Eosinopenia in COVID-19 Patients: A Retrospective Analysis

Objectives: The aim of this study was to assess the diagnostic role of eosinophils count in COVID-19 patients. Methods: Retrospective analysis of patients admitted to our hospital with suspicion of COVID-19. Demographic, clinical and laboratory data were collected on admission. Eosinopenia was defined as eosinophils < 100 cells/mm3. The outcomes of this study were the association between eosinophils count on admission and positive real-time reverse transcription polymerase chain reaction (rRT-PCR) test and with suggestive chest computerized tomography (CT) of COVID-19 pneumonia. Results: A total of 174 patients was studied. Of those, 54% had positive rRT-PCR for SARS-CoV-2. A chest CT-scan was performed in 145 patients; 71% showed suggestive findings of COVID-19. Eosinophils on admission had a high predictive accuracy for positive rRT-PCR and suggestive chest CT-scan (area under the receiver operating characteristic—ROC curve, 0.84 (95% CIs 0.78–0.90) and 0.84 (95% CIs 0.77–0.91), respectively). Eosinopenia and high LDH were independent predictors of positive rRT-PCR, whereas eosinopenia, high body mass index and hypertension were predictors for suggestive CT-scan findings. Conclusions: Eosinopenia on admission could predict positive rRT-PCR test or suggestive chest CT-scan for COVID-19. This laboratory finding could help to identify patients at high-risk of COVID-19 in the setting where gold standard diagnostic methods are not available.


Introduction
At the end of December 2019, the first case of COVID-19 (Coronavirus disease 2019) caused by a novel human coronavirus, called SARS-CoV-2, was reported in Wuhan, China [1]. This virus is an encapsulated β-coronavirus, which contains a single-stranded RNA as a nucleic material, with a size ranging from 26 to 32 Kbs in length [2]. This virus is most closely related to SARS-CoV, another acute-lung-injury causing coronavirus of zoonotic origin and shares a large proportion of its genome also with the Middle-East Respiratory Syndrome (MERS)-associated virus [3]. COVID-19 showed a

Study Population
We enrolled patients admitted to our hospital from March 10 (i.e., first admitted patient) to 31 March 2020 with the following inclusion criteria: (a) age of 18 years or older; (b) tested using rRT-PCR assay on the nasopharyngeal swab and/or BAL specimens for suspicion of COVID-19; (c) available laboratory exams on the admission to the hospital. Exclusion criteria were patients already admitted in our institution for other reasons (i.e., intrahospital acquisition of  and patients transferred from other hospitals.

Data Collection
We collected demographics, comorbidities, patients' characteristics on admission as well as results of rRT-PCR and chest thin-section CT-scan. Chest CT evaluation was performed by one radiologist (PAG), who reported whether the CT-scan was suggestive or not for COVID-19 (i.e., ground-glass opacities, consolidation or crazy-paving patterns) [11,13]. For clinical assessment on admission, we computed a simplified score based on admission symptoms; one point was attributed to each clinical feature, so that it could range from 0 (i.e., no symptoms) to 5 (i.e., all symptoms were present). We considered the following clinical features: (1) dyspnea; (2) flu-like symptoms (i.e., at least one among fever cough, cold, sore throat, myalgia and fatigue); (3) gastrointestinal symptoms (i.e., at least one among vomiting, diarrhea, dysgeusia and nausea) or anosmia; (4) oxygen saturation <94% breathing room air; (5) respiratory rate > 30 breaths per minute. Results of the blood sample tests on admission, including complete blood count (CBC), coagulation and inflammatory markers (i.e., Fibrinogen; D-Dimers; prothrombin time, PT; C-Reactive Protein, CRP; Procalcitonin, PCT; Lactic Dehydrogenase, LDH; Ferritin) as well as microbiological samples, whenever present, were collected. Eosinopenia was defined as eosinophils <100 cells/mm 3 .

Outcome Assessment
The primary endpoint of this study was to assess the association of eosinophils count on admission and a positive rRT-PCR test. Secondary outcomes included the association of eosinophils count on admission and a suggestive chest CT-scan and the identification of predictors of positive rRT-PCR test and suggestive CT-scan. In addition, we assessed whether eosinopenia was able to predict disease severity on admission, according to WHO criteria (i.e., Mild/Moderate vs. Severe/Critical) [14].

Statistical Analysis
Descriptive statistics were computed for all study variables. A Kolmogorov-Smirnov test was used, and histograms and normal-quantile plots were examined to verify the normality of distribution of continuous variables. Discrete variables were expressed as count (percentage) and continuous variables as mean ± SD or median (25th-75th percentiles), as appropriate. Demographics and clinical differences between groups (rRT-PCR positive vs. negative test; suggestive vs. nonsuggestive chest CT-scan for COVID-19) were assessed using a chi-square test or Fisher's exact test for categorical variable, and Student's t-test or Mann-Whitney U-test for continuous variable, as appropriate. The discriminative ability of eosinophils to predict a positive rRT-PCR or a suggestive CT-scan was evaluated using receiver operating characteristic (ROC) curves with the corresponding area under the curve (AUROC). Youden's index was computed to assess the optimal cut-off of eosinophils for sensitivity and specificity to predict positive rRT-PCR or suspected chest CT-scan findings. Multivariable logistic regression analysis with positive rRT-PCR as the dependent variable was performed in all patients. Collinearity between variables (i.e., an index of collinearity higher than 0.3) was excluded prior to modelling; only variables associated with positive rRT-PCR in the univariate analysis (p < 0.1) were included in the multivariate model. Odds ratios (OR) with 95% confidence intervals (CI) were computed. A similar approach was used to perform the multivariate analysis with suspected CT-scan as the dependent variable. A p < 0.05 was considered as statistically significant. Statistical analyses were performed using the SPSS 25.0 for Macintosh (IBM, Armonk, NY, USA).

Availability of Data and Materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Study Population
Of the 264 patients fulfilling the inclusion criteria, 90 patients were excluded as already hospitalized or transferred from other hospitals ( Figure 1); as such, a total of 174 patients were included in the final analysis. Median age of the study population was 58 [43-71] years, with 52% of men and a body mass index (BMI) of 27.5 [23.5-31.6]; the most common reported comorbid disease was arterial hypertension. The median value of the clinical score on admission was 2 [2,3], and the most frequent symptoms were cough (62%), fever (55%) and dyspnea (52%). Among all the tested patients, 94 (54%) had positive rRT-PCR test for SARS-CoV-2 assay on nasopharyngeal swab and/or BAL. The main diagnosis for patients with negative rRT-PCR test are reported in the Table S1. Five patients with an initial negative rRT-PCR test showed positivity to a further test performed at 72 h after admission. Chest CT-scan was performed in 145 (83%) patients; 103/145 (71%) of them showed suggestive findings of COVID-19 related pneumonia.

Eosinophils and RT-PCR
Eosinophils count on admission was 10 [0-30]/mm 3 and 134/174 (77%) patients presented eosinopenia. Eosinophils were significantly lower in patients with positive rRT-PCR than others (0 [0-10] vs. 80 ; p < 0.01, Table 1); also, eosinopenia was more frequently observed in patients with positive rRT-PCR than others (89/94 (95%) vs. 45/80, 56%; p < 0.01). Patients with anosmia also had more frequently eosinopenia than others, although this was not statistically significant (Table S3).  Patients with positive rRT-PCR had a higher BMI, higher values of CRP, PT and LDH and lower white blood cells count, albumin levels and lymphocyte to CRP ratio than others (Table 1). In particular, all subtypes of white blood cells were significantly lower in patients with a positive rRT-PCR when compared to those with negative tests. Predictive accuracy of white blood cells subtypes to predict a positive rRT-PCT is shown in Table S2; eosinophils count on admission had the highest predictive value (AUROC 0.84 (95% CIs 0.78-0.90); p < 0.01- Figure 2), for positive rRT-PCR among all of them. The Youden's index identified the threshold of eosinophils of 45/mm 3 for the best combination of sensitivity (94%) and specificity (63%) to predict positive rRT-PCR. The highest positive predictive value (PPV; i.e., 86%) was observed for an eosinophils count < 5/mm 3 , which was associated with a specificity of 86%, a sensitivity of 59% and a negative predictive value (NPV) of 64%. The highest NPV (i.e., 88%) was observed for an eosinophils count > 100/mm 3 , which was associated with a specificity of 44%, a sensitivity of 95% and a PPV of 66%. In the multivariate analysis, eosinopenia and elevated LDH were independent predictors of positive rRT-PCR (OR 13.42 (95% CIs 4.11-43.77) p < 0.01 and OR 1.01 95% [CIs 1.00-1.01] p = 0.02, respectively) ( Table 3). The eosinophils count on admission was similar between patients with mild/moderate and severe/critical COVID-19 (10 [0-40] vs. 10 [0-40]/mm 3 ; p = 0.94).
Patients with a suggestive chest CT-scan were more frequently male, had hypertension and a higher BMI and were more frequently admitted to the ICU. In addition, suggestive chest CT-scan was more frequently associated with a positive rRT-PCR (83/103, 81%) vs 7/42, 17%), p < 0.01; OR 16.46 (95% CIs 6.49-41.77), p < 0.01) and higher levels of CRP, LDH, PT, fibrinogen and LCR, as well as lower albumin levels. As for RT-PCR, all subtypes of white blood cells were significantly lower in patients with a suggestive chest CT-scan when compared to others. Predictive accuracy of white blood cells subtypes to predict a suggestive chest CT-scan is shown in Table S4; eosinophils count on admission had the highest predictive value (AUROC 0.84 (95% CIs 0.77-0.91); p < 0.01, Figure 3) to predict a suggestive chest CT-scan among all of them. The Youden's index identified the threshold of eosinophils of 25/mm 3 for the best combination of sensitivity (82%) and specificity (74%) to predict a suggestive chest CT-scan. The highest positive predictive value (PPV; i.e., 95%) was observed for an eosinophils count <5/mm 3 , which was associated with a specificity of 93%, a sensitivity of 55% and a NPV of 46%. The highest NPV (i.e., 68%) was observed for an eosinophils count >100/mm 3 , which was associated with a specificity of 45%, a sensitivity of 91% and a PPV of 80%.

Discussion
In this study, eosinopenia on admission was a common finding in patients with a positive rRT-PCR as well as those with a suggestive chest CT-scan for COVID-19. Eosinophils count had a high predictive accuracy value and was independently associated with both positive rRT-PCR and suggestive chest CT-scan. As such, the combination of eosinophils count on admission and other clinically relevant variables may help physicians to identify patients at risk of COVID-19 if laboratory or imaging resources are lacking.
Eosinophils are circulating and tissue-resident leukocytes, which show proinflammatory, immunoregulatory and antiviral effects [15]. Eosinopenia has been reported in different medical conditions such as bacterial infections, acute inflammation, stress or corticosteroid therapy [16][17][18]. In the setting of COVID-19, eosinopenia may be related to the migration of these cells into the peripheral tissues or to a decreased production of eosinophils into the bone marrow, due to the inflammatory state [16][17][18][19]. Lymphopenia could be another potential explanation for eosinopenia. A large number of patients with COVID-19 present a decreased lymphocytes count, and Th2-subtype lymphocytes are responsible for the production of IL-5, which stimulates the activation and the production of eosinophils [20,21]. In addition, as the eosinophils count can be significantly influenced by the endogenous secretion of glucocorticoids [21], one can argue that increased cortisol levels observed during COVID-19 may result in temporary eosinopenia. Finally, activation of the Fas pathways can induce eosinophil apoptosis (9), which was enhanced by the presence of the Th-1 cytokines, such as IFN-γ and/or TNF-α [22]. We did not specifically assess the causes of eosinopenia and further studies are necessary to investigate these hypotheses in COVID-19 patients.
Recently, Zhang et al. reported that half of the 140 COVID-19 patients admitted to the hospital had eosinopenia (i.e., < 20/mm 3 ) on admission 13 ; similar results were also reported in another study [23]. In one study (n = 55) including confirmed COVID-19 patients, eosinophil count was associated with poor prognosis [24]. Du et al. reported that in 85 deceased patients with severe COVID-19, 81% of them had absolute eosinophil counts below 20/mm 3 [25]. In another study, eosinopenia on admission was frequent but eosinophil levels progressively improved over time in all patients prior to discharge, suggesting that resolution of eosinopenia may be an indicator of improving clinical status [26]. A lower eosinophil count on admission was also associated with a significantly longer SARS-CoV-2 positivity on repeated swabs [27]. In another study, Tanni et al. compared two cohort of patients with diagnosis of COVID-19 or Influenza infection and showed that eosinopenia on admission was a good predictor of COVID-19 [28]. Another study also suggested that eosinopenia has an important role to distinguish patients with SARS-CoV-2 infection from other causes of pneumonia [29].
Besides the association of eosinopenia with the presence of COVID-19 or the severity of the disease, one relevant question for the management of COVID-19 is the triage of patients, which has been particularly challenging because of the large number of suspected patients admitted to the hospital and the long waiting time for rRT-PCR results and chest CT-scan availability. Therefore, it is necessary to develop more effective and accessible methods to optimize the triage process of these COVID-19 patients. In one study, the AUROC of eosinophils on admission to predict positive rRT-PCR was 0.77 [95%CI (0.641-0.886); p< 0.001], which was lower than the neutrophils-to-lymphocytes ratio and the monocyte-to-lymphocytes ratio [30]; also, the lowest value of eosinophils was observed in severe COVID-19 patients requiring ICU admission. Nevertheless, the authors did not specifically select only patients with available laboratory tests on admission (i.e., also included transfer from another hospital or intrahospital acquisition), which might have influenced final results. In a recent study (n = 989) conducted in Wuhan, eosinopenia (i.e., defined as < 20/mm 3 ) was observed in 75% of COVID-19 patients and more frequently (i.e., 31%) than in control patients [31]; eosinopenia had a AUROC of 0.72 to predict positive rRT-PCR. In addition, the combination of eosinopenia and elevated CRP levels increased the predictive accuracy to 0.73. Despite this study not being entirely comparable to our cohort as hospital admission and triage strategy between China and Belgium could have substantially differed, these studies suggest that early eosinophils count assessment, together with other available clinical data, might be useful to identify patients with a possible diagnosis of COVID-19. Eosinophils count is available in routine blood tests worldwide and could facilitate a rapid triage for priority of radio-diagnostic and/or rRT-PCR tests.
Very few diagnostic scores are available to rapidly identify patients with a high risk of positive rRT-PCR testing during COVID-19 pandemic. In one study, history of close contact, fever, neutrophil-to-lymphocyte ratio, high temperature and gender were included into a "COVID-19 early warning score", which had a AUROC to diagnose a positive rRT-PCR of 0.96 [32]; however, chest CT-scan was also included in the final model. In a systematic review evaluating diagnostic models to detect COVID-19 infection in patients with symptoms, ten prognostic models including clinics, routine laboratory tests and other clinical information on admission were identified, with variable predictive accuracy and calibration [33]; their applicability in clinical practice remains therefore to be further evaluated. In addition, no reliable or useable clinical score currently exists to predict outcomes or inform decisions regarding hospital admission for COVID-19 patients.
This study has several limitations. First, this is a retrospective, single-centre study; as such, the results may suffer from selection biases. Moreover, we did not calculate the sample size to test the diagnostic accuracy of eosinopenia to predict a positive rRT-PCR test as, at the moment the study was initiated, the data on eosinopenia in COVID-19 were very limited and the prevalence of the positive rRT-PCR test in patients presenting in our institution was unpredictable. In particular, for a prevalence estimated around 50% of positive rRT-PCR testing, a minimum sample size of 365 patients would have been required to achieved a power of 80% in order to detect a change in the sensitivity from 0.80 to 0.90 and a change in the specificity from 0.6 to 0.7, with a significance level of 0.05. Future and multicentric studies are warranted to confirm our results using larger cohorts of patients. Second, some relevant laboratory variables might have been missed. Third, the sample size was not calculated and external validation not provided. Fourth, rRT-PCR could yield false negative results, mainly due to an insufficient viral load below the detection limit or inadequately collected nasopharyngeal swabs. Fifth, the clinical score was used locally and has not been validated in larger cohorts. Finally, the "pre-test" prevalence of COVID-19 in the analysed cohort could be a significant confounder to extrapolate the validity of our findings in clinical practice. Giving the high prevalence of COVID-19 among patients (i.e., the study was carried out during the epidemic phase), it is difficult to conclude that eosinopenia is always related to a positive RT-PCR test for SARS-Cov-2. Indeed, these data should be re-evaluated during a phase of lower COVID-19 prevalence, where other inflammatory diseases could also be associated to initial eosinopenia and contribute as false positive. Moreover, we evaluated only hospitalized patients and the conclusions of our study cannot be applied to mild COVID-19 forms.

Conclusions
In this study, eosinopenia was frequently observed in COVID-19 patients and could be helpful to identify patients with positive rRT-PCR testing or suspected chest CT-scan findings. This easily accessible parameter should be considered in the triage process to speed up the diagnosis and treatment of COVID-19 patients.