Abstract
Ambient air pollution increases the risk of respiratory mortality, but evidence for impacts on lung function and chronic obstructive pulmonary disease (COPD) is less well established. The aim was to evaluate whether ambient air pollution is associated with lung function and COPD, and explore potential vulnerability factors.
We used UK Biobank data on 303 887 individuals aged 40–69 years, with complete covariate data and valid lung function measures. Cross-sectional analyses examined associations of land use regression-based estimates of particulate matter (particles with a 50% cut-off aerodynamic diameter of 2.5 and 10 µm: PM2.5 and PM10, respectively; and coarse particles with diameter between 2.5 μm and 10 μm: PMcoarse) and nitrogen dioxide (NO2) concentrations with forced expiratory volume in 1 s (FEV1), forced vital capacity (FVC), the FEV1/FVC ratio and COPD (FEV1/FVC <lower limit of normal). Effect modification was investigated for sex, age, obesity, smoking status, household income, asthma status and occupations previously linked to COPD.
Higher exposures to each pollutant were significantly associated with lower lung function. A 5 µg·m−3 increase in PM2.5 concentration was associated with lower FEV1 (−83.13 mL, 95% CI −92.50– −73.75 mL) and FVC (−62.62 mL, 95% CI −73.91– −51.32 mL). COPD prevalence was associated with higher concentrations of PM2.5 (OR 1.52, 95% CI 1.42–1.62, per 5 µg·m−3), PM10 (OR 1.08, 95% CI 1.00–1.16, per 5 µg·m−3) and NO2 (OR 1.12, 95% CI 1.10–1.14, per 10 µg·m−3), but not with PMcoarse. Stronger lung function associations were seen for males, individuals from lower income households, and “at-risk” occupations, and higher COPD associations were seen for obese, lower income, and non-asthmatic participants.
Ambient air pollution was associated with lower lung function and increased COPD prevalence in this large study.
Abstract
In one of the largest analyses to date, ambient air pollution exposure was associated with lower lung function and increased COPD prevalence, with stronger associations seen in those with lower incomes http://bit.ly/2DLBPA6
Introduction
Ambient air pollution increases the risk of respiratory mortality, but evidence for impacts on lung function and obstructive lung disease is less well established. Recent studies and reviews have reported suggestive evidence linking outdoor air pollution and lung function and chronic obstructive pulmonary disease (COPD) [1–4]. Recently, the European Study of Cohorts for Air Pollution Effects (ESCAPE) project showed that higher ambient nitrogen dioxide (NO2) and exposure to particulate matter with a 50% cut-off aerodynamic diameter of 10 µm (PM10), as well as higher traffic load on roads near residences were associated with impaired lung function in adults, using a meta-analysis across five European cohorts [5]. A separate meta-analysis of four of these same cohorts found positive but nonsignificant associations between chronic exposure to ambient air pollution and COPD [6]. Other large studies have not shown consistent evidence of long-term air pollution exposure on adult-onset COPD [7, 8]. Inconclusive findings have been, in part, attributable to lack of statistical power to detect small effects. Sample size limitations have also curtailed exploration of associations among population subgroups.
The objectives of this cross-sectional study were to examine whether air pollution was associated with lung function and COPD using a very large UK study. Secondly, potential vulnerability factors of the relationships between air pollution and lung function and COPD were explored.
Methods
Study participants
We used baseline questionnaire, anthropometric measures and spirometry data from the UK Biobank collected in 2006–2010. UK Biobank is a national cohort study of half a million participants aged 40–69 years, largely in urban areas of England, Wales and Scotland recruited from the UK National Health Services register [9]. Full study sampling methods are described elsewhere [9, 10]. A research protocol for our study obtained all necessary approvals from the UK Biobank's review committees.
Lung function and COPD
Trained healthcare technicians and nurses in UK Biobank assessment centres performed pre-bronchodilation lung function tests using the Vitalograph Pneumotrac 6800 spirometer (Maids Moreton, UK). Contraindications were chest infection in the past month; history of detached retina, heart attack or surgery to eyes, chest or abdomen in past 3 months; history of a collapsed lung; pregnancy (1st or 3rd trimester); or currently on medication for tuberculosis. Two blows were recorded for each participant and a third blow was administered if the differences between both forced vital capacity (FVC) and forced expiratory volume in 1 s (FEV1) of the first two blows were >5%. Acceptability of spirometry data was assessed by quality appraisal of a sample of manoeuvres, as described previously [11]. The highest values for both FVC and FEV1 from acceptable blows were used in analyses. Participants reporting having smoked or used an inhaler within an hour prior to spirometry testing were excluded.
To adjust for normative ageing effects as well as variations according to sex, height and ethnicity, we defined COPD outcomes using the Global Lung Function Initiative (GLI) 2012 reference values for lower limit of normal (LLN) [12], which were computed using the GLI R macro [13]. Individuals with a FEV1/FVC ratio below the LLN were classified as having COPD [14].
Air pollution estimates
Land use regression (LUR)-based estimates of NO2, PM10, fine particles with diameter <2.5 μm (PM2.5), and coarse particles with diameter between 2.5 μm and 10 μm (PMcoarse) for 2010 were generated as part of ESCAPE [15, 16] and linked to geocoded residential addresses of UK Biobank participants. Predictor variables used in final pollutant-specific LUR models and model R2s are presented in supplementary table S1. Leave-one-out cross-validation, where each site is left out sequentially while the included variables of the models are left unchanged, showed good model performance for PM2.5, PM10 and NO2 (cross-validation R2=77%, 88% and 87%, respectively) and a moderate performance for PMcoarse (cross-validation R2=57%). Evaluation of ESCAPE LUR estimates was conducted by comparing model predictions to the UK's Automatic Urban and Rural Network monitoring data [17]. The LUR NO2 model predicted measured concentrations reasonably well (R2=0.67), while the LUR PM10 model predicted concentrations moderately well in central and southern areas of the UK (R2=0.53), but R2 values were <0.5 in northern England or Scotland, so these areas were dropped from PM analyses [18].
Confounder and effect modifier variables
Sociodemographic and behavioural confounders and potential effect modifiers were identified a priori through literature search. Age was derived from birth date and date of baseline assessment. Body mass index (BMI) was constructed from measured height and weight. The UK Biobank five-level before-tax household income variable was dichotomised to “less than” or “equal to or above” GBP 31 000 categories, being closest to the UK median gross household income in 2009/2010 (GBP 27 789) [19]. Educational attainment was defined as “lower vocational qualification or less” versus “higher vocational qualification or more”. The smoking status variable classified participants as never, previous or current tobacco smokers. Passive smoking exposure was defined as exposure ≥1 h·week−1 to other people's tobacco smoke at home. Asthma status was based on a self-reported doctor diagnosis of asthma. Lastly, participants reporting current employment for one of 14 jobs associated with an increased risk (prevalence ratio ≥1.30) for COPD identified by De Matteis et al. [11] (supplementary table S2) were classified as having an “at-risk” occupation.
Statistical analyses
We performed descriptive analyses followed by cross-sectional linear regression analyses for lung function and logistic regression analyses for COPD. The associations between baseline FEV1, FVC and FEV1/FVC ratio and annual average air pollutant concentrations at place of residence were adjusted for age, age-squared, sex, height, BMI (kg·m−2), household income, education level, smoking status and passive smoking exposure. Associations for COPD (FEV1/FVC <LLN) at baseline were adjusted for age, sex, BMI, household income, education level, smoking status and passive smoking exposure. In order to allow direct comparison with previous ESCAPE studies on air pollution and lung function impairment and COPD [5, 6, 20], all associations were reported per 5 μg·m−3 increase of PM2.5 and PMcoarse and per 10 μg·m−3 increase of PM10 and NO2. To allow interquartile range (IQR) comparison of pollutant effects in the UK Biobank population, results were also reported per IQR increase of air pollutant. Sensitivity analyses were conducted by restricting analyses to individuals living at the same address for ≥10 years, to minimise exposure misclassification. In addition, we investigated whether pro-inflammatory characteristics modified the relationship between PM2.5 and NO2 air pollution and lung function and COPD. Stratified analyses were conducted for sex (male versus female), age (<65 versus ≥65 years), obesity (non-obese versus obese), smoking status (never versus current or past smoker), household income (<GBP 31 000 versus ≥GBP 31 000), asthma status (never versus ever diagnosed) and occupational status (at-risk versus not at-risk occupation) (supplementary table S2). Lastly, we calculated attributable fraction of COPD prevalence due to PM2.5 exposure above World Health Organization (WHO) air quality guideline levels (>10 μg·m−3), current/past smoking and passive smoking exposure at home.
All statistical analyses were limited to participants with complete exposure and model covariate data and were performed in R Statistical Software, version 3.4.4 [21].
Exclusions and missing data
Study population, exclusions, and missing data are outlined in figure 1. Of the 502 655 UK Biobank participants, 36 had withdrawn from the cohort prior to beginning analyses. 48 818 participants had not completed spirometry tests, and an additional 67 823 participants were excluded due to invalid spirometry measures (n=59 850) or having smoked or used an inhaler within an hour of lung function test (n=7973), resulting in 385 978 participants with valid FEV1 and FVC measures. 82 091 participants had missing data for at least one covariate in fully adjusted models, leaving 303 887 participants with complete covariate data and valid lung function measures. The COPD outcome variable was available for 303 183 participants, as 704 individuals had no data for the ethnicity variable used to calculate the LLN threshold. After excluding participants with missing air pollution metrics, our final samples for lung function analyses were 299 537 (NO2 population) and 278 228 (PM population). COPD analyses included 298 848 and 277 567 participants for NO2 and PM populations, respectively.
Study population, exclusions and missing data. FEV1: forced expiratory volume in 1 s; FVC: forced vital capacity; LLN: lower limit of normal; COPD: chronic obstructive pulmonary disease; NO2: nitrogen dioxide; PM: particulate matter.
Results
Characteristics for participants with complete data in fully adjusted lung function models and for excluded participants due to incomplete data are presented in table 1. The mean age of participants with complete data was 56 years and ∼53% of participants were female. The majority of participants were overweight (43%) or obese (24%), had higher education qualifications (48%) and came from households earning more than GPB 31 000 annually (55%). Three out of five participants were lifetime nonsmokers, only 3% were current smokers and 5% reported exposure to tobacco smoke at home. ∼11% of study subjects had been diagnosed with asthma and 2% were currently employed in an occupation associated with an increased COPD risk. Lastly, LLN-defined COPD prevalence was 7% in our final sample. Significant differences between individuals with complete data and those with incomplete data (n=203 082) were found for all variables, except asthma status. Notably, the incomplete data subset had a considerably lower percentage of individuals with higher educational qualifications (38% versus 48%) and from higher income households (42% versus 55%), and higher proportions of current smokers (22% versus 3%) and individuals in occupations at risk of COPD (4% versus 2%).
Population characteristics
Table 2 shows the distribution of residential ambient air pollution concentrations. Mean±sd annual estimates of PM2.5, PM10, PMcoarse and NO2 were 9.94±1.04 µg·m−3, 16.18±1.90 µg·m−3, 6.41±0.90 µg·m−3 and 26.31±7.49 µg·m−3, respectively. NO2 concentrations were highly correlated with PM2.5 (r=0.87), but less so with other PM metrics. PM10 and PMcoarse were also highly correlated (r=0.81).
Pollutant descriptive statistics and correlation matrix#
Higher exposure to all pollutants showed significant associations with lower lung function (table 3). In adjusted models, a 5 µg·m−3 increase in PM2.5 exposure was associated with lower FEV1 (−83.13 mL, 95% CI −92.50– −73.75 mL), FVC (−62.62 mL, 95% CI −73.91– −51.32 mL) and FEV1/FVC ratio (−9.68, 95% CI −10.81– −8.56). For each 10 µg·m−3 increase in NO2, lower FEV1 (−33.85 mL, 95% CI −36.34– −31.36 mL), FVC (−33.47 mL, 95% CI −36.47– −30.46 mL) and FEV1/FVC ratio (−2.27, 95% CI −2.57– −1.96) were observed. Furthermore, results showed negative associations between PM10 and PMcoarse concentrations and lung function, with stronger effects on FVC than FEV1. The FEV1/FVC ratio showed no association with ambient PM10 exposure and a small positive association with PMcoarse (1.34, 95% CI 0.04–2.63, per 10 µg·m−3). In the main analyses for COPD prevalence, a significant association was observed for PM2.5 (OR 1.52, 95% CI 1.42–1.62, per 5 µg·m−3), PM10 (1.08, 95% CI 1.00–1.16, per 10 µg·m−3) and NO2 (1.12, 95% CI 1.10–1.14, per 10 µg·m−3), but not for PMcoarse (table 4). Associations per IQR increase in exposure are presented in supplementary tables S3 (for lung function) and S4 (for COPD). When compared to associations with smoking status, lower levels of FEV1 observed per 5 µg·m−3 increase in PM2.5 represented 65% and 29% of FEV1 loss associated with being a former and current smoker, respectively (supplementary table S5). Furthermore, the odds of COPD per 5 µg·m−3 increment of PM2.5 was equivalent to more than half the odds of COPD associated with passive smoking exposure at home (supplementary table S5). Sensitivity analyses restricted to those who had lived in the same place for the past 10 years did not substantially change lung function or COPD associations (supplementary tables S6 and S7). Finally, attributable fraction of COPD prevalence for residential ambient PM2.5 exposure above WHO guidelines was almost half (5.6%) that of current/past tobacco smoking (12.1%) in the cohort and over four times that of passive smoking exposure at home (1.2%).
Associations of lung function and ambient air pollution exposure#
Associations of chronic obstructive pulmonary disease and ambient air pollution exposure#
Results of PM2.5 and NO2 subgroup analyses for lung function and COPD are shown in tables 5 and 6, respectively. FEV1-stratified analyses showed stronger PM2.5 and NO2 associations among males, participants from lower income households and individuals with at-risk occupations. The same effect modification patterns were observed for FVC-stratified analyses, with never-smokers showing significantly lower FVC per PM2.5 and NO2 increase. Individuals from lower-income households had approximately twice as low FEV1 and FVC levels compared to higher-income participants and individuals with at-risk for COPD occupations showed three-fold lower FEV1 and FVC levels compared to individuals not in these occupations, per unit increase in PM2.5 or NO2 (table 5). Age, obesity, smoking status and household income, but not at-risk occupations modified the relationship between the FEV1/FVC ratio and PM2.5 and NO2, with stronger adverse associations for older, obese, current/past smokers and lower-income individuals. In COPD subgroup analyses (table 6), PM2.5 and NO2 associations were stronger among obese, lower income and non-asthmatic participants. Again, household income especially influenced the exposure–outcome relationship, with over three times stronger associations between COPD and each pollutant among lower- compared to higher-income individuals.
Lung function subgroup analyses for fine particulate matter with diameter <2.5 μm (PM2.5) and nitrogen dioxide (NO2)#
Chronic obstructive pulmonary disease subgroup analyses for fine particulate matter with diameter <2.5 μm (PM2.5) and nitrogen dioxide (NO2)#
Discussion
Ambient concentrations of particulate matter and NO2 air pollution were associated with lower lung function and increased COPD prevalence in this very large UK cohort. Given the size of the study, we were able to investigate interactions, finding evidence for effect modification, with larger impacts of air pollution on 1) lung function in males, individuals from lower-income households and individuals with at-risk occupations; and 2) COPD in obese, lower-income and non-asthmatic participants.
Lung function is a good indicator of respiratory morbidity and mortality, especially among COPD patients [22]. Given an average FEV1 loss of 32–46 mL·year−1 after the age of 30 years [12], the associations per 5 µg·m−3 exposure of PM2.5 found in our study are approximately equivalent to an additional 2 years of normal loss of lung function in healthy individuals if results in this cross-sectional study are confirmed in future longitudinal follow-up. We found significant reductions on lung function, even at a relatively low levels of ambient PM2.5, thereby echoing the need for more actions to be taken to control air pollution [23].
Comparison with studies using the same air pollution estimates
The current study replicated cross-sectional analyses in ESCAPE, the previous largest European study to date, using a single cohort with >10-fold higher numbers and the same models to estimate air pollutant exposures [15, 16] and similar covariate adjustment. Findings from two ESCAPE meta-analyses [5, 6] and from a Dutch study using ESCAPE air pollution estimates [20] are presented in figure 2. Our large sample size resulted in much smaller confidence intervals, with more statistically significant results and stronger evidence of an adverse effect of air pollution (figure 2). We found stronger (more negative) effects on lung function than in the studies by Adam et al. [5] or de Jong et al. [20] for each of the four air pollutants studied (PM2.5, PM10, PMcoarse, NO2). For COPD, our confidence intervals were much tighter than (but overlapped with) those in the study conducted by Schikowski et al. [6], but unlike that study, we found significant associations with both PM2.5 and NO2.
Air pollution exposure, a) forced expiratory volume in 1 s (FEV1), b) forced vital capacity (FVC) and c) chronic obstructive pulmonary disease (COPD): comparison with other studies using European Study of Cohorts for Air Pollution Effects (ESCAPE) air pollution estimates. Cohorts included in each study: Adam et al. [5]: European Community Respiratory Health Survey (ECRHS), French Epidemiological Study on Genetics and Environment of Asthma (EGEA), National Survey of Health and Development (NSHD), Study on the Influence of Air Pollution on Lung Function, Inflammation and Aging (SALIA) and Swiss Cohort Study on Air Pollution and Lung and Heart Diseases in Adults (SAPALDIA); de Jong et al. [20]: Lifelines Cohort Study; Schikowski et al. [6]: ECRHS, NSHD, SALIA and SAPALDIA. PM2.5: fine particulate matter with a 50% cut-off aerodynamic diameter <2.5 μm; PM10: particulate matter with a 50% cut-off aerodynamic diameter <10 μm; PMcoarse: coarse particulate matter with diameter 2.5–10 μm; NO2: nitrogen dioxide.
The mean and range of estimated annual NO2 concentrations in our study were similar to those of studies included in the original ESCAPE meta-analyses [5, 6], whereas mean PM concentrations were generally lower in our study, with the exception of the British National Survey of Health and Development, which were comparable (supplementary tables S8 and S9). The range of air pollutant concentrations used in de Jong et al. [20] was smaller than in our own study (supplementary table S10). Using similar air pollution models to those used in past ESCAPE studies means that differences in lung function and COPD associations are less likely to be due to differences in exposure estimates [24]. However, given that the original ESCAPE meta-analyses by Adam et al. [5] and Schikowski et al. [6] back-extrapolated air pollution estimates to date of lung function measurement by up to two decades for some participating cohorts, the larger effect size seen in our study may in part relate to reduced air pollution exposure misclassification. Finally, the same spirometers and spirometry protocols were applied in UK Biobank, whereas this was not the case across original ESCAPE studies. This may have also contributed to more precise estimates in our study.
Comparison with studies using other air pollution estimates
Our results are consistent with the small number of studies investigating PM10, PM2.5 and NO2 in relation to lung function, but few studies have investigated PMcoarse. In a study of UK residents, Forbes et al. [25] reported comparable results for FEV1, showing a 92 mL and 22 mL decrease per 10 µg·m−3 increase in PM10 and NO2, respectively. A study of 9651 healthy never-smokers in the Swiss Study of Air Pollution and Lung Disease in Adults (SAPALDIA) also found negative effects of both NO2 and PM10 exposure on FEV1 and FVC [26]. An analysis of Framingham Heart Study participants by Rice et al. [3] found significant negative associations of residential PM2.5 exposure with both FEV1 and FVC levels, and a faster decline in lung function levels.
Our findings of associations between PM2.5 and airflow obstruction and COPD are consistent with a recent study of 285 000 Taiwan residents showing significant associations between ambient PM2.5 and reduced FEV1/FVC ratio and risk of COPD [4], and with findings from the German Study on the Influence of Air pollution on Lung, Inflammation and Aging (SALIA) cohort in relation to NO2 and FEV1/FVC ratio and spirometrically defined COPD [27]. A separate analysis of SALIA participants also showed a decline in COPD with reduced NO2 concentrations [28]. However, in contrast to our findings, no associations of NO2 or PM2.5 exposure with FEV1/FVC were reported by Forbes et al. [25] or Rice et al. [3].
Effect modifiers of air pollution
We observed considerably stronger associations for lung function and COPD among individuals from lower-income households. The greater vulnerability of lower-income individuals to the respiratory health effects of air pollution exposure is in line with previous studies [18, 29, 30], and is probably due to numerous factors, including more childhood respiratory infections, poorer housing conditions and indoor air quality, poor nutrition and occupational exposures [31].
Our study found that occupational status in a job judged at risk of COPD to be an important effect modifier of associations between air pollution exposure and lung function, but not its associations with COPD. The latter may be due to the “healthy worker” effect, whereby those with COPD are less likely to be employed in an at-risk job; we did not have information on past occupation. Few studies are available for comparison, but the Harvard Six Cities Study found higher relative risks of death per unit of PM2.5 among individuals reporting workplace exposure to dust or fumes [32].
We observed stronger PM2.5 and NO2 associations with FEV1 and FVC among males and stronger associations between PM2.5 exposure and COPD for females. Equivocal evidence has been found regarding effect modification of sex in associations between air pollution exposure and lung function and COPD in adults [33]. In studies reporting stronger effects in males, work-related exposures leading to greater predisposition to airway disease, and more time spent outdoors potentially resulting in higher exposures for a given concentration have been suggested as potential sources of differential effects [34, 35]. Hypotheses for a larger impact of air pollution on lung health among females include more time spent at home leading to better accuracy of residential air pollution exposure assignment, as well as biological factors such as greater airway reactivity [36, 37].
We also found significant effect modification by obesity, with higher air pollution associations with COPD risk and reduced lung function for obese individuals, which is consistent with other studies using ESCAPE air pollution estimates [5, 20]. Mechanistic studies have shown greater than additive effects of excess body fat and air pollutant exposure on systemic inflammation and oxidative stress [38, 39], suggesting an enhanced response to inflammatory stimuli [39], resulting in airway damage and inflammation in obese individuals.
Stronger negative effects of air pollution on respiratory disease among never-smokers have been reported previously [6, 8, 40], which our analyses also found. As smoking might already reduce pulmonary function through inflammatory pathways, any additional impact of air pollution on respiratory abnormalities could be smaller or harder to detect in this subgroup.
In addition, we found that asthma status modified the associations between PM2.5 and NO2 and COPD prevalence, with significantly stronger associations in non-asthmatics. This may be related to treatment in asthmatics, modifying adverse impacts of air pollution or alternatively, avoidance in that asthmatics aware of impact of air pollution on symptoms may choose to live in less polluted areas.
Strengths and limitations
The large sample size of our study provided good statistical power to assess effects of air pollution, even in relatively small subgroups such as individuals in occupations with increased COPD risk. An additional strength of our study was the use of a single well-respected cohort with a rigorously defined protocol.
A potential major limitation of the study is the large number of participants with missing data for covariates included in our final regression models. This did not appear to be missing at random (therefore difficult to address using imputation), but gave us a wealthier and healthier cohort. This does not invalidate findings, but may affect generalisability. Given our findings of interactions with lower socioeconomic status individuals, we might expect this would underestimate associations of air pollution and lung function and COPD in a general population.
Another limitation is that while COPD should be classified using post-bronchodilator spirometry tests, only pre-bronchodilator measures were available, similar to the ESCAPE five-cohort analysis [5]. The extent that air pollution affects FEV1 and FEV1/FVC ratio could potentially have been mitigated if assessed post-bronchodilator.
Common to most other ambient air pollution studies, we used place of residence to estimate air pollution exposure, which will result in exposure misclassification. Furthermore, annual air pollution estimates at recruitment address were modelled to a single year (2010), which may differ by up to 4 years from when lung function was measured. We made a reasonable assumption that the spatial contrast in air pollution exposures will have been relatively stable in the UK over these years [41], but cannot exclude the possibility of exposure misclassification. Finally, the cross-sectional relationship between air pollution and lung function and COPD demonstrated in our study show associations, but are prone to the influence of confounders and do not allow us to examine temporal patterns between air pollution exposure and respiratory outcomes. Longitudinal analyses of future follow-up data in large cohorts such as UK Biobank are needed to strengthen inferences regarding causal relationships between air pollution and respiratory disease, particularly among vulnerable subpopulations.
In conclusion, this is one of the largest analyses to date to examine associations between ambient air pollution and lung function and COPD. Air pollutant concentrations were clearly associated with lower lung function and increased COPD prevalence with higher impacts in males, individuals from lower income households, those in occupations with adverse respiratory exposures and those who were obese.
Supplementary material
Supplementary Material
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Supplementary material ERJ-02140-2018.Supplement
Footnotes
This article has supplementary material available from erj.ersjournals.com
Author contributions: D. Doiron, K. de Hoogh and A.L. Hansell proposed the study; all authors contributed to development of the study design; D. Doiron conducted the statistical analyses and wrote the first draft of the paper; all authors commented on results and contributed to the manuscript.
Conflict of interest: D. Doiron has nothing to disclose.
Conflict of interest: K. de Hoogh has nothing to disclose.
Conflict of interest: N. Probst-Hensch has nothing to disclose.
Conflict of interest: I. Fortier has nothing to disclose.
Conflict of interest: Y. Cai has nothing to disclose.
Conflict of interest: S. De Matteis has nothing to disclose.
Conflict of interest: A.L. Hansell has nothing to disclose.
Support statement: This research has been conducted using the UK Biobank Resource under application number “9946”. Y. Cai is supported by an MRC Early Career Research Fellowship awarded through the MRC-PHE Centre for Environment and Health (grant number MR/M501669/1).
- Received November 8, 2018.
- Accepted April 22, 2019.
- Copyright ©ERS 2019
References
