This is a retrospective study performed at the People’s Hospital of Xinjiang Uygur Autonomous Region (PHXUAR) in Urumqi, Xinjiang, China.

The study period spanned from January 1, 2020, to December 31, 2023. Patients with a post-bronchodilator FEV1/FVC ratio <0.7 (30), as determined by spirometry, were initially considered for inclusion. The baseline was defined as the date of the first occurrence of FEV1/FVC < 0.7 at the hospital. The baseline period was defined as the three-month interval immediately before and after this date, during which baseline characteristics were collected. Participants were required to have had at least one in-person hospital visit after the end of the baseline period. Exclusion criteria included: (1) age under 18 years, (2) missing outcome, (3) absence of residential address information or a transient residence at the registered address, and (4) a short observation duration (<1 year). Consequently, 660 participants were deemed eligible for analysis. A flow chart detailing the inclusion and exclusion process was provided in Figure 1.

AEs were identified through participants’ medical records at the hospital. The diagnosis of an AE of COPD was established by physicians based on a combination of patients’ clinical signs and symptoms, spirometry results, chest imaging findings, and other pertinent clinical data. The outcome in this study was defined as occurrence of AE within the first year after baseline period.

Air pollutant data for this study were obtained from the National Urban Air Quality Real-time Publishing Platform¹, which compiles hourly measurements from all fixed-site monitoring stations across China. These data were published by the China National Environmental Monitoring Center under the administration of China’s Ministry of Ecology and Environment. In this study, missing data for specific hours were estimated by calculating the average of the data from the immediately preceding and succeeding hours.

For each participant in our study, we extracted their permanent addresses in Xinjiang and geocoded these addresses to obtain longitude and latitude coordinates. Subsequently, we matched each participant’s residential address to the nearest fixed-site monitoring station to retrieve the most relevant pollution data. The distance between each participant’s location and the nearest monitoring station was calculated using the “geodesic” function from “geopy” library in Python (Version 3.9). We then calculated the average concentrations of PM_2.5, PM₁₀, O₃, NO₂, SO₂, and CO during each participant’s respective baseline period and used these averages as a proxy for personal exposure in this study. For participants without a specific address on record, the average pollution levels for their respective city were utilized to estimate their exposure.

We systematically collected characteristics from medical records that have been previously associated with AEs, including GOLD grade and severity, age, gender, ethnicity, occupation, education level, marital status, residential setting, tobacco smoking history, body mass index (BMI), comorbid asthma, hypertension, diabetes, dyslipidemia, cardiovascular disease, and a history of AEs in the past year.

Based on the GOLD spirometry classification (30), the severity of COPD was categorized as follows: FEV1 ≥ 80% of predicted value was designated as GOLD 1 (mild), 50% ≤ FEV1 < 80% of predicted value as GOLD 2 (moderate), 30% ≤ FEV1 < 50% of predicted value as GOLD 3 (severe), and FEV1 < 30% of predicted value as GOLD 4 (very severe). BMI was categorized as follows (31): a BMI of less than 18.5 was classified as underweight, 18.5 to 23 as normal weight, and greater than 23 as obese.

Socioeconomic factors, including occupation, education level, and marital status, were considered in their potential association with AEs of COPD. However, data on occupation and education level were missing for over 30% of the participants, and the vast majority (over 95%) were married, which limited the contribution of these factors in the analysis. Consequently, these socioeconomic variables were not included as confounders in the study.

Baseline characteristics of continuous variables were reported as means (standard deviations), and categorical variables were reported as counts (percentages). Air pollutant concentrations were described by maximum values, minimum values, and quartile values. Differences across cohorts were compared using T, Mann–Whitney U, or χ² tests, as appropriate. The Kolmogorov–Smirnov test was used to assess the normality of the variable distributions.

In the initial analysis, the occurrence of AEs within the first year was examined as the outcome variable in univariate logistic regression models. By this approach, associations between AEs and all variables, including air pollutants and the confounders, were explored. Subsequently, we developed single-pollutant models for each pollutant. Adjusting for age, gender, ethnicity, residential setting, smoking status, and BMI, as well as asthma, diabetes, dyslipidemia, CVD, history of AEs in the past year, and FEV1% predicted at baseline, six multivariable logistic regression models were built to describe the association of each air pollutant with incidence of AEs. GOLD grade, FEV1/FVC, season of baseline, and hypertension were removed for multicollinearity. Occupation and education level were removed for high missing rate. Risk estimates were expressed as odds ratios (ORs), with 95% confidence intervals (CIs) and corresponding p-values. The multivariable regression model was constructed using the “glm” function.

To examine the robustness of observed associations, sensitivity analyzes were performed. We extended our analysis beyond single-pollutant models by fitting two-pollutant models for each pair of the six pollutants, with the aim to control for potential confounding effects due to co-exposures. The pair of PM_2.5 and PM₁₀ were excluded as PM_2.5 is inherently a component of PM₁₀ measurements.

Utilizing the extensive individual-level features, a series of stratified analyzes were conducted to elucidate the specific associations between air pollution and AEs of COPD in subgroups. These analyzes were stratified by and disease severity (GOLD grade 1 vs. grade 2/3), age (≤65 years vs. >65 years), gender (male vs. female), ethnicity (Uyghur vs. Han), residential setting (rural vs. urban), smoking status (no smoking history vs. previous or current smoking history), frequency of AEs of COPD in the past year (0 vs. ≥1), a history of CVD (yes vs. no), a history of asthma (yes vs. no), and season of baseline (spring vs. winter).

All statistical analyzes were performed using R-Studio version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria). Two-sided p-values of less than 0.05 were considered statistically significant.

As shown in Figure 1, of the initial 2,211 patients diagnosed with COPD based on an FEV1/FVC ratio of less than 0.7, all were above 18 years of age. Exclusion criteria led to the removal of 1,054 patients who did not report their outcome of AEs in the following year. Furthermore, 140 participants were excluded due to missing residential addresses or not residing for at least 3 years at their registered addresses. Additionally, 357 individuals who were observed for a short duration were also excluded. Consequently, a total of 660 patients were included in the statistical analysis.

Table 1 presents the baseline characteristics of the eligible participants. The FEV1% predicted values were nearly normally distributed, with a median of 102.66 and a mean of 102.73. The majority of participants were categorized as having mild COPD. Notably, those in the AE group exhibited significantly lower FEV1% predicted and FEV1/FVC ratios at baseline. The majority of participants were admitted during the winter months, accounting for 61.8% of the total admissions, while the remaining 38.2% were admitted in the spring. The average age was 64.8 ± 10.5 years, spanning from 30 to 88 years. The majority of participants were male (64.1%). Ethnic distribution was 54.1% Uyghur, 36.8% Han Chinese, and 9.1% other ethnicities. Urban residence was reported by 62.7% of the participants. 39.5% had a smoking history. 6.6% of participants were classified as low body weight. Concurrent COPD and asthma diagnoses were present in 56.4% of the participants. Additionally, the prevalence of hypertension, diabetes, dyslipidemia, and CVD was 64.4, 30.5, 20.0, and 65.3%, respectively. Regarding history of AEs of COPD, 263 (39.8%) participants did so once, and 195 (29.5%) had two or more within the previous year.

We preformed exploratory analysis on the associations between these factors and AEs (Supplementary Table S1). Urban residency and a history of AEs in the previous year were significantly associated with AEs in the univariate analysis (ORs of 1.091 [95% CI 1.018–1.169, p = 0.014] and 1.113 [95% CI 1.035–1.196, p = 0.004], respectively).

Figure 2 illustrates the average concentrations of air pollutants across fifteen cities in Xinjiang from 2010 to 2023. Over this period, the annual mean concentration of NO₂ declined by 13% (from 27.9 μg/m³ in 2017 to 24.2 μg/m³ in 2023), CO concentration declined by 26% (from 0.91 mg/m³ to 0.67 mg/m³), and SO₂ declined by 27% (from 9.1 μg/m³ to 6.7 μg/m³). Whereas O₃ concentration increased by 5.9% (from 86.9 μg/m³ to 92.0 μg/m³). Regarding particulate matter, both PM_2.5 and PM₁₀ concentrations remained relatively stable, with the exception of the year 2021 when COVID-19 restrictions reduced industrial activities, leading to a temporary dip in concentration levels.

The baseline dates for the cohort ranged from 2020 to 2023. We defined the baseline period as a six-month interval centered by the baseline date, during which we collected data on air pollutants. Table 2 summarizes the six-month average concentrations of air pollutants to which the cohort was exposed. The median levels of these pollutants were as follows: 45.6 μg/m³ for PM_2.5, 84.3 μg/m³ for PM₁₀, 33.7 μg/m³ for NO₂, 0.9 mg/m³ for CO, 7.9 μg/m³ for SO₂, and 87.9 μg/m³ for O₃. The distribution of these pollutants by study outcome was detailed in Supplementary Figure S1.

The relationships between each air pollutant and the occurrence of at least one AE during the first year of follow-up are detailed in Table 3. In the initial univariate logistic regression analysis, no significant associations were observed between air pollutant exposure and the occurrence of AEs. After adjusting for confounders in the single-pollutant multivariable analysis, exposure to O₃ was found to be associated with the risk of AEs (OR = 1.007 [95% CI 1.000–1.013], p = 0.046). By contrast, exposure to PM_2.5, PM₁₀, NO₂, SO₂, and CO showed no association with AEs in the entire cohort, with ORs [95% CI] as 1.000 [0.997–1.002], 1.000 [1.000–1.001], 0.996 [0.991–1.000], 0.902 [0.746–1.091], and 0.997 [0.982–1.012], respectively.

The results of two-pollutant models are presented in Figure 3 and Supplementary Table S2. We found that the associations between all pollutant factors and AEs in the following year were robust after adjusting for co-exposure. For O₃, the associations were consistently significant in all the models except for adjustments for NO₂, with ORs of 1.006 [95% CI 1–1.013, p = 0.065]. For PM_2.5, PM₁₀, NO₂, SO₂, and CO, the associations remained insignificant, regardless of adjustments for any other air pollutant.

Subgroup analyzes were conducted to determine whether adjusted associations between AEs of COPD and O₃ varied across different individual characteristics. By this approach, we identified specific sub-populations that were more susceptible to the effects of O₃ (Figure 4). Our findings revealed that the associations of O₃ with AEs of COPD were relatively larger in male patients (OR of 1.009 [95% CI 1–1.017], p = 0.046), those aged over 65 years (OR of 1.012 [95% CI 1.002–1.021], p = 0.014), those of Han ethnicity (OR of 1.019 [95% CI 1.007–1.032], p = 0.003), those with a history of AEs in the previous year (OR of 1.008 [95%CI 1–1.017], p = 0.048), and those without a concurrent diagnosis of asthma (OR of 1.014 [95%CI 1.004–1.025], p = 0.009). No differences were observed between subgroups defined by varying disease severity, residential settings, smoking history, or the presence of CVD history. Furthermore, for PM_2.5, PM₁₀, NO₂, SO₂, and CO, the associations remained insignificant in sub-populations analyzed (Supplementary Table S3).