Immune Cell-Based versus Albumin-Based Ratios as Outcome Predictors in Critically Ill COVID-19 Patients

Introduction

The hallmark of critically ill patients with infection, regardless of the pathogen, is exacerbated, life-threatening immune dissonance with devastating consequences.¹ Patients with severe forms of COVID-19 do fulfill SEPSIS 3.0 criteria for viral septic shock with specific immunopathogenesis.² Recent efforts to provide tailored immunomodulation have focused on analyzing observed heterogeneity into sepsis subtypes, keeping in mind that novel immunomodulatory therapies have not been successful.³ Even before the COVID-19 pandemic, trends in continuous septic shock mortality rates over time as well as regional disparities suggested that there was still a need for improvement in standard sepsis treatment.⁴ Therefore, it is of paramount importance to identify those patients who are at higher risk of mortality, even more so in a state of medical emergency like a COVID-19 pandemic. In this specific clinical setting, low-cost, readily available parameters would be useful. Quite a few of them were investigated as outcome predictors in critically ill patients with bacterial or viral sepsis, often with contradictory results.^5–7

From the very beginning of the COVID-19 pandemic, efforts have been made to establish the predictive role of specific laboratory parameters as biomarkers regarding severity and outcome. A review of multiple, mostly nonspecific biomarkers (C-reactive protein, ferritin, serum amyloid A, and procalcitonin) and T-lymphocyte markers emphasized that interpretation of the data is limited for the most part due to the small sample size.⁸ That same year, in 2020, interesting research regarding plasma albumin levels in critically ill COVID-19 patients was published.⁹ The authors demonstrated that decreased levels of albumin could, at an early stage, be used as an independent predictor of lethal outcome. During 2021, as more features of COVID-19 became known, attempts were made to assess their prognostic value regarding lethal outcome. One interesting study included 38 critically ill COVID-19 patients in order to assess ICU admission levels of various endothelial biomarkers as predictors of mortality.¹⁰ The authors showed that elevated levels of soluble E-selectin, angiopoietin 2, and soluble intercellular adhesion molecule 1 in nonsurvivors had a significant prognostic value, with an AUC/ROC of > 0.85. Authors of another study demonstrated that elevated levels of mid-regional pro-adrenomedullin on the ICU admission had predictive capacity of mortality in 95 critically ill COVID-19 patients, with an AUC/ROC of 0.73.¹¹ In 2022, larger sample size research regarding inflammatory markers as early predictors of disease severity and outcome in critically ill COVID-19 patients emerged. One study performed on 100 patients showed that a triad of C-reactive protein, ferritin, and interleukin-6 at admission may predict the disease severity early in the course.¹² Another study included 380 patients and focused on the role of biomarkers in predicting the outcome in this patient population. Out of several investigated parameters, the authors found that ferritin was the only significant biomarker with higher values in nonsurvivors.¹³ Several other parameters were found to be of significant predictive value for the mortality in critically ill COVID-19 patients: IL-6 values and IL-6/lymphocyte ratio (sample size of 117 patients)¹⁴ and lactate dehydrogenase (sample size of 168 patients).¹⁵ Although COVID‐19 was declared no longer to be a global health emergency on May 5, 2023, interest in biomarkers as predictors of mortality remained. That same year, an interesting study of 340 critically ill COVID-19 patients was published, focusing on obesity as well as the prognostic values of biomarkers regarding mortality.¹⁶ The authors demonstrated that mortality in obese patients was lower than in the normal weight and overweight groups, yet it did not reach statistical significance. They also found that ferritin and neutrophil/lymphocyte ratio were independent predictors of lethal outcome in obese patients. As an indicator of continued interest in predictive biomarkers of COVID-19 prognosis, just a few months ago, two studies focusing on serum levels of various cytokines were published.^17,18 It is next to impossible to overestimate the importance of routine blood values as predictors of severity and outcome in general; this is particularly true for critically ill patients. Low cost and short turnaround time (TAT) make routine blood parameters very useful in this clinical setting. We have previously investigated several immune cell-based ratios as predictors of lethal outcome in a few subgroups of critically ill patients with secondary bacterial sepsis.⁵ Here, we wanted to expand this approach with both immune cell-based and albumin-based ratios in a cohort of critically ill COVID-19 patients. In the literature available to us, we did not find any research with this unique approach and such a wide array of investigated parameters in a large cohort of patients.

The aim of our retrospective, single-center study was to assess the prognostic value of immune cell-based as well as albumin-based ratios regarding lethal outcome in 612 critically ill COVID-19 patients. The outcome measure was hospital mortality. Immune cell-based ratios included Neutrophil-to-Lymphocyte Ratio (NLR), Monocyte-to-Lymphocyte Ratio (MLR), Platelet-to-Lymphocyte Ratio (PLR), Mean Platelet Volume (MPV), Mean Platelet Volume-to-Platelet count Ratio (MPV/PC), derived NLR ratio, Neutrophil to Lymphocyte x Platelet ratio (N/LP), C-reactive protein (CRP)-to-Lymphocyte Ratio (CLR) and CRP-to-Platelet Ratio (CPR), and Lactate dehydrogenase (LDH)-to-Lymphocyte Ratio (LLR). We also calculated Systemic Immune Inflammation Index (SII) as well as Systemic Inflammation Response Index (SIRI). Albumin-based ratios included CRP-to-Albumin Ratio (CAR), Lactate-to-Albumin Ratio (LAR) and LDH-to-Albumin Ratio (LDH/ALB).

Materials and Methods

Study Design

We performed a single-center retrospective analysis of 612 adult critically ill COVID-19 patients who were admitted to the intensive care unit (ICU) of military hospital Karaburma, which is an integral part of the tertiary university hospital (Military Medical Academy, Belgrade, Serbia) during 32 months (April 2020–November 2022). Blood measurement encompassed Complete Blood Count (CBC), IL-6, C-reactive protein (CRP), albumin, lactate, lactate dehydrogenase (LDH), serum bicarbonate, arterial base deficit/excess (BD/E) and D-dimer. Blood samples were collected on admission to the ICU. All of the measured and calculated parameters were compared according to the outcome between survivors and nonsurvivors, with the outcome measure being hospital mortality. Approval in concordance with the Declaration of Helsinki was obtained from the local ethics committee on March 22, 2024 (No. 33/2024).

Qualified and trained phlebotomists drew the patient’s venous blood. If several measurements were available on the day of enrollment in the study for some patients, the first one at the time point of diagnostic criteria fulfillment was always used to maximize consistency among the patient population. We placed the blood samples into BD Vacutainer K2 EDTA tubes and analyzed them within 2 hours of the venepuncture. The Siemens Advia 120 hematology system, a flow cytometry-based system, determined a complete blood count. Differentiation of white blood cells is done by the peroxidase and basophil channels. More details of blood sample laboratory analyses were explained and published elsewhere.⁵

Study Population

Adult patients with COVID-19, confirmed by positive SARS-CoV-2 detection tests, who required non-invasive or invasive mechanical ventilation in the ICU were enrolled. There were 723 patients who fulfilled the inclusion criteria. The exclusion criteria included malignant disease of any origin and/or preexisting immunodeficiency (65 patients). Also, patients with liver failure (12 patients) and those on renal replacement therapy (34 patients) were excluded from the study because of the impact on biomarker concentrations. Finally, out of 723 patients, 111 were excluded from the study. The remaining 612 patients were included; regarding the outcome there were 241 survivors and 371 nonsurvivors. Figure 1 represents a study flowchart.

Figure 1 Study flowchart.

The study population’s vaccination status and comorbidities have been determined. There was no statistically significant difference regarding vaccination status (partially or fully vaccinated; booster shot) between survivors and nonsurvivors. Of all comorbidities, cardiac dysfunction, obesity, and cerebrovascular disease were more frequent in nonsurvivors (Table 1).

Table 1 Vaccination Status and Comorbidities of the Study Population According to Outcomes

Lung involvement was determined using a chest CT severity score. The total CT score is measured by the sum of the individual lobar scores and can range from 0 (no involvement) to 25 (maximum involvement), when all five lobes show more than 75% involvement. All patients in our study had scores > 15, which indicated severe involvement on the CT severity grading scale. There was no statistically significant difference between survivors and nonsurvivors in this regard.

Calculation of Various Ratios

Immune cell-based ratios included Neutrophil-to-Lymphocyte Ratio (NLR), Monocyte- to-Lymphocyte Ratio (MLR), Platelet- to-Lymphocyte Ratio (PLR), Mean Platelet Volume (MPV; fl), Mean Platelet Volume-to-Platelet count Ratio (MPV/PC; fL 10⁻⁵μL⁻¹), Derived (d-)NLR ratio (neutrophil count divided by the result of white blood cell (WBC) count – neutrophil count); Neutrophil count x 100/Lymphocyte count x Platelet count to calculate N/LP ratio; measured CRP (mg/l) for CRP-to-Lymphocyte Ratio (CLR) and CRP-to-Platelet Ratio (CPR; mg/10⁶ cells), as well as measured LDH (U/l) for LDH-to-Lymphocyte Ratio (LLR). We also calculated Systemic Immune Inflammation Index (SII) using following formula: platelet x neutrophil/lymphocyte count as well as Systemic Inflammation Response Index (SIRI) using following formula: neutrophil x monocyte/lymphocyte count.

Measured albumin-based ratios included CRP-to-Albumin Ratio (CAR), Lactate-to-Albumin Ratio (LAR) and LDH-to-Albumin Ratio (LDH/ALB).

Data Collection

Designated members of our research team gathered and synthesized a variety of demographic, clinical, and laboratory data from hospitalized patients’ medical records. The patients’ confidentiality was protected.

Statistical Analysis

Data are presented using the descriptive statistical methods, for continuous variables, median and interquartile range, Q1-Q3 (first and third quartile) and for categorical variables, absolute and relative frequencies. The normality of the data distribution was confirmed by the Shapiro–Wilk tests.

Differences in examined variables between survivors and nonsurvivors were assessed using Mann–Whitney test and Chi-square test.

The area under the receiver operating characteristic (ROC) curve (AUC) with 95% confidence intervals was calculated for predicted lethal outcomes. The optimal cut-off value was expressed as its sensitivity, specificity and Youden index is shown for all biomarkers.

A large number of independent variables with a statistical significance level of 0.05 in the univariate analysis influenced the careful selection of the independent variables to be used in the multivariate regression analysis model. In order to minimize the effect of multicollinearity, all variables that had a high correlation were excluded from the model, while making sure that the excluded variables were contained in the remaining variables. The multivariate model contained 16 independent variables for outcome prediction (dependent variable).

All statistical analysis was performed using IBM SPSS Statistics 22 (IBM Corporation, Armonk, NY, USA). The criterion for statistical significance was p <0.05.

Results

Baseline Characteristics of the Study Population

During a 32-month study period, 612 critically ill COVID-19 patients, predominantly male (73%) and older (median age of 70 years), were enrolled. Overall mortality was 60.6%. Comparison of baseline characteristics of the study population according to outcome revealed that female mortality rate of 68.9% was significantly higher (χ2 = 6.535, p=0.011) than male mortality rate of 57.5% (Table 2).

Table 2 Baseline Characteristics of the Study Population According to Outcomes

Also, nonsurvivors were significantly older (U=493.5, p=0.037). Hospital stay was significantly longer in survivors (U=224.0, p<0.001) while duration of ICU stay did not differ significantly between survivors and nonsurvivors (U=548, p=0.131).

Laboratory Characteristics of the Study Population on Admission to the ICU

Immune Cell-Based Ratios According to Outcome

White blood cell (WBC) (p<0.001) and neutrophil counts (p<0.001) were significantly higher, while lymphocyte (p<0.001) and platelet counts (p<0.001) were significantly lower in nonsurvivors. Thrombocytopenia was significantly more frequent in nonsurvivors (p=0.003). MPV, MPV/PC, NLR, derived NLR, MLR, N/LP, CRP, LDH, CPR, CLR, LLR, SII, and SIRI values were significantly higher in nonsurvivors (p value for CRP was 0.004; for all other aforementioned parameters, p value was <0.001). So, all differences reached high statistical significance. Monocyte count (p=0.937) and PLR values (p=0.136) did not differ significantly between survivors and nonsurvivors (Table 3).

Table 3 Immune Cell-Based Ratios According to Outcomes

Albumin-Based Ratios, Interleukin(IL)-6, Bicarbonate, Lactate, Arterial B/DE, and D-Dimer According to Outcome

Lactate, CAR, LAR, LDH/ALB, IL-6, and D-dimer values were significantly higher (p value for IL-6 was 0.001; for all other aforementioned parameters, p value was < 0.001). Albumin (p<0.001), bicarbonate (p=0.003), and BD/E (p=0.001) values were significantly lower in nonsurvivors. So, as with most immune cell-based ratios, the differences in all investigated parameters reached high statistical significance (Table 4).

Table 4 Albumin-Based Ratios, IL-6, Bicarbonate, Lactate, Arterial B/DE and D-Dimer According to Outcomes

Clinical Accuracy of Immune Cell-Based and Albumin-Based Ratios, IL-6, Bicarbonate, Lactate, Arterial B/DE, and D-Dimer in Predicting Lethal Outcomes

The clinical accuracy of immune cell-based ratios in predicting lethal outcomes was investigated on admission to the ICU. Most of these parameters demonstrated insufficient discriminative power regarding outcomes, despite statistically highly significant differences. Values of Area under the Receiver Operating Characteristic (ROC) Curve (AUC/ROC) for all cells and most ratios were between 0.49 and 0.60. On the other hand, levels of LLR (AUC/ROC 0.68), N/LP (AUC/ROC 0.66) and NLR (AUC/ROC 0.65) higher than cut-off values are moderate predictors of lethal outcomes (Table 5).

Table 5 Clinical Accuracy of Immune Cell-Based Ratios in Predicting Lethal Outcomes

The same analysis was performed on albumin-based ratios, IL-6, bicarbonate, lactate, arterial B/DE, and D-dimer on admission to the ICU. Apart from LDH/ALB (AUC/ROC 0.66), LDH (AUC/ROC 0.64), IL-6 (AUC/ROC 0.63), and D-dimer (AUC/ROC 0.63), all other parameters demonstrated insufficient discriminative power regarding outcomes. Levels of LDH/ALB, LDH, IL-6, and D-dimer higher than cut-off values are moderate predictors of lethal outcomes (Table 6 and Figure 2).

Table 6 Clinical Accuracy of Albumin-Based Ratios, IL-6, Bicarbonate, Lactate, Arterial B/DE and D-Dimer in Predicting Lethal Outcomes

Figure 2 Receiver operating characteristic (ROC) curves of Albumin-Based Ratios, IL-6, bicarbonate, lactate, arterial B/DE and D-dimer in predicting lethal Outcomes.

Combination of Various Parameters in a Composite Bioscore on Admission to the ICU

To evaluate whether combining LLR, NLR, IL-6, and D-dimer into a single composite bioscore would enhance their predictive performance for fatal outcomes, individual values were assigned a score of 1 or 0 according to how they related (above or below) to previously defined AUC/ROC cut-off levels. The Bioscore I point range is 0–4. It was demonstrated that composite Bioscore I is a statistically highly significant predictor of lethal outcomes with good discriminative power (AUC/ROC 0.70; cut-off value >1 point), better than individual components. When these four parameters were expanded with MLR, LAR, and LDH/ALB and the same type of analysis was performed, composite Bioscore II was created, with a point range of 0–7. However, its predictive performance regarding lethal outcomes did not improve (AUC/ROC 0.672; cut-off value >3 points), although it was statistically highly significant (Table 7 and Figure 3).

Table 7 Clinical Accuracy of Composite Bioscore I and II in Predicting Lethal Outcomes

Figure 3 Receiver operating characteristic (ROC) curves of composite Bioscore I and II in predicting lethal outcomes.

Independent Prognostic Significance of Immune Cell-Based Ratios, Albumin-Based Ratios, IL-6, Bicarbonate, Lactate, Arterial Base Deficit/Excess (B/DE), and D-dimer in Predicting Lethal Outcomes

The study employed logistic regression analyses to investigate potential associations between individual variables and lethal outcomes. To identify the possible predictors of lethal outcomes independent of potential confounders, a forward stepwise multivariate logistic regression analysis was run (Table 8). Multivariate logistic regression model was statistically highly significant (χ2 = 152.386, p<0.001). Independent predictors of lethal outcomes by multivariate logistic regression analysis were age [β=0.120, p<0.001; OR (95% CI)=1.13 (1.07–1.19)], ICU stay [β=0.295, p= 0.001, OR (95% CI)=1.34 (1.16–1.55)], LDH/ALB [β=0.169, p=0.007, OR (95% CI)=1.18 (1.05–1.34] as well as hospital stays [β=−0.455, p<0.001, OR (95% CI)=0.64 (0.54–0.75)]. In general, older patients with longer ICU stay and higher LDH/ALB were more likely to die, while those with longer hospital stays were more likely to survive.

Table 8 Multivariate Logistic Regression Analyses of Variables for Predicting Lethal Outcomes

Discussion

From the beginning of the COVID-19 pandemic, clinicians and researchers alike made considerable efforts to investigate possible prognostic biomarkers. Given the fact that the immune response is crucial in the development of severe forms of COVID-19, attention has been focused on the interplay between various cells and mediators of innate and adaptive immunity, as well as cascades of coagulation.¹⁹ COVID-19 is a heterogeneous disease spectrum; therefore, timely risk stratification would be beneficial for selecting adequate therapies and for efficiently utilizing intensive care resources to save as many lives as feasible.²⁰ As early as 2021, an intriguing review regarding the possible implementation of infection-mediated biomarkers for COVID-19 electrical biosensing platforms was published.²¹ Apart from immunoinflammatory biomarkers (various immunocompetent cells and mediators) and coagulopathy biomarkers (d-dimer), markers for end-organ injury (lactate dehydrogenase - LDH and lactate) are of interest in this clinical setting.^22,23 Polypeptide inflammatory biomarkers such as CRP, procalcitonin, LDH, and serum albumin (ALB), as one of the most important proteins in human physiology, emerge as the focus of COVID-19 biomarker research.²⁴ In addition to examining the serum ALB level, recent research has also looked into the prognostic value of clinical indices that take albumin into account as a variable.²⁵ While many parameters generated from blood tests were rapidly incorporated into standard clinical practice, other suggested circulating biomarkers did not, despite their potential value. This was mostly because these biomarkers were more expensive and had limited availability in general hospital settings during medical emergencies like the COVID-19 pandemic.²⁶

Because of the disease’s intricate pathophysiology, there is undoubtedly no single biomarker that captures the essence of every remarkable feature of COVID-19. We previously demonstrated that several immune cell-based ratios can be used as prognostic parameters in critically ill patients who developed secondary sepsis.⁵ In our present study, our aim was to investigate a variety of immune cell-based and albumin-based ratios as outcome predictors on admission to the ICU, in 612 critically ill COVID-19 patients. As far as immune cell-based ratios, ie, NLR, d-NLR, MPV, MPV/PC, MLR, N/LP, CRP, LDH, CPR, CLR, LLR, SII, and SIRI values, all were higher in nonsurvivors, with differences reaching high statistical significance. On the other hand, monocyte count, and PLR values did not differ significantly between survivors and nonsurvivors. Contrary to our results, Ullah et al reported that there was no statistically significant difference in NLR and LCR values between COVID-19 survivors and nonsurvivors on day 1.²⁷ They concluded that high NLR and decreased LCR values predicted higher odds of mortality, but on day 7. Also, these authors included hospitalized COVID-19 patients, unlike our cohort of critically ill ICU patients. Another study regarding older COVID-19 patients evaluated NLR and PLR as predictors of mortality on hospital admission.²⁸ It was shown that higher values of both ratios are associated with a higher risk of mortality. This is partially in accordance with our results; higher NLR values were associated with lethal outcomes, but PLR values did not differ between survivors and nonsurvivors. In hospitalized COVID-19 patients, Regolo et al demonstrated that NLR is an independent predictor of mortality and ICU admission, unlike PLR.²⁹ In our study, NLR was a highly significant predictor of lethal outcomes only in univariate logistic regression analysis. It has to be noted that a possible explanation for this lack of independence in multivariate analysis might be that, in our investigation, only ICU COVID-19 patients were included. Two interesting studies included only critically ill COVID-19 patients admitted to the ICU, as we did in our investigation. The first one included 272 patients. Authors investigated NLR, d-NLR, PLR, and MLR as predictors of lethal outcome at time 0 (admission to the ICU), in addition to other variables at different time points.³⁰ Clinical accuracy of NLR and d-NLR was higher in our study (0.65 and 0.64, respectively, vs 0.62 and 0.60, respectively), while AUCs for PLR, MLR, and SII were between 0.5 and 0.6 in both studies. These authors investigated the dynamic changes of the listed parameters between time 0 and after 48 hours, which is a good approach. However, we opted to include more variables at one time point, ICU admission, in our pragmatic, real-life study. The second study included 348 patients.³¹ Several immune cell-based (NLR, PLR, MLR, MPV, MPV/PC, SII) and albumin-based (LDH/ALB, CAR) parameters overlapped with those investigated in our study. In accordance with our results, all these parameters were significantly higher in nonsurvivors, except PLR, which was not statistically different between groups. However, the results of multiple logistic regression analysis showed that independent predictors of lethal outcome in this study were age and NLR, while our study demonstrated that independent predictors of death were age and LDH/ALB.

Given that CPR is an emerging risk factor that can indicate both coagulation state and inflammation, it is surprising that there is so little research on it in sepsis settings.

An interesting cross-sectional study regarding the clinical value of CPR in neonatal sepsis revealed that higher CPR values are an independent predictor of severe neonatal sepsis.³² These results are in line with our findings of significantly higher levels of CPR in nonsurvivors.

CLR has lately gained popularity as an early sepsis marker³³ as well as an early biomarker to distinguish sepsis from pneumonia;³⁴ both of these recently published studies focused on neonates. In a multicenter, retrospective French investigation, the predictive value of CLR was examined with respect to COVID-19 severity and mortality at the emergency department (a cohort of around 1000 hospitalized patients with moderate to severe disease).³⁵ Contrary to our findings, the authors reported that only lymphocyte count was significantly linked with mortality in multivariate analysis. However, only one-quarter of this large French cohort were ICU patients. Regarding the clinical accuracy of CLR to predict mortality, French authors reported 159.5 as a cut-off value (AUC/ROC of 0.60; sensitivity 48%, specificity 70%). We found the same AUC/ROC, with very similar sensitivity and specificity (0.60, 47.1%, and 71.7%, respectively), but with a much higher best threshold of 230.8, presumably because of our different cohort (only ICU patients).

The prognostic ability of LLR was investigated in 145 severe COVID-19 patients.³⁶ The authors reported that LLR had the highest accuracy (AUC/ROC of 0.86) compared with other scores (NLR, CLR, CK-lymphocyte ratio, PT-lymphocyte ratio, APTT-lymphocyte ratio, and D-dimer-lymphocyte ratio). This is consistent with our results, which demonstrated the highest AUC/ROC of LLR (0.68) compared to other immune cell-based and albumin-based ratios, IL-6, bicarbonate, lactate, arterial B/DE, and D-dimer. Yet, there are differences between these two studies. Our study showed a lower LLR AUC/ROC value but was established on a more than 4-fold larger sample size. In another study,³⁷ a significant positive correlation was found between LLR and IL-6 levels in 94 hospitalized COVID-19 patients, contrary to our results.

A few years ago, it was reported that high SII is an independent predictor of lethal outcome in patients with metastatic colorectal cancer.³⁸ Recently, in a large cohort of 2218 hospitalized patients in the Department of Internal Medicine, SIRI quartiles 3 and 4, as well as SII quartiles 3 and 4, were independently associated with mortality.³⁹ SII, as a novel predictor of intubation requirement and mortality, was retrospectively investigated in 224 hospitalized COVID-19 patients. The authors reported that the severity and outcome of COVID-19 could be reliably predicted at admission by calculating this marker, with an emphasis on routinely assessed and feasible clinical laboratory values.⁴⁰ We agree with this approach, which can be particularly useful to improve patient management during a pandemic high-influx crisis. Our study demonstrated that both SI and SIRI values were significantly higher in nonsurvivors. However, their clinical accuracy in predicting lethal outcomes was not impressive (AUC/ROCs of 0.58 and 0.61, respectively), although it was statistically highly significant.

Similar to CLR, CAR has recently emerged as an early prognostic biomarker for neonatal sepsis with or without pneumonia.^7,41 The authors of recently published interesting research created a novel LASSO-COX-based prognostic nomogram regarding the association between CAR and 30-day mortality in ICU sepsis patients.⁴² They concluded that the CAR-based model is highly accurate in predicting the risk of hospital mortality in this patient population. In a study of 116 COVID-19 patients, CAR was an independent risk factor for 30-day mortality.⁴³ However, of 116 enrolled COVID-19 patients, severe form of disease had only 17 and there were only 8 nonsurvivors. So, regarding mortality, there is a highly uneven size of two groups (survivors 108 vs nonsurvivors 8, 13.5-fold larger group). The June 2024 review of biomarkers as outcome predictors in sepsis found that CAR outperformed CRP or albumin alone in terms of sensitivity and specificity.⁴⁴ Contrary to these findings, in our cohort of critically ill COVID-19 patients, CRP, ALB, and CAR performed similarly in predicting lethal outcome (AUC/ROCs of 0.57, 0.61, and 0.59, respectively). A retrospective Turkish study enrolled 619 patients with severe COVID-19; the authors analyzed several albumin ratios [blood urea nitrogen (BUN)/ALB ratio (BAR), D-dimer/ALB ratio (DAR), neutrophil/ALB ratio (NAR), as well as CAR] regarding their prognostic ability.⁴⁵ Amazingly, all four studied ratios (BAR, NAR, DAR, and CAR) were found to be clinically accurate in predicting lethal outcomes with rather high AUC/ROSs (0.81, 0.82, 0.77, and 0.80, respectively). The authors of another similar large-scale study (611 COVID-19 patients; mortality rate of 73.2%) investigated the ferritin/ALB ratio (FAR) as a predictor of lethal outcomes.⁴⁶ They revealed even more impressive results: FAR is not only an independent predictor of mortality, but it also has an astonishing AUC/ROC of 0.95. Keeping in mind how protein catabolism is a major issue in critically ill patients, an interesting study revealed a new albumin-based marker, the urea/ALB ratio (UAR).⁴⁷ UAR can reflect inflammation and protein catabolism simultaneously, and the authors demonstrated that this marker is a very good predictor of lethal outcomes (AUC/ROC of 0.77).

LDH is the final marker enzyme in the metabolic pathway of anaerobic glycolysis with the end product of lactate from glucose; therefore, it is the generator of lactate. This enzyme is also a prognostic biomarker for immune surveillance.⁴⁸ During inflammatory activation, innate immune cells release a lot of lactates, implying that lactate metabolism plays a significant role in inflammation and appears to be a regulator of immune cell metabolism. Despite the most severe COVID-19 patients meeting the Sepsis-3.0 criteria for septic shock, there is a growing recognition that the absence of hyperlactatemia often results in an underestimation of illness severity and mortality risk. In severe COVID-19 patients, the Sepsis 3.0 criteria of septic shock may exclude approximately one third of patients with a similarly high risk of a poor outcome and mortality rate, which should be equally addressed.² This is in line with our findings, demonstrating that, although lactate levels were significantly higher in non-survivors, as shown in other studies,⁴⁹ the maximal value was 2.1 mmol/l, which is barely borderline hyperlactatemia in critically ill COVID-19 patients. Nevertheless, lactate level was a statistically highly significant yet modest predictor of lethal outcomes, with an AUC/ROC of 0.60.

LAR, as a prognostic biomarker at ICU admission, has been investigated in sepsis patients; the authors of one study revealed significantly higher 28-day mortality in sepsis patients with elevated LAR.⁵⁰ This is in accordance with our results showing that nonsurvivors had significantly elevated LAR at ICU admission. Regarding the clinical accuracy of LAR in predicting lethal outcomes in critically ill COVID-19 patients, we found that the AUC/ROC was 0.62, so LAR is in the group of statistically highly significant yet modest predictors. This is in line with results from the study of a total of 282 critically ill COVID-19 patients.⁵¹ The authors revealed much higher AUC/ROCs values for lactate and LAR (0.79 and 0.82, respectively), while the AUC/ROC of 0.64 for albumin was slightly higher than in our study (the AUC/ROC of 0.61).

Of all investigated immune cell-based and albumin-based ratios, the only independent predictor of lethal outcomes by multivariate logistic regression analysis was LDH/ALB, along with age and hospital stay, in our cohort of 612 critically ill COVID-19 patients. LDH levels may reflect the extent of cellular damage, and ALB is a well-known negative acute-phase protein. This interesting ratio was investigated in a large cohort of 583 ICU sepsis patients as a predictor of outcome in a very interesting Korean study.⁵² The authors used the same approach as we did, and they also investigated several albumin-based ratios. Regarding the clinical accuracy of LDH/ALB, LDH, ALB, CAR, and LAR in predicting lethal outcomes, their results were comparable to ours: LDH/ALB AUC/ROC of 0.64 vs 0.66 in our study; LDH AUC/ROC of 0.58 vs 0.64 in our study; ALB AUC/ROC of 0.63 vs 0.61 in our study; CAR AUC/ROC of 0.53 vs 0.59 in our study; LAR AUC/ROC of 0.61 vs 0.62 in our study. In accordance with our results, these authors demonstrated that the LDH/ALB ratio was independently associated with in-hospital mortality, and they concluded that it is useful as a prognostic factor in critically ill patients with sepsis. It has to be noted that we had a different patient population in terms of the causative organism; our patients had viral sepsis. Two studies recently published focused on the LDH/ALB ratio as a prognosticator of outcome in critically ill septic patients. In both of them, authors used data from the Medical Information Mart for intensive Care IV (MIMIC-IV) database for retrospective cohort analysis. First study included 4265 septic patients with a 28-day mortality rate of 51.9%.⁵³ It was demonstrated that patients with higher LDH/ALB levels had higher mortality. Also, AUC/ROC values for predicting mortality were 0.65 for LDH/ALB, 0.64 for LDH, and 0.55 for ALB. This is consistent with our results. The second study included 6059 septic patients from the MIMIC-IV database.⁵⁴ The authors performed retrospective cohort analysis with propensity score matching. Here also, patients with a high LDH/ALB ratio had a significantly higher mortality rate. The clinical accuracy of LDH/ALB in predicting lethal outcomes was comparable with our results; the AUC/ROC of 0.68 was slightly higher than our value of 0.66.

In the aforementioned study investigating LDH/ALD in COVID-19 patients,³¹ contrary to our results, LDH/ALB was not an independent predictor of lethal outcomes. Another retrospective cross-sectional study evaluated LDH/ALB as an early outcome predictor in 477 hospitalized COVID-19 patients.⁵⁵ The authors revealed a higher AUC/ROC value of 0.76, with sensitivity and specificity of 72% and 70%, respectively. These values in our cohort of only critically ill COVID-19 patients were lower (0.66, 61%, and 65.8%, respectively). The prognostic value of LDH/ALB was investigated in 206 hospitalized COVID-19 patients older than 65 years.⁵⁶ Again, here the authors reported higher clinical accuracy of this ratio in predicting lethal outcomes (AUC/ROC of 0.81, sensitivity 75.9%, and specificity 76.3%) in comparison with our results obtained in a cohort of critically ill COVID-19 patients. The LHD/ALB ratio also proved to be useful in predicting COVID-19 severity.^57,58 There is one interesting study investigating LDH/ALB To-Urea Ratio (LAU) as a novel prognostic marker in COVID-19 patients.⁵⁹ The authors demonstrated that patients with high LAU have increased mortality rates (AUC/ROC of 0.67).

In an attempt to enhance the predictive value of investigated parameters regarding lethal outcomes, we created two composite bioscores. Bioscore I (LLR, NLR, IL-6, and D-dimer) and Bioscore II (to four existing parameters, MLR, LAR, and LDH/ALB are added). The clinical accuracy of Bioscore I in predicting fatal outcomes was better than that of individual components with an AUC/ROC of 0.7; Bioscor II with an AUC/ROC of 0.67 did not improve performance in comparison with individual parameters. For the most part, individual parameters as well as bioscores are moderate predictors, although statistically highly significant ones. In our patient population, however, both of the composite bioscores we created were better than the well-known SII and SIRI immunoinflammatory scoring systems (AUC/ROC of 0.70 and 0.67 vs 0.58 and 0.61, respectively). Our group used a similar approach in previous research, where composite bioscores were superior to individual parameters in predicting mortality in adult critically ill patients with secondary sepsis.^5,60

The authors of a fairly recent comprehensive review of sepsis biomarkers concluded that most of them have not been well studied and that the clinical role of these biomarkers needs to be better evaluated.⁶¹ Therefore, finding the perfect biomarker remains an eternal quest.⁶² Biomarkers can help clinicians to prognosticate critically ill patients with sepsis of any origin; they should be useful in making informed decisions regarding patient care. Generally, the clinical context should determine the interpretation of biomarkers, never using them as a stand-alone test. However, biomarkers do remain conceptually attractive.

Cause of death in critically ill COVID-19 patients is a very interesting issue. In 2021, compelling COVID-19-association-dependent categorization of death causes in 100 autopsy cases was published.⁶³ There were three types of associations: (1) “strong” (n=57), where COVID-19 was the primary cause of death; (2) “contributive” (n=27), where a pre-existing condition unrelated to COVID-19 was the primary cause of death; and (3) “weak” (n=16), where COVID-19 was either barely responsible for death or not at all. For both the “strong” and “contributive” categories, the leading cause of death was lung involvement, which showed up as diffuse alveolar damage (DAD) on the autopsy. Several post-mortem studies demonstrated that microthrombi were a major cause of cardiac injury in COVID-19.^64,65 In our patient population, in the early phase of treatment for critically ill COVID-19 patients, many fatalities were attributed to sudden cardiac death of unknown cause or cardiac arrest due to pulmonary thromboembolism, as these patients have a high risk of immunothrombosis. In contrast, cerebrovascular incidents accounted for a relatively small proportion of deaths among critically ill COVID-19 patients. The majority of deaths, however, resulted from refractory respiratory insufficiency, which typically occurs later in the disease progression. Refractory respiratory insufficiency is often a consequence of pulmonary fibrosis or complications from prolonged mechanical ventilation, such as ventilator-induced lung injury (VILI). Additionally, bacterial superinfections leading to sepsis were also significant contributors to mortality during this stage of the illness. In general, late deaths can be attributed to chronic critical illnesses, manifesting as various clinical endotypes such as persistent inflammation, immunosuppression, and catabolism syndrome (PICS), for instance.⁶⁶ Consistent with temporal patterns and clinical outcomes in our study are findings regarding an immunological time course analysis including COVID-19 patients, who had persistent inflammation and immunosuppression over the time and suffered worse clinical outcomes, especially when SARS-CoV-2 infection was followed by secondary bacterial infection.⁶⁷

There are several limitations to the current investigation. Since this was a retrospective, single-center study, it was challenging to avoid any possible residual confounding. We cannot extrapolate our findings to the entire population of critically ill patients who are infected because the study was conducted during the COVID-19 pandemic in a very unique clinical setting. Therefore, these data should be interpreted with caution before applying them to the altered reality of 2023 and beyond (vaccination rates, variants, reinfections). Also, it must be acknowledged that all these investigated markers of inflammation are not specific for COVID-19; they are important parameters for other inflammatory diseases as well. Another drawback of our research is the single measurement of the investigated parameters.

Conclusion

Our study aimed to assess the prognostic value of an array of biomarkers regarding the likelihood of a lethal outcome in critically ill COVID-19 patients. We took advantage of the opportunity to classify them into immune cell-based and albumin-based parameters with a reasonable turnaround time in a critical care setting. The only independent predictor of lethal outcomes at ICU admission is the albumin-based LDH/ALB ratio. Most of the other parameters were moderate, although highly significant predictors of mortality in critically ill COVID-19 patients: levels of LLR, N/LP, NLR, LDH, IL-6, and D-dimer higher than cut-off values are moderate predictors of lethal outcomes.