The influence of individual characteristics and non‐respiratory diseases on blood eosinophil count

Abstract Background Blood eosinophil (B‐Eos) count is an emerging biomarker in the management of respiratory disease but determinants of B‐Eos count besides respiratory disease are poorly described. Therefore, we aimed to evaluate the influence of non‐respiratory diseases on B‐Eos count, in comparison to the effect on two other biomarkers: fraction of exhaled nitric oxide (FeNO) and C‐reactive protein (CRP), and to identify individual characteristics associated with B‐Eos count in healthy controls. Methods Children/adolescents (<18 years) and adults with complete B‐Eos data from the US National Health and Nutritional Examination Surveys 2005–2016 were included, and they were divided into having respiratory diseases (n = 3333 and n = 7,894, respectively) or not having respiratory disease (n = 8944 and n = 15,010, respectively). After excluding any respiratory disease, the association between B‐Eos count, FeNO or CRP, and non‐respiratory diseases was analyzed in multivariate models and multicollinearity was tested. After excluding also non‐respiratory diseases independently associated with B‐Eos count (giving healthy controls; 8944 children/adolescents and 5667 adults), the independent association between individual characteristics and B‐Eos count was analyzed. Results In adults, metabolic syndrome, heart disease or stroke was independently associated with higher B‐Eos count (12%, 13%, and 15%, respectively), whereas no associations were found with FeNO or CRP. In healthy controls, male sex or being obese was associated with higher B‐Eos counts, both in children/adolescents (15% and 3% higher, respectively) and adults (14% and 19% higher, respectively) (p < 0.01 all). A significant influence of race/ethnicity was also noted, and current smokers had 17% higher B‐Eos count than never smokers (p < 0.001). Conclusions Non‐respiratory diseases influence B‐Eos count but not FeNO or CRP. Male sex, obesity, certain races/ethnicities, and current smoking are individual characteristics or exposures that are associated with higher B‐Eos counts. All these factors should be considered when using B‐Eos count in the management of respiratory disease.

factors should be considered when using B-Eos count in the management of respiratory disease.

K E Y W O R D S
airway inflammation, blood eosinophils, eosinophilic inflammation, respiratory diseases

| INTRODUCTION
Airway type-2 inflammation is a feature of common phenotypes of asthma 1-3 and chronic obstructive pulmonary disease (COPD). 4,5 The most studied biomarkers used to characterize patients with type-2 inflammation are blood eosinophil (B-Eos) count and fraction of exhaled nitric oxide (FeNO). 6 In asthma, elevated B-Eos count is associated with poor disease control, accelerated lung function decline, increased risk of severe exacerbations, and re-hospitalizations. [7][8][9][10][11] Also, it has been reported a dose-response effect between inhaled corticosteroids (ICS) and the reduction in B-Eos levels. 12 In COPD patients, higher B-Eos levels are associated with an improved response to ICS in preventing exacerbations. [13][14][15][16] During exacerbations, higher B-Eos levels predict greater response to oral corticosteroids, while lower levels are associated with worse outcomes. 17,18 Furthermore, exacerbations are associated with an increased decline in lung function among COPD patients with elevated B-Eos and without ICS. 18 However, the overall reported odds ratios are low, and since B-Eos count could be significantly influenced by different co-factors, we suggest that the association between B-Eos count and respiratory morbidity could be improved by adjusting for these factors.
In asthma, persistent type-2 inflammation indicated by, for example, elevated FeNO or B-Eos count, may identify patients with poor responsiveness to ICS, despite adherence to treatment. 19 Furthermore, biomarker-directed risk stratification to identify patients suitable for different biological treatments, including FeNO and B-Eos count, have been proposed, 20,21 and this combination of biomarkers seems to provide additive predictive information on asthma morbidity and risk. 22,23 FeNO is affected by individual factors such as age, height and sex, 24,25 and cigarette smoke exposure, 26 but very few studies have analyzed possible determinants of B-Eos count. Nevertheless, individual characteristics may also affect B-Eos values, which could hamper the clinical interpretation of a B-Eos count, especially in the initial assessment of patients with respiratory symptoms. Elevated B-Eos levels have been described to be associated with smoking [27][28][29] and increasing age. 27,29,30 However, evidence on the influence of individual factors on B-Eos count is inconsistent, and even less data exists for subjects without respiratory disease. 27 Moreover, there is a great controversy regarding the optimal B-Eos cutoff that is more strongly associated with various disease outcomes, which could make the interpretation of a B-Eos count inconsistent. 31 Identification of factors not related to respiratory disease that should be considered when interpreting a B-Eos count may be useful for a targeted and personalized approach in the clinical management of respiratory diseases. Moreover, in addition to the type of inflammation, its location is also important for the disease assessment. To this end, it may be necessary to have a comparison with other systemic markers and local type-2 markers, such as C-reactive protein (CRP) and FeNO, respectively. 32,33 The aim of this investigation was to (a) evaluate the influence of non-respiratory disease on B-Eos count, FeNO, and CRP, and (b) to identify individual characteristics that are associated with B-Eos count in healthy individuals.

| Data source and study subjects
We analyzed the data publicly available from six 2-year surveys

| Inflammatory biomarkers
B-Eos count was analyzed using a quantitative hematologic analyzer and leukocyte differential cell counter, Beckman Coulter HMX (Beckman Coulter). Complete details on blood collection procedures, quality assurance, and control procedures are described elsewhere. 35 FeNO was measured following the ATS/ERS recommendations 36 using a handheld device with an electrochemical sensor, NIOX MINO

| Variables
Demographic characteristics, such as age, sex, race/ethnicity, body mass index (BMI), and smoking status were analyzed. Two age groups were defined: children/adolescents (if < 18 years) and adults (≥18 years).
Race/Ethnicity was categorized as non-Hispanic white, non-Hispanic black, Mexican/Hispanic, and other (multi-racial). BMI was categorized according to international recommendations, both for children/adolescents 37 and adults, 38 into underweight, normal weight, overweight, and obese. Smoking status was defined as: never smokers, current smokers, and former smokers. Children/adolescents were considered as nonsmokers (see Data S1 for details).
Respiratory diseases were considered when having a self-reported diagnosis of asthma and/or hay fever (in children/adolescents), or self-reported diagnosis of asthma and/or hay fever, and/or other respiratory diseases (in adults). Asthma was defined as an affirmative response to the question "Has a doctor ever told you that you had asthma?" Hay fever was defined if the subject answered positively to either: "Has a doctor ever told you that you had hay fever?" or "During the past 12 months, have you had an episode of hay fever?" Other respiratory diseases were defined if the subject answered positively to "Has a doctor or other health professional told you that you had emphysema and/or chronic bronchitis?" Healthy control groups were obtained by excluding children/adolescents with any of the respiratory diseases described above. In adults, healthy controls were obtained by excluding subjects having: any respiratory disease and at least one of the non-respiratory diseases (arthritis, heart diseases, stroke, cancer, hypertension, diabetes, hypercholesterolemia, and metabolic syndrome) that were significantly associated with B-Eos after adjustment for individual characteristics (see below). Details on the definition of the nonrespiratory diseases are described in the Data S1.
The two respiratory disease groups were defined as: (a) children/ adolescents with respiratory disease; and (b) adults with respiratory disease and without any significant non-respiratory disease associated with elevated B-Eos counts ( Figure 1).

| Statistical analysis
Statistical analyses were conducted in Stata/IC 15.1 (Stata Corp), and a statistical significance level was set at p < 0.05. In all analyses, the complex multistage sampling and sampling weights were considered using the svy package.
B-Eos count and FeNO were log-transformed because of their highly skewed distribution and described using geometric means and 95% confidence intervals (GM [95%CI]). Also, we categorized both B-Eos count and FeNO measurements by using the most commonly used cut-offs. [39][40][41] Simple descriptive statistics were used to describe the study population. To explore the association between B-Eos count and each disease factor we performed multivariate linear regression modeling. Separate univariate and multivariate models F I G U R E 1 Flowchart of the National Health and Nutritional Examination Surveys (NHANES) participants. *Respiratory diseases included previous diagnosis of asthma (n = 1984) and/or hay fever (n = 1349); ¥ Respiratory diseases included previous diagnosis of asthma (n = 4744), hay fever (n = 3244) and/or other respiratory diseases (emphysema n = 653; chronic bronchitis: n = 1748) AMARAL ET AL. were built using each biomarker (B-Eos, FeNO, and CRP) as the outcome variable. All variables with p < 0.20 were considered for inclusion in the final model. Adjustments were made for co-variables: sex, age, race, smoking status, height, and BMI, and weight instead of BMI, in all models. Diagnostic for multicollinearity was performed using the variance inflation factor (VIF) test. A VIF above 5 was considered to indicate a high degree of correlations among the predictor variables. 42 Coefficients (ß) with 95% confidence intervals (95% CI) were presented, and the model fit was assessed using the svygof function. Sensitivity analysis was performed by multiple imputation of missing values ( Figure 1) using the MI command.
Additionally, percentage change was calculated by the ratio between the amount of change and the original value, multiplied by 100.
To analyze the prevalence of the non-respiratory disease factors, we divided age into tertiles (18-35; 36-57; ≥58 years old). Spearman correlation coefficients were calculated between B-Eos count and other markers (FeNO and CRP), in both healthy control and respiratory disease groups (in children/adolescents and adults, respectively).

| Participant characteristics
NHANES included a total of 60,936 participants from 2005 to 2016, and 82% had complete data on B-Eos count across all ages ( Figure 1).
Participant characteristics for children/adolescents and adults, respectively, are described and compared in Table 1 of females was slightly higher in adults compared to the young group.
Furthermore, the proportion of non-Hispanic whites was larger among adults (Table 1). Adults reported significantly more respiratory diseases than children/adolescents, mainly due to the reporting of chronic bronchitis and emphysema as well as more hay fever, whereas the proportion reporting asthma was similar in the two age groups (Table 1). Children/adolescents showed higher B-Eos counts (12%), whereas adults presented with higher FeNO (13%) and CRP levels (150%).

| Effect of non-respiratory diseases on B-Eos count
In adults without respiratory disease (Figure 1), the prevalence of all the non-respiratory disease factors across age tertiles is shown in Table S1. Univariate analyses showed that B-Eos counts were higher in individuals reporting arthritis, heart disease, stroke, hypercholesterolemia, diabetes, hypertension, or metabolic syndrome than in subjects without these disorders (Table 2). Similarly, participants reporting these diseases had higher FeNO levels (except for stroke), and higher CRP levels (except for hypercholesterolemia).
After multivariable adjustment, reporting heart disease, stroke, and/or metabolic syndrome were associated with elevated B-Eos counts, independently of individual characteristics (age, sex, smoking status, race/ethnicity, BMI, and height; Figure 2). B-Eos count was 13%, 15%, and 12% higher, respectively, in subjects with any of these non-respiratory diseases compared to subjects without. Using the same variables in the model, that is, both individual characteristics and reported non-respiratory diseases, no independent associations were found between having a non-

| Effect of individual characteristics on B-Eos count in healthy control groups
After excluding children/adolescents with any respiratory disease, and adults with any respiratory disease and at least one nonrespiratory disease independently associated with B-Eos count (heart diseases, stroke, and/or metabolic syndrome), we obtained healthy control groups with children/adolescents (n = 8944) and adults (n = 5667), respectively (Figure 1).
The overall distribution of B-Eos count in all the participants of the healthy control groups, ranging from 1 to 85 years and stratified by sex, is illustrated in Figure 3. Males show higher B-Eos counts compared to females across the whole age range. A decrease in B-Eos count was seen with increasing age in individuals <18 years, followed by a stabilization up to around 70 years, regardless of sex. Furthermore, children/adolescents had significantly higher B-Eos counts than adults regardless of individual characteristics such as sex, BMI, and race/ethnicity, except in obese individuals (Additional file 1: Table S2).
In univariate analyses, males had significantly higher B-Eos count than females among both children/adolescents and adults when analyzing absolute ( Figure 4) and relative (Table 3) differences.
Furthermore, differences in B-Eos counts were seen between different groups of race/ethnicity; the most prominent difference was between non-Hispanic whites and other race/ethnicity. In adults, being overweight or obese and being former or current smoker was associated with higher B-Eos counts.
In multivariate analyses, male sex, age (p = 0.025; p < 0.001 in 0-11 years and p = 0.041 in 12-17 years), being obese and being of non-Hispanic black or other race/ethnicity, were all independently associated with higher B-Eos counts in children/adolescents (Table 3). In adults, the same factors, except age, were independently associated with higher B-Eos count, and being overweight or current smoker were also independently associated with higher B-Eos counts. Furthermore, non-Hispanic blacks showed lower B-Eos count than non-Hispanic whites in adults. Height was a nonsignificant factor in both age groups, and similar results were obtained using weight instead of BMI in all models (data not shown).
Of note, in adults where only respiratory diseases had been excluded (n = 15,010; see Figure 1), age was independently associated with higher B-Eos count (p < 0.001). When introducing heart diseases, stroke, and metabolic syndrome as independent factors in this model, the age effect was no longer significant (p = 0.290).

F I G U R E 2
Regression coefficients (ß) and 95% confidence interval of the non-respiratory disease factors for blood eosinophil (B-Eos), fraction of exhaled nitric oxide (FeNO) and C-reactive protein (CRP), in adults without respiratory diseases.

| Association between inflammatory markers
In both children/adolescents and adults, B-Eos count and FeNO were higher in those with respiratory disease, while CRP levels were not significantly different in the two age groups (Additional file 1: Table S3).
B-Eos count correlated weakly with CRP in adults, both in the healthy control group and the respiratory disease group, whereas no correlation was found in children/adolescents (Additional file 1: Table S4). On other hand, FeNO correlated moderately with B-Eos count in children/adolescents and weakly in adults, both in the healthy control group and the respiratory disease group. In adults these correlations became stronger after excluding current smokers (Additional file 1: Table S4)

| DISCUSSION
In a large population-based study of US participants without respiratory disease we found that having heart disease, stroke, and/or metabolic syndrome independently increased B-Eos levels by 12%-15%, after adjusting for covariables. No significant association was found between any non-respiratory disease and FeNO or CRP.
Furthermore, sex, overweight/obesity, race/ethnicity, and smoking status were also related to B-Eos count. Age was independently associated with B-Eos count only in children and adolescents.
To the best of our knowledge, this is the first study to explore the effects of non-disease-related factors and non-respiratory diseases separately, performed in a large multiethnic population-based sample. Furthermore, we compared B-Eos count with FeNO and CRP regarding the association with non-respiratory disease.
Eosinophils play a major role in mediating allergic inflammation.
However, they have also been implicated in many other disorders, such as specific organ damage in localized infiltrative eosinophilic entities. [43][44][45] In our study, having heart disease, stroke, and/or metabolic disease were independently associated with B-Eos count in adults, similar to previous findings. 27 This indicates that other diseases than respiratory diseases must be considered when using B-Eos count in the management of asthma and COPD.
In spite of the correlation with B-Eos and being well-known and widely used in population-based studies as a marker of systemic inflammation, CRP did not independently associate with nonrespiratory diseases in our study, whereas B-Eos count did. This suggests that an increased level of B-Eos may better reflect the total systemic burden of inflammation. 46,47 Although elevated CRP levels have been shown to be associated with metabolic syndrome, 48,49 and to be an independent risk factor for coronary artery disease, 50 rating that there was a relationship but no co-linearity between nonrespiratory diseases and age. These results suggests that the age effect seen in previous studies was related to non-respiratory diseases commonly seen in middle-aged and older adults. Our results are also consistent with observations in a recent study that analyzed a different population setting. 27 The same authors also found a similar age trend in children and adolescents, namely that subjects in early life (≤12 years) presented the highest B-Eos counts with an increasing trend with decreasing age, regardless of sex. The reason for higher B-Eos counts in children should be studied further.
Eosinophilic inflammation has also been found to be associated with higher BMI among adults. 56 In our study, obese subjects had higher B-Eos values than those with normal weight. Also, current smoking was independently associated with higher B-Eos levels, compared to never and former smokers. This is in line with several studies that demonstrated a significant increase in B-Eos counts by smoke exposure. 57 Moreover, an association between higher B-Eos counts and serum cotinine levels was previously reported in healthy subjects, and even passive smoke exposure was shown to cause elevated B-Eos counts. 26 In the same study, B-Eos count was higher in presently non-exposed (serum cotinine below lower limit of detection) former smokers compared to never smokers. However, this could be explained by the lack of adjustment for non-respiratory diseases and BMI.
To our knowledge, the effect of ethnicity on B-Eos count is poorly evaluated in the literature. In our study, we found a race/ ethnicity influence among healthy control groups. Specifically, non-Hispanic white children and adolescents, and non-Hispanic black adults presented with the lowest B-Eos levels. However, further studies are needed to explore these results.
Our estimated B-Eos levels of the adult healthy control group were similar to those recently obtained in the same setting, 29 and in other populations without respiratory diseases, 58,59 but higher than those presented in the Hartl et al. study. 27 The latter study included subjects from a different ethnical setting than in our study, and differences in B-Eos count may also be explained by differences in the cell counting methodology.
The strengths of this study were a large number of participants, the inclusion of a broad age range and different races/ethnicities, and that we used B-Eos as a continuous value, rather than using predefined cutoffs or median as in a previous study. 27 Furthermore, we were able to study the effects of individual characteristics, respiratory disease, and non-respiratory disease separately by forming the corresponding subgroups. We also used other clinically used inflammatory markers as a benchmark when studying associations with non-respiratory diseases.
The study also has some limitations. The cross-sectional design of the study, and the fact that most of the diseases were self-reported, may have limited the ability to support the predictive properties of the markers. However, we used broad definitions of respiratory disease to reduce the risk of including individuals with true disease in healthy control groups. Moreover, the definitions of respiratory disease that we applied have been commonly used in NHANES reports. 60,61 However, further crosssectional and/or longitudinal studies that include a medical diagnosis of the analyzed respiratory diseases, including, for example, lung function tests, are needed. Also, the lack of data regarding atopy, allergic sensitization, nasal polyps, urticaria, parasitic infection, inflammatory bowel diseases, eosinophilic drug reactions and circadian variation prevented the adjustments for these variables.
Although this is out of scope of our study aim, it should be further explored.
In conclusion, several individual characteristics, and nonrespiratory diseases, should be considered when interpreting B-Eos counts. Although FeNO has long been recognized to be influenced by several individual factors, this marker did not associate with any non-respiratory disease. The individual factors that were found to influence B-Eos count are readily available in the clinic and could be incorporated into novel reference equations to obtain individualized cutoffs that would support a targeted and personalized approach in the clinical management of chronic respiratory diseases. With the development of new biologics that target eosinophilic airway AMARAL ET AL.