Introduction

Chronic obstructive pulmonary disease (COPD) is an obstructive airway disease with both overlapping and distinctive features as with asthma1. Asthma is characterized by eosinophilic inflammation2, whereas COPD is predominantly associated with neutrophilic inflammation in the airways3. Growing evidence suggested that neither characteristic was immutably ingrained in either disease. This difference in cellular composition of induced sputum may, if ever, be indistinguishable between these disease groups2. Increased sputum eosinophils has been reported in both stable3 and exacerbation phase4 of patients with COPD, implying the potential role of eosinophils in the pathogenesis of COPD2.

Eosinophilia is generally defined as greater or equal to 2% eosinophils in either blood or sputum3,5,6,7. Alternatively, an absolute blood eosinophil count of 0.34 × 109 cells per liter can be used as a threshold for risk stratification7. Peripheral blood eosinophil count is highly associated with eosinophilia of the respiratory tract5. This blood biomarker has also been shown to reflect submucosal eosinophilia of the lung and reticular basement membrane thickening8. Given this context, we considered that patients with COPD who had more than 2% of eosinophils, either in the blood or sputum, as eosinophilic COPD.

Acute exacerbation of COPD significantly increases symptoms, deteriorates pulmonary function, increases rate of hospitalization and lengthens hospital stay further impairing functional capacity and quality of life (QOL) imposing additional burden to healthcare system9,10,11. The in-hospital mortality can reach 30% or more12. Seeking for predictive biomarkers for clinical outcome in this population is thus of high priority.

Numerous studies have evaluated eosinophilia in relation to exacerbation risk5,7,13, length of hospital stay14,15,16, in-hospital mortality12,17,18, and response to steroidal and bronchodilator therapies9,10,11 but the evidence is conflicting. Some studies have reported a higher risk for exacerbation in patients with eosinophilic COPD13,19. Conversely, a retrospective study suggested that a higher level of eosinophils protected against disease aggravation16. Other research teams failed to detect any association5,7,20.

We conducted a systematic review and meta-analysis of clinical outcome measures comparing patients with COPD who had eosinophilia and those without eosinophilia.

Results

Of 3,131 abstracts identified by the initial search, 1,710 and 1,323 articles were removed, respectively, because of irrelevance or overlaps. After exclusion, 37 studies involving 99,122 patients published between 1998 and 2016 were included for qualitative synthesis (Fig. 1). Of these, 14 studies were included in meta-analysis. The number of entries derived from different search terms has been summarized in Table 1. The mean age of the subjects was 66.95 years with the proportion of male subjects ranging from 455 to 100%21. On average, each subject had a 46 pack-year smoking history. The mean forced expiratory volume in one second (FEV1) ranged from 0.96 L to 1.62 L. A total of 21 studies explored the role of blood eosinophilia in COPD. The remaining articles detected eosinophils in sputum and bronchial fluid after treatment with bronchodilators or steroidal therapy. The description of studies is summarized in Table 2. More than half of the included studies were either conducted in the United Kingdom1,9,10,11,13,17,18,22,23,24,25,26,27 or other European countries2,3,4,21,28,29,30,31. Eleven studies were originated from the Asia-Pacific region5,6,32,33,34,35 and the North America19,20,36,37,38. There was only a single relevant publication from the Middle East12.

Figure 1
figure 1

Flow diagram of literature search and selection of studies.

Full size image
Table 1 Number of entries by different search terms.
Full size table
Table 2 Description of the included studies.
Full size table

Overall, included studies fell into low to moderate quality (Supplementary Tables 1 and 2). Of 24 non-randomized observational studies evaluated by Newcastle-Ottawa scale, the mean score was 4.5 out of nine (range: 2–6). Five studies scored six or above in a nine-point scale, indicating high study quality6,7,11,22,30. In 13 randomized control trials assessed by Cochrane Collaboration Risk of Bias tool, seven studies were rated as low risk in terms of allocation concealment, blinding of participants and personnel, blinding of outcome assessment and incomplete outcome data9,10,11,20,24,26,27. Notably, two studies were ranked as high risk for randomization, blinding, and selective reporting4,32.

Eight populations of six studies5,7,13,16,19,20 were pooled for risk analysis. Overall, no association was observed between eosinophilia and risk for exacerbation warranting hospital admission in 12 months (OR = 1.07, 95% CI 0.86–1.32, P = 0.55, I 2 = 73%). This null effect remained in sub-group analysis of studies involving hospitalized COPD patients13,16,19,20. Interestingly, in patients with stable COPD as defined as having no hospitalization in the previous 12 months, eosinophilia appears to increase the risk for exacerbation by 18% (OR = 1.18, 95% CI 1.03–1.34, I 2 = 0%) (Fig. 2 ).

Figure 2
figure 2

Forest plots of studies comparing the risk for exacerbation in 12 months in COPD patients with or without eosinophilia. Vedel-Krogh (2015) subgroup A, clinical COPD; Vedel-Krogh (2015) subgroup B, COPD cohort in general population; Pavord (2016) subgroup A, COPD patients on fluticasone propionate and salmeterol; Pavord (2016) subgroup B, COPD patients on fluticasone propionate.

Full size image

Pooled estimate of five studies12,14,16,17,18 did not indicate an association between eosinophilia and in-hospital mortality, though approaching statistical significance (P = 0.08). Of note, a single largest study published in the Lancet26 did not identify any association between clinical outcomes and eosinophilia. Although pooled estimate of the other studies12,14,17,18 showed that eosinophilia was a protective factor against in-hospital mortality (OR = 0.38, 95% CI 0.17–0.86, I 2 = 35%), these studies have to be interpreted with cautions due to potential risk of bias. Patients with eosinophilic COPD had 1.2 days shorter hospital stay than non-eosinophilic individuals. Given moderate to high heterogeneity of overall estimates, sensitivity analysis was performed. Except for in-hospital mortality, no single study substantially altered the pooled estimates (Figs 3 and 4).

Figure 3
figure 3

Forest plots of studies comparing the risk for in-hospital mortality in COPD patients with or without eosinophilia.

Full size image
Figure 4
figure 4

Forest plots of studies comparing the mean difference of the length of hospital stay.

Full size image

Subsequent to concurrent treatments with bronchodilators and steroids the pooled estimate revealed slight improvement in change of FEV1 (SMD = 0.52, 95% CI 0.33–0.71) (Fig. 5). Sub-group analysis has also shown that outpatients with eosinophilic COPD exhibited improvement in pulmonary function. For outpatient groups, the combined mean differences for FEV1 and percentage of predicted FEV1 were 0.11 L (95% CI 0.09–0.13, P < 0.001) and 1.64% (95% CI 0.05–3.23, P < 0.001), respectively (Figs 5 and 6).

Figure 5
figure 5

Forest plots of studies comparing the mean difference of the change of FEV1 in COPD patients after therapy. Bafadhel (2012) subgroup A, clinical outcomes in 2 weeks after therapy. Bafadhel (2012) subgroup B, clinical outcomes in 6 weeks after therapy.

Full size image
Figure 6
figure 6

Forest plots of studies comparing the mean difference of the change of % FEV1 predicted in COPD patients after therapy. Bafadhel (2012) subgroup A, clinical outcomes in 2 weeks after therapy. Bafadhel (2012) subgroup B, clinical outcomes in 6 weeks after therapy. Pavord (2016) subgroup A, COPD patients on fluticasone propionate and salmeterol; Pavord (2016) subgroup B, COPD patients on fluticasone propionate; Pavord (2016) subgroup C, COPD patients on salmeterol.

Full size image

Of the three studies comparing reported QOL in patients with COPD, chronic respiratory disease questionnaire (CRQ)9,10 and St George’s respiratory questionnaire (SGRQ) were used11. The eosinophilic group consistently reported a higher QOL score subsequent to therapy. For studies using CRQ, a standardized mean difference of 0.85 (95% CI 0.56–1.14) was observed. For studies using SGRQ, an improved quality of life was also reported (SMD = 3.14, 95% CI 2.93–3.36). The pooled analysis is presented in Fig. 7.

Figure 7
figure 7

Forest plots of studies comparing the standardized mean difference of the change of quality of life scores in COPD patients after therapy. Pavord (2016) subgroup A, COPD patients on fluticasone propionate and salmeterol.

Full size image

Discussion

Overall, eosinophilia in COPD patients does not contribute to exacerbation risk, in-hospital mortality, and length of hospital stay. However, higher eosinophil count in the outpatient sub-group demonstrated an increased risk of exacerbation by 18%. On the other hand, eosinophilic COPD patients appeared to be more responsive to therapeutic interventions.

In previous investigation of hospitalized COPD patients with severe exacerbation, eosinophilia lacked association with more than three-fold increased risk for re-admission in 12 months19. Retrospective analysis of COPD population with a post-bronchodilator FEV1/forced vital capacity (FVC) ratio below 0.7 did not identify significant difference in exacerbation risk amongst the eosinophil dominant group22. These were in contrast to a Turkish study in which a greater risk for re-admission was demonstrated in the eosinophilic group16. In a Dutch general population study, eosinophilia was found to increase risk for acute exacerbation of COPD7. Consistently, we found 18% increased risk for disease aggravation in outpatients. Exacerbation has been linked to airway inflammation characterized by eosinophilia4,6,24 and imbalance of metalloproteinases23. Higher level of eotaxin, an eosinophil chemotactic factor, is elevated in pulmonary lavage37. It has been suggested that frequency and severity of COPD exacerbation was a result of impaired macrophage efferocytosis of eosinophils36. Marked eosinophilia was observed in virus-induced exacerbations30.

Our pooled analysis showed that eosinophilia is associated with reduced length of hospital stay. This is consistent with previous studies including severely exacerbated COPD patients14,18. Conversely, peripheral blood eosinopenia increased in-hospital mortality by up to five-fold12,17. The disparity may be attributable to the timing of blood specimen collection. For hospitalized patients, samples were collected at the time of admission12,14,16,17,18. The time for collection in the outpatient group varies across studies and included at the screening stage11, at exacerbation10, and at 24 h after bronchodilator therapy9. In addition, recent hospitalization histories of these outpatients were uncertain9,10,11. In other words, they may have never been hospitalized or had follow-up at clinics soon after discharge. It has been suggested that airway eosinophilia facilitated responsiveness to bronchodilator and steroidal therapies26,33. The better response to therapy in this patient population may explain the consistently shorter length of stay and lower mortality.

Eosinophilia has been suggested to indicate individual responsiveness to bronchodilator and steroidal therapies9,10,11,13,15,25,26,34. Post-hoc analysis confirmed that level of eosinophil correlates with the response to bronchodilators27. Specifically, post-bronchodilator FEV1 and sputum eosinophil level had a high correlation of 0.8231. After oral prednisolone therapy, sputum eosinophil count changed accordingly along with interleukin-525. Blood eosinophils were also found to be associated with changes in pulmonary function after inhaled corticosteroids10,11,13,20. In our meta-analysis, although the predicted %FEV1 changed by 1.64%, this may represent a substantial improvement given these subjects were considered as severe COPD with baseline predicted %FEV1 less than 50%9,10. However, the addition of hospitalized patients nullified the effect. This suggested that disease severity may be a significant confounder in the observed relationship.

The overall risk of bias in the included randomized control trials ranged from low to moderate. The inferior quality was mostly attributed to unclear sequence generation and likelihood of selective outcome reporting4,32,34,35,37. Eight of the studies applied allocation concealment, and blinding of participants and outcome assessors9,10,11,18,20,24,27,35. In quasi-experimental studies, the potential risks of bias included self-reporting for outcomes, insufficient follow-up period and unclear relationship between loss of follow-up and outcome of interest. In addition, appropriate adjustments were not performed for previously reported confounders associated with eosinophil level and clinical outcome of COPD38. The majority of the included population was originated from the United Kingdom and other European countries; only a few studies were conducted in the Continent of Asia and the America. This racially skewed population may preclude the generalizability of the evidence.

We performed this systematic review according to a pre-defined data abstraction form. Minor alterations were made to facilitate data pooling. There were missing data on some of the outcome measures of our interest, reducing the number of eligible studies. Given the limited number of included studies for each outcome comparison, neither funnel plot nor Doi plot were conducted to examine publication bias. Our sensitivity analysis revealed that, except for in-hospital mortality, the pooled estimates remained stable.

Given no consensus on definition of eosinophilia, there may be mixing of eosinophilic and non-eoinophilic groups of COPD patient, diluting the effect size. The estimation of eosinophil level varies with the type of specimens. Within the same patient group, bronchial biopsies yielded lower eosinophil count than induced sputum29. Importantly, the temporal variation of eosinophilia in COPD was largely ignored in the included studies. Longitudinal study of 1,483 patients with COPD revealed that 49% of the subjects had variable eosinophil counts39. Only 37% and 14% of the individuals were persistently eosinophilic and eosinopenic, respectively39. The level of this cellular marker can increase considerably soon after sputum induction40. In this connection, spotshot sampling may lead to misclassification of case and control.

The moderate to high heterogeneity of the pooled estimates suggests the presence of unknown confounders in association with eosinophilia and COPD. This may be attributed to a range of severity of COPD patients included in the studies and the timing of blood collection. Other potential confounding variables may include, but not limited to, specimen type, baseline characteristics of the study population, study quality and unknown pre-existing co-morbidities. Cross-sectional analysis of 948 COPD patients revealed that eosinophilic group was associated with lower rate of heart attack and anemia38. If these contributed to different clinical outcome of this sub-group remained equivocal. The use of steroidal therapy may interfere with the risk for exacerbation. Given the lack of accessibility to information on individual exposure, it was impossible to control for the factor of steroidal therapy in the pooled estimate of exacerbation risk.

In conclusion, eosinophilia is associated with a better improvement of pulmonary function and reported QOL subsequent to therapy in outpatients. Given its association with eosinophil level in the airway, blood eosinophil count may be a predictive biomarker in patients with stable COPD for response to steroidal and bronchodilator therapies.

Methods

Searching strategy

This systematic review was performed in accordance with the guidelines on Preferred Reporting Items for Systematic Reviews and Meta-analyses: The PRISMA Statement 200941. Original articles published in PubMed (MEDLINE), ISI Web of Knowledge, EMBASE, and Scopus database were identified using Medical Subject Heading (MeSH) or Title/ Abstract keywords from inception up to December 2016. The MeSH search terms include a combination of eosinophil, blood, sputum, pulmonary disease, chronic obstructive, and/or airway disease. The number of entries retrieved from each database is summarized in Fig. 1. Two authors (JH and WH) performed the literature search and selected the relevant studies independently. Disagreements in terms of study selection were resolved by discussion with senior authors.

Inclusion and exclusion criteria

Included studies were primary research articles comparing patients with and without eosinophilic COPD in terms of exacerbation risk, mortality, morbidity, length of hospital stay, and response to corticosteroids and bronchodilators. Quasi-experimental studies and randomized controlled trials were included. Pre-clinical studies, review articles, editorials, commentaries, conference abstracts and book chapters were excluded.

Data extraction

Relevant data were extracted according to a pre-defined data abstraction form. Information on sample size, baseline characteristics, incidence of exacerbation in the past 12 months, length of hospital stay, in-hospital mortality, QOL, and pulmonary function were extracted by one researcher (JH) and verified by a second researcher (WH).

Quality assessment and statistical analysis

The methodological quality of the included randomized controlled trials and quasi-experimental studies was evaluated by the Cochrane Risk of Bias Tool42 and the Newcastle-Ottawa scale43 respectively. The former tool indicates studies with high, low or unclear risk according to five domains: selection bias, performance bias, detection bias, attrition bias, and reporting bias. The latter scale evaluates the quality of studies in three attributes, namely selection of cohort, comparability, and outcome. In this review, a high-quality study is defined as having >6 points whereas a low-quality study as having ≤5 points.

Meta-analysis compared patients with eosinophilic and non-eosinophilic COPD in terms of exacerbation risk, length of hospital stay, in-hospital mortality, and change of pulmonary function and QOL in response to medical interventions. Heterogeneity across studies was determined by the I 2 statistic using Cochrane Review Manager 5.344. An I 2 values ≥ 25, 50 and 75% were considered as mild, moderate, and high degree of heterogeneity, respectively. For pooled outcome measures with I 2 > 50%, a random-effect model was used to evaluate the overall effect of a given comparison. Studies were weighted by inverse of variance. Categorical data was presented as odds ratio (OR) in 95% confidence interval (CI). For continuous variables, the pooled estimates were compared by mean difference (MD) or standardized mean difference (SMD), as appropriate. In the occasion when the remaining studies appeared to be different from the overall estimate, sub-group analysis was performed.