as IL-10, downregulation of HLA-DR receptors on monocytes, and the upregulation of immune checkpoint molecules such as PD-(L)1, leading to decreased lymphocyte numbers and functionality [4,5].Nevertheless, also an overzealous or persistent anti-inflammatory response is associated with adverse outcomes.This phenomenon, called 'sepsis-induced immunosuppression', is related to secondary infections and mortality [6,7].
A multitude of therapies aimed to restore the dysregulated immune response in sepsis have been investigated in clinical trials over the last decades.However, up till now, all of these immunotherapeutic approaches have failed to reduce mortality.This is likely due to profound interindividual heterogeneity in the immune dysfunction in patients with sepsis [8].Therefore, stratification based on inflammatory biomarkers is warranted, as these provide information on underlying immune dysregulation.Such a strategy is known as 'predictive enrichment' and will be instrumental in successful application of immunotherapy for individual patients [9].Nevertheless, next to predictive enrichment, early recognition of severe disease and, related to that, identification of patients with increased risk of mortality who need intensive monitoring and/or intensive care is also of utmost importance.This is known as 'prognostic enrichment' [9].Scores or biomarkers with prognostic value also allow for fair comparisons of study populations and, along the same lines, can be used as covariates in statistical comparisons of (treatment) groups.Lastly, they are key in facilitating the initial enrichment step in intervention trials employing a predictive enrichment approach, because selecting a study population with a high baseline mortality risk increases the chances of demonstrating clinical benefit of any intervention.Up till now, mainly clinical scores are used for prognostication in clinical practice and trials.For instance, an increase in the sequential (sepsis-related) organ failure assessment (SOFA) score was shown to have good discriminatory power for survivors versus non-survivors [10][11][12].Furthermore, the APACHE II score has been successfully applied as a prognostic predictor of mortality in patients with an infection admitted to the intensive care unit (ICU) [13][14][15].However, these clinical scores insufficient address the heterogeneity in the underlying dysregulated pathways involved in the host response to acute infections.Therefore, inflammatory biomarkers reflecting (de)activation of immunological pathways might, apart from their predictive enrichment properties, also have value for mortality risk prediction among patients with acute infections.Furthermore, although the recent COVID-19 pandemic caused an increased interest in inflammatory biomarkers, only limited research compared the prognostic potential of these biomarkers in bacterial and viral infections.
In this narrative review, we present an overview of the value of various inflammatory biomarkers for mortality prediction in bacterial and viral infections.To this end, we searched PubMed for studies addressing the prognostic value of biomarkers in patients with acute bacterial and viral infections.We only included articles that quantified the prognostic value of biomarkers by describing the area under the receiver operator curve (AUC) or univariate odds ratios (OR), hazard ratios (HR), or relative risks (RR) for mortality (28-day, 30-day, 90-day, or in-hospital).Furthermore, the study population had to consist of at least 100 adult patients.Cut-off values are reported if based on the Youden's J index.Systematic reviews were considered if pooled analyses were provided.The results of this literature review are presented below, and a comprehensive overview is listed in Tables 1 and 2.

Inflammatory biomarkers 2.1.1. Adipocytokines
Leptin is secreted by adipocytes and is therefore called an adipocytokine.Increased concentrations of leptin are observed in obesity and relate to reduced satiety, but also pro-inflammatory effects of leptin have been described [16,17].Leptin concentrations are generally considered to be increased in patients with bacterial sepsis [18], but similar and lower leptin concentrations in bacterial sepsis compared to healthy controls or non-septic ICU patients have also been described [19,20].Subsequently, leptin is not of prognostic relevance in bacterial sepsis (AUC of 0.56) [21].Regarding viral infections, significantly higher leptin concentrations have been found in COVID-19 patients admitted to the ICU compared to healthy controls, but, similar to bacterial infections, leptin concentrations did not discriminate between surviving and non-surviving COVID-19 patients (AUC of 0.55) [22].
Another adipocytokine secreted by adipocytes is adiponectin.Adiponectin is present in the circulation of healthy individuals in relatively large amounts (up to 10 μg/mL), whereas lower circulating concentrations have been reported in individuals with obesity as well as in patients with inflammation [23].Adiponectin is involved in glucose metabolism and insulin sensitivity and also functions as an antiinflammatory mediator [24].Adiponectin concentrations have some prognostic value in patients with bacterial sepsis, reflected by an AUC of 0.68 to predict mortality, but it needs to be acknowledged that these patients displayed similar adiponectin concentrations as healthy controls or critically ill patients without sepsis [25].Moreover, mixed results have been observed for adiponectin concentrations in COVID-19 patients and no data regarding the prognostic value of adiponectin in COVID-19 have been reported up till now [26,27].
Resistin is an adipocytokine that was shown to play a role in insulin resistance in mice [28].In humans however, it was found to be secreted by macrophages and not by adipocytes.Pro-inflammatory characteristics of resistin in humans were described and a potential role in infections and sepsis was suggested [29,30].Increased concentrations of resistin in sepsis patients compared to healthy controls and non-septic patients have been reported [31,32].Furthermore, resistin concentrations appear to be associated with mortality in ICU patients with bacterial sepsis [32,33] although with a modest prognostic value (AUC of 0.68) [21].Similarly, resistin has shown discriminatory value for COVID-19-related mortality, reflected by an AUC of 0.76 [34].

Complement system
The complement system is initiated early on in the response to infection.Activation of the terminal complement pathway leads to the formation of the terminal complement complex (TCC), which induces cell lysis [35].The anaphylatoxin C3a, which induces phagocytosis and chemotaxis, is formed during complement activation [35].The prognostic value of the complement pathway in bacterial sepsis has been assessed based on plasma concentrations of the anaphylatoxin C3a, albeit only in small patient populations (<50 patients) [36][37][38].These studies have shown increased concentrations of C3a in patients that eventually die, but prognostic accuracy was not investigated.In patients with COVID-19, activation of the complement pathway, reflected by increased C3a and soluble (s)TCC plasma concentrations, was associated with ICU admission, respiratory failure, as well as with mortality [39][40][41].However, when assessing the prognostic value of individual proteins of the complement pathway, C3a (AUC of 0.67), but not sTCC (AUC of 0.60) discriminated between survivors and non-survivors [42].

C-reactive protein
C-reactive protein (CRP) is one of the acute phase proteins produced by hepatocytes during systemic inflammation and induced by the cytokine IL-6.In patients with bacterial sepsis, CRP had poor prognostic value for 28-day mortality, with an AUC of 0.59 [43,44].Other studies reported that CRP is not predictive for mortality at all in patients with bacterial sepsis [13,45,46].In accordance, CRP did not provide additional prognostic value over diagnosis of shock or organ dysfunction [47].Therefore, the prognostic value of CRP for mortality in bacterial infection is considered to be very limited at best.Regarding viral infections, CRP was shown to be associated with ICU admission in patients with severe influenza, reflected by an AUC of 0.82 [48].However, the authors report that because of a small number of deaths, a ROC analysis for mortality was not feasible.Another study including patients with severe influenza found that CRP is not independently associated with ICU mortality, albeit individual data on ROC analyses was not reported [49].In COVID-19, significant prognostic values of CRP have been reported, but studies vary considerably with regard to study populations and mortality rates.This is reflected in a wide range of reported AUC values (0.56 to 0.89), making it difficult to draw clear conclusions [50][51][52][53][54][55][56][57][58][59].

Ferritin
Ferritin is a protein involved in the regulation of iron metabolism.Besides, ferritin is released from hepatocytes upon inflammatory stimulation and is therefore also classified as an acute phase protein.
Recently, ferritin has been recognized as a marker for hyperinflammation, with diagnostic value for the hyperinflammatory entity called macrophage activation-like syndrome (MALS; AUC of 0.78) [60].The optimal cut-off of ferritin for detection of MALS was 4420 ng/mL.The prognostic value of ferritin to predict mortality in patients with pneumococcal bacteraemia was demonstrated for ferritin concentrations above 1125 μg/L (AUC of 0.75) [61].In patients with viral

HMGB-1
High mobility group box-1 protein (HMGB-1) is released into the circulation by activated immune cells upon infection and is a so-called danger-associated molecular pattern (DAMP).It acts as a late mediator of inflammation by promoting the secretion of pro-inflammatory cytokines [67].HMGB-1 is increased in patients with bacterial sepsis [68].Mixed results regarding its association with mortality were found, as both higher, lower, and similar HMGB-1 concentrations in survivors and non-survivors were described [68][69][70].As the discriminatory or prognostic power of HMGB-1 is not reported, it appears plausible that it is limited in bacterial infections.When assessing HMGB-1 in patients with COVID-19, moderate discriminatory value between survivors and nonsurvivors based on plasma HMGB-1 concentrations was demonstrated (AUC of 0.69) [71].

IL-6
IL-6 is a cytokine with both pro-and anti-inflammatory properties that, among others, stimulates the secretion of acute phase proteins by hepatocytes.It has been observed that patients with a bacterial infection have higher concentrations of IL-6 compared to patients with a viral infection [72].Moreover, IL-6 was associated with the development of sepsis in hospitalized patients with community-acquired pneumonia [73].When assessing the prognostic value of IL-6, studies in patients with bacterial sepsis have identified discrimination between survivors and non-survivors with AUC values in the range of 0.61 to 0.85 for patients admitted to either the ED or the ICU [13,44,46,74].In patients with COVID-19, IL-6 was found to be associated with critical disease defined by respiratory failure or the need for ICU care and subsequently with mortality [51,75].Discriminatory power of IL-6 for outcome in COVID-19 was indicated early on in the pandemic based on an AUC of 0.90 [76].As optimal cut-off values are either not reported or highly variable (30 to 237 pg/mL), the optimal concentration of IL-6 for discrimination between survivors and non-survivors remains elusive.Furthermore, it is important to note that the prognostic value of IL-6 in COVID-19 is abolished by the now standard use of dexamethasone (which rapidly decreases IL-6 plasma concentrations) and tocilizumab (which enhances IL-6 plasma concentrations).
Other inflammatory cytokines, like IL-8, IL-10, IL-18, IL-1RA, and TNF were found to be elevated in patients with bacterial infections or COVID-19, especially in more severely ill patients and those that eventually do not survive, but the prognostic value for mortality has not been comprehensively investigated [66,73,76,77].

Neopterin
Neopterin is released by monocytes and macrophages upon stimulation by IFN-y excreted by T-cells.Therefore, it reflects activation of cellular immunity [78].Because of the well-established link with     inflammation and its stable kinetics, neopterin was suggested to be a potential prognostic biomarker in patients with infections.A small study investigating neopterin in patients with bacterial sepsis showed increased concentrations in non-survivors compared to survivors [79].Moreover, neopterin concentrations were shown to be increased in nonsurviving patients with S. pneumonia bacteraemia [61].Besides these observations, the prognostic value was not sufficiently investigated in patients with bacterial infections.In COVID-19, associations with disease severity were found [80].Furthermore, when neopterin concentrations were compared between 298 surviving and 76 non-surviving COVID-19 patients, a very good discriminatory power (AUC of 0.94) was observed with a corresponding optimal cut-off value of 53 nmol/L [81].

Pentraxin-3
Pentraxin-3 (PTX-3) is a pattern recognition molecule of the long pentraxin subfamily that is released by leukocytes following inflammatory stimulation [82].It exerts multiple immunological functions, including regulation of complement activation and enhancement of phagocytosis.PTX-3 has been assessed as a prognostic biomarker in bacterial infections, both in ED and ICU settings.Prediction of severe and fatal disease in patients with bacterial sepsis at the ED based on PTX-3 plasma concentrations was demonstrated in two studies that nevertheless found very different optimal cut-off values (7.7 ng/mL and 26.90 ng/mL) [45,74].A study in ICU patients with bacterial sepsis showed some discriminatory power for mortality, represented by an AUC of 0.69.However, the authors subsequently used a cut-off value based on the median value (20.9 ng/mL) for further analyses, while not reporting the optimal cut-off value based on Youden's J index [43].In COVID-19 patients, PTX-3 was higher in COVID-19 patients that eventually died and displayed an AUC of 0.84 for prediction of mortality.Furthermore, it was reported that a doubling in PTX-3 concentration relates to an OR of 2.61 for mortality, underlining its prognostic relevance in patients with COVID-19 [50].

Presepsin
Presepsin, also known as soluble CD14 subtype (sCD14-ST), is shed from the monocyte surface after binding of lipopolysaccharide (LPS) to CD14.In patients with bacterial sepsis admitted to the ED, presepsin differentiated to some extent between survivors and non-survivors (AUC values of 0.67 and 0.66) [83,84].Prognostic value of presepsin was also identified in infectious patients admitted to the ICU (AUC values of 0.64 and 0.72) [46,85].For patients with COVID-19 admitted to the ED, presepsin demonstrated AUC values of 0.85 to predict in-hospital mortality with an optimal cut-off value of approximately 870 pg/mL [86,87].Similar discriminatory capacity with AUC values of 0.75 and 0.84 was observed for hospitalized patients with COVID-19, of which a third were admitted to the ICU [53,88].

Procalcitonin
Procalcitonin is produced by the thyroid as a precursor of calcitonin.In healthy individuals, only calcitonin is released into the circulation.However, LPS and inflammatory cytokines such as IL-6 and TNF can induce the release of procalcitonin [89,90].Multiple studies presented moderate prognostic power of procalcitonin to predict mortality in patients with bacterial sepsis (AUC values in de range of 0.59 to 0.79) [44][45][46]74,83,84,[91][92][93].Two other studies showed poorer predictive values (AUC values of 0.57 and 0.55), potentially due to the high mortality rate and preselection of patients enrolled in the cohorts [13,85].In patients with COVID-19, procalcitonin was found to discriminate between nonsurvivors and survivors with AUC values ranging from 0.63 to 0.89 [54,59,62,87].Moreover, the optimal cut-off values varied between studies in both bacterial and viral infections (0.47 ng/mL to 1.11 ng/mL for bacterial infections and 0.0027 ng/mL to 1.0 ng/mL for viral infections).Altogether, whether procalcitonin has prognostic value in acute infections largely remains elusive.

sTREM-1
Triggering Receptor Expressed on Myeloid cells 1 (TREM-1) is an innate immune receptor that plays an important role in the amplification of the innate immune response to infection by stimulating the release of pro-inflammatory cytokines [94].The soluble form of this receptor, sTREM-1, is released from the cell membrane and secreted into the circulation during infection.Higher plasma concentrations of sTREM-1 have been reported in non-surviving patients with bacterial sepsis compared to surviving patients [92,93,95].Moreover, sTREM-1 showed prognostic value with AUC values in the range of 0.64 to 0.86, albeit the reported optimal cut-off was in the wide range of 250 to 950 pg/mL [92,93,95].Discrimination between survivors and non-survivors based on sTREM-1 was also shown to be relatively effective in COVID-19 patients (AUC values of 0.73 and 0.75) [55,96].

suPAR
Soluble urokinase plasminogen activator receptor (suPAR) is the soluble form of the cell-bound urokinase plasminogen activator receptor (uPAR) that is present on the surface of various activated innate immune cells.uPAR is known to be upregulated in inflammation and exerts pleiotropic functions in the host response to infection [97].Infection and inflammation cause increased secretion of suPAR into the blood.Therefore, suPAR is thought to reflect the degree of immune activation.Concentrations of suPAR are significantly higher in patients with bacterial sepsis that eventually die and therefore distinguishes nonsurvivors from survivors (AUC of 0.79), with suPAR concentrations of 6.4 ng/mL or higher predicting death [98].For patients with COVID-19, suPAR levels of >3.91 ng/mL associate with the development of ARDS and the need for ICU admission [99], whereas a doubling in suPAR concentrations in ICU patients showed a strong association with mortality (HR of 24) [58].However, the discriminatory and prognostic power for mortality of suPAR concentrations at admission requires further investigation.

Immunosuppression biomarkers 2.2.1. Monocytic HLA-DR expression
Human leukocyte antigen DR (HLA-DR) is a protein expressed on the surface of monocytes and is important for activation of the immune system.Patients with sepsis often display downregulation of monocytic (m)HLA-DR expression, corresponding to immunosuppression and subsequently to the incidence of secondary infections and mortality [100,101].Expression of mHLA-DR is reported either as the percentage positive monocytes or as the number of receptors per cell, with the latter representing a standardized measurement which has been validated across different centres [102].For diagnosis of immunosuppression, a cut-off value of 50% of positive monocytes or 5000 antibodies per cell (Ab/cell) have been indicated [102].In the first 48 h after diagnosis of bacterial sepsis, mHLA-DR expression is inadequate to predict mortality [101].However, in the following days, an increase in mHLA-DR expression was observed in surviving patients, whereas non-surviving patients displayed persistently low mHLA-DR expression [101].A percentage of 30% of positive cells was indicated as the best cut-off for prediction of mortality at days 3-4 [101].Decreased expression of mHLA-DR is also observed in patients with COVID-19, albeit to a lesser extent than in bacterial sepsis [103].Nevertheless, mHLA-DR expression has prognostic value in COVID-19 at 7-10 days after admission (AUC of 0.85 with optimal cut-off of 4672.5 Ab/cell), which was better than the prognostic value at days 0-3 post-admission (AUC of 0.64 with optimal cut-off of 11,312.5 Ab/cell) [104].As this marker reflects immunosuppression, which may develop later on during the disease course, it is not surprising that its prognostic value is higher at later timepoints.

PD-L1
Programmed cell death ligand-1 (PD-L1) is a checkpoint molecule expressed on monocytes and is the ligand for the checkpoint molecule programmed cell death 1 (PD-1) expressed on T-lymphocytes.The PD-1/ PD-L1 pathway is involved in immunosuppression by regulating T-cell proliferation and stimulating lymphocyte apoptosis [4].In patients with bacterial sepsis, the percentage of monocytes expressing PD-L1 at ICU admission has discriminatory value for mortality with AUC values in the range of 0.77 to 0.85 and optimal cut-off values of approximately 45% [105,106].However, as PD-L1 expression is measured by flow cytometry, lack of standardization procedures associated with this technique may hamper implementation in clinical practice.Furthermore, monocytic (m)PD-L1 expression was not predictive for mortality in patients with bacterial sepsis at the emergency department (AUC of 0.49) [83].This discrepancy may indicate that mPD-L1 has only prognostic capacity when assessed later on during the disease course of patients with bacterial sepsis.For patients with COVID-19, only results on soluble (s)PD-L1 concentrations have been reported, a molecule for which only limited data is available across the literature.Nevertheless, concentrations of sPD-L1 were higher in non-surviving patients with COVID-19 and a RR of 1.46 (per pg/mL) for mortality was observed in ICU COVID-19 patients [107].

Adrenomedullin
Adrenomedullin (ADM) is a hormone with blood pressure-lowering effects, including vasodilation, diuresis, natriuresis, increasing capillary permeability, and aldosterone inhibition.Circulating biologically active ADM (bio-ADM) concentrations have prognostic value in patients with bacterial sepsis based on a HR of 2.3 per quartile [108].Midregional pro-adrenomedullin (MR-proADM) is formed in the synthesis of adrenomedullin (ADM).MR-proADM is not biologically active but is a stable metabolite which reflects ADM concentrations [109,110].Concentrations of MR-proADM were shown to have value in diagnosing sepsis [111][112][113].Moreover, if ICU patients present with MR-proADM concentrations above 0.88 nmol/L, a 11.2-fold higher change for mortality was found compared to patients with MR-proADM below 0.88 nmol/L [91].In patients with COVID-19, MR-proADM shows relevant prognostication illustrated by AUC values in the range of 0.79 to 0.85 and similar optimal cut-off points of approximately 1 nmol/L [114][115][116].A slightly higher optimal cut-off value of 1.57 nmol/L to predict mortality was observed in ICU patients with COVID-19 [59].

D-dimer
Before the COVID-19 pandemic, D-dimer was not considered to be a useful biomarker for mortality in acute infections.However, early studies reported high D-dimer concentrations in COVID-19 patients, which were associated with severe disease and mortality, while the increased risk for thromboembolic events was not recognized at that moment in time [117,118].Later on, the high rates of thromboembolic complications were noticed and related to adverse outcomes [119,120].Multiple studies have reported on the prognostic value of D-dimer at admission (before the administration of thromboprophylaxis) for COVID-19-related mortality with an overall good discriminatory power (AUC values of approximately 0.80) and cut-off values of approximately 1 to 2 μg/mL [51,52,117,118,121].

Lactate
Increased blood lactate is a surrogate marker of tissue hypoperfusion and is frequently elevated in critically ill patients.Research investigating the prognostic value of lactate in bacterial sepsis found AUC values in the range of 0.66 to 0.75, indicating moderate discriminatory power [10,74,91,92,95,122].In patients with COVID-19 either admitted to the ED or the ICU, a similar prognostic value for mortality has been reported, with AUC values of 0.65 and 0.68, respectively [59,87].

Lactate dehydrogenase
Lactate dehydrogenase (LDH) is an enzyme important for cellular energy production.LDH plasma concentrations increase upon cell damage, for example due to hypoxia or inflammation.In COVID-19 patients, elevated levels of LDH have also been observed [56,123].LDH was shown to discriminate relatively well between survivors and non-survivors of COVID-19 (AUC values of 0.78 and 0.92) [51,52].

Neutrophil-to-lymphocyte ratio
The neutrophil-to-lymphocyte ratio (NLR) was suggested to reflect the balance between inflammation and immunity based on the blood counts of neutrophils and lymphocytes, respectively.It is associated with overall mortality in the general population [124].In patients with bacteraemia, an NLR above 7 was associated with mortality with a HR of 2.74 compared to a NLR below 7 [125].Furthermore, its prognostic value for mortality was established in patients with bacterial sepsis based on an AUC of 0.78 [44].From the beginning of the COVID-19 pandemic, NLR was also investigated as a potential prognostic indicator of COVID-19-related mortality.Two studies performed in Wuhan observed the prognostic value of NLR with an AUC of 0.81 and 0.95 with corresponding optimal cut-off values of 3.16 and 11.75, respectively [54,126].Later on, more research on NLR in COVID-19 was performed and combined in a meta-analysis including 2967 patients from 10 studies.Pooled analyses showing good prognostic value of NLR based on an AUC of 0.90, although the included studies used different cut-off values (ranging from 3.0 to 11.8) [127].

sACE2
Angiotensin converting enzyme 2 (ACE2) is an important molecule in the renin-angiotensin-aldosterone system (RAAS) that regulates blood pressure and electrolyte balance.Furthermore, it is the primary receptor used by SARS-CoV-2 to enter cells [128].The ACE2 receptor is expressed by cells in the nasal and oral mucosa and by type II alveolar pneumocytes in the lungs [129,130].The soluble form of ACE2 can be released into the circulation, is enzymatically active, and maintains the ability to bind SARS-CoV-2.As there is no clear link to its pathophysiology, this biomarker has not been assessed in bacterial sepsis.In COVID-19 patients, increased plasma concentrations of soluble (s)ACE2 have been reported [131,132].Moreover, patients with severe COVID-19 display higher concentrations of sACE2 compared to moderately ill COVID-19 patients [133].The association between sACE2 and mortality was established based on a OR of 1.032 per mU/L increase in sACE2 concentration [134].

Viral load
A viral infection is diagnosed by measuring viral genetic material in the patient's blood by reverse transcription-polymerase chain reaction (RT-PCR).This is quantified by the Ct-value, which stands for the number of cycles that have to be performed in order to detect the virus' genetic material.Therefore, lower Ct-values reflect higher viral loads.Generally, a patient sample with a Ct-value of 25 or lower is considered as positive.The viral load and the Ct-value can also be of use in prognostic prediction, which was investigated in COVID-19.ORs in the range of 1.5 to 2.9 for mortality were found when Ct-values are below 25 [135,136].

Omics-based prognostication
Other promising results in the area of prognostication are increasingly generated using '-omics-technologies', including transcriptomics and metabolomics.These techniques use large amounts of data, leading to the identification of potentially prognostic signatures.Specific endotypes based on blood transcriptomes were related to mortality in ICU patients with community acquired pneumonia (HR of 1.86 if belonging to the Mars1 endotype) [137].Moreover, metabolome studies revealed the prognostic capacity of expression of death-related metabolic pathways, which showed an AUC of 0.88 for mortality prediction in prospective validation, with a sensitivity of 80.4% and a specificity of 78.8% [138].The prognostic value of transcriptomic signatures obtained from circulating leukocytes was also shown for viral infections, both COVID-19 and non-COVID-19, based on AUC values of 0.76 and 0.90, respectively [139,140].

The impact of observed mortality rate on prognostic value
As the mortality rate of the investigated cohort might influence the observed prognostic accuracy of a biomarker, we present the association between the observed mortality rate and the AUC of various biomarkers we discussed in Fig. 1.In general, mortality rates of the study population did not show an association with the observed AUC of the biomarker.Therefore, the difference in mortality rate between studies did not likely introduce bias in the reported prognostic accuracy of the biomarkers.

Discussion
The present review describes the prognostic potential of multiple inflammatory biomarkers for predicting mortality in bacterial and viral infections, yielding several relevant observations.Overall, AUC values for mortality prediction were slightly lower in bacterial infections compared to viral infections for most of the biomarkers, which could possibly be explained by differences in heterogeneity among study populations.Studies investigating bacterial infections included mostly patients with sepsis due to a variety of bacterial infections, whereas the studies investigating viral infections focused on a single viral infection, such as influenza or COVID-19.This distinctive heterogeneity among patients with bacterial infections could account for the more limited prognostic value observed.Furthermore, several biomarkers clearly showed better prognostic performance in COVID-19 patients compared to patients with a bacterial infection, for example presepsin and procalcitonin.As both presepsin and procalcitonin are induced by bacterial LPS, this may appear counterintuitive.Perhaps it could be hypothesized that in patients with COVID-19, a bacterial superinfection or translocation of bacteria/LPS from the gut causes increased concentrations of these biomarkers and therefore show stronger associations with mortality.Ferritin, NLR, and MR-proADM may also have relevant prognostic value in COVID-19 patients but less so in patients with bacterial infections, which requires further evaluation.Several other biomarkers showed relevant prognostic value for bacterial infections, whereas they are insufficiently investigated in patients with viral infections.For example, PTX-3, suPAR, and PD-L1 show moderate prognostic power in bacterial infections, but whether this also accounts for patients with COVID-19 or other viral infections requires further investigation and validation.Studies on neopterin show promising results, but more research is warranted for ascertainment of its prognostic potential in both bacterial and viral infections.Likewise, data on complement, mHLA-DR, and adipocytokines are sparse or based on small study populations.In general, it must be acknowledged that the prognostic value and the reported optimal discriminatory cut-off values were highly variable across studies, precluding definitive conclusions.
The prognostic relevance of a biomarker could be influenced by the observed mortality rate of the assessed cohort.However, as presented in Fig. 1, the observed AUC values appear not to be affected by the mortality rates observed in the different studies, although the number of studies investigating the same biomarker was relatively limited.Moreover, differences in the degree of immune activation between bacterial and viral infections could impact the prognostic value of inflammatory biomarkers.Along these lines, a small study investigated the host immune response in COVID-19 and bacterial sepsis patients and showed that circulating inflammatory markers were higher in patients with bacterial sepsis compared to COVID-19 patients at ICU admission [141].The same pattern was also observed in other studies [142,143].However, while the plasma concentrations of these cytokines decreased over time in bacterial sepsis patients, they remained high or even showed an increase over time in COVID-19 patients [141].Moreover, both populations also presented with an immunosuppressive profile, illustrated by low mHLA-DR expression and high sPD-L1 concentrations.While this resolved in patients with bacterial infections, patients with COVID-19 had persistent immunosuppression after 1-3 weeks.Nevertheless, also a less pronounced suppression of mHLA-DR (and thus less severe immunosuppression) compared to bacterial sepsis patients has been described in COVID-19 patients both at ICU admission and one week later [103,144].These differences could potentially affect the prognostic value of these markers.Furthermore, the administration of immunomodulatory drugs majorly influences plasma concentrations of inflammatory biomarkers and therefore also have a large impact on their prognostic capacity.Some studies report on this potential influence.For instance, findings on the prognostic value of IL-6 and PTX-3 were validated in a cohort of patients treated with dexamethasone [50] and another study corrected the RR for mortality of IL-6 and PD-L1 for steroid and tocilizumab treatment [107].However, as most of the studies only included patients before the introduction of these treatments, the value of the assessed inflammatory biomarkers in COVID-19 needs reappraisal.
In this review, multiple biomarkers are listed with moderate prognostic power (AUC values ranging from 0.70 to 0.80).Therefore, these biomarkers lack accuracy to guide clinical decisions for individual patients with infections.Furthermore, the additional value of these biomarkers over the clinically used disease severity scores like SOFA and APACHE II is rarely reported.Although some studies have indicated an improvement in prognostication when lactate, procalcitonin, PD-L1 and/or IL-6 were added to the SOFA or APACHE II score [106,[145][146][147][148], it remains elusive to what extent this would potentially facilitate clinical decision-making.Therefore, the most optimal combination of scoring systems and biomarkers requires further study.In any case, we can conclude that, in general, single biomarkers are inadequate for prognostication of individual patients with bacterial of viral infections.Below, we propose several possible explanations for the limited prognostic power of these biomarkers.
First, the host response to an infection is complex and highly heterogeneous.Multiple dysregulated systems are contributing to organ dysfunction that may result in death, however, also other factors, such as age, comorbidities, are of relevance.Therefore, the degree of inflammation, reflected by inflammatory biomarkers only partially explains a patient's risk of mortality.Moreover, multiple inflammatory pathways could become activated upon an infection, with large interindividual differences.On the one hand, a potentially involved pathway will not be expressed in every patient with a high mortality risk, whereas on the other hand, a patient with a high mortality risk will not express all pathways.This makes that prognostication based on a single inflammatory biomarker is not likely to be reliable.Combinations of inflammatory biomarkers show improved prognostic relevance compared to use of single biomarkers.For example, the combination of IL-6 and NLR resulted in adequate mortality prediction with an AUC of 0.90, a sensitivity of 72%, and a specificity of 94% [44].Moreover, combinations of five pro-inflammatory and anti-inflammatory markers resulted in an increase in the AUC to 0.94, whereas AUC values of the individual biomarkers were approximately 0.70 [149].In a cohort of COVID-19 patients, the combination of CRP, D-dimer, ferritin, and IL-6 (defined as the SCOPE score) was reported to adequately predict disease progression (AUC of 0.81) and could stratify patients for early immunotherapeutic treatment [150].Since these observations are from single cohorts and the combinations of biomarkers varied, further research into optimal combinations of biomarkers is warranted.An approach that could potentially overcome this limitation is the use of the described omics-technologies for identification of sets of biomarkers with adequate prognostication.Next, the findings of these comprehensive analyses should be translated to clinically available tests and extensive validations in multiple cohorts are warranted.
Second, the timing of biomarker measurement is highly relevant in its prognostic accuracy.Most studies have assessed the prognostic potential of biomarkers at hospital or ICU admission, where the clinical relevance of mortality prediction is highest.However, the time to death is variable per patient and due to differences in (the number of) dysregulated systems.If death occurs later, the causative dysregulated pathways potentially evolve during hospital or ICU stay and might not be present at admission.Therefore, biomarkers identifying these pathways might not be prognostic at admission but could potentially adequately prognosticate when assessed later on during the disease course.For some biomarkers, the progression and kinetics in the days after admission have shown additional value.For example, prognostication based on mHLA-DR expression was most effective when persisting low expression was observed several days after hospital admission [101,104,151].For ferritin, a less pronounced decrease in serum concentrations in the days following hospital admission was associated with a higher mortality [60].
Nevertheless, inflammatory biomarkers clearly have value.They facilitate stratification of groups of patients with a high versus a low risk of mortality and therefore allow group comparisons.Moreover, identification of prognostic traits could spark investigations into new treatment targets related to this trait.
There are several limitations of this literature review that need to be acknowledged.First, there is a potentially relevant impact of publication bias in the assessment of prognostic biomarkers from the literature, as positive studies tend to be published more.Second, this review only focuses on prognostic value for mortality and did not assess prognostic value for other adverse outcomes, as reporting of these is highly variable across different studies.Third, there is a large variability in disease severity across studies, as different inclusion criteria were used, mostly based on either the Sepsis-2 or Sepsis-3 definition.Fourth, research on prognostic biomarkers in viral infections used to be limited and was mainly performed in small study populations.This changed with the COVID-19 pandemic.Therefore, results on prognostic biomarkers in viral infections mainly rely on data obtained in COVID-19 patients.Finally, it is important to mention that this narrative review does not include all potential prognostic biomarkers and therefore is not allinclusive.
In conclusion, immunological biomarkers illustrating dysregulation of biological processes can be used for predictive enrichment, but deviations from normal values may also be related to patient outcomes and can therefore be used to give a better estimation of prognosis.This narrative review focused on this prognostic value of inflammatory biomarkers in both bacterial and viral infections.Although some biomarkers perform slightly better in bacterial compared to viral infections and vice-versa, this review shows that no distinct patterns in the prognostic performance of biomarkers between infections of bacterial or viral origin are present.Furthermore, we demonstrate that the observed mortality rate is not clearly related to the predictive value of the biomarkers.The predictive value of single biomarkers was mostly moderate.Therefore, it is important to emphasize that caution is required when translating the prognostic value of these inflammatory biomarkers to clinical practice, and that clinical decisions should not solely rely on these biomarkers.Combinations of multiple biomarkers and biomarkers with clinical disease severity scores appear superior for mortality prediction.Such strategies have only been sparsely studied but hold great promise for accurate prognostication.

Table 1
Prognostic biomarkers in bacterial infections.

Table 2
Prognostic biomarkers in viral infections.
(continued on next page)A.H. de Nooijer et al.