Eosinophil Recovery in Hospitalized COVID-19 Patients is Associated with Lower Rates of ICU Admission and In-Hospital Mortality: An Observational Cohort Analysis

Background Admission eosinopenia (<100 cells/μL) is associated with poor clinical outcomes in hospitalized COVID-19 patients. However, the effects of eosinophil recovery (defined as reaching ≥50 eosinophils/μL) during hospitalization on COVID-19 outcomes have been inconsistent. Methods The study included 1,831 patients admitted to UCLA hospitals between February 2020 and February 2021 with PCR-confirmed COVID-19. Using competing risk regression and modeling eosinophil recovery as a time-dependent covariate, we evaluated the longitudinal relationship between eosinophil recovery and in-hospital outcomes including ICU admission, need for mechanical ventilation, and in-hospital mortality. All analyses were adjusted for covariates including age, BMI, tobacco smoke exposure, comorbidities known to be risk factors for COVID-19 mortality, and treatments including dexamethasone and remdesivir. Results Eosinophil recovery was evaluated in patients with <50 eosinophils/μL on admission (n=1282). These patients cumulatively amassed 11,633 hospital patient-days; 3,985 of those days qualified as eosinophil recovery events, which were represented by 781 patients achieving at least one instance of eosinophil recovery during hospitalization. Despite no significant difference in the rate of mechanical ventilation, eosinophil recoverers had significantly lower rates of in-hospital mortality (aHR: 0.44 [0.29, 0.65], P=0.001) and ICU admission (aHR: 0.25 [0.11, 0.61], P=0.002). Conclusion Trending eosinophil counts during hospitalization is simple and can be performed in resource-limited healthcare settings to track the inflammatory status of a patient. Lack of eosinophil recovery events can identify those at risk for future progression to severe COVID.

While multiple risk factors for severe COVID-19 have been identified, the search continues for additional biomarkers and better prognostic tools to better identify patients likely to become critically ill with COVID-19.
Peripheral blood eosinophil counts have garnered the interest of researchers as a prognostic biomarker for COVID-19 since eosinopenia (≤50 cells/μL) appears to be a unique feature of COVID-19 12 . A study in Italy estimated approximately 75% of patients presenting with SARS-CoV-2 infection had undetectable levels of eosinophils (absolute eosinopenia) on hospital admission, while similar findings were observed in China during the outset of the pandemic 13,14 .These persistent findings inspired some researchers to use admission eosinophil counts as a relatively specific diagnostic marker for SARS-CoV-2 infection when PCR kits were in short supply 15 . While no unifying rationale explains these observations, multiple studies have demonstrated an inverse relationship between admission blood eosinophil counts and COVID-19 severity 16,17 .
Less is known about tracking changes in peripheral blood eosinophil counts throughout disease progression as a predictor for clinical outcomes in COVID-19 patients. Isolated reports in Spain have suggested eosinophil recovery during hospitalization to be associated with improved survival 18 . Although little is known about the relationship between eosinophils and COVID-19 progression, laboratory studies have postulated that eosinophils may play a role in viral mitigation through direct, cell-mediated manners or antigen presentation [19][20][21][22][23] . Therefore, it stands to reason that changes in blood eosinophil count throughout hospitalization may reflect disease evolution and can serve as an easily attainable biomarker capable of predicting and tracking disease severity in hospitalized COVID-19 patients.
Our aim was to evaluate the longitudinal relationship between eosinophil recovery (defined as reaching 50 cells/μL) and time-dependent in-hospital outcomes including intensive care unit (ICU) admission, mechanical ventilation, and in-hospital mortality. We hypothesized that eosinophil recovery in COVID-19 patients presenting with eosinopenia (<50 cells/μL) would be independently associated with positive clinical outcomes such as overall survival, less need for invasive ventilation, and decreased ICU admission for patients with COVID-19.

Patients and Study Design
We performed a retrospective cohort analysis of 1,831 patients admitted to both UCLA

Measurements
Prior comorbidities, patient demographics, tobacco exposure history, and lab values were available for analysis along with full, detailed hospital course (Table 1). Comorbidities were determined through documented International Classification of Diseases (ICD)-10 diagnosis per chart review (Supplemental Table 2). Routine laboratory results (i.e. metabolic panels, complete blood counts, etc.) were considered -admission‖ values if they were obtained during initial patient intake. Non-routine laboratory values (i.e. IL-6, procalcitonin, C-reactive protein, Ddimer, and other inflammatory markers) were considered admission values if they were collected during intake or within 48 hours of admission, whichever came first. We categorized the 1,651 patients with available blood eosinophil data into three groups based on their admission eosinophil counts: absolute eosinopenia (0 cells/μL), marked eosinopenia (<50 cells/μL but >0 cells/ μL), and non-eosinopenia ( 50 cells/μL). Cutoffs for each group were chosen to maximize equal distribution of patients into each group; absolute eosinopenia was an exception given its predominance in our cohort (n=910; 55.1% of cohort). Eosinophil recovery, defined as reaching an eosinophil count 50 cells/μL, was assessed daily for each patient during hospitalization. We used 50 cells/μL as our cutoff for eosinophil recovery because it reflected our previous choice for delineating patients with marked eosinopenia from -non-eosinopenic‖ patients. Each patient was also assigned a score based on the NIH COVID-19 Severity Scale (Supplemental Table 1), an eight-category ordinal scale used by the Adaptive COVID-19 Treatment Trial to classify disease severity from 1 (not hospitalized, no limitations) to 8 (in-hospital death) at time of outcome 24 .
Main outcomes included ICU admission, need for mechanical ventilation, and in-hospital mortality. Other metrics for COVID severity included length of hospitalization, NIH Severity Scores (Supplemental Table 1), and need for COVID-specific treatments such as remdesivir and dexamethasone. Mechanical ventilation was defined by invasive ventilation requiring endotracheal intubation, whereas non-invasive mechanical ventilation was defined by the use of high-flow nasal-cannula systems or positive airway pressure (PAP) ventilation with bi-level or continuous PAP. Criteria for ICU admission was multifaceted and included, but not limited to, need for ventilation, severe myocardia depression, neurological monitoring, and acute renal support.

Statistical Analysis
Summary statistics stratified by baseline eosinophil status (non-eosinopenia, marked eosinopenia, and absolute eosinopenia) were tabulated for key variables of interest with categorical variables analyzed by Chi-Squared test and continuous variables by Kruskal-Wallis or Mood's median tests. We utilized Fine and Gray competing risk regression to evaluate the relationship between baseline eosinophil status and time-dependent in-hospital outcomes (ICU admission, intubation, and in-hospital mortality) with adjustments for potential confounders, including age, sex, tobacco exposure, history of diabetes with end-organ damage, chronic pulmonary conditions, coronary artery disease, liver disease, chronic kidney disease, and COVID-specific treatments remdesivir and dexamethasone (hereafter referred to as -covariates‖). Competing risk models for ICU admission and mechanical ventilation were performed to account for competing events, including in-hospital death and hospital discharge.
Meanwhile, competing risk analysis for in-hospital death was adjusted for hospital discharge. We also evaluated the relationship between eosinophil recovery and in-hospital outcomes among patients with low eosinophils at baseline (defined by <50 cells/μL) using a cause-specific competing risk model that modeled eosinophil recovery as a time-varying covariate. Noneosinopenic patients were excluded from this analysis to highlight the effect of eosinophil recovery in patients with marked or absolute eosinopenia.
A non-parametric linear regression model with bootstrapping was used to evaluate the relationship between baseline eosinophil status and mean NIH Severity Score. Logistic models were used to evaluate the associations between baseline eosinophil status and in-hospital treatments, including dexamethasone, remdesivir, systemic antibiotics, and oxygen therapy. To control for potential confounding and moderating effects associated with prior immunosuppression, a sub-analysis was performed to compare outcomes in immunosuppressed transplant recipients (n=181) stratified by eosinophil status. This comparison was adjusted using propensity-score matching due to small sample size. Additional sub-analysis for eosinophil recovery was also performed on hospitalized patients treated with any systemic corticosteroid (n=1011) and adjusted accordingly with competing risk regression given sufficient sample size.
All models were fit using complete case analysis. We used a threshold of 15% missingness on key variables to trigger multiple imputation, which was not met for any of our analyses. Patients who were transferred to another hospital were treated as censored observations in competing risk models, accounting for 252 patients (15.3% of cohort).

Baseline characteristics of the study population
Of the 1,651 hospitalized COVID-19 patients in our cohort with eosinophil data, 910 eosinopenia when compared to the other eosinophil groups, the difference was likely not clinically significant. Current tobacco use was more prevalent among patients with absolute eosinopenia (7.9% vs 3.5% and 3.3%, P<0.001), while previous history of tobacco use was similar in all three groups (P=0.242). Patients in all three groups also had similar rates of comorbidities apart from congestive heart failure, which was found to be less prevalent among patients with absolute eosinopenia (P=0.01).
Baseline inflammatory state was also evaluated. In Table 1, non-specific markers for inflammation, including C-reactive protein, ferritin, IL-6, and procalcitonin, were increased in patients with absolute eosinopenia. Neutrophil-lymphocyte ratio (NLR) was also significantly higher in the absolute eosinopenia group (P<0.001). However, this was driven predominantly by lymphocyte depletion rather than neutrophil expansion. Table 2, describes associations of eosinophil statuses with clinical outcomes. Rates of ICU admission, mechanical ventilation, and in-hospital mortality were higher in patients admitted with absolute eosinopenia before adjusting for any covariates. Other metrics for COVID severity, including length of hospitalization and need for COVID-specific treatments, were also increased in patients with absolute eosinopenia. After adjusting for pre-specified covariates and competing risks in Table 3

Eosinophil recovery as a predictor for specific outcomes
Finally in Table 4, we pooled patients with admission eosinophil counts <50 cells/μL (n=1,282) and found that they cumulatively amassed 11,633 hospital days; 3985 of those days  Table 4).

Discussion
In this retrospective analysis of adults hospitalized with COVID-19 at a large academic institution between Feb 2020 and Feb 2021, recovery from admission eosinopenia was associated with improved overall survival and decreased need for ICU admission independent of age, major comorbidities, and in-hospital corticosteroid use. In addition, absolute eosinopenia upon admission was found to be an independent risk factor for ICU admission, mechanical ventilation, and in-hospital death. Our findings suggest that eosinophils are associated with risk reduction in developing severe COVID-19.
To our knowledge, this is the first study to evaluate the significance of peripheral blood eosinophil count recovery in SARS-CoV-2 infected hospitalized patients using a model that encapsulates dynamic eosinophil changes throughout hospitalization. Previous retrospective analysis performed on COVID patients admitted across 147 Spanish hospitals also assessed the association of eosinophil recovery with live discharge 18 . The latter study, however, assessed eosinophil recovery seven days after hospital admission. A limitation of measuring eosinophil recovery within a fixed timeframe is data skewing that may result from placing more weight on the initial days of hospitalization, when in fact many patients may only begin to decompensate at a later point during their hospital stay 25 . Indeed, stratification of our own cohort based on NIH Severity Scores showed duration of hospitalization to be directly correlated with COVID disease severity (Supplemental Fig. 1). Using fixed time intervals to assess eosinophil recovery as a predictor for positive outcomes also fails to consider medication administration that can alter peripheral blood immune cell counts. As an example, corticosteroids cause fluctuations in leukocyte count by inhibiting neutrophil margination and extravasation 26,27 . We attempted to correct for this potential confounder by examining eosinophil recovery in a subset of patients treated with systemic corticosteroids during hospitalization. Hazard ratios for in-hospital mortality, ICU admission, and mechanical ventilation in this sub-analysis were unchanged from our results in our initial analysis ( Table 4), suggesting that the relationship between eosinophil recovery and improved clinical outcomes was irrespective of corticosteroid treatment. Although prospective studies are needed, our initial findings support the use of eosinophil recovery as a favorable prognostic factor for hospitalized COVID-19 patients. With eosinophil count evaluation being a simple, inexpensive, and routinely available lab test, trending eosinophils can help clinicians identify COVID patients at risk for deterioration. This is particularly helpful in resource-limited settings, where identification of at-risk patients is crucial for proper allocation of beds, mechanical ventilators, and healthcare staff 35 .
Our study has limitations. Assessing outcomes other than mortality, ICU admission, and mechanical ventilation was beyond the scope of this retrospective analysis. Finer details on the time between symptom onset and admission eosinophil counts was also unreliable given that data collection started only upon hospital admission, and reported symptoms were not universally documented on initial patient encounter. The nature of retrospective analysis lends itself to inherent biases and unmeasured confounding that could influence our overall findings.
Prior comorbidities were also based on ICD-10 coding with its inherent flaws of underreporting diagnoses 36 . But given the scope and length of our study, missing data from underreporting was likely random and unable to cause any significant confounding. Medication history prior to hospitalization was also often unknown, limiting our ability to directly evaluate drug-induced eosinopenia on clinical outcomes. Our attempt to control for this potential confounder by assessing eosinophil recovery in immunosuppressed transplant recipients was also likely suboptimal given the relatively small sample size of transplant recipients. Even so, eosinophil recovery among steroid-treated patients during hospitalization continued to demonstrate protective effects against SARS-CoV-2 infection. However, corticosteroid data lacked dosage and schedule of steroid administration for each patient, thus limiting our ability to evaluate the relationship between corticosteroid use and eosinophil counts. Additionally, the relationship of blood eosinophil recovery with the recovery of other cell lines and overall relevance of the role of Th-2 immune pathway in the clinical recovery from COVID-19 is unclear. Trending specific immunologic and inflammatory markers such as IL1, IL12, IL6, IL5, IL17 or TNF-alpha could offer a better insight into the mechanisms of cellular immunity recovery in COVID-19, nevertheless, none of these specific biomarkers has been routinely monitored and such data is missing here. Lastly, our pre-selected cutoffs for eosinophil recovery and our differing eosinophil groups were based off observation at our institution rather than clinical guidelines.
We did not utilize any cutoffs from prior studies because there did not seem to be a universally accepted cutoff for eosinopenia in the context of SARS-CoV-2 infection.
Nevertheless, several strengths should be emphasized. This report is based on data from a large and well-characterized database from a large quaternary academic center from the early phase of the COVID-19 pandemic. Medical and supportive management were mandated to follow current CDC guidelines, and our medical centers did not have any major shortages or system limitations that could impact outcomes. In addition to comparing outcomes of patients with or without eosinophil recovery, we controlled for in-hospital corticosteroid use and correlated positive outcomes with eosinophil recovery beyond the first few days of hospitalization. This study, therefore, expands upon previous studies and further supports using eosinophil recovery as a readily available marker for improving disease that can be measured at all stages of hospitalization.
In conclusion, eosinophil recovery is associated with improved survival and decreased ICU admission rate in the setting of COVID-19. Peripheral blood eosinophil count, therefore, is a clinically relevant biomarker whose dynamic behavior may predict COVID-19 disease course and assist with disease risk stratification. Further studies are needed to determine whether eosinophils themselves should be potential targets of interest for future therapeutic approaches in COVID-19 as it is unclear whether eosinophils play an active role in viral mitigation or whether they are byproducts of underlying biological mechanisms that have yet to be elucidated.

Author contributions
DM, IZB, JAF, and PDY had full access to the data and take responsibility for the integrity of the data and accuracy of the analysis. PDY, DM, RGB, DPT, and IZB were involved in the design of the analysis. PDY, DM, RGB, and IZB contributed to the drafting of the manuscript; and DPT, RHY, and RSR were involved in the editing of the manuscript. All authors approved the manuscript for submission.

Role of the sponsors
Industry sponsors had no role in the design of the study, the collection and analysis of the data, or the preparation of the manuscript.

Other contributions
The authors thank the UCLA participants and participating physicians, investigators, and staff for making this research possible.

Ethics Approvals
This study was reviewed and approved by the UCLA Institutional Review Board .

Disclosures
Drs. Buhr and Tashkin report personal consulting fees from Theravance Biopharma/Viatris, unrelated to this work. Drs. Buhr, Fulcher, and Salehi-rad are employed by the Veterans Health Administration. The views and positions in this manuscript do not necessarily represent those of the United States Government.
Drs. Hixson, LeMaster, and Shover, Ms. Markovic, and Mr. Yan report nothing to disclose. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper