Stability of blood eosinophils during acute exacerbations in patients with chronic obstructive pulmonary disease Authors: Qidong Chen, Wei Ji, Hongyu Wang, Jiahui Lin, Tanwei Liu, Jialin Lin, Xu Chen, Xiaoyang Chen Published: 2025 ABSTRACT AbstractBackgroundBlood eosinophils are recommended by GOLD as a biomarker for inhaled corticosteroid (ICS) therapy in COPD, yet their stability during acute exacerbations of COPD (AECOPD) remains uncertain.MethodsWe retrospectively analyzed 3,311 hospitalized AECOPD patients (2018–2023). Blood eosinophil distribution, longitudinal trends, and within-patient variability were assessed. Linear mixed-effects models estimated overall trajectories of ln(1 + EOS) over standardized hospital days (SHD). Variability was quantified by coefficient of variation (CV%), within-patient SD (wSD), and median absolute change between consecutive tests. Robustness of thresholds (100 and 300 cells/µL) was examined by crossing and reclassification rates, with simulations of 20–30% overall decline. Near-threshold “decision bands” were derived from the empirical median change (≈ 50 cells/µL).ResultsBlood eosinophils showed a right-skewed distribution with a median of 100 cells/µL (IQR 20–200). Mixed-effects models indicated only a mild overall increase (+ 2.16% per + 1 SD in SHD). In patients with repeated tests (n = 1,548), within-patient variability was substantial (median CV% 76.6%; wSD 57 cells/µL; median |Δ| 50 cells/µL). Threshold crossing and first–last reclassification were frequent (≥ 100: 45.7% and 32.9%; ≥300: 22.7% and 16.0%), and declined only slightly in simulated steroid-reduction scenarios. Approximately 13% of patients fluctuated between < 100 and ≥ 300 cells/µL during one admission.ConclusionEOS should be interpreted dynamically, with retesting or trend assessment recommended to guide clinical decisions during acute exacerbations.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12890-025-04063-4.Keywords: COPD, Exacerbation, Eosinophil, Biomarkers, Distribution, Stability Background Chronic Obstructive Pulmonary Disease (COPD) is a heterogeneous lung condition characterized by chronic respiratory symptoms due to abnormalities of the airways and/or alveoli that cause persistent, often progressive, airflow obstruction [1]. COPD is statistically the most common cause of death from chronic respiratory diseases, with a rate of 41.9 deaths per 100,000 people, and is responsible for more than 3 million deaths globally each year [2]. COPD is usually thought to be a neutrophilic inflammatory disease, but in recent years studies have found that blood eosinophils play an important role in COPD [3, 4]. Combination therapy with ICS can significantly improve lung function and clinical symptoms in patients with COPD, as well as reduce the frequency of acute exacerbations [5–7]. GOLD identifies blood eosinophil count as an important biomarker to assess which patients with stable COPD are most likely to benefit from ICS-containing therapy. When blood eosinophil counts are ≥ 300 cells/µL, GOLD strongly recommends the use of ICS in COPD patients; when blood eosinophil counts are < 100 cells/µL, ICS is not recommended; for values in between, the use of ICS is supported [8]. The GOLD guidelines recommend the systemic use of glucocorticoids in patients with acute exacerbations of COPD, as they can improve lung function and oxygenation [9, 10], reduce the risk of early relapse and treatment failure [11], and shorten hospital stays [12]. However, some studies have observed that long-term oral glucocorticoid therapy during acute exacerbations may increase the risk of pneumonia and mortality [13]. Other studies have used blood eosinophil counts during acute exacerbations as a biomarker to guide systemic glucocorticoid therapy, suggesting that this strategy is non-inferior to standard therapy while reducing the duration of systemic corticosteroid exposure [14, 15]. Having high reproducibility is an important requirement for a good biological indicator [16]. Though some studies have explored the stability of eosinophil levels in COPD patients, most have focused on stable COPD patients, and there remains a high degree of variability in this data [17–20]. Fewer studies have addressed the stability of acute exacerbations in COPD patients, and there is heterogeneity methodology and relatively small sample sizes [21, 22]. Emanuel Citgez et al. explored changes in blood eosinophils between two acute exacerbations and concluded that blood eosinophils fluctuate between acute exacerbations [21]. Yanan Cui et al. enrolled 530 patients and examined the stability of COPD during the same hospitalization.They studied the stability of blood eosinophils in patients with acute exacerbations of COPD during the same hospitalization, but only a cut-off value of 300 cells/µL was taken and only the values of routine blood counts from both admission and discharge were used [22]. Therefore, further research is needed to investigate the quantitative changes in the stability of blood eosinophil counts in patients with acute exacerbations of COPD during the same hospitalization. Meta-analyses have reported the median blood eosinophil counts in patients with COPD and indicated that blood eosinophils in this population follow a right-skewed distribution [23]. However, some studies have described blood eosinophil levels in COPD patients using arithmetic means rather than geometric means or medians [24]. Studies by Sylvia Hartl et al. [25]and Yunus Çolak et al. [26] have shown that, in the general population, blood eosinophil counts are higher in men than in women. Whether this characteristic also applies to patients with COPD remains to be further explored. In this study, we aimed to quantify the short-term variability of blood eosinophil counts in patients hospitalized with AECOPD and, on this basis, to evaluate the applicability and practicality of the GOLD-recommended thresholds in this population. Methods Study design and study population This was a retrospective cohort study of patients admitted to the Second Hospital of Fujian Medical University with a primary diagnosis of AECOPD from 1 January 2018 to 31 December 2023. The diagnostic criteria for COPD were 1) according to the GOLD guidelines (including evaluation of symptoms, risk factors, and the presence of non-fully reversible airflow obstruction (FEV1/FVC < 0.7 post-bronchodilation) measured by spirometry confirms the diagnosis of COPD) diagnosis of COPD [8]. Acute exacerbation was defined as worsening of symptoms requiring additional treatment [22, 27]. Exclusion criteria were as follows: (1) patients < 40 years of age; (2) missing data in visit records, routine blood tests, and prescriptions; and (3) patients with one of the following disorders: asthma, allergic diseases, pneumothorax, pulmonary embolism, lung cancer, systemic fungal infections, severe psychiatric disorders, and pneumonic autoimmune or hematological disorders [28, 29]. Demographic characteristics, admission and discharge dates, and all complete blood count (CBC) results during hospitalization were extracted from the hospital information system. For each CBC record, the absolute eosinophil (EOS) count (×10^9/L) and the corresponding “days since admission” were recorded; values were converted to cells/µL when necessary (conversion factor: 1 × 10^9/L = 1000 cells/µL). The first EOS measurement during hospitalization was designated as the “initial test,” while subsequent measurements were categorized as “follow-up tests.” To reduce the influence of extreme values, EOS data were winsorized at the 1 st and 99th percentiles, and ln(1 + EOS) transformation was applied for modeling. The time variable was centered and standardized based on days since admission (Standardized Hospital Days, SHD). The objectives of this study were to quantify the overall distribution and short-term within-patient variability of eosinophil counts (EOS) during hospitalization for AECOPD, and to evaluate the practicality of the 100 and 300 cells/µL thresholds in this setting. For patients with repeated measurements, within-patient variability was assessed by calculating the coefficient of variation (CV%), within-patient standard deviation (wSD, cells/µL), and the median of absolute differences between consecutive measurements (median adjacent |Δ|). In addition, ordinary least squares regression was applied to estimate individual slopes of ln(1 + EOS) over time, and heterogeneity across the population was illustrated by combining CV% with these slope estimates. The primary longitudinal analyses were conducted using linear mixed-effects models. Model 1 was a random-intercept model, with ln(1 + EOS) as the dependent variable, standardized hospital days (SHD) as the independent variable, and patients as the random effect. Model 2 extended this framework by incorporating patient-level random slopes for time (fitted only among patients with ≥ 3 measurements). Fixed-effect estimates with 95% confidence intervals (CIs) were reported, and the relative change per + 1 standard deviation in time (% change per + 1 SD) was used as the effect size. Variance components of the random effects and the intraclass correlation coefficient (ICC) were also presented. Threshold-related supplementary analyses included the definition of the “threshold-crossing rate” (whether a patient crossed a threshold during hospitalization) and the “reclassification rate between first and last measurements” (proportion of patients whose first and last values fell on opposite sides of a threshold). In addition, “upper-bound simulations” (assuming an overall 20% or 30% decline after the first measurement) were conducted to evaluate threshold robustness. Based on the overall median absolute difference between adjacent measurements (≈ 50 cells/µL), “near-threshold zones” were defined around the 100 and 300 cells/µL cutoffs (50–150 and 250–350 cells/µL, respectively) to propose 24–48 h retesting recommendations and to construct a “threshold–action suggestion” chart. Sensitivity analyses were performed according to the timing of the first measurement (≤ 1 day and ≤ 12 h after admission). Data analysis Statistical analyses were performed in Python 3.13 (pandas, numpy, statsmodels). Continuous variables are summarized as median (IQR) or mean (SD), and categorical variables as n (%). Group comparisons used the chi-square test for categorical variables and the Mann–Whitney U test for continuous variables; two-sided α = 0.05. Longitudinal change was modeled with linear mixed-effects models (statsmodels MixedLM) using ln(1 + eosinophils [cells/µL]) as the outcome and standardized hospital day (SHD; centered and scaled so 1 unit = + 1 SD) as the predictor. We fitted a random-intercept model in the full cohort and, among patients with ≥ 3 measurements, an additional random-slope model. Fixed effects are reported as % change per + 1 SD = (exp(β) − 1) × 100%, with Wald z, 95% CI, and two-sided normal-approximation P values; variance components are shown for random effects. Within-patient variability was quantified per patient by CV% (SD/mean×100), within-patient SD (wSD), and the median absolute change between time-ordered consecutive tests (|Δ|). Threshold analyses evaluated crossing and first–last reclassification at 100 and 300 cells/µL, with decline simulations (− 20%/−30%) applied to all post-first values. Decision bands around thresholds were defined as ± 50 cells/µL (50–150; 250–350), chosen from the cohort median consecutive |Δ| ≈ 50 cells/µL; a ± 40 sensitivity analysis showed similar patterns. Results From January 1, 2018, to December 31, 2023, a total of 3,311 patients with AECOPD who had at least one complete blood count were admitted, with a mean age of 70.7 ± 9.8 years, and 92.3% were male. Among these patients, 1,548 had at least two blood eosinophil measurements during the acute exacerbation, including 1,440 men and 108 women. The baseline characteristics of the patients are presented in Table 1. Table 1Patient characteristics of the study populationPatient characteristicN = 3311Sex Male, n (%)3057(92.3) Female, n (%)254(7.7)Age (years), n (%) 40–4958(1.8) 50–59389(11.7) 60–691047(31.6) 70+1817(54.9)Mean age, years (SD)70.7(9.8)Blood values, mean (SD) White blood cells, (×109/L)8.9(6.5) Neutrophils%70.0(17.8) Neutrophils count, (×109/L)6.7(4.7) Platelets, (×109/L)241.8(95.5) Hemoglobin, g/L125.6(24.3) Total cholesterol, mmol/L4.3(1.1) Total protein, g/L63.1(7.8) Albumin, g/L36.9(5.7)Comorbidities, n (%) Hypertension1292(39.0) Arrhythmia205(6.2) Ischemic heart disease373(11.3) Congestive heart failure194(5.9) Diabetes386(11.7) Previous tuberculosis208(6.3) Patient characteristics of the study population Distribution of blood eosinophils The distribution of blood eosinophils is shown in Fig. 1, demonstrating a right-skewed pattern among patients with AECOPD. Blood eosinophil levels at various cutoff points are presented in Table S1. In the AECOPD population, the median blood eosinophil count was 100 cells/µL (IQR: 20–200), with a median of 100 cells/µL (IQR: 20–200) in men and 100 cells/µL (IQR: 20–192) in women, showing no statistically significant difference between the two groups (p = 0.950). Blood eosinophil distribution in AECOPD patients. Histogram of eosinophil counts (cells/µL) at the first available in-hospital measurement for all included patients Overall variability A linear mixed-effects model was fitted with standardized hospital days (SHD) as a fixed effect and patients as a random effect. The results showed that ln(1 + EOS) increased only mildly over time. In the full-sample random-intercept model, the coefficient for SHD was β = 0.021 (SE = 0.0019, z = 11.52, P < 0.001), corresponding to a 2.16% increase in the expected value of (1 + EOS) per + 1 SD of hospitalization time (95% CI: 1.78%–2.51%) (Fig. 2). Fixed-effect estimates are presented in Table 2. Trajectories of blood eosinophil counts during hospitalization. Trajectories of ln(1 + eosinophils [cells/µL]) over standardized hospital day (SHD). Light lines show individual patients; the dark line shows the mixed-model fixed effect Table 2Fixed effects from mixed-effects models of ln(1 + eosinophils) vs. standardized hospital day (SHD)ModelTermBeta (SE)95% CIP-value% change per + 1 SDRandom intercept onlyIntercept0.126 (0.003)0.120–0.132< 0.001–Random intercept onlyStandardized hospital days0.021 (0.002)0.018–0.025< 0.0012.16%Random intercept + slopeIntercept0.109 (0.002)0.105–0.113< 0.001–Random intercept + slopeStandardized hospital days0.016 (0.007)0.003–0.0290.0141.61%Outcome = ln(1 + eosinophils [cells/µL]). SHD is centered and standardized; one unit equals + 1 SD. “% change per + 1 SD” = (exp(β) − 1) × 100%. Wald z and 95% CI are reported. The random-slope model includes patients with ≥ 3 measurements and SHD variability Fixed effects from mixed-effects models of ln(1 + eosinophils) vs. standardized hospital day (SHD) Outcome = ln(1 + eosinophils [cells/µL]). SHD is centered and standardized; one unit equals + 1 SD. “% change per + 1 SD” = (exp(β) − 1) × 100%. Wald z and 95% CI are reported. The random-slope model includes patients with ≥ 3 measurements and SHD variability In the random-intercept plus random-slope model (subsample of patients with ≥ 3 measurements), the fixed-effect coefficient was β = 0.016 (SE = 0.0065, z = 2.46, P = 0.014), corresponding to a 1.61% increase (95% CI: 0.33%–2.88%). Both models showed consistent directions with small effect sizes, suggesting that overall mean eosinophil counts changed little with hospitalization time. The distribution of patient-specific slopes is illustrated in Fig. 3. Distribution of patient-specific slopes in the random slope model. Distribution of patient-specific slopes β (per + 1 SD in SHD) from the random-slope model; dashed line indicates β = 0 Within-patient variability and heterogeneity Among patients with repeated measurements, EOS exhibited substantial short-term fluctuations. Patient-level variability metrics are summarized in Table 3. The median within-patient coefficient of variation (CV%) was 76.6% (IQR: 41.2–122.4%); when restricted to those with the first test performed within ≤ 1 day of admission, the value was 78.1% (IQR: 44.7–114.7%). The median within-patient weighted standard deviation (wSD) was 57 cells/µL (57 cells/µL for first test ≤ 1 day), and the median absolute difference between two consecutive measurements was 50 cells/µL (55 cells/µL for first test ≤ 1 day). Table 3Within-patient variability of eosinophil counts (overall and early-sampling subgroups)MetricOverallEarly ≤ 1 dayEarly ≤ 12 hCV% median (IQR)76.6 (41.2–122.4)78.1 (44.7–125.9)77.1 (44.7–114.7)wSD median (cells/µL)575757Median |Δ| (cells/µL)505555Crossing ≥ 100 (%)45.746.646.2Crossing ≥ 300 (%)22.722.522.9First–last reclassification ≥ 100 (%)32.933.332.6First–last reclassification ≥ 300 (%)16.016.417.6ICC (intercept-only)0.520.520.42CV% = SD/mean × 100%. wSD = within-patient SD (cells/µL). “Median |Δ|” is the median absolute difference between time-ordered consecutive in-hospital measurements per patient. “Crossing” = any in-hospital value moving from < threshold to ≥ threshold (or vice versa) at 100 or 300 cells/µL. “First–last reclassification” = change in stratum between the first and last in-hospital measurements. “Early ≤ 1 day/≤12 h” refer to timing of the first test after admission. ICC from an intercept-only mixed model Within-patient variability of eosinophil counts (overall and early-sampling subgroups) CV% = SD/mean × 100%. wSD = within-patient SD (cells/µL). “Median |Δ|” is the median absolute difference between time-ordered consecutive in-hospital measurements per patient. “Crossing” = any in-hospital value moving from < threshold to ≥ threshold (or vice versa) at 100 or 300 cells/µL. “First–last reclassification” = change in stratum between the first and last in-hospital measurements. “Early ≤ 1 day/≤12 h” refer to timing of the first test after admission. ICC from an intercept-only mixed model The random-intercept model yielded an intraclass correlation coefficient (ICC) of 0.52 (0.52 for first test ≤ 1 day), indicating that between-patient differences and within-patient variability contributed comparably. The distribution of CV% is shown in Fig. 4. A scatterplot depicting heterogeneity with patient-specific slopes versus CV% demonstrated that most patients clustered in the region of “slopes near zero but relatively high CV%” (Figure S1). Within-patient variability of eosinophil counts (CV%). Within-patient coefficient of variation (CV%) of eosinophil counts during hospitalization Threshold crossing and reclassification Using 100 and 300 cells/µL as thresholds, both “threshold crossing” during hospitalization and “first-to-last reclassification” were found to be common. At baseline, 45.7% of patients crossed the 100 cells/µL threshold, and 22.7% crossed the 300 cells/µL threshold; the proportions of first-to-last reclassification were 32.9% and 16.0%, respectively (Table S3).In “upper-bound” scenario simulations (assuming all post-initial measurements decreased by 20% or 30%), these proportions declined only slightly: with a 20% decrease, threshold crossing was 42.9% for 100 cells/µL and 18.2% for 300 cells/µL, while reclassification was 30.6% and 13.3%, respectively. With a 30% decrease, the corresponding values were 41.9% and 16.2% for threshold crossing, and 30.0% and 12.3% for reclassification (Figures S2–S3).These findings suggest that single EOS measurements close to commonly used thresholds are subject to considerable instability. Threshold–action recommendation bands Given the observed instability, we developed threshold–action decision bands (Figure S4) to highlight near-threshold risk zones. Near-100 is defined as 50–150 cells/µL and near-300 as 250–350 cells/µL (i.e., ± 50 around 100/300, chosen from the median consecutive |Δ| ≈ 50 cells/µL). Values within these bands should be re-tested in 24–48 h. Values ≥ 350 and ≥ 500 cells/µL indicate high and very high ranges, respectively, and should be interpreted in clinical context. This figure operationalizes retesting based on the clustering of reclassification near cut-offs; it does not propose new diagnostic thresholds. Discussion This study quantified the distribution characteristics and short-term dynamics of blood eosinophils in hospitalized patients with AECOPD. The main findings were as follows: (1) EOS demonstrated a right-skewed overall distribution; (2) linear mixed-effects models indicated only a mild increase in overall levels with hospitalization days, with a small effect size; (3) within-patient variability was substantial, with a median CV% approaching 80%, and the intraclass correlation coefficient suggested that both between-patient differences and within-patient fluctuations contributed considerably; (4) threshold crossing and first-to-last reclassification were common, and these phenomena decreased only slightly even under simulated 20–30% overall reductions in EOS levels. These results suggest that, during acute exacerbations, a single EOS measurement is insufficient to serve as a reliable basis for patient stratification. First, this study confirmed that the distribution of blood EOS in patients with AECOPD is right-skewed, consistent with previous findings in population-based studies and in stable COPD [23]. For such distributions, the use of arithmetic means may overestimate overall levels; therefore, medians or geometric means provide a more accurate representation of EOS characteristics in the population. Second, using linear mixed-effects models, we found that the overall trend of EOS levels during hospitalization showed little change. This indicates that, at the population level, EOS remained relatively stable on average; however, short-term within-patient fluctuations were highly pronounced. Measures including CV%, wSD, and absolute differences between consecutive tests all demonstrated that most patients experienced substantial variability during hospitalization, with some crossing the commonly used thresholds of 100 or 300 cells/µL within just a few days. Consistent with the findings of Citgez and Cui et al., our results further suggest that EOS should be regarded as a dynamic marker rather than a fixed dichotomous threshold in the acute setting [21, 22]. Third, this study employed an “upper-bound simulation” approach to further examine the impact of systemic corticosteroids on EOS variability. Previous evidence has demonstrated that corticosteroids can markedly reduce EOS levels; however, even under hypothetical scenarios assuming an overall 20% or 30% decline after the first measurement, threshold-crossing and reclassification rates remained common and decreased only slightly. This suggests that the instability of EOS cannot be fully explained by corticosteroid use alone, but may also be influenced by disease progression, infection status [30], and individual biological variability. Fourth, at the clinical level, these findings suggest that treatment decisions regarding long-term ICS use should not be based on a single EOS measurement during acute exacerbations, particularly when values are near the 100 or 300 cells/µL thresholds. Instead, trends or short-term retesting should be considered. Our supplementary analyses also developed a “threshold–action recommendation band,” providing a visual reference for clinical practice: when EOS values fall within near-threshold zones, retesting within 24–48 h is recommended to reduce misclassification caused by short-term variability. This study has several limitations. First, as a single-center retrospective study, the conclusions require validation in other regions and prospective cohorts. Second, we were unable to precisely capture the dosage and initiation timing of corticosteroid use in all patients. Although the upper-bound simulations suggested that corticosteroid therapy alone cannot fully explain the observed variability, the impact of pharmacologic interventions warrants further investigation. Finally, this study focused only on short-term dynamics during hospitalization, and the findings cannot be directly extrapolated to the stable phase or to long-term follow-up. Conclusion Blood eosinophil levels in hospitalized AECOPD patients show limited overall change but marked within-patient variability, making single measurements at common thresholds unreliable. EOS should be interpreted dynamically, with retesting or trend assessment recommended to guide clinical decisions during acute exacerbations. Supplementary Information Supplementary Material 1. Acknowledgements The authors thank all participants for contributing to this work. Abbreviations Inhaled corticosteroids Chronic Obstructive Pulmonary Disease Acute exacerbations of chronic obstructive pulmonary disease Interquartile spacing Standard Deviation Forced Expiratory Volume in one second Forced Vital Capacity Authors’ contributions Chen QD and Ji W contributed equally to data analysis and manuscript writing. Wang HY was responsible for the analysis of results. Lin JH and Liu TW contributed to the revision and proofreading of the manuscript. Lin JL and Chen X were responsible for data collection and organization. Chen XY designed the study.All authors read and approved the final manuscript. Funding The study was supported by The Natural Science Foundation of Fujian Province (2023J01741); Joint funds for the innovation of science and technology, Fujian province(2023Y9262). Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study was approved by the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (Ethics approval number: 536, 2024). The study was conducted in accordance with the principles of the Declaration of Helsinki. Prior to completing the survey, participants were informed that participation was voluntary and that their data would remain anonymous. Participants provided informed consent for their participation in the survey and for the publication of the results. Informed consent was obtained from all subjects and/or their legal guardians. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Qidong Chen and Wei Ji contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.