Association between acute exacerbation of chronic obstructive pulmonary disease and short-term exposure to ambient air pollutants in France

Abstract

Background

Ambient air pollution is recognized as a major risk factor for chronic obstructive pulmonary disease (COPD) which is the third leading cause of death worldwide. We examined whether variations in daily outdoor air pollutants levels were associated with excess hospital emergency room visits (ERV) for acute exacerbation of COPD (AECOPD).

Methods

This two-center ecological cohort study was conducted in Amiens, France. We collected all consecutive ERV for AECOPD throughout 2017 and developed single pollutant models to assess the association between AECOPD and nitrogen dioxide (NO₂), ozone (O₃), or particulate matter (PM_2.5 and PM₁₀) levels, while adjusting for temperature, hygrometry, influenza circulation and pollen allergy risk. For a subgroup of patients, we also applied geographical modeling to analyze annual exposure to outdoor air pollutants.

Results

We recorded 240 ERV among 168 COPD patients in 2017 and identified 9 peaks of ERV. There was a statistically significant positive correlation between the daily ERV for AECOPD and the daily average concentrations of PM_2.5 (RR = 1.06 (95%CI = [1.00–1.11]), p = 0.049), but no correlation with NO₂, O₃ or PM₁₀ (p = 0.073, p = 0.114 and p = 0.119, respectively). Our geographical modeling study revealed that long-term exposure to any of the four outdoor air pollutants was not associated with more frequent AECOPD.

Conclusion

Even though the pollution levels measured generally remained below or near the 2021 short-term air quality guidelines issued by the World Health Organization, significant aggregate-level associations were found between severe AECOPD leading to ERV and daily concentrations of PM_2.5.

Clinical trial registration

NCT03079661.

Background

Chronic obstructive pulmonary disease (COPD) is the third leading cause of death worldwide, it caused 3.23 million deaths in 2019 [1]. Smoking and air pollution are the most common causes of COPD. Tobacco smoking accounts for over 70% of COPD cases in high-income countries whereas household air pollution is a major risk factor for COPD in low- and middle-income countries [1, 2]. Moreover, The World Health Organization (WHO) estimates that ambient air pollution was responsible for 4.15 million premature deaths in 2019 [3]. Outdoor air pollution is increasingly recognized as a cause of acute exacerbation of COPD (AECOPD) [4]. A meta-analysis by Li et al. [5], while underlining substantial heterogeneity across studies, reported a significant association between short-term exposure and AECOPD for all gaseous pollutants (mainly nitrogen dioxides (NO₂) and ozone (O₃)) and small-diameter particulate matter pollutants (PM_2.5−10 ; particles with aerodynamic diameters of less than 2.5–10 μm).

In 2021, the WHO issued air quality guidelines (AQG) recommending maximal concentrations levels for various pollutants [6]. Nevertheless, for individuals, the respiratory risk of ambient air pollution is dose-dependent with no apparent “safe” thresholds. Even in countries with low outdoor air pollution levels, chronic exposure to small-diameter particulate matter (PM_2.5−10) and nitrogen dioxides (NO₂) significantly impairs lung growth in children, accelerates lung function decline in adults and increases the risk of developing COPD, particularly among those with additional risk factors for this disease. Poor air quality due to outdoor air pollution also increases the risk of AECOPD, hospitalizations and mortality [4, 7,8,9].

Study objectives

The primary objective of this study was to assess whether variations in daily outdoor air pollutants levels (NO₂, O₃, PM_2.5, PM₁₀) in Amiens, France, over the course of 2017 were associated with an increase in emergency room visits (ERV) for AECOPD. A secondary objective was to investigate whether long-term annual exposure to outdoor air pollutants was linked to a higher frequency of AECOPD episodes.

Methods

Study design

We conducted a two-center ecological cohort study from January 1st, 2017, to December 31st, 2017 to analyze the association between time series of daily air quality and daily ERV related to AECOPD, at an aggregate level for both parameters. The two centers involved (University Hospital Center Amiens-Picardie and the Europe Clinic) house the only two emergency rooms in the Amiens urban area, all ERV over the study period were thus recorded. The city of Amiens is located in the north of France, 100 km south of Lille and 115 km north of Paris. Consecutive ERV for AECOPD were prospectively collected regardless of whether patients required hospital admission. Each patient could be included several times if they had more than one AECOPD episode during the study period. COPD and AECOPD diagnoses were based on the 2017 report of the Global Strategy for Diagnosis, Management, and Prevention of COPD (GOLD), with AECOPD defined as an acute worsening of respiratory symptoms that require additional therapy, once alternative diagnoses such as pneumonia, heart failure, and pulmonary embolism have been ruled out [10].

Air quality was assessed using data from the air quality monitoring network Atmo Hauts-de-France, which corresponds to daily data on outdoor pollutants, as measured by three stations located in the Amiens urban area. The sites of the stations were chosen to cover the city with various exposures to ambient pollutants: one urban station located in the city center, one station located close to a major traffic lane and one peri-urban station located in a suburb 6 km away from Amiens city center. Daily average concentrations were computed for O₃, NO₂, PM_2.5, PM₁₀ (only the urban station collected data on all four pollutants) and were compared to the short-term AQG levels from 2021 (100 µg/m³, 25 µg/m³, 15 µg/m³ and 45 µg/m³, respectively) [6]. One station also collected data on pollen counts, and a pollen allergy risk index was calculated for each week of 2017 based on the number of pollen grains counted and the allergenic potential of the species sampled [11]. The period covering the influenza epidemic was defined based on incidental cases of respiratory samples testing positive for influenza nucleic acid at our institution.

Variables

In addition to air quality data, we collected data on patients’age, gender, home and work addresses, vaccinal status and COPD characteristics (GOLD stage, smoking status, exacerbations status). For each patient included, we recorded the number of AECOPD during the study period as well as any hospital admissions. Based on the average number of ERV in 2017, we defined an AECOPD peak as a threefold increase in daily ERV over a consecutive 3-day period.

Statistical analysis

No calculation was performed to determine the required number of patients/exacerbations, as we chose to include all AECOPD cases admitted to emergency departments consecutively throughout 2017.

We first conducted a descriptive analysis of our population and the air quality data, expressing quantitative variables as mean (± standard deviation) or median (interquartile range) and qualitative variables as percentage. To address our main objective, we used time series with a generalized linear autoregressive integrated moving average (ARIMA) Poisson model (GLARMA) using the Fisher scoring iteration algorithm and considering a lag effect based on data for the seven previous days. These calculations allowed us to account for autocorrelation between the data, which is likely to occur with daily AECOPD counts. The dependent variable was daily AECOPD counts and the independent variables were air quality variables (O3, NO2, PM2.5, PM10). We evaluated stationarity of the AECOPD count time series using a Dickey-Fuller test. Goodness of fit was assessed for observed versus fitted plots, applying the PIT (Probability Integrated Transform) histogram, Q-Q plot of randomized residuals, ACF (autocorrelation function) and PACF (partial autocorrelation function) of randomized residuals. PIT is a goodness of fit measure that compare empirical probabilities from the fitted model with a uniform distribution. A uniform histogram indicates a good fit with the model. The analysis was adjusted for temperature, hygrometry, influenza circulation and pollen allergy risk. Temperature and hygrometry were entered in the model as continuous variables while influenza circulation and pollen allergy risk were incorporated as categorical variables. Relative risk (RR) of AECOPD was calculated with a 95% confidence interval [95%CI].

To determine wether outdoor air pollution was associated with an increased risk of AECOPD, we modeled the mean long-term (one-year) exposure at each subject’s residential address. Information on the spatial variability of the pollutants were obtained with the ADMS urban 4.0 (Atmospheric Dispersion Modelling System), developed by the United Kingdom’s Met Office and Cambridge Environmental Research Consultants (CERC). This three-dimensional air quality model simulates the Gaussian dispersion and transformation of pollutants in the atmosphere, accounting for meteorological conditions, topography, and local emissions from various sources, including line sources (road, rail, and river traffic), point sources (industrial chimneys) and area sources (residential heating, agriculture). Additionally, background concentrations were incorporated using stations measurements from outside the model domain. The emission data used were extracted from the regional spatial inventory of pollutant emissions (INV A2012 M2012) and further refined with the latest available local data for traffic (2014) and industrial emission (2016/2017), corresponding to the various sources and their specific locations throughout the region. Finally, we used a non-parametric Mann-Whitney test to compare two groups of patients – those who experienced two or more exacerbations over the course of 2017 and those who experienced only one exacerbation – based on their mean annual exposure to outdoor air pollutants. To control the false discovery rate associated with multiple testing, we used the Benjamini-Hochberg method to adjust the observed p-values. These adjusted p-values were then compared against a significance threshold of α = 0.05 to identify statistically significant results.

All statistical analyses were performed using R software version 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria; www.r-projet.org) via the RStudio interface, Version 2023.06.2 – © 2009–2023.

Ethical considerations

This study was approved by the research ethics committee Nord-Ouest on November 11th, 2016 (number RNI2016-42). In line with the French legislation on non-interventional studies, the research ethics committee waived the need for written informed consent. After receiving oral and written information on the study, patients gave their oral consent for participation. The study database was registered with the French National Data Protection Commission (Commission Nationale de l’Informatique et des Libertés, Paris, France). Clinical trial registration: NCT03079661.

Results

During the one-year study period, 240 ERV for AECOPD were recorded for a total of 168 COPD patients (114 patients treated at the Amiens University Hospital and 54 at the Amiens Europe Clinic), representing an average of 0.66 ERV per day. According to our predefined criteria, an AECOPD peak was defined as at least 6 ERV occurring over a consecutive 3-day period. Demographic and clinical data of the subjects are presented in Table 1. Patient characteristics were fairly similar between the two centers. Recent pulmonary function tests (< 1 year before or after ERV) were available for 146 of the patients included (86.9%). The majority of patients had mild or severe COPD: 6.8% were classified as GOLD Stage I; 24.7% as GOLD Stage II; 36.3% as GOLD Stage III; and 32.2% as GOLD Stage IV. During the study period, 152 COPD patients were hospitalized at least once (90.5%) and 28 patients had two or more ERV for AECOPD (16.7%). Of the 240 ERV, 210 were followed by hospitalization (87.5%).

Table 1 Characteristics of the study population

Nine ERV peaks due to AECOPD were noted in 2017 during weeks 2, 6, 9, 10, 14, 16, 26, 49 and 52. The daily average concentrations of outdoor pollutants exceeded short-term (24-hour) AQG levels for PM₁₀ on three occasions during the study period (weeks 3, 4 and 6), while the threshold for PM_2.5 concentrations was exceeded for 24 weeks, mainly during the first half of the year. High O₃ concentrations were only observed during week 25, which corresponded to a heatwave. Conversely NO₂ levels exceeded the AQG levels almost continuously throughout the whole year. A high index of pollen allergy risk was observed from week 14 to week 15 and from week 21 to week 26, whereas the influenza epidemic lasted from week 1 to week 8 and from week 47 to week 52. The distribution of the peaks of ERV for AECOPD and the recorded concentrations of the different pollutants during 2017 are shown in Fig. 1. A statistically significant positive correlation was found at the aggregate level between the daily ERV for AECOPD and the daily average concentrations of PM_2.5 recorded by the urban station (RR = 1.06 (95%CI = [1.00–1.11]), p = 0.049). No significant correlation was observed with NO₂, O₃ or PM₁₀ levels (respectively, RR = 1.02 (95%CI = [1.00–1.05]), p = 0.073; RR = 1.01 (95%CI = [1.00–1.03]), p = 0.114 and RR = 0.96 (95%CI = [0.91–1.01]), p = 0.119). These results suggest that higher concentrations of PM_2.5 are associated with an increased risk of ERV for AECOPD at the population level, with an RR of 1.06 indicating a 6% increase in the number of ERV for AECOPD per each additional unit of PM_2.5 concentration (in µg/m³). Detailed results from this multivariate analysis are presented in Table 2. We assessed potential lagged effects, but our results indicated that past values contribute minimally to the observed AECOPD counts. Additionally, sensitivity analyses varying the lag parameter from 1 to 7, did not alter the model’s coefficients. Detailed information on the modeling can be found in the additional file [see Additional file 1].

Fig. 1

Concentrations of the different pollutants recorded during 2017 and peaks of emergency room visits for acute exacerbation of COPD. Panels A, B, C, and D show data for O₃, NO₂, PM_2.5, and PM₁₀, respectively. Peaks in emergency room visits are indicated by blue shading. The influenza epidemic period is indicated by the violet arrows, and periods with a high pollen allergy risk index are marked with brown arrows

Table 2 Multivariate analysis on the daily risk of emergency room visits for acute exacerbation of COPD

Of the total of 168 patients, 58 could be included in the geographical modeling (30 patients provided no or incomplete addresses, whereas the remaining 80 patients lived outside the perimeter studied). The distribution of the mean annual exposure to the different outdoor air pollutants is presented in Tables 3 and 4. Except for O₃, the mean annual exposure to pollutants recorded in 2017 exceeded the long-term AQG levels from 2021 [6]. Figure 2 presents a modeling map showing the mean annual concentration of NO₂ in the study area, along with the localization of subjects’ residences. Of the 58 patients, 13 experienced at least two AECOPD episodes during 2017. No significant difference was observed between these patients and those with only one AECOPD when comparing annual concentrations of the ambient pollutants monitored (p = 0.532, p = 0.532, p = 0.532 and p = 0.532 for PM₁₀, PM_2.5, O₃ and NO₂, respectively). Considering the small size of our subgroup and the limited number of patients experiencing two or more AECOPD episodes, we did not adjust for patients’ characteristics.

Table 3 Mean annual exposure to outdoor air pollutants at the home address of the 58 subjects included in the modeling study

Table 4 Mean annual exposure to outdoor air pollutants at the home address of the 58 subjects included in the modelling study, grouping participants according to their COPD severity

Fig. 2

Modeling map showing the mean annual concentration of NO2 in the study area, along with the localization of subjects’ residences Subjects’ residences are indicated by small circles. The black arrow in the color scale represents the French long-term quality objective for NO2 (40 µg/m³)

Discussion

Our results indicates that daily variations in PM_2.5 significantly correlate with a higher risk of severe AECOPD leading to ERV. No such association was found with daily variations in NO₂, O₃ or PM₁₀. Given that our model was adjusted for several confounders (temperature, hygrometry, influenza circulation and pollen allergy risk index), we are confident in the robustness of our results. These findings suggest that daily variations may provide a more accurate prediction of AECOPD risk at the population level than fixed thresholds. Indeed, the relationships between outdoor air pollutants and health impacts (all-cause mortality, respiratory mortality, AECOPD, and asthma) has been shown to follow nearly linear concentration-response curves for NO₂, PM_2.5, PM₁₀ or O_3, but without clear-cut thresholds [8, 12, 13].

Few studies have been conducted in France to examine the relationship between ambient air pollution and respiratory health [12, 14]. A recent study in Berlin, Germany, with a fairly similar range of daily air pollutant concentrations to ours, reported an increased risk of COPD and asthma exacerbations leading to hospitalization. Interestingly, in the German study, NO₂ was the only pollutant for which a 24-hour increase in concentration was associated with a significant uptick in COPD hospitalizations and morbidity (by 12% per 10 µg/m³) [15]. An Italian study reported similar findings, showing a significant link between NO₂ exposure and AECOPD risk, and no association between PM₁₀ and such a risk [16]. In contrast, our study showed a non-significant increased risk of ERV for AECOPD with NO₂, but a clear association with daily variations in PM_2.5. Similarly, Liu et al. reported that the relationships between respiratory mortality and PM concentrations were slightly stronger with PM_2.5 than with PM₁₀ in most countries and regions [12]. Our findings add evidence that PM_2.5 accounts for a large proportion of the effects of PM₁₀. The stronger threat of PM_2.5 is supported by the abundant evidence that this particulate fraction contains more small particles that can absorb toxic components from the air and carry them deep into the lungs [17].

Long-term exposure to air pollution is a major contributor to global disease burden [18]. Cohen et al. reported that PM_2.5 air pollution accounted for 24.7% of lower respiratory infection, and 27.1% of COPD mortality in 2015, while ozone exposure contributed to an estimated 8.0% of global COPD mortality [7]. A systematic review of PM_2.5 and cause-specific mortality reported a meta-analytic effect estimate of RR of 1.08 [95% CI: 1.06–1.09] per 10 µg/m³ PM_2.5, assuming a linear relationship [19]. The authors indicated a supralinear relationship, suggesting that risk increased more steeply at lower exposure levels. In another meta-analysis, Huangfu et al. reported that respiratory mortality, COPD-related mortality, and lower respiratory infection-related mortality were positively associated with long-term exposure to NO₂ [20]. Conversely, our modeling study of long-term exposure to outdoor air pollutants found no association with more frequent AECOPD episodes. One possible explanation for this apparent discrepancy is that annual average pollutant exposure for all of the patients in our study was relatively low, only slightly above the 2021 long-term AQG levels, with little variability between individuals (Table 3). Furthermore, long-term exposure appeared to be similar across different COPD severity levels in our subgroup analysis (Table 4). Another factor is the limited statistical power of our cohort, as only 58 patients living within the study area were included in this secondary analysis. Additionally, the constraints of our modeling approach may have impacted the findings. Specifically, pollutant exposure at patients’ home address may not accurately reflect their daily personal exposure, as we were unable to account for their movements throughout the day. Relying on residential address as a proxy for individual-level exposure may therefore introduce uncontrolled measurement errors, limiting the strength of the conclusions that can be drawn from our results.

Our study has several other limitations. First, we did not measure individual exposure to outdoor pollutants in our main analysis. As a result, while our results may be valid at the population level, potential cross-level bias prevent us from concluding that they apply at the individual level. Since we collected data from three stations representing different sources of pollution depending on their location, we are confident that our measurements are representative of the overall pollution in the study area. As illustrated in Fig. 1, variations in the concentrations of outdoor pollutants were almost identical across the three stations. Secondly, we had no data on exposure to indoor air pollution, a factor that is known, even in high-income countries, to contribute to increased respiratory morbidity and annual lung function decline among current and former smokers, with or without COPD [21, 22]. Thirdly, our single-pollutant model was unsuitable for determining whether the observed associations for each pollutant were independent of the others, a hypothesis supported by previous studies [8, 12]. Finally, the choice of our definition of an AECOPD-related ERV peak is open to debate. To our knowledge, there is no consensus in the literature on how to define such a peak. After analyzing ERV data from the year preceding the study, we proposed a threshold of three times the average daily number of visits for AECOPD over a period of three consecutive days. Although somewhat arbitrary, this stringent definition has the advantage of minimizing the risk of considering minor fluctuations in numbers of visits as significant.

Our results have important implications as they provide evidence that variations in outdoor pollutants levels may lead to an increased risk of AECOPD, even in areas with fairly low overall levels. As AECOPD and consequent hospitalization generate significant costs for the healthcare system and are associated with an increased decline in lung function and mortality risk, it is crucial to mitigate the impact of daily variations in outdoor pollutant exposure for COPD patients. Currently, interventions are mainly limited to recommending that patients avoid going outside and limit their physical activities during pollution alerts. However, this type of recommendation raises questions: How is the population informed when thresholds are exceeded? Which pollutants thresholds should trigger population alerts? What about exposure to indoor air pollution (which is known to have negative respiratory impacts [23])? And what about the negative consequences of limiting physical activity for COPD patients [24]? A study is currently being undertaken to explore whether individual-level interventions can impact PM exposure, respiratory function, respiratory symptoms, quality of life, and the risks of exacerbation, hospitalization, and death in a population of 120 COPD patients in Korea [25].

Conclusion

Our study is among the few conducted in France to investigate the relationship between daily variations in outdoor air pollutants levels and AECOPD risk. Even though the levels measured remained generally below or near the short-term AQG levels from 2021, we demonstrated significant aggregate-levels associations between severe AECOPD leading to ERV and daily concentrations of PM_2.5; no such association was found with PM₁₀, NO₂, and O₃. These results have important implications for public health, highlighting the need for interventions to mitigate the impact of daily pollutant exposure on COPD patients.