The NHANES employs a sophisticated multi–tiered sampling methodology to select a representative group from the US civilian non–institutionalized population. We utilized data from NHANES interviews, questionnaires, and laboratory tests. The National Centre provided NHANES Health Statistics data for 1999–2018, including mortality data until December 31, 2019 (21). All analyses of NHANES 1999–2018 data used mobile examination center (MEC) exam data weights to consider stratification and clustering, employing weights wtmec4yr and wtmec2yr at different years, respectively.

To accurately identify our target population, we used three different diagnostic criteria that can be implemented in the NHANES database. These identification strategies have been widely used in some literature (22, 23), and our constraints added smoking history and COPD–related medicine use to make the population more accurate. Here, we describe each selected criterion: (1) Forced Expiratory Volume in the first 1.0 second/Forced Vital Capacity <0.7 after administration of bronchodilators (1); (2) Diagnosed with COPD by a doctor or other health professional; (3) Participants ≥40 years of age taking COPD–related therapeutic drugs (including phosphodiesterase–4 inhibitors, mast cell stabilizers, bronchodilator combinations, etc.) and have any of the following conditions: have smoked at least 100 cigarettes in life, or diagnosed with emphysema or chronic bronchitis, or still have chronic bronchitis.

We gathered data on blood eosinophil count, neutrophil count, lymphocyte count, NLR, use of ICS in patients with COPD, and patient survival status. Absolute cell counts were presented as data. Blood eosinophil count reference guide values were stratified from low to high into four categories, i.e., <100, 100–200, 200–300, and ≥300 cells/µL (1), with <100 cells/µL used as the reference group. For continuous variables such as blood eosinophil count, lymphocyte count, and NLR, these were divided into quartile 1 to quartile 4 (Q1–Q4) groups based on quartile levels from low to high, and grouped as necessary for discussion; quartile 1 was used as a reference. The primary endpoint in this study was all–cause mortality.

Moreover, several socio–demographic attributes and other covariates were gathered as potential confounders, including age (40–50; 50–60; 60–70; 70–80; ≥80 years), gender, race, smoking, medical health status (in order from good to poor: Excellent, Very good, Good, Fair, or Poor), and prescribed medication usage (ICS).

Demographic characteristics and laboratory data of COPD participants with varying outcomes were described and compared. A t–test was utilized to examine the differences in continuous variables, including age, follow–up time, blood counts of interest, and NLR, expressed as mean (standard error, SE). The χ² test was employed to examine differences in categorical variables, including sex, race, smoking status, health status, blood eosinophil count subgroup, and usage of ICS, presented as unweighted sample size (n) and weighted percentages (%).

Kaplan–Meier survival curves were utilized to depict participant survival (24). Follow–up durations were computed from the interval between the interview date and the last follow–up time (December 31, 2019), or the date of death. To validate the association between leukocyte–based inflammatory markers and all–cause mortality, or specific mortality events in patients with COPD, a series of multivariate weighted Cox proportional hazard regression analyses were performed, applying stepwise adjustment for potential confounders for sensitivity analysis (25), with results expressed as Hazard ratio (HR) and 95% CI. Additionally, p for trend was calculated for the cell count subgroup.

The restricted cubic spline (RCS) was executed using the R package “rms” (version 6.3–0). This can depict non–linear relationships and trends between the leukocyte–associated counts and COPD death risk. Log (loge) hazard was used to illustrate the relationship between the level of leukocyte–based inflammatory markers and the risk of death. The population was categorized based on the application of ICS, and separate RCSs were produced to explore the optimal response values of the relevant biomarkers to ICS.

We also carried out a series of sensitivity analyses: (1) Blood eosinophil count levels were divided into four groups based on GOLD guideline recommendations for more accurate clinical comparison; (2) Necessary analyses were performed by quartile grouping of the remaining relevant leukocyte counts to characterize the overall sample better; (3) To examine the association of leukocyte–based inflammatory markers with any observed risk of mortality outcomes, further adjustments were made for age, gender, race, smoking status, and medical health status.

All analyses were conducted using R software (version 4.2.1), and a two–sided p value of <0.05 was considered statistically significant.

All pertinent data were sourced from the open NHANES database, and no recruitment of patients with COPD was necessary. The NHANES protocol received approval from the Institutional Review Board of the NCHS and CDC, and informed consent was obtained from all participants. For further information on NHANES procedures, methodologies, and the approval of the NCHS Ethics Review Board, see (https://www.cdc.gov/nchs/nhanes/index.htm, accessed October 7, 2023). Hence, this study did not require an ethical review.

As shown in Figure 1 , the final weighted sample for analysis encompassed 1,715 COPD participants, aged between 40 and 85 years, representing an extrapolated total of 6,976,232 non–institutionalized US residents. Of these participants, a total of 556 all–cause deaths were recorded in the survival data and 716 were on ICS medication, underlining the significance of ICS in managing COPD.

Weighted data were used for statistical analyses and calculations. The average follow–up duration was 84.88 months, and the mean age was 62.09 years, with males constituting 50.77%. A significant portion of these COPD participants, comprising 83.38%, were smokers. 40.28% of smokers have a regular smoking frequency, and the average smoking volume is 34.69 pack–years. It was reported that the majority of COPD participants (35.39%) experienced good overall health in the past 30 days, based on recent illness, indicating a median health status among most participants.

The survival and death groups showed significant differences (p < 0.0001) in age, smoking status, health status, and follow–up duration, as presented in Table 1 . Specifically, among smokers with a regular smoking frequency, the death group exhibited a significantly higher average number of cigarettes smoked (41.60 pack–years, SE=3.64) compared to the survival group (32.85 pack–years, SE=1.21; p = 0.02). Regarding blood cells, although no significant difference was found in the overall blood eosinophil count, notable differences were present in the subgroups, and other leukocyte–related counts, including NLR, showed significant differences.

Medication–wise, a smaller proportion of participants in the survival group used ICS compared to the death group (36.78% vs 51.41%, p < 0.0001), and the proportion of mortality in the ICS use group was significantly higher compared to the non–ICS use group (33.14% vs 21.42%, p < 0.0001). However, participants using ICS had significantly longer follow–up (92.49 months, SE=3.62 vs 79.67 months, SE=2.59; p = 0.003) and higher NLR (2.79, SE=0.08 vs 2.52, SE=0.05; p = 0.01) compared to those not using ICS. Notably, the ICS use group also displayed a higher proportion of participants over 70 years, with 31.39% versus 25.52% in the non–ICS use group (p < 0.0001).

We generated Kaplan–Meier survival curves with risk tables, and the risk tables present weighted survival rates with a follow–up time of 200 months as the endpoint, excluding extreme data.

In the Kaplan–Meier survival curves, the death risk among the COPD population was lower for patients with blood eosinophil counts ≥100 cells/μL compared to those with <100 cells/μL (53% vs 18%) ( Supplemental File 1 ). Population survival increased with rising lymphocyte counts and decreased progressively with increasing neutrophil counts and NLR ( Supplemental File 1 ). The crossing of these Kaplan–Meier survival curves after grouping indicates that survival time was not solely influenced by different cell count values, suggesting the presence of confounding factors and other significant influences not considered or leading to baseline incomparability. We analyzed this issue by employing multivariate weighted Cox proportional hazard regression models. Moreover, only the two more characteristic cell subgroups, blood eosinophils and NLR were investigated in the subsequent study.

We performed a univariate weighted Cox proportional hazards analysis for age, gender, race, smoking, health status, and ICS application in the included populations, and calculated HRs, 95% CIs, and p values. After taking into account follow–up time and survival ending, the risk of fatal events increased by 8% for each additional year of age in patients with COPD. There was no statistical difference in the risk of death between those with and without ICS application, demonstrating that patient application of ICS was not an independent risk factor ( Table 2 ).

Weighted Cox regression analysis was adjusted. Three models were developed: the Crude Model for the univariate Cox proportional hazards analysis; Model 1, which adjusts for age, gender, and race; and Model 2, which additionally adjusts for smoking and health status. The Crude Model showed significant differences in blood eosinophil count subgroups with large values compared to the <100 cells/μL group (p< 0.001). For NLR, using the quartile 1 group as a reference value, the risk of death increased by 179% (HR=1.79; 95% CI=1.28–2.52; p<0.001) and 286% (HR=2.86; 95% CI=2.06–3.95; p<0.0001) in the quartile 3 and quartile 4 group, respectively ( Table 3 ).

Adjusted analyses using a multivariate weighted Cox proportional hazards regression model showed that neutrophil count, and NLR all had statistically significant changes in risk of death as unit values changed in Models 1–2, and lymphocyte count regained significance to all–cause mortality after adjustment in Model 2. Each subgroup of blood eosinophil count demonstrated the ability to reduce the risk of death with increasing levels of values in Model 2, almost reaching or exceeding a 50% reduction for each group compared with the reference group, with an overall linear trend for the groups (p for trend=0.025). Changes in HR for quartiles Q1–Q4 of NLR in model 2 (1, 0.99, 1.39, 1.82) indicate a trend of progressively increasing risk of death with increasing NLR values (p for trend <0.0001).

Moreover, we analyzed the top three COPD–related causes of death: chronic lower respiratory diseases (124 cases), malignant neoplasms (124 cases), and heart diseases (116 cases). These three groups accounted for 65.47% of the overall causes of death. Through the models, we found that blood eosinophil count, neutrophil count, and NLR were independent risk factors for mortality from chronic lower respiratory diseases in patients with COPD. Neutrophil count was an independent risk factor for mortality from heart disease in patients with COPD ( Supplemental File 1 ).

We produced the RCS to describe the relationship between continuous accurate type variables and the risk of outcome occurrence. To fit the curve model well, we selected the number of nodes to range from 3 to 5 (p for non–linearity <0.05) and adjusted for age, gender, ethnicity, smoking, health status and follow–up time and grouped according to the application of ICS.

In the blood eosinophil counts ( Figure 2A ), five RCS nodes were selected (p for non–linearity <0.02). With ICS administration, the mortality risk curve first showed a decline, reaching the lowest point at 120 cells/μL, then rose to a peak at 240 cells/μL, and then plateaued at around 500 cells/μL, reflecting the levels seen at 100 and 150 cells/μL. Conversely, the risk of death in COPD participants not on ICS was lowest at around 240 cells/μL, and then continued to rise with increasing counts.

The non–linear relationship in neutrophil count followed a reverse L–shape, with the trend changing after 3,800 cells/μL, switching from a decreasing trend in risk of death to an increasing trend. The ability of ICS to reduce the risk of death also became apparent after 3,800 cells/μL ( Figure 2B ). The lymphocyte count was non–linearly L–shaped, declining slowly until 2400 cells/μL and then essentially flat ( Figure 2C ). The ability of ICS to slow down the trend in the risk of death was observed at all stages of increased risk of death in these cell counts, although no significant improvement was seen compared to the non–ICS use group.

For NLR ( Figure 2D ), the number of RCS nodes selected was four (p for non–linearity <0.01). The ICS application curve shows a reduced risk in the lower range, followed by a significant increase when reaching a minimum risk ratio of about 1.75. However, beyond a value of approximately 5, the curve shows a decreasing trend. In the non–ICS use group, the mortality risk consistently increased with increasing values, with no notable deceleration. A statistically significant difference between the two groups emerged after the value of 4.79 (p = 0.03), with ICS use associated with a 51% reduction in mortality risk (HR =0.49; 95% CI =0.27–0.86; p = 0.014).

For the RCS plots showing the relationship between leukocyte–related counts and the risk of death in COPD participants not grouped by ICS use can be seen in Supplemental File 1 .