Hematological Parameters and Procalcitonin as Discriminants between Bacterial Pneumonia-Induced Sepsis and Viral Sepsis Secondary to COVID-19: A Retrospective Single-Center Analysis

Bacterial and viral sepsis induce alterations of all hematological parameters and procalcitonin is used as a biomarker of infection and disease severity. Our aim was to study the hematological patterns associated with pulmonary sepsis triggered by bacteria and Severe Acute Respiratory Syndrome–Coronavirus–type-2 (SARS-CoV-2) and to identify the discriminants between them. We performed a retrospective, observational study including 124 patients with bacterial sepsis and 138 patients with viral sepsis. Discriminative ability of hematological parameters and procalcitonin between sepsis types was tested using receiver operating characteristic (ROC) analysis. Sensitivity (Sn%), specificity (Sp%), positive and negative likelihood ratios were calculated for the identified cut-off values. Patients with bacterial sepsis were older than patients with viral sepsis (p < 0.001), with no differences regarding gender. Subsequently to ROC analysis, procalcitonin had excellent discriminative ability for bacterial sepsis diagnosis with an area under the curve (AUC) of 0.92 (cut-off value of >1.49 ng/mL; Sn = 76.6%, Sp = 94.2%), followed by RDW% with an AUC = 0.87 (cut-off value >14.8%; Sn = 80.7%, Sp = 85.5%). Leukocytes, monocytes and neutrophils had good discriminative ability with AUCs between 0.76–0.78 (p < 0.001), while other hematological parameters had fair or no discriminative ability. Lastly, procalcitonin value was strongly correlated with disease severity in both types of sepsis (p < 0.001). Procalcitonin and RDW% had the best discriminative ability between bacterial and viral sepsis, followed by leukocytes, monocytes and neutrophils. Procalcitonin is a marker of disease severity regardless of sepsis type.


Introduction
Sepsis is defined as a life-threatening condition associated with variable degrees of organ dysfunction in a host with an aberrant response to infection [1]. Although bacteria, viruses and parasites can be the causative agent of sepsis, in clinical practice bacterial sepsis prevails [2,3]. The Coronavirus Infectious Disease-19 (COVID-19) pandemic had a major impact on how we approach patients with viral sepsis, and was also an opportunity to deepen our understanding regarding its mechanisms.
Whether bacterial or viral, sepsis is associated with profound disturbances of the normal homeostasis secondary to increased production of pro-inflammatory cytokines-the associated with the degree of organ involvement and one should cautiously use its value when diagnosing and treating a suspected bacterial infection [22][23][24].
Many sepsis subtypes are described based on advanced laboratory tests, but the actual clinical practice is still far from using them on a daily basis [1,2,5,25]. Thus, our aim was to study the effects of bacterial and viral sepsis on the hematological parameters routinely measured in critically-ill patients (absolute count of blood cells, derived hematological indices) and procalcitonin level and describe the main differences. Another aim of this research was to study the discriminative ability of the aforementioned parameters.
pneumonia are conflicting and some authors suggest that increased levels of procalcitonin are highly associated with the degree of organ involvement and one should cautiously use its value when diagnosing and treating a suspected bacterial infection [22][23][24].
Many sepsis subtypes are described based on advanced laboratory tests, but the actual clinical practice is still far from using them on a daily basis [1,2,5,25]. Thus, our aim was to study the effects of bacterial and viral sepsis on the hematological parameters routinely measured in critically-ill patients (absolute count of blood cells, derived hematological indices) and procalcitonin level and describe the main differences. Another aim of this research was to study the discriminative ability of the aforementioned parameters.

Discriminative Ability of the Hematological Parameters and Procalcitonin between Bacterial and Viral Sepsis
ROC curves were plotted in order to study the discriminative ability of the hematological parameters and procalcitonin between bacterial and viral sepsis. Among the hematological parameters (Table 2), RDW% had a very good discriminative power with an area under the curve (AUC) of 0.87 (95% CI: 0.82-0.91, p < 0.001) (Figure 2A). A RDW% cut-off value of > 14.8 (95% CI: 14.6-15.2) had a Sn of 80.7% (95% CI: 72.6-87.2) and a Sp of 85.5% (95% CI: 78.5-90.9) with a +LR of 5.56 (95% CI: 3.68-8.42) and a -LR of 0.23 (95% CI: 0.16-0.33) for the diagnosis of bacterial sepsis. No other hematological parameter had an AUC value higher than 0.8. The ROC curves plotted for white blood cells, monocytes, neutrophils and eosinophils performed in a similar manner (Figure 2A), having good discriminative ability, with AUCs between 0.72 and 0.78 (p < 0.001). No significant difference between AUCs values were observed (p > 0.05). The identified cut-off values for the diagnosis of bacterial sepsis together with Sn, Sp, +LR and -LR are found in Table 2.

Discussion
Our study describes a comparative analysis of the hematological changes induced by two different types of pulmonary sepsis in 262 patients admitted to ICU. We demonstrated that some hematological parameters (the absolute count of white blood cells, neutrophils, monocytes) have good discriminative ability between viral and bacterial sepsis, while the best biomarker differentiating these two entities is procalcitonin. Suprisingly, RDW% also had a very good power in identifing patients with bacterial sepsis. Derived hematological indices did not perform well compared with other hematological parameters and have low sensitivity and specificity. From our knowledge, this is the first study to test in an extensive manner the discriminative ability of these parameters between COVID-19 sepsis and bacterial sepsis secondary to bacterial pneumonia.

Discussion
Our study describes a comparative analysis of the hematological changes induced by two different types of pulmonary sepsis in 262 patients admitted to ICU. We demonstrated that some hematological parameters (the absolute count of white blood cells, neutrophils, monocytes) have good discriminative ability between viral and bacterial sepsis, while the best biomarker differentiating these two entities is procalcitonin. Suprisingly, RDW% also had a very good power in identifing patients with bacterial sepsis. Derived hematological indices did not perform well compared with other hematological parameters and have low sensitivity and specificity. From our knowledge, this is the first study to test in an extensive manner the discriminative ability of these parameters between COVID-19 sepsis and bacterial sepsis secondary to bacterial pneumonia.
In the study presented herein, patients with bacterial sepsis were older compared with COVID-19 patients. Severe bacterial pneumonia and subsequently, sepsis, appear more often at the extremes of age, in patients living in nursing homes or with higher degrees of chronic disease severity [26]. In our cohort, 27.4% of patients presented with HCAP and overall, patients with bacterial sepsis had significantly higher Charlson Comorbidity Index (CCI) value. One should take into account the age variable in the CCI calculation and interpret cautiously this difference given the older population in the bacterial sepsis group.
Our results regarding the hematological alterations are in line with data reported by Moser et al., except for lymphocytes [4]. In their study population, patients with bacterial sepsis secondary to pneumonia had a significantly higher proportion of neutrophils (p = 0.025), but lower lymphocytes (p = 0.033) and platelet numbers (p < 0.001). Moreover, procalcitonin (p < 0.001) and IL-6 (p = 0.036) were higher in the bacterial sepsis group. Lastly, higher eosinophil counts are found in bacterial sepsis patients compared with COVID-19 sepsis, but no difference was observed for the basophil count [27].
Wu et al., specifically compared the characteristics of patients with COVID-19 sepsis and patients with bacterial sepsis secondary to carbapenem-resistant Klebsiella pneumoniae (CrKp) pneumonia [28]. Quantitative alterations of the hematological parameters are similar with the ones reported by this study. Patients with CrKp had significantly higher median values for leukocytes, neutrophils, monocytes, NLR and procalcitonin (p < 0.05). Furthermore, for the median value of lymphocytes and PLR, the same observation as in our analysis was made. Subjects with CrKp sepsis had lower median PLR values (p < 0.05), while COVID-19 sepsis subjects had significantly lower lymphocytes values (p < 0.05) [28].
Hematological parameters with good discriminative ability were leukocytes, monocytes and neutrophils with AUCs between 0.76 and 0.78 (p < 0.001). Given that both types of sepsis are associated with leukocytosis and neutrophilia, the cut-off values of >16 × 10 3 /mm 3 for WBC and >14.1 × 10 3 /mm 3 , although seem high, they differentiate between two types of distinctive inflammatory response [4,12]. The cut-off value of >0.69 × 10 3 /mm 3 for monocytes, even with a value found in the normal range, is explained by monocytosis' higher frequency in the bacterial sepsis group, while monopenia was observed more often in the COVID-19 group. Quantitative alterations are mostly non-specific, but are used on a daily basis for the assessment of patients with sepsis. From a pathophysiological point of view, they may seem less important compared with the qualitative alterations induced by these two conditions [2,4,5]. On the other hand, laboratory tests studying the qualitative alterations are not widely used, nor widely available. Thus, the identified cut-off values we reported, bring more clinical insight into the differential diagnosis of bacterial and viral sepsis.
Furthermore, Perschinka et al. found that procalcitonin, IL-6 and C-reactive protein were significantly higher in patients with bacterial sepsis, even when compared between bacterial pneumonia and COVID-19 pneumonia [29]. Their results showed that two distinct phenotypes were present and steroid therapy did not represent a confounder in their analysis. This was because these differences were maintained between bacterial and COVID-19 sepsis in patients regardless of steroid treatment. Moreover, in their opinion, the high lactate values seen in bacterial sepsis made the diagnosis of sepsis in COVID-19 infection questionable [29]. The reduced degree of organ involvment, other than the lung, is still debated and is considered by others a solid argument to consider COVID-19 a single-organ disease [30]. These discrepancies can be explained, in part, by the different mechanisms through which the systemic responses are generated and propagated during disease evolution [2,5,[31][32][33][34]. Bacteria are capable of producing higher levels PAMPs (pathogen-associated molecular patterns) causing direct injury in different organs. This, in turn, leads to increased production of DAMPs (damage-associated molecular patterns) which will further maintain the inflammatory response and lead to organ failure [33,34]. On the other hand, SARS-CoV-2 primarly manifests with alveolar thrombosis and pulmonary involvement. After the virus enters the blood, there is evidence of organ involvement (other than the lungs), but to a milder extent in terms of organ dysfunction [32,33,[35][36][37]. As seen in our study, patients with viral sepsis had lower SOFA scores, mainly based on the high value of the respiratory subscore. This observation was made in a previous study and raised questions regarding the utility of SOFA score in COVID-19 diagnosis, risk stratification and prognosis. This led to the development of a new score by our research group, the COVID-SOFA score, powered to predict 28-day all-cause mortality. The score had significantly higher discriminative ability when compared with the SOFA score alone [38].
Although not specific, RDW%, was found to be elevated in bacterial sepsis through changes into rheological characteristics of the erythrocytes [39]. These alterations are related to disturbances in the structure and function of the red blood cell (RBC) membrane and cytosol [39][40][41][42]. Thus, these will lead to increased RBC sphericity and abnormal deformability. The final result is increased RBC anisocytosis [39,41]. Lastly, the bacterial insult leads to increased release of erythroblasts into circulation [42]. Increased RDW% was found to be associated with mortality risk in COVID-19 patients [43,44] and progression to septic shock in patients with COVID-19 with an AUC of 0.77 [43]. Results from these studies are difficult to be compared with ours because patients were classified as high or normal RDW%. The high RDW group was considered at a value >14.6% [44]. In our study, the cut-off value for RDW as a discriminant between bacterial and COVID-19 sepsis was >14.8%. This value is higher than the upper limit of normality, suggesting that bacterial sepsis has direct and indirect effects on RBC morphology more than COVID-19 sepsis. Thus, we consider RDW% a valuable tool in the differential diagnosis of bacterial sepsis and also, an indicator of disease severity and prognosis. Caution should be taken when using RDW as a discriminant, given that its value can be increased in different other conditions and is dependent on RBC transfusion [45][46][47].
Although observed in both types of sepsis, the stronger correlation between procalcitonin and disease severity in the viral sepsis group was an interesting finding of our study. In COVID-19 patients, higher procalcitonin levels are correlated with the degree of organ damage and subsequently, disease severity [51,52]. Moreover, higher procalcitonin values predict the development of secondary infections during ICU stay [49,50]. Lastly RDW% values were also associated with disease severity in bacterial sepsis group. Our observations are similar with other reports for both RDW% and procalcitonin [53,54].
The cut-off value for procalcitonin in our study was higher than the ones reported in the aforementioned studies. Given that increased procalcitonin levels were associated with disease severity in COVID-19, we suggest that procalcitonin should be used with caution.
Higher procalcitonin values should not exclusively be considered secondary to bacterial pneumonia, unless viral etiologies were excluded.
Our study has some limitations. Firstly, the retrospective nature of this manuscript prones our results to selection bias. Secondly, only 262 patients from a single center were included in this study, henceforth our findings can not be extensively applied to different types of population, increasing the risk of geographic bias. Thirdly, C-reactive protein measurement was not available in the bacterial sepsis group, nor was measurement of cytokines such as IL-1, IL-6 or TNF-α as these biomarkers are extensively studied in sepsis, regardless of etiology. Furthermore, although patients with bacterial sepsis presented with different types of pneumonia (CAP, HCAP, VAP), in the viral sepsis group, only patients with COVID-19 sepsis were analyzed. This limits the extension of our results to all types of viral sepsis. Moreover, patients with COVID-19 and bacterial co-infection were not introduced in this study.

Study Population
We conducted a retrospective, observational and comparative analysis on a cohort of 262 patients with sepsis secondary to bacterial pneumonia and viral sepsis secondary to COVID-19, admitted to the intensive care unit of a tertiary center (Elias Univeristy Emergency Hospital of Bucharest). Two study periods were chosen: (i) patients admitted with sepsis secondary to bacterial pneumonia, during a period of 24 months (January 2017-December 2018), (2) patients admitted with viral sepsis due to COVID-19 during a period of 16 months (August 2020-December 2021). Local ethics committee of Elias University Emergency Hospital approved this study. Sepsis was defined according to Sepsis-3 criteria [1]. The inclusion criteria for patients with bacterial sepsis were age >18 years, bacterial pneumonia confirmed based on clinical findings and chest X-ray or CT scan and (i) positive bacterial cultures (qualitative or quantitative assessment) or (ii) without an identified pathogen but with clinical response to antibiotic therapy. Exclusion criteria were patients with confirmed viral pneumonia and bacterial co-infection and patients admitted in ICU initially with sepsis secondary to an infection other than pneumonia. Inclusion criteria for patients with viral sepsis were age > 18 years and COVID-19 confirmed through real-time polymerase chain reaction (RT-PCR) and chest X-ray or CT scan. The exclusion criteria for patients with viral sepsis were bacterial co-infection (positive qualitative or quantitative bacterial cultures regardless of site, clinical or radiological findings suggestive of bacterial co-infection), patients admitted in our ICU but mechanically ventilated for > 48 h in a different ICU. In both groups, patients with ongoing radio-, chemo-or immunotherapy, patients with end-stage organ disease (cardiac, kidney, lung, liver), patients with end-stage cancer or severe hematological diseases, patients with missing data and those not meeting Sepsis-3 criteria were excluded.

Data Collection and Analysis
The following data were collected: demographic (age, gender), associated diseases (cardiac, respiratory, kidney, liver, diabetes mellitus, obesity) and Charlson Comorbidity Index values. Moreover, the following data were collected at the moment of sepsis diagnosis: mechanical ventilation and vasopressor drugs requirement, hematological parameters (absolute count of white blood cells, neutrophils, lymphocytes, monocytes, platelets and immature granulocytes, red blood cell distribution width (RDW%) and platelet distribution width (PDW%)), P/F ratio, procalcitonin values and SOFA (Sequential Organ Failure Assessment) score.

Statistical Analysis
Firstly, data was tested for normality of distribution using the Kolmogorov-Smirnov test. Categorical variables were expressed as absolute (number) and relative (percentage) frequency. After crosstabulation, categorical data were compared based on the Chi-square test. Continous data was expressed as median and interquartile range [IQR: Q1-Q3]. The Mann-Whitney U test was used to asses the differences between continous and not normally distributed data across two independent groups. Also, bivariate analysis using Spearman's rho coefficient was performed in order to study the rang correlation between two continous variables. The results from the bivariate analysis were plotted as scatter diagrams, including trend lines with a LOESS smoothing span of 99%. Receiver Operating Characteristics (ROC) analysis was performed to test the discriminative ability of hematological parameters and procalcitonin between bacterial and viral sepsis. For every variable, a ROC curve was plotted and the area under the curve (AUC) was calculated. The cut-off value was identified based on the Youden index. Sensitivity (Sn%), specificity (Sp%), positive likelihood ratio (+LR) and negative likelihood ratio (-LR) together with 95% confidence intervals (95% CI) were calculated for the identified cut-off value. The 95% CI were also reported for the AUC, Youden index, cut-off values and Spearman's rho coefficient of correlation. ROC curves were compared using the DeLong method [55]. The level of significance was established at an alpha level <0.05.
The statistical analysis was conducted using IBM Statistical Package for Social Sciences (SPSS) for Windows ® version 20.0 (IBM Corporation, Armonk, NY, USA) and MedCalc for Windows, version 20.106 (MedCalc Software ® , Ostend, Belgium).

Conclusions
In conclusion, bacterial and viral sepsis secondary to bacterial pneumonia and COVID-19, respectively, have different hematological patterns, as it was demonstrated by the good discriminative ability of leukocytes, neutrophils, monocytes, PLR and eosinophils. Other hematological parameters were fair (MLR, platelets, basophils, lymphocytes and NLR) or no (dNLR, SII) discriminants at all. Moreover, the particular involvement of other organs was once again outlined in this study since patients with bacterial sepsis presented higher degrees of non-respiratory organs dysfunction, while in COVID-19 sepsis, the mainstay of organ involvement was the severe acute respiratory failure. RDW% is a very good discriminant between bacterial and COVID-19 sepsis, reiterating the effects of bacterial sepsis on red blood cells. Lastly, the best discriminant between them remains procalcitonin, but one should keep in mind that a definitive threshold is not set and that in both types of sepsis, procalcitonin level is strongly correlated with disease severity.

Institutional Review Board Statement:
The study protocol was approved by the Local Ethics Committee of Elias Emergency University Hospital of Bucharest, Romania (10301-1/3 January 2023) and in accordance with the ethical guidelines of the World Medical Association Declaration of Helsinki. The requirement for informed consent from individual subjects was waived due to the retrospective and noninterventional nature of the study.

Informed Consent Statement:
The requirement for informed consent from individual subjects was waived due to the retrospective and noninterventional nature of the study.

Data Availability Statement:
The data presented in this study are available on reasonable requestfrom the corresponding author.