Revisit Eosinopenia as a Biomarker for Diagnosis of Sepsis: A Meta-Analysis

Sepsis is a life-threatening and time-critical medical emergency, therefore, the early diagnosis of sepsis is essential to timely treatment and favorable outcomes for patients susceptible to sepsis. Eosinopenia has been identied as a potential biomarker of sepsis in the past decade. So, we performed a meta-analysis to assess the diagnostic ecacy of eosinopenia for sepsis. Method We searched PubMed, EMBASE, Web of Science, and Cochrane Central Register of Controlled Trials CENTRAL databases to identify studies that met the inclusion criteria. Two authors performed data extraction independently. The pooled outcomes were calculated by TP (true positive), FP (false positive), FN (false negative), TN (true negative) by using bivariate meta-analysis model in STATA 14.0 software. Seven studies were included in the present study with a total number of 3842 subjects. The pooled sensitivity, specicity, positive likelihood ratio, negative likelihood ratio and diagnostic odds ratio were 0.66 (95%CI [0.53-0.77]), 0.68 (95%CI [0.56-0.79]), 2.09 (95%CI [1.44-3.02]), 0.49 (95%CI [0.34-0.71]) and 4.23 (95%CI [2.15-8.31]), respectively. The area under the summary receiver operator characteristic curve (SROC) was 0.73 (95%CI [0.68-0.76]). Meta-regression analysis revealed that no single parameter accounted for the heterogeneity of pooled outcomes. For each subgroup of different eosinopenia cutoff values (50, 40, <=25, 100), the sensitivity was 0.61, 0.79, 0.57, 0.54, and the specicity was 0.61, 0.75, 0.83, 0.51, respectively. Our ndings suggest eosinopenia no in Further large clinical trials are needed control patient (SIRS/non-infection), sample size ( ≥ 200/<200), type of study (prospective/retrospective), country (Europe and America/others). A sensitivity analysis was conducted to examine the inuence of each study on the meta-analysis, by calculating the pooled outcomes after omitting one study at a time. The impact of eosinopenia cutoff used in each study was further investigated by performing a subgroup analysis. the results showed that the examined covariates could not account for the heterogeneity. We also conducted a sensitivity analysis to detect the inuence of each study on the pooled DOR and overall outcomes (Suppl. Table. 2). This Leave-one-out analysis indicated that no single study exerted signicant inuence on the results of the meta-analysis. Because of the internal heterogeneity and small number of including studies, our results show only weak association between eosinopenia and the pathophysiological status of sepsis. Future researches are supposed to focus on the following two aspects: (1) whether eosinopenia could contribute to the diagnosis of sepsis caused by some specic pathogens; (2) whether eosinopenia show a signicant advantage in the era of sepsis-3.0, given that the studies included in our meta-analysis were almost conducted before the emergence of sepsis-3.0 criteria[1].


Abstract Background
Sepsis is a life-threatening and time-critical medical emergency, therefore, the early diagnosis of sepsis is essential to timely treatment and favorable outcomes for patients susceptible to sepsis. Eosinopenia has been identi ed as a potential biomarker of sepsis in the past decade. So, we performed a metaanalysis to assess the diagnostic e cacy of eosinopenia for sepsis.

Method
We searched PubMed, EMBASE, Web of Science, and Cochrane Central Register of Controlled Trials CENTRAL databases to identify studies that met the inclusion criteria. Two authors performed data extraction independently. The pooled outcomes were calculated by TP (true positive), FP (false positive), FN (false negative), TN (true negative) by using bivariate meta-analysis model in STATA 14.0 software.

Results
Seven studies were included in the present study with a total number of 3842 subjects. The pooled sensitivity, speci city, positive likelihood ratio, negative

Conclusions
Our ndings suggest that eosinopenia shows no superiority in the diagnosis of sepsis. Further large clinical trials are needed to re-evaluate eosinopenia as a biomarker of sepsis.

Background
As "a life-threatening organ dysfunction caused by a dysregulated host response to infection" [1], sepsis has been recognized as a major threat to global health causing millions of deaths per year globally according to WHO, and a true time-critical medical emergency susceptible to death [2]. Therefore, an accurate and quick diagnostic testing is essential to improve outcomes in the patients with sepsis. Although working as the gold standard for identifying infectious conditions, microbial culture has obvious limitations such as time-consuming delay and and considerable false negative results [3]. Thus, seeking timely, sensitive and speci c biomarkers has become the research focus in this eld, with C reaction protein (CRP), procalcitonin (PCT) and white blood cell count as potential candidates, while the e ciency is far below satisfaction [3,4] Eosinophils, comprising 1%-3% of circulating leukocytes, play a role in host defense against helminths and the propagation of allergic conditions, and are considered as an integral part of immune and in ammatory network, and homeostatic regulation as well in recent years [5]. The reduction of circulating eosinophil count (namely eosinopenia) in response to acute infection was rstly described in 1893 by Zappert et al. [6], and is now acknowledged as a signal of acute infection [7,8]. The feasibility of eosinopenia in diagnosis of sepsis was rstly tested by Abidi et al, who showed that eosinopenia is good at discriminating between non-infection and infection, but is rather weak in differentiating sepsis from systemic in ammatory response syndrome (SIRS) [9]. Given that there is a signi cant correlation between sepsis and eosinopenia, several studies carried out in addressing the value of eosinopenia to diagnose sepsis recently [9][10][11][12][13][14][15] and to predict the prognosis of sepsis [16][17][18] with pretty much inconsistence. With emerging of new studies on eosinopenia as a marker of sepsis, we determine to perform the present meta-analysis to reevaluate whether eosinopenia is a good biomarker for diagnosis of sepsis.

Study design and Literature search
The present meta-analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement [19]. The protocol for this meta-analysis is available in PROSPERO (CRD42020197664). Studies that investigated the diagnostic e cacy of eosinopenia for sepsis were searched for inclusion. Discrepancies were resolved by the consensus of all authors.
A comprehensive electronic search of the PubMed, EMBASE, Web of Science, and Cochrane Central Register of Controlled Trials CENTRAL databases was performed. Then we reviewed the reference lists of included articles to obtain additional relevant articles. No language restriction or publication date restrictions were applied, and the date of our search was until 9 July 2020.
We used the keywords "sepsis" and "eosinopenia" to search articles, and the process of study selection was schematically presented in the PRISMA ow diagram. The detailed search strategy is showed as in Suppl. Table 4.

Selection criteria
Studies were selected based on the following inclusion criteria: (1) enough data to calculate the outcome data (true positive(TP), false positive(FP), false negative(FN), true negative(TN)); (2) the gold standard for diagnosis of sepsis was de ned in the study; (3) prospective or retrospective study design; (4) only adult patients were involved. The literature was excluded if they are (1) reviews, case reports, editorials and animal experiments; (2) only healthy people were used as controls.

Data extraction
Two authors (LY, RJB) independently collected data referring to study and patient characteristics. A third author (ZZC) independently assessed these data in case of inter-reviewer discrepancies. The extracted information from each study included rst author name, year of publication, country, study design, sample size, control patients, cutoff value, prevalence, male, mean age, timing of eosinophils counts assessment, reference standard and outcome data (TP, FP, FN, TN). For studies providing multiple eosinopenia cutoff values, the outcome data of all cutoff values were extracted.

Quality assessment
The methodological quality of the included studies was tested with the QUADAS-2 tool by two authors. QUADAS-2 consists of four sections: patient selection, index text, reference standard, and ow and timing [20]. The included studies were ranked as low risk, high risk, or unclear risk. We performed the quality assessment using Review Manager (RevMan), version 5.3 (Cochrane Collaboration)

Statistical analysis
The pooled sensitivity, speci city, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), AUC, and corresponding 95% credible interval (CI) were calculated by TP, FP, FN, TN using a bivariate regression model using STATA 14.0 software. Deek's funnel plot was used to detect publication bias, with P < 0.05 indicating publication bias. A Fagan plot was assembled for the visual presentation of diagnostic performance.
We used Spearman correlation coe cient to detect threshold effects by using Meta-DiSc software (version 1.4), and P value < 0.05 indicates a signi cant threshold effect. We used I 2 to describe the heterogeneity in the meta-analysis. I 2 value ≥ 50% is considered as a signi cant heterogeneity. Then, the possible sources of heterogeneity were explored by conducting a meta-regression analysis and sensitivity analysis. The examined parameters in meta-regression included cutoff (≥ 50/<50), control patient (SIRS/non-infection), sample size (≥ 200/<200), type of study (prospective/retrospective), country (Europe and America/others). A sensitivity analysis was conducted to examine the in uence of each study on the meta-analysis, by calculating the pooled outcomes after omitting one study at a time. The impact of eosinopenia cutoff used in each study was further investigated by performing a subgroup analysis.

Result
Included studies 7 studies that represent 12 trials were included in this meta-analysis (using different cutoff values in the same studies was regarded as different trials), with a total number of 3842 subjects. Among which, 1152 were diagnosed as sepsis, while the rest of 2690 were considered as non-sepsis. A summary of the characteristics ( rst author name, year of publication, country, study design, sample size, control patients, cutoff value, prevalence, male, mean age, timing of Eosinophils counts assessment) of the 7 included studies is outlined in Table 1. The data used for the construction of the 2 × 2 table was presented in Suppl. Table 1, along with the cutoff values of each study. A ow diagram was assembled to describe the details of the study selection process (Fig. 1)

Quality assessment
We used the QUADAS-2 tool to evaluate the quality of included studies, and results are illustrated in Fig. 2 and summarized in Suppl. Figure 1. High risk of bias was indicated in the section of index test, as 5 studies did not use pre-de ned cutoff values to calculate sensitivity and speci city, but rather used the optimal ones, and 1 study did not mention whether it used a pre-de ned value. When considering applicability concerns, 3 studies showed potential problems in patient selection.
We used I 2 to measure the heterogeneity in our meta-analysis. For the pooled sensitivity, speci city, PLR, NLR, DOR, the I 2 values were 94.32%, 97.32%, 91.75%, 92.12%, 100%, which indicates a signi cant heterogeneity in the majority of analysis. To determine the sources of heterogeneity, we brought cutoff, control patient, sample size, study design, country into meta-regression analysis ( Fig. 5 and Suppl. Figure 6), and the results showed that the examined covariates could not account for the heterogeneity. We also conducted a sensitivity analysis to detect the in uence of each study on the pooled DOR (Fig. 6) and overall outcomes (Suppl. Table. 2). This Leave-one-out analysis indicated that no single study exerted signi cant in uence on the results of the meta-analysis.

Discussion
Sepsis is common and life-threatening in intensive care unit, and early diagnosis and early treatment are essential for the favorable outcome of patients with sepsis. Multiple biomarkers, especially PCT and CRP, have been widely studied for early diagnosis of sepsis in past decade, showing only moderate diagnostic performance [4]. Many other novel biomarkers, such as neutrophil CD64, presepsin, IL-27, cfDNA and miRNAs, also have been intensely examined, showing no de nite and satisfactory diagnostic e cacy individually, while combination may provide better sensitivity and speci city [21].
Eosinopenia has been identi ed as a sign of acute infection for a long time [6][7][8] and its correlation with sepsis was rstly reported by Abidi et.al in 2008 [22].
Eosinopenia is a response to acute in ammation, displaying a decreasing number of eosinophils in circulation, an accumulation of eosinophils at the in ammatory sites, and an inhibition of eosinophils production in bone marrow [23]. Eosinophils are increasingly recognized to play a critical role in modulating local and systemic immune and in ammatory responses [5]. Given that sepsis is accompanied by dysregulated immune responses and in ammatory cascades, it is reasonable to regard eosinopenia as an indication of pathophysiological status during sepsis. In the past decade, eosinopenia has aroused much attention in its diagnostic and prognostic values for bacterial infection and sepsis, with some studies showing promising results. [10-15, 17, 18, 24, 25] However, the results of currently available studies on the diagnostic performance of eosinopenia are quite different, which is manifested by a range of sensitivity from 23.2-92.5% and a range of speci city from 28.57-91% [9][10][11][12][13][14][15]. Therefore, we conducted a meta-analysis to integrate all results from different studies in order to get the pooled outcomes The present meta-analysis covers studies from 2008 to 2016 on the value of eosinopenia for diagnosis of sepsis in adults, which has not been performed previously. Our primary nding is that eosinopenia is not a good biomarker for the diagnosis of sepsis, as shown the area under the SROC curve was 0.73, the DOR was 4.23, and the pooled sensitivity and speci city was 0.66 and 0.68 respectively. Eosinopenia is inferior to the two most widely used biomarkers, PCT and CRP, when being used to diagnose sepsis. According to a latest meta-analysis of the use of PCT and CRP for the diagnosis of sepsis, PCT exhibited an AUROC of 0.85, a DOR of 12.5, a sensitivity of 0.80 and a speci city of 0.77, and CRP exhibited an AUROC of 0.73, a DOR of 6.89, a sensitivity of 0.80 and a speci city of 0.61 [26]. Similar result has been found by Jose Garnacho-Montero and his colleagues when compared the diagnostic value of eosinopenia to CRP and PCT in critically ill patients with suspicion of sepsis [11].
However, after a further examination of the literature used for this meta-analysis, we found that there is signi cant statistical heterogeneity, which has a potential causal relationship with country, sample size, study designs, control setting, and cut-off values etc. Firstly, the 7 included studies were conducted in 7 different countries (suppl. table 2) from Asia, Africa, Europe, and America, where medical levels and ethnic characteristics vary greatly. Secondly the sample size ranges from 68 to 692 as we know that inappropriate sample size may contribute to an unreliable result. However, according to the results of the sensitivity analysis, no individual study that was removed could signi cantly change the pooled DOR estimated by the remaining studies. Thirdly, we included all observation studies (prospective and retrospective), and it is universally acknowledged that prospective studies have a higher credibility than retrospective studies. Fourthly, some included studies assessed only SIRS patients as controls while the other studies assessed SIRS patients and infected patients without SIRS as controls or just use non-infectious patients as controls (without determining whether they have SIRS). It is more di cult to differentiate sepsis from SIRS in clinical practice because of the similar clinical signs. Fifthly, the cutoff values of eosinopenia among these included trials range from 10 to 100 (cells/mm 3 ), which may also be the sources of heterogeneity, and then we performed a subgroup analysis. When using the cutoff value = 40 cells/mm 3 , eosinopenia showed the best diagnostic performance with a sensitivity of 0.79 and a speci city of 0.75, although the insu cient number of included trials in some subgroups is a big limitation of the subgroup analysis. However, no individual factors could account for the heterogeneity except the parameter of country can interpret the heterogeneity of speci city alone. Most likely there are undiscovered factors contribute to the heterogeneity. Furthermore, according to this meta-regression analysis, study design exerted no observable in uence on pooled sensitivity and speci city, while selecting SIRS patients as controls exactly exhibited lower sensitivity and speci city, although there is no statistical signi cance.
To further explore the potential value of eosinopenia for diagnosis of sepsis in clinical practice, we constructed a Fagan nomogram, which showed that testing eosinopenia could help increase the post-probability to 47% and reduce the post-probability to 17%, with setting a pre-test probability of 30%. In summary, our results did not support eosinopenia as a clinically useful tool for diagnosis of sepsis.

Limitation of the study
Firstly, the number of included studies is small, in spite of the extensive search of literature. Secondly, a signi cant heterogeneity exists in our analysis, and we nd no single factor to account for it. Thirdly, these included studies span from 2008 to 2016, and was conducted in 7 different countries across 4 continents, which involves complicated factors such as healthcare system, detection sensitivity, antibiotic utilization, and ethnic characteristics.

Implications for future researches
Because of the internal heterogeneity and small number of including studies, our results show only weak association between eosinopenia and the pathophysiological status of sepsis. Future researches are supposed to focus on the following two aspects: (1) whether eosinopenia could contribute to the diagnosis of sepsis caused by some speci c pathogens; (2) whether eosinopenia show a signi cant advantage in the era of sepsis-3.0, given that the studies included in our meta-analysis were almost conducted before the emergence of sepsis-3.0 criteria [1].
Collectively, our results reveal little possibility of eosinopenia as a good biomarker for sepsis due to its low sensitivity and speci city. This condition could be due to the insu cient number of included studies, thus more comprehensive clinical studies should be included to make a de nite conclusion. Moreover, future studies can focus on the combination of several biomarkers instead of one single biomarker to diagnose sepsis.

Consent for publication
Not applicable. The manuscript does not contain any personal data in any form.

Availability of data and materials
All data generated during this study are included in this published article and its supplementary information les.

Competing interests
The authors declare that they have no competing interests.

Funding
This research received no speci c grant from any funding agency.
Authors' contributions LY had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. LY and RJB contributed to the study design, data collection, interpretation, and statistical analysis. LY contributed to the manuscript preparation. ZZC contributed to the study supervision and manuscript revision. All authors read and approved the manuscript.     Summary receiver operator characteristic plots with 95% CI of sensitivity against speci city of eosinopenia for diagnosing sepsis.