Introduction Outdoor air pollutants are associated with respiratory morbidity and mortality, but little longitudinal work has been undertaken in this area in chronic obstructive pulmonary disease (COPD). Patients with α-1-antitrypsin deficiency (AATD) typically exhibit faster decline of lung function than subjects with usual COPD and thus represent a group in whom studies of factors influencing decline may be more easily clarified.
Methods Decline of FEV1 and KCO in subjects of the PiZZ genotype from the UK AATD registry were studied. Pollution levels (PM10, ozone, sulphur dioxide, nitrogen dioxide) during the exposure window were extracted from GIS maps, matching the measurement to each patient's home address. Clinical predictors of decline were sought using generalised estimating equations, and pollutants added to these subsequently. Single pollutant models were used due to multicollinearity.
Results In the FEV1 decline analysis, higher baseline FEV1 was associated with rapid decline of FEV1 (p<0.001). High PM10 exposure predicted more rapid decline of FEV1 (p=0.024). In a similar analysis for KCO decline, higher baseline KCO predicted rapid decline (p<0.001) as did higher exposure to ozone (p=0.018). High PM10 exposure also showed a trend towards this effect (p=0.056).
Conclusions Exposure to ozone and PM10 predicts decline of lung function in AATD.
- Air pollution
- α-1-antitrypsin deficiency
Statistics from Altmetric.com
What this paper adds
Outdoor air pollution is associated with respiratory morbidity and mortality, but longitudinal epidemiological studies have been hampered by the use of crude outcome measures.
Exposure to outdoor air pollution relates to decline of both gas transfer and FEV1 in subjects with α-1-antitrypsin deficiency.
Effects may segregate between airways and parenchymal disease, and by gender.
Adverse health effects have been linked to air pollution by both epidemiological and toxicological studies. Mortality effects have been seen in both long term and time series studies for particulate matter,1–4 while for ozone5 6 and sulphur dioxide (SO2)3 7 this effect has been seen in time series studies. However, studies of outdoor air pollution in chronic lung disease are fraught with methodological issues, including which outcome measure to use and what exposure window is most appropriate. A sensitive method accounting for this would be to look at decline of lung function as an outcome measure, with exposure window constituting the period over which decline is calculated.
α-1-Antitrypsin deficiency (AATD) is a genetic disorder that predisposes to the development of chronic obstructive pulmonary disease (COPD)8 due to the relatively unopposed action of neutrophil elastase, which degrades lung elastin9 and leads to a greater disease burden in those exposed to inflammatory stimuli, predominantly cigarette smoke.8 Air pollution also induces lung inflammation10 and we have shown that ozone exposure related to lung function in a cross-sectional study of subjects with AATD in the UK.11 Other environmental agents, predominantly relating to occupation, may also influence lung function in AATD,12 13 but it is not known whether they influence lung function decline in this group.
Decline of lung function over time occurs due to age, but may be faster in those with AATD, with a mean loss of 81 ml/year in FEV114 compared to 55 ml/year in usual COPD15 and 30 ml/year in healthy individuals. Decline of FEV1 may be influenced by cigarette smoking,14 exacerbation frequency,16 baseline lung function17 and age.17 Information pertaining to all these covariates is available for subjects within the UK AATD registry, together with longitudinal data on lung function. The current study documents the decline in lung function over time in a well characterised group of AATD subjects and explores the impact of a variety of factors including ambient air pollution on this decline.
All PiZZ subjects from the UK national registry for AATD who had completed a baseline assessment and at least one follow-up appointment by the year 2006 were studied, giving a total of 401 individuals. Of these 220 had at least 4 years of lung function recorded. The study was approved by the local research ethics committee and all patients gave informed consent. All patients had a serum α-1-antitrypsin (AAT) level of <11 μM and PiZZ genotype confirmed by specific PCR (Heredilab, Salt Lake City, Utah, USA). Lung function tests were performed as described previously,18 at our centre in Birmingham. Exacerbations were defined by standard Anthonisen criteria,19 the data being obtained by questionnaire, and those events with symptoms consistent with type 1 or 2 exacerbations included. Active smoking was quantified using the standard pack-year measure. None of the subjects had ever received AAT replacement. Two subjects were excluded because they had a lung transplant during the period of follow-up. Patients' addresses throughout the time period used to calculate decline were mapped to Ordnance Survey (OS) coordinates using the National Statistics Postcode Directory.20 The census was also used to obtain the Carstairs deprivation index at the ward level for each subject's address.
To assess any relationship between occupational exposure and risk of lung damage, an occupational hygienist created a job exposure matrix21 using Standard Occupational Classifications (SOC 2000) generated using the CASCOT22 system. Job codes were divided into three groups according to the likelihood and probable intensity of exposure to agents known to cause occupational lung disease with potential to affect lung function. Patients who were working at any time during the period used to calculate lung function decline in a job where the exposure intensity was likely to be less than 30% of the workplace exposure limit23 were classed as low risk, with intensities above this being deemed high risk. Those who were not working or working in a non-exposure prone job were classed as zero risk. Where no workplace exposure limits were available, and there were no published data regarding the risk associated with the profession, subjects were classed as having missing data for this field.
Annual mean data for SO2, nitrogen dioxide (NO2) and PM10 were obtained from geographical information system (GIS) maps for each of the years patients had been followed up. As annual mean ozone is not available at the level of resolution of GIS maps, an alternative metric was used for this pollutant. AOT40 (ozone μg/m3.h) is the parameter used to represent accumulated ozone dose and is the sum of the differences between the annual hourly mean ozone concentration and 40 ppb for each hour when the concentration exceeds this limit during daylight hours. The maps detail pollutant levels on a 1 km×1 km grid across the UK, using data from the National Atmospheric Emissions Inventory and a combination of modelling methods, including dispersion kernel approaches and weighted regression analyses to calculate effects from major industrial sources. This methodology and its validity is described in detail elsewhere.24 25 Pollutant levels were mapped to patients using the OS coordinates from the grid. The cumulative dose of each pollutant per patient over the period used to ascertain decline was calculated from this data.
Data were analysed using SPSS v 15. Age, gender, smoke exposure, level of occupational risk and baseline lung function were added to multivariate models, each of which included a single pollutant. The multivariate models used generalised estimating equations (GEES), with lung function decline as the outcome measure. We chose FEV1 and KCO as the specific lung function parameters for which to calculate decline; these were chosen because FEV1 is a standard measure of severity of COPD, and KCO a good measure of parenchymal lung disease, deemed sensitive in AATD in a recent randomised controlled trial.26 In SPSS GEES can be structured to include a time function, thus accounting for repeated measures of both lung function and pollution for any given individual. Multi-pollutant models were not used due to multi-collinearity of the pollutant data.
Subgroup analyses were carried out based on (i) more or less than 4 years of lung function data available and (ii) index status, where lung index cases were the first case of lung disease related to AATD within a family, and non-index cases those diagnosed through family screening.
The clinical characteristics of the subjects are shown in table 1. The only significant difference between the whole and the subgroup with at least four lung function data points was that the subgroup tended to work in higher risk professions (p<0.01). The majority of the population were working; a breakdown of commonly observed risk professions is shown in the online supplementary material, together with information pertaining to their geographical distribution across the UK.
All measured pollutants showed significant variability over time (all p<0.0001), the trend being to lower levels of SO2 and NO2, but with no clear trends for either particles or ozone. Annual mean levels in the group and the mean per year during the period over which decline was calculated are shown in table 2.
In the GEES analyses, pollution exposure and how this related to lung function decline was the main outcome of interest. However, other clinical predictors were included in the models and it is useful to consider their relationship to decline, and some of the interactions between variables. Baseline FEV1 was the only significant predictor of FEV1 change over time (p<0.001). Similarly, baseline KCO was a predictor of KCO change over time (p<0.001). Neither gender, smoke exposure, age or occupational risk predictors were significant predictors of the change in FEV1 or KCO. Of the pollutants studied, higher PM10 exposure was associated with FEV1 decline (p=0.024) and tended towards an association with KCO decline (p=0.056). High ozone exposure was the most significant predictor of KCO decline (p=0.018).
Gender differences in response to environmental agents, such as cigarette smoke, are recognised. For this reason we sought gender interactions as well within both the FEV1 and KCO decline analyses. These analyses showed some gender differences: in the FEV1 decline analysis, female subjects in low or high risk jobs exhibited greater lung function deterioration than male subjects in such jobs (p=0.038 and p=0.03, respectively). No interaction with pack-years smoked or outdoor air pollution was seen. In the KCO decline analysis, female gender exhibited a significant interaction with pack-years smoked, such that female subjects declined faster (p=0.021). For occupational risk, females in low risk jobs relative to zero risk jobs declined less than similar males (p=0.043), but as there was no clear dose–response relationship across all levels of at-risk occupations, this may be a chance finding. No gender interactions with pollution were seen.
The GEES analyses generate a B value, similar to a regression co-efficient, which can be used to express the change in each lung function parameter per year for a 1 unit increase in the predictor, where this is quantitative, such as pack-years, age or pollutants. For categorical predictors, such as gender, the B value can be used to derive the change in lung function over time when compared to a specified default category. Thus male and female gender and different categories of occupational risk can be compared. Table 3 shows all assessed predictors for FEV1 and KCO analyses, the mean change in each lung function parameter and the associated 95% CI.
This table shows the difference in FEV1 and KCO changes over time for females in comparison to males with the same environmental exposure. For instance, the second row shows a decline in FEV1 of 151 ml/year in females in low risk jobs compared to males in low risk jobs, after adjustment for other covariates, such as age and cigarette smoke exposure.
In the subgroup analyses, PM10 remained associated with a greater degree of FEV1 decline in those with four or more years of lung function measurements (mean (95% CI) −3 (−0.2 to −5), p=0.033), but when considering only lung index cases or non-index cases, the significance was lost (both p>0.1). The ozone association with change in KCO did not alter when considering only those with four or more years of lung function measurements (mean (95% CI) −0.009 (−0.02 to −0.001), p=0.028) nor when considering only lung index cases (−0.02 (−0.03 to −0.003), p=0.008). There were too few non-index cases to substratify effectively in a similar ozone analysis.
The current study assesses factors influencing the decline in FEV1 and KCO with particular reference to ambient air pollution. In contrast to a previous paper from our group in this area,11 this study has been able to take account of window of exposure, and actual cumulative exposures, thus strengthening confidence in the associations seen.
In order to interpret the data, the methods used to calculate decline need consideration. FEV1 has a variability of about 100 ml between repeated measures,27 which exceeds the average annual decline of FEV1 in AATD. Thus, calculation of decline from two annual measures might be insufficiently sensitive. However, the SE of the measure is inversely proportional to the number of years over which it is measured, such that over 3 years the error of the slope reduces to 33.3 ml/year—less than the decline described in AATD previously.28 In the current study, the mean decline was similar to this, but the addition of two more data points reduces the error further (to approximately 11.1 ml/year), so that calculated decline should be reliable for individuals in the subgroup with at least 4 years' measurement. Gas transfer shows a greater degree of variability in its measurement than FEV128 and is a more sensitive measure of decline in AATD than FEV1.26 Although no data are available on intra-measurement variability, unlike for FEV1, it is likely that similar principles will apply when considering decline.
The subgroup with four or more data points from which to calculate decline is by definition smaller in number than the whole group, thus there is loss of power, and selection bias could have been an issue. It was for these reasons that the whole group was used, and then subgroup comparisons made. There was a difference in the risk level of occupations between the whole group and the subgroup, but no other feature differed. This is suggestive of a degree of selection bias, albeit not affecting the main pollution analyses within the paper but a subgroup. It is interesting to speculate why those in the subgroup who had at least four data points from which to calculate decline tended to work in higher risk professions that the whole group—it may be that those working in high risk professions are more motivated to continue follow-up, perhaps because they are used to regular assessments as part of occupational health follow-up in their workplace, or because of greater concern regarding their respiratory health.
It is also relevant to consider the pollution data used to derive estimates of exposure. Sophisticated statistical modelling methods that account for meteorological effects, dispersion from large sources and roads in addition to background pollution levels, has allowed the generation of GIS maps which resolve pollution levels on a 1 km×1 km grid across the UK. Although this is the greatest resolution of pollution possible in large scale geographical work, there is some potential for misclassification of exposure which could not be resolved without the use of prospective individualised measurements. The modelling data have been validated by those responsible for map generation, published elsewhere,24 25 and their inter-relationships discussed in our previous work.11 Extensive longitudinal data exist for some, but not all, pollutants and the lack of available longitudinal data pertaining to ozone may be of particular importance in AATD, since AAT provides approximately 80% of the lung's protection against ozone induced inflammation29 and was associated with disease severity in our previous study.11
FEV1 decline was associated with higher baseline FEV1, consistent with the fact that decline tends to be fastest in those with moderately impaired, rather than severely impaired, FEV1.30 Increasing exposure to PM10 was associated with greater decline in FEV1, consistent with a previous population based study of FEV1 decline,31 which determined change in FEV1 by the use of two spirometric measures 10 years apart, such that the measure was accurate to within 10 ml/year during the follow-up period.31 However, differences in FEV1 decline attributable to PM10 were less than 10 ml, so it is possible the results represent natural variation. Similarly in the current study, differences in FEV1 decline attributable to PM10 were small (averaging 3 ml/year), such that their clinical significance could conceivably depend on the duration of exposure. Further longitudinal studies will be required to clarify this.
KCO decline was predicted by baseline KCO, which is perhaps intuitive and in line with the association of FEV1 decline with its baseline value. Of the outdoor air pollutants studied, only ozone was associated with increased KCO decline, although PM10 also showed a strong trend in this direction. In AATD the predominant pathology is parenchymal disease,8 such that changes in KCO may reflect disease progression better than FEV1. In our previous cross-sectional study of pollution effects in AATD, an association was shown between ozone and markers consistent with parenchymal lung disease.11 The earlier study was smaller because we considered only subjects who had never moved from their place of birth, in order to provide a cross-sectional model of lifelong pollution exposure. Also, although we demonstrated linear relationships between current and past pollution exposures to validate our previous cross-sectional model, this could be considered flawed, since it could not account for true cumulative pollution exposure, unlike the longitudinal model shown here. The results for KCO decline support our previous work in concluding that ozone is important in determining lung function in AATD, but also suggest that PM10 could have an influence. Since PM10 is associated with airway inflammation,10 acute changes in FEV132 and mortality,1 2 while ozone is associated with hospital admissions for respiratory disease and mortality,6 33 our results are consistent with current knowledge of pathophysiology and epidemiology for both pollutants.
We have also looked at gender differences in lung function after exposure to each of the environmental agents studied, by using a gender interaction term with each agent in the GEES analyses. Some interesting differences were seen with regard to occupation and cigarette smoke exposure. Many studies, reviewed recently elsewhere,34 have suggested gender differences in the effect of cigarette smoke, with women appearing more susceptible. Our observation of increased KCO decline in females for a similar smoke exposure is consistent with this. Explanations proposed for this association include altered immune responses to infection, altered nicotine metabolism and a relative under-estimation of smoke exposure in women, who tend to be exposed more to passive smoke than men.34 The lack of association with FEV1 decline may reflect the fact that KCO decline is a more sensitive measure than FEV1 decline in AATD, as discussed previously, or indicate differences in the subtype of COPD exhibited by females. The rapid decline in FEV1 exhibited by women in exposure prone jobs could also relate to this. In usual COPD cohorts there is a relative predominance of airway disease rather than parenchymal disease in females,35 a feature generally thought to be reflected by FEV1. The lack of consistent association of occupational risk with KCO decline in women may simply be an effect of sample size as indicated by the wide confidence limits. The small numbers of women in the highest risk professions and earlier age of retirement (hence fewer current female workers) may also have influenced the analysis. There is little current evidence to suggest mechanisms of differential occupational risk in women, or indicate if gender differences in response to outdoor air pollution exist. We have not been able to prove the latter here, but it is conceivable that shared inflammatory mechanisms after any inhaled exposure could be influenced by sex hormones, as is thought to be the case with cigarette smoke.
This study is strengthened by the detailed clinical assessments made, and the ability to adjust for age, gender, social deprivation, cigarette smoking and occupational exposure. It is, however, limited by its size, particularly with regard to assessments of occupational risk and of social deprivation, which would usually require much larger numbers for adequate study. When assessing occupational risk we used an arbitrary workplace exposure limit of 30% and graded risk from their current job only. More detailed quantification would require contemporaneous measurement of risk agents in each work environment. For large scale studies, as needed to assess pollution effects in the general population, this would be prohibitively complex and expensive. A final point to note is that our analyses have not been corrected for multiple statistical testing. This is because both FEV1 and KCO decline correlate with one another, as do the pollutants studied. A Bonferroni correction would not be appropriate under these circumstances. We restricted decline analyses to two measures of lung function, and did not consider single symptom changes over time, in part to avoid testing overlapping outcome measures, and in part because of the risk of false positive results when performing multiple statistical tests. FEV1 and KCO decline are the most well recognised, and validated, outcomes in an AATD population and as such the proper focus for our study.
In summary, these data show a significant association between outdoor air pollution in general and disease progression in AATD, which concurs with population based work and highlights the importance of KCO in the monitoring of patients at risk of COPD.
The authors would like to thank Andrew Kent at AEA Energy and Environment for provision of the pollution data and Peter Nightingale, statistician at the Wellcome Trust Clinical Research Facility, for advice on statistical analysis.
Competing interests None.
Ethics approval This study was conducted with the approval of the South Birmingham LREC 3359.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.