Ethics and approvals

All the participants of the Oslo Health Study gave their written consent. The participants' names and personal ID numbers were omitted when data were used. The Norwegian Data Inspectorate approved the study and the Regional Committee for Medical Research Ethics evaluated it.

Dependent variable: self reported air pollution problems (APP)

Self reported air pollution problems (APP) was derived from the question "Are you (in your local environment) troubled by air pollution from traffic?", with the response categories 'very troubled' and 'somewhat trouble' collapsed into yes, and 'not troubled' indicating no. There may be some ambiguity in the participants' interpretation of what this question actually measures. Ideally, the participants' answers to this question should be based on a reflection of the local levels of traffic related air pollution levels, which would be a direct way to measure the correspondence between modeled and perceived air pollution. However, the question is formulated so that the participants were asked to rate to what extent they feel troubled by a given level of air pollution. This may vary from person to person, and thus is not the same as considering the amount of air pollution levels per se. Instead we consider it an approach that indirectly captures the perceived pollution levels, as the likelihood to feel troubled by air pollution is likely to increase with increased levels of air pollution. The question was chosen from a list of several local items that the participants were asked to consider whether they were troubled by in their local environment (e.g. traffic noise, noise from neighbours, poor drinking water). In this context we believe it was clear to the study participants that the question was related to an assessment of the quality of local area features, not whether or not these features were affecting their health.

Modeled exposure assessment

Indicator of ambient air pollution exposure at the participants' home addresses was nitrogen dioxide (NO₂). NO₂ was chosen as indicator of air pollution exposure as it is considered to be a good marker for traffic related air pollution, has been shown to have spatial variability, and comparisons between modeled and measured levels have shown reasonable agreement [8, 9].

Air pollution data was estimated using the GIS-based Air Quality Information System (AirQUIS) developed at the Norwegian Institute of Air Research (NILU) [10, 11]. This model combines data on meteorology, emissions, background air pollution concentrations and topography. The model calculates the ambient exposure levels according to home addresses on a km² grid and a large number of receptor points close to busy roads.

On the day the study participants were examined they were given a supplementary questionnaire in which they should answer the APP-question. The response rate of this questionnaire was 84% (that is 37% of all invited). Most participants returned the questionnaire after 2–4 weeks. Based on this, we calculated the average NO₂ exposure at the participants' home addresses in the four weeks after examination. This modeled exposure would reflect the time period in which the participants answered the APP-question. NO₂ values were quintilized for the analyses. Initial analyses with yearly average values of NO₂ and analyses where NO₂ was measured continuously were performed with results quite similar to the 4-weeks variable (figures not shown). We therefore chose to proceed with the 4-weeks variable for two reasons. Firstly, to minimize the possibility of migration bias and secondly to better handle the possibility that the participants' reporting of APP could be affected by the air pollution levels at the time of answering the questionnaire, rather than being a reflection of what it is like in the area of living during the year as a whole.

We adjusted for season of examination by categorizing the months of examination into seasons; December, January and February (winter); March, April and May (spring); June, July and August (summer); September, October and November (autumn) [12].

Independent variables

We adjusted for two indicators of housing quality that may affect the likelihood to feel troubled by outdoor air pollution. Dwelling damp, draughts or cold was derived from the question "Are you (in your home) troubled by damp, draughts or cold?" with the response categories 'very troubled' and 'somewhat trouble' collapsed into yes, and 'not troubled' indicating no. Similarly, poor indoor climate was derived from the question "Are you (in your home) troubled by other forms of poor indoor climate?" with the same responses collapsed. Self reported pollution due to wood or oil heating, factory etc. was derived from the question "Are you (in your local environment) troubled by air pollution due to wood or oil heating, factory etc.?" with the response categories 'very troubled' and 'somewhat trouble' collapsed into yes, and 'not troubled' indicating no. This question is rather similar to APP in the sense that there may be some ambiguities in the rationales of the respondents' answers cf. our comments in reference to the APP-question.

Area deprivation was a composite measure of five items, a method that has been shown to have stronger independent effects on health than any one variable on its own [13]. The items were area percentages of population affected by social aid, being unemployed, receiving disability pension, having no academic college or university education, and average taxable income in the areas. A rank score for Oslo's 25 administrative areas was calculated and quintilized for the analyses [14]. Each participant was assigned a value for area of living in year 2000. The Oslo City Council provided these data.

In addition, we adjusted for depression because there is a possibility that it may be present in some of the chronic diseases and therefore related to the reporting of air pollution. Depression was derived from the question "Have you during the last two weeks felt depressed?" with the response category 'no' indicating no, 'a little' indicating some, and the categories 'quite much' and 'very much' indicating yes.

Asthma, diabetes, fibromyalgia and osteoporosis were derived from the question "Do you have any of these illnesses, or have you suffered from any of them in the past?" Self rated health was derived from the question "How would you describe your present state of health?" dichotomised into good (good or very good) and poor (not very good or poor). Coronary heart disease (CHD) was measured by The Rose Questionnaire of angina pectoris [15], and chronic obstructive pulmonary disease (COPD) by a modified version of the Medical Research Council's questionnaire with three items [16].

Analyses

The statistical analyses were conducted using the SPSS 11.0 software program. The chi-square test and independent samples t-test were used in comparing the excluded (i.e. those with missing values) and the included respondents. Both logistic and ordinary least square regression was used in the main analyses. In order to investigate a potential over-reporting of APP among people with a chronic disease, we first conducted a logistic regression analysis with APP as the outcome. In this analysis, NO₂ was included as an independent variable. In a second analysis (ordinary least square regression), NO₂ measured as a continuous variable was modeled as the outcome (APP was included as an independent variable). The purpose of this was to explore whether health was associated differently with APP and NO₂. Theoretically, if APP is a good indicator of air pollution levels, then we would expect rather similar associations between health variables and APP/NO₂-levels. Modelling NO₂ as the outcome is an unusual strategy, but in this cross-sectional study we were not looking for causal effects, only statistical associations. If health is associated with APP, and not NO₂, we then believe it indicates an over-reporting of air pollution among people with a disease. An association between health and NO₂, on the other hand, may provide support for the presence of causal air pollution effects on health. Therefore, in order to test for over-reporting it is important to compare how both APP and NO₂ are related to health.

In both analyses each health variable was adjusted for age, sex, employment status, education, smoking, area deprivation and season of examination. In the logistic regression analysis we also investigated a model in which dwelling conditions, depression and self reported pollution due to wood or oil heating, factory etc. were removed. The rationale for this was that while we believe these variables are important contributors to the construct of APP, they are poorly measured, and may obscure our results.

Study limitations

There are several limitations to our study. Our health measures were self reported. Thus we cannot say whether our results indicate that those with a disease over-report air pollution problems or whether those likely to report disease in a questionnaire are also over-reporting air pollution problems.

We were unable to account for two conditions that may be important. First, daily mobility, in order to capture some of the potential for exposure misclassification in those that reside in the city but work elsewhere as those who work close to their homes would be most likely to have an accurate exposure assessment [17]. Second, objective measures of dwelling conditions, as our data on these aspects were self reported and may be biased by disease. Previous studies have found that those who rent (compared to those who own) their dwellings are significantly more affected by noise, hazards, vibration, cold and dampness [18]. Indoor exposures may be as important as outdoor exposures [19], although there is evidence that indoor pollution can be directly related to outdoor pollution [20]. Other dwelling factors that righteously may shape reporting of air pollution problems (floor level, whether windows are against road or backyard, etc.) were not available.

Thus, APP is a concept that reflects much more than just exposure to different levels of air pollution, e.g. NO₂. It may therefore not be a very good agreement between the two. We identified several other factors that were independently associated with APP, but not with NO_2.Interestingly, we found that the exclusion of some of the dwelling factors led to increasingly stronger associations between health variables and APP, which indicates that APP does not only reflect air pollution in the local area, but may also reflect air quality in the home itself.

The items we used to measure dwelling factors (damp, draughts, cold or poor indoor climate), as well as our item measuring depression, are of some concern. They are all self reported, not validated, and are somewhat ambiguous. In lack of better data, we still considered it useful to include them in our analyses, as they are intended to measure assumingly important contributors to APP. Our analyses proved this to be right. These variables were independently associated with APP, and they did not attenuate any other associations except the health variables. This indicates that they capture some aspects that are important for the reporting of APP. This became especially clear when we analysed their associations with NO₂ and found that none of them were significantly associated.

Contributors to the construct of self reported air pollution problems

Different environmental impacts may result from at least two important pathways: differential exposures or differential susceptibilities [21]. The incorporation of social stratification as a health effect modifier is well established in air pollution epidemiology [17, 22–24]. Evidence have been provided that lower socio-economic status areas [25] and households [19] experience the worst air quality, described as examples of the 'inverse air law' [19], analogous to the inverse care law, i.e. that people with the worst lung function tend to live in areas with the worst air quality. This may trigger 'the triple jeopardy', which begins with increased exposure among lower status groups, is augmented by the pre-existing burden of poor health that accompanies low status, and is confounded by an interaction between the two conditions [26].

Over and above modeled air quality many aspects of the environment have been identified in shaping people's perception of the air quality [27–29], indicating that perceptions are socially and culturally constructed [30]. Individuals' perceptions of whether or not they have the ability to bring about change through their behaviour may also influence the perception of local air quality or degree of air pollution problems [27]. For example, when the analyses of Heinrich et al. [1] were restricted to participants with asthma or hay fever, the subjective assessments of air pollution from traffic were increasingly overestimated compared to the modeled levels. Over-reporting of pollution exposure among people with respiratory disease (or any family members) has been found in other studies [31, 32], and it has been argued that this over-reporting could be caused by the publicity/media coverage given to air pollution and respiratory health [31]. Therefore, it does not seem unreasonable to claim that the associations between self reported air pollution/traffic and disease may be severely biased by the concern and awareness of having a disease that the lay public link to air pollution. For example, Petrie et al. [33] identified how certain modern health worries were associated with having a disease that most people would be likely to associate with it (e.g. tainted food concerns and gastrointestinal problems, or toxic intervention concerns and pseudoneurology complaints). Seemingly, this logic does not apply to people with diabetes, in which social inequalities are not so marked [34]. The distribution is less stratified by social position and area of living in Oslo, compared to other diseases. The patients themselves often regard their condition as partly a result of obesity, but not from social circumstances, and not from pollution which is more prevalent in the deprived areas of Oslo. Thus, they may be less apt than other chronically ill to be concerned about pollution [34].

It has been claimed that more educated people have longer time horizons than the poorly educated [35], which could explain why well-educated people behave differently as they are more concerned about (and possibly more aware of) the long-term consequences of day-to-day activities [36]. It could be that those with diseases commonly related to air pollution are more likely to report air pollution problems due to a greater concern of the long term-consequences of such exposure. This assumption was tested in our analyses by including both respiratory and non-respiratory diseases, and our conclusion is that the over-reporting of air pollution problems seen for all diseases indicates an over-reporting from chronic disease in general, and not only from those diseases commonly related to air pollution.

The association between air pollution and respiratory health seem well established [37, 38], although there is unclear evidence of impacts of outdoor air pollution and asthma incidence [38–40]. Thus, claiming that self reported air pollution/traffic intensity is predictive to health seems plausible. But such a presumption is normally not tested against diseases for which the etiological evidence in reference to air pollution is scarce or missing. This is especially problematic in those studies that do not have any objectively measured equivalents to the self reported problems, as they are then unable to differentiate the part of the association into that of true exposure and that of a socially or culturally constructed bias. Three such biases were tested in our analyses; that of disease, that of living in deprived areas, and that of depression. After the partitioning of NO₂and disease/area deprivation/depression was done in our analyses, our results indicated, despite the cross-sectional design, that health status affects the reporting of air pollution problems, making the latter an unreliable variable in studies on air pollution effects on health.