Lanzhou is located in the interior of northwest China; it is the capital city of the Gansu Province. The city has four main districts (Chengguan, Qilihe, Xigu, Anning). The geographical location, urban area, population, and climate are reported in our previous articles (18, 19).

In 2015, most hospitals in Lanzhou City had a complete electronic medical record system. However, in 2020, upon the emergence of the novel coronavirus epidemic, hospitalization data were abnormal. Hence, data on daily COPD hospital admissions from 1 January 2015 to 31 December 2019 were collected from 25 hospitals, which have complete electronic medical record systems. Hospitals’ locations are shown in Figure 1. Data included the following: age, gender, residential address, date of hospital admission, and principal diagnosis. Inclusion criteria include: Patients diagnosed with COPD by the attending physician, diagnosed according to the Chronic Obstructive Pulmonary Disease (GOLD) criteria, with risk factors and/or clinical symptoms, and lung function indicating persistent airflow limitation (FEV1/FVC < 0.70 after bronchodilator use). The International Classification of Diseases, 10th Revision (ICD-10) codes is J44. Exclusion criteria include: (1) patients with a residential address outside of the urban districts in Lanzhou and (2) incomplete records. A total of 18,275 eligible subjects, aged 28 to 104, were included.

Data on hourly air pollution from 1 January 2015 to 31 December 2019 were collected from five national environmental monitoring stations in the city of Lanzhou, including PM_2.5, PM₁₀, SO₂, NO₂, CO, and O₃. Figure 1 shows the locations of the five monitoring stations for air pollutants. Our previous studies reported the calculated daily 24-h mean concentrations of PM_2.5, PM₁₀, SO₂, NO₂, CO, and the maximum daily 8-h moving average of O₃ (O₃8h) (18, 19).

The daily average temperature (T, °C) and daily average relative humidity (RH, %) were obtained from the China Meteorological Data Service Center.¹

According to previous studies (10, 20), the association between air pollution and COPD outcomes is potentially non-linear and lagged, so time series analysis of the effects of air pollutants on COPD using the distributed lag non-linear model (DLNM), which can describe both non-linear exposure–response curves and lag–response relationships (21).

The constructed model is shown below: Log[E(Yt)] = α + βXt,l + ns(Time, df = 7/year) + ns(Temt, df = 3) + ns(RHt, df = 3) + Dow + Hol

t: the time observed (days); E(Yt): the expected mean of number of COPD hospital admissions at t day; Xt,l: cross-basis matrix obtained by series of transformations for daily mean concentration of a certain air pollutant, l: the maximum lag days; β: the coefficient of the matrix; ns: the function of natural cubic spline, the ns was used to control the non-linear confounders such as long-term and seasonal trends of the daily COPD hospital admissions, the average temperature, and the average relative humidity; Time: the time variable, df: the degree of freedom; Temt: the average temperature on day t; RHt: the average relative humidity on day t; Dow: the day of the week effect, and Hol: the public holiday effect.

We analyzed the relationship between air pollutants and the total COPD hospital admissions population by using DLNM, and also stratification was analyzed by gender (male and female), age (<65 years and ≥ 65 years old), and season (warm season and cold season). The cold season is from November to March (the heating season of Lanzhou City), and the warm season is from April to October (the non-heating season). Moreover, to test the robustness of the model, the lag day with the largest estimated effect in the single-pollutant model was chosen to perform the sensitivity analysis. (1) The two-pollutant models, except PM_2.5 and PM₁₀. (2) The df of the ns function of Time was changed from 6 to 10. The relative risk (RR) and corresponding 95% confidence interval (CI) were calculated to estimate the effect of each 10 μg/m³ (CO, 1 mg/m³) increase for each air pollutant on COPD. The basic statistical analysis was conducted using SPSS software, version 22.0. The regression analysis of DLNM was conducted using R software version 3.6.3, with its “dlnm” package. Details of the statistical analysis are reported in our previous articles (18, 19).

Tables 1, 2 show the descriptive statistics on COPD hospital admissions and the air pollutants in Lanzhou City during 2015–2019. A total of 18,275 COPD hospital admissions were collected, including 11,850 males and 6,425 females, 3,516 cases aged <65 years, and 14,759 cases aged ≥65 years. The average daily hospitalization of COPD in the cold season was 11.1 ± 6.9 cases/day, which was much higher than in the warm season (9.0 ± 5.8 cases/day). Meanwhile, the average concentrations of PM_2.5, PM₁₀, SO₂, NO₂, and CO in cold seasons were also higher than in the warm seasons (all p-values<0.0001). Whereas, there was no statistical difference in the average concentrations of O_3-8h between the cold and warm seasons (p = 0.894).

The Spearman rank correlation (rs) between daily COPD admissions, air pollutants, and meteorological factors is shown in Table 3. Daily COPD admissions were significantly positively correlated with daily PM, SO₂, NO₂, and CO concentrations, but significantly negatively correlated with T and RH. Except for SO₂ and O_3-8h, there were significant correlations among air pollutant concentrations, with the highest rs between PM_2.5 and PM₁₀ (0.876) being the highest. Air pollutants other than O_3-8h were negatively and significantly correlated with T, as well as RH.

Figure 2 and Supplementary Table S1 show the lag-response plots of the association between air pollutants and COPD hospital admissions. In addition to O_3-8h, all air pollutants had a significant positive correlation with COPD hospital admissions. The greatest effect of air pollutants on COPD admissions was found at lag07. For 10 μg/m³ increase in PM_2.5, PM₁₀, SO₂, NO₂, and 1 mg/m³ increase in CO at lag 7 day, the RR95%CI of COPD admissions were 1.048 (1.030, 1.067), 1.008 (1.004, 1.013), 1.091 (1.048, 1.135), 1.043 (1.018, 1.068), and 1.160 (1.084, 1.242). However, for 10 μg/m³ increase in O_3-8h, the RR95%CI was 0.983 (0.970, 0.996) at lag 7, and 1.023 (1.001, 1.045) at lag 5. This was obvious that air pollutants (except O_3-8h) may had an adverse effect on COPD hospital admissions, and the adverse effect of PM_2.5 was higher than that of PM₁₀, with CO having the strongest harmful effect.

Figure 3 shows the exposure–response curves of the effects of air pollutants on COPD hospital admissions. Based on the results of Figure 2 and Supplementary Table S1, we focused on COPD admissions related to air pollutants at lag 7 day (O_3-8h at lag 5 day). The six exposure–response curves were approximately linear, with the curves of PM_2.5, PM₁₀, SO₂, NO₂, and CO showing a significant positive relationship and no thresholds, but the exposure–response curve of O_3-8h was not statistically significant.

Figure 4 shows the RRs (95% CI) of COPD admissions associated with air pollutants stratified by age. In those aged <65 years, only PM_2.5, PM₁₀, NO₂, and CO were significantly and positively associated with COPD hospital admissions. In those aged ≥65 years, COPD hospital admissions were negatively correlated with O_3-8h but positively with other air pollutants. Simultaneously, the harmful effect of particulate matter on those aged <65 years was slightly higher than those aged ≥65 years; however, the effect of the gaseous pollutants on those aged <65 years was slightly lower than those aged ≥65 years.

Figure 5 shows a gender-stratified analysis of the effects of air pollutants on COPD. As shown in the figure, all air pollutants had an impact on COPD hospital admissions of different genders. In addition to O_3-8h, there was a weakly negative correlation between O_3-8h and COPD hospital admissions of different genders, and other pollutants were positively correlated with COPD of different genders. That is, air pollutants (except O_3-8h) have adverse effects on COPD of different genders, and the impact on females was slightly higher than that on males.

Stratified analyses according to season (Figure 6; Supplementary Tables S2, S3) revealed that air pollutants (except O_3-8h) had statistically adverse effects on COPD hospital admissions during the cold season (heating period), while no statistical effects were observed in the warm season. COPD had a significant positive correlation with O_3-8h at lag 3–4 days, but a negative correlation with O_3-8h at lag 6–7 days during the cold season, and there was no correlation in the warm season. In conclusion, the influence of air pollutants on COPD in cold seasons was greater than that in warm seasons.

After further adjustment of the df for calendar time (6–10 df per year) (Figure 7; Supplementary Table S4), the estimated effects of air pollutants (except O_3-8h) on COPD hospital admissions changed slightly but the effects were still significant. Two-pollutant models (Figure 8; Supplementary Table S5) showed that, for PM, although the estimated effects changed after the introduction of other air pollutants, the change was small, and the impact on COPD hospital admissions was still significant. For SO₂ and NO₂, after further adjustment of other air pollutants (except PM_2.5 and CO), the relationships between SO₂, NO₂, and COPD hospital admissions were consistent with the main model, but the estimated effects changed slightly. For CO, the effects on COPD hospital admissions were still significant after the introduction of other air pollutants (except PM_2.5 and SO₂). The main model (single-pollutant model) was statistically significant, whereas some two-pollutant models were insignificant; this may be related to the co-linearity between different air pollutants (22), such as the high rs (Table 3) between PM_2.5 and SO₂ (rs = 0.649), PM_2.5 and CO (rs = 0.698), and SO₂ and CO (rs = 0.777). For O_3-8h, after further adjustment of other air pollutants, the weakly positive correlation between O_3-8h and COPD hospital admissions was insignificant. Overall, the above results suggested that DLNM was robust, and the results between COPD hospital admissions and air pollutants (except O_3-8h) were stable.

Previous studies reported a similar negative correlation between PM and COPD. Such as, an ecological study (10) in Pakistan, found that PM_2.5 contributes 31.41% attributable-proportion to COPD in adults age 30+. Another nationwide ecological study in Korea (23), that used generalized additive models, found significant associations between COPD (emergency department visits or hospital admissions) and PM; the RR95%CI for COPD for every 10 μg/m³ increase in PM_2.5 and PM₁₀ were 1.013 (1.004, 1.022) and 1.014 (1.008, 1.020), respectively. A study in Poland, using DLNM as in this paper (24), also found that the proportions (95%CI) of COPD hospitalizations attributable to air pollution were 7.61% (1.27, 13.49%) for PM_2.5 and 9.08% (3.10, 15.08%) for PM₁₀. An ecological study in Guangdong, China showed that declined attributable hospital admissions for COPD may be associated with the reduction in PM concentrations. Other studies with different methods in different regions have found similar results, such as a cohort study in the UK (11) and in southern China (25), which were based on 265,506 and 580,757 adults, respectively, and found that PM exposure could affect the progression from free of chronic lung disease to incident COPD morbidity and death. A case-crossover study (26) showed that short-term PM_2.5 exposure was associated with increased odds of exacerbations in Canadians with COPD, further heightening the awareness of non-infectious triggers of COPD exacerbations. A meta-analysis (27) of 37 studies showed that a 10 μg/m³ increase in PM_2.5 was associated with a 1.6–3.4% increase in COPD-related emergency department visits and hospital admissions. A randomized controlled trial (28) in the United States showed that a reduction in PM_2.5 caused clinically significant improvement in respiratory health for individuals with COPD. It is reported that PM_2.5-induced cytokine release and oxidative stress are the main mechanisms leading to COPD. Meanwhile, the bacteria, eukaryotes, and viruses in PM may directly cause neutrophilic inflammation or break the microorganism balance, contributing to the development and exacerbation of COPD (29). On the other hand, air pollution may make COPD patients more susceptible to respiratory viral infections, especially influenza virus A (30). Mice experiments also showed that acute PM2.5 exposure increased airway inflammation (31). A human in vitro cell culture experiment (32) showed that epithelial remodeling and dendritic cell dysfunction might accelerate inflammation after PM exposure in COPD.

SO₂ mainly comes from the combustion of fossil fuels or emissions from industrial activities. The rats experiment has revealed that (33) SO₂ affects the airway inflammatory and immune responses of the rats and enhances the susceptibility to ovalbumin by aggravating inflammatory responses in the lungs, upregulating proinflammatory cytokine expression, and causing a Th1/Th2 imbalance, which might contribute to the increased risk of respiratory system disease. SO₂ can penetrate deep into lung tissue and be converted to hydrosulfite, which interacts with sensory receptors, causing respiratory irritation, bronchoconstriction, mucus production, increasing the sensitivity of the respiratory tract to sensitizing substances, etc., and can also induce inflammation and oxidative stress in the whole body, including the lung (34). Prospective cohort data from the Australian Longitudinal Study on Women’s Health also showed that, after controlling for covariates, air pollutants (PM, SO₂, NO_X, and CO) modeled individually were significantly associated with the risk of COPD. When modeled together, only SO₂ remained significantly associated with COPD (35). Evidence from Guangdong, China, showed that a decrease in hospital admissions for COPD is associated with a reduction in concentrations in Guangdong (36). This suggests that improving ambient air quality, a modifiable risk factor, has been shown to reduce COPD hospitalizations. In recent years, with the large-scale deployment of flue gas desulfurization at power plants, strict control of SO₂ industrial emissions, as well as engine technology innovation and the use of clean fuels and other effective measures in Lanzhou City (37, 38) has been enforced. As a result of this, from 2015 to 2019, SO₂ concentration in Lanzhou City showed a significant decline trend; furthermore, the average concentration was 19.96 μg/m³. Compared with some megacities and heavily polluted cities, such as Arak, Iran (54.83 μg/m³) (39), Taiyuan (69.34 μg/m³) (40), and Shenyang (55.0 μg/m³) (41), China. The level of SO₂ in Lanzhou is low and meets the national air quality standard level I limit (the average annual standard for SO₂ is 20 μg/m³). However, the harmful effect of SO₂ on COPD was still found in this study, and the harmful effect was stronger than that of PM. It is suggested that the “low exposure, high risk” effects of SO₂ on health may need to be considered in the formulation of environmental health policies, and more stringent air pollution intervention policies to protect human respiratory health should be implemented.

This study found that every 10 μg/m³ increase in NO₂ corresponded to an RR (95%CI) =1.043 (1.018, 1.068) increase in COPD hospital admissions, and the harmful effect was higher than that of PM₁₀. Like SO₂, NO₂ is also an oxidizing gas. It was reported that exposure to NO₂ can increase the amount of proinflammatory cytokine, thus aggravating the original inflammatory response of the respiratory system (42). It means that exposure to NO₂ can further worsen symptoms associated with respiratory infections, such as wheezing and difficulty breathing, forcing infected patients to be hospitalized (43). When NO₂ reaches the alveoli, the nitric acid and its salts formed with the water on the surface of the alveoli can be stimulated and result in the corrosion of the lung tissue, which results in the destruction of the alveolar epithelium and capillary epithelium. The phagocytotic ability of the alveolar phagocyte are reduced, which enhances the susceptibility of lung tissue to infection (34). The Canadian Cohort Obstructive Lung Disease study (44) showed that NO₂ exposure was associated with lower lung function, even at low concentrations, especially those at the extremes of dysanapsis. We also found a statistically significant negative connection between NO₂ and the risk of COPD inpatients. Similar to the results in Canada (26), Warsaw, Cracow, and Tricity, Poland (24), Belluno, Italy (45), Berlin, Germany (46), European Cohorts in the European Study of Cohorts for Air Pollution Effects (ELAPSE) Project (12), a nationwide study in Korea (23). Jinan, China (13), Qingyang, and China (47), all results showed that short-or long-term NO₂ exposure was associated with increased risks of daily COPD admissions.

O₃ can cause the muscles in the airways to constrict, trapping air in the alveoli and leading to wheezing and shortness of breath. It can also make the lungs more susceptible to infection (48). However, the results of studies on the relationship between O₃ and COPD are still inconsistent. This study found that for every 10 μg/m³ increase in O₃, the RR95%CI was 0.976 (0.955, 0.997) at lag 7 and 1.041 (1.005, 1.079) at lag 4. However, the exposure–response curve of the effects of O₃ on COPD hospital admissions was not statistically significant. Sensitivity analyses revealed that the results between COPD hospital admissions and O₃ were not stable. That is, O₃ is not associated with COPD hospital admissions in this study. This is consistent with studies conducted in Berlin, Germany (46), Beijing (49), Shijiazhuang (50), and Jinan (13), China. However, there are also studies showing that exposure to O₃ can increase (23, 36) or decrease (12, 26, 47) the risk of hospitalization for COPD. An early national multicity study in the United States (51) showed that exposure to O₃ was negatively associated with COPD admissions on the same day (lag0), but was positively associated with COPD on lag 1. Studies have shown that variability of summer apparent temperature (51) and the use of air conditioning (51, 52) can modify the health effect of O₃. A recent study on the association between long-term exposure to O₃ and hospital admissions for four cardiovascular and respiratory outcomes among the Medicare population in the United States (53) suggests that NO₂ and O₃ confound each other’s health effects or may be confounded by an unknown or unmeasured variable, and the negative relationship between the NO₂, O₃, and health effects is more likely to be affected by unknown or unmeasured confounding factors, while the adverse effects of the two on health are less likely to be affected by confounding factors. So, more studies are needed to ascertain the relationship between O₃ and COPD hospitalizations.

This study found that every 1 mg/m³ increase in CO corresponded to an RR (95%CI) =1.222(1.108, 1.348) increase in COPD hospital admissions, and the harmful effect was stronger than that of PM. This result was similar to a cohort study in the UK (11), a nationwide study in Korea (23), and studies conducted in Ganzhou (20) and Qingyang City (47) in China. The toxicity of CO is due to its stronger affinity to hemoglobin than oxygen (34), and a review of CO-triggered health effects has shown that CO exposure changed the ability of hemoglobin to transport oxygen to peripheral tissues, induced pulmonary edema, immune cell infiltration, and lung integrity disruption, probably caused respiratory injury, and worsened the symptoms of respiratory diseases (54). A time series study based on clinical randomized controlled trials also showed that exposure to CO reduces lung function (55). In this study, the CO concentration is 1.21 ± 0.72 mg/m³, which is below the national air quality standard level I limit of 4 mg/m³. This suggests that, like SO₂, even at lower concentrations, harmful effects of CO on COPD hospitalizations can be observed at the population level.

The exposure–response curves: In this study, the exposure–response curves of the effects of air PM_2.5, PM₁₀, SO₂, NO₂, and CO on COPD hospital admissions showed a significant positive relationship and were approximately linear without thresholds. However, for O_3-8h, the exposure–response curve was not statistically significant. The results were similar to those of most studies on the exposure–response curve (10, 20, 56). Such as, a study in Lahore City, Pakistan (10) showed that the exposure–response curves of PM_2.5 on COPD mortality were approximately linear with no indication of thresholds. A study in Ganzhou, China (20) also revealed a positive linear relationship between CO exposure and COPD hospitalization risk, and they did not observe any obvious threshold concentration below which CO has no effect or negative effect on the risk of COPD hospitalization. However, other studies have shown that the exposure–response curve between air pollutants and COPD is non-linear (23, 47). For instance, a time-stratified case-crossover study in Qingyang, China, analyzed the relationship between air pollutants and acute exacerbation of COPD hospitalizations, which found that the exposure–response curves were approximately linear for PM_2.5 and that the curves for PM₁₀, SO₂, and CO were complicated, but only if the statistically significant part was considered. There was an associated “V” shape for PM₁₀, a “U” shape for NO₂, and an inverted “V” for SO₂, CO, and O₃. This may be related to the different concentrations of pollutants in different regions. Furthermore, different study designs (including different cities, population structures, inclusion, and exclusion criteria for study subjects), social institutions, and cultures may also account for the inconsistencies (20).

In this study, the harmful effect of PM on those aged <65 years was slightly higher than aged ≥65 years; however, the effect of gaseous pollutants on those aged <65 years was slightly lower than on those aged ≥65 years, which is inconsistent with other similar studies conducted in China (13, 20, 25, 47). A large population cohort study in southern China (25) reported that long-term PM exposure increased the risk of COPD mortality and that it seems to be more pronounced among older adult participants. An ecological study in Jinan (13) found that short-term exposure to NO₂ and SO₂ was associated with increased risks of daily COPD admissions, especially for the older adult. A time-stratified case-crossover study in Qingyang (47) showed that the harmful effects of PM_2.5, PM₁₀, SO₂, NO₂, and CO on acute exacerbation of COPD hospitalization were stronger in aged 15–64 years. While a time series study in Ganzhou (20) reported that there was no significant effect modification of age on CO exposure-associated hospitalizations for COPD. The older adult are more susceptible to the impact of air pollutants, which may be due to their normal pathological aging and weakened immune function; meanwhile, the older adult demographic usually suffers from preexisting chronic diseases and more advanced COPD (25). While people aged <65 years spend more time outdoors due to work, education, social interactions, and other reasons, which significantly increases their exposure to air pollutants (47). In addition, people aged <65 years are more likely to expose themselves to risk factors for respiratory diseases such as smoking and alcohol consumption, making them more susceptible to pollutants.

In this study, the adverse effects of air pollutants on female COPD hospitalization were slightly higher than that on males. These findings were supported by existing evidence (13, 20, 25, 47). However, there are a few papers with inconsistent results: One study (57) found that the adverse effects of air pollutants on males were higher than on females and one study (58) showed that the effect of air pollutants on males and females was not significantly different. This inconsistency may be related to the differences in genetic, anatomical, and physiological characteristics of the different genders. Compared with males, female lungs, and trachea are smaller, and under the same pollution conditions, female lungs and tracheal tissue experience greater pressure. Furthermore, female lung immunity may be weaker (13, 25). On the other hand, it may also be related to gender differences in occupational distribution, personal characteristics (such as lifestyle, smoking, and other hobbies), or socioeconomic status: males are more engaged in industrial or taxi driving occupations or outdoor activities, have poor awareness of occupational protection, and have unhealthy habits such as smoking and drinking, making them more susceptible to the impact of air pollutants.

At present, there are few reports on the seasonal effects of air pollutants on COPD hospitalization. In this study, the influence of air pollutants on COPD in the cold season was greater than that in the warm season. This is consistent with Cheng’s (13) research findings in Jinan, but contrary to Song’s (20) findings (the CO exposure-associated hospitalization risk for COPD was stronger during the warm season). Our previous research also showed that the influence of air pollutants on respiratory diseases (19) and pneumonia (18) in the cold season was greater than that in warm seasons. This may be related to the higher concentration of pollutants in Lanzhou during the cold season.

In this study, the COPD hospital admissions data of 25 hospitals in Lanzhou City were used to analyze the impact of air pollutants on COPD hospitalizations. The lag effect and exposure–response curves were conducted, as well as a stratified analysis by age, gender, and season. However, the research still has some limitations. First, this is an ecological study. Although confounding factors such as temperature and humidity were adjusted during model fitting, there are still some potential factors (such as the use of air conditioners and purifiers) that may affect the relationship between air pollutants and COPD. At the same time, genetics, early-life events, respiratory infections, individual characteristics (inhalation of tobacco, drugs, and other combustible substances), and occupational exposure are also associated with COPD (7). However, these potential factors and individual characteristics cannot be obtained in an ecological study; hence, their impact on the relationship between air pollutants and COPD was not considered, which may reduce the accuracy of the conclusion. In the future, analytical epidemiological methods and molecular epidemiological methods can be applied to more accurately analyze the impact of air pollutants on COPD and explore its mechanism based on fully considering individual characteristics and potential factors. Second, this paper analyzed the effects of conventional air pollutants on COPD only in Lanzhou City. The spatial heterogeneity of air pollutants on health in multiple cities is worthy of further study. At the same time, the health effects of particulate matter components (such as copper, nickel, manganese, zinc, and other elements), different particle sizes (such as PM1 and PM4), and new pollutants (such as atmospheric microplastics) are more worthy of further study in the future.