Article Text

PDF

Does deprivation index modify the acute effect of black smoke on cardiorespiratory mortality?
  1. M Carder1,
  2. R McNamee2,
  3. I Beverland3,
  4. R Elton4,
  5. G R Cohen5,
  6. J Boyd6,
  7. M Van Tongeren7,
  8. R M Agius7
  1. 1Occupational and Environmental Health Research Group, Faculty of Medical and Human Sciences, The University of Manchester, Oxford Road, Manchester, UK
  2. 2Biostatistics Group, Health Methodology Research Group, Division of Epidemiology and Health Sciences, Faculty of Medical and Human Sciences, The University of Manchester, Oxford Road, Manchester, UK
  3. 3Department of Civil Engineering, John Anderson Building, University of Strathclyde, Glasgow, UK
  4. 4Public Health Sciences Section, University of Edinburgh, Edinburgh, UK
  5. 5Emmes Corporation, Rockville, Maryland, USA
  6. 6Information and Services Division, NHS National Services Scotland, Gyle Square, Edinburgh, UK
  7. 7Occupational and Environmental Health Research Group, Faculty of Medical and Human Sciences, The University of Manchester, Oxford Road, Manchester, UK
  1. Correspondence to Melanie Carder, Occupational and Environmental Health Research Group, Faculty of Medical and Human Sciences, The University of Manchester, Oxford Road, Manchester M13 9PL, UK; melanie.carder{at}manchester.ac.uk

Abstract

Objectives To investigate whether deprivation index modifies the acute effect of black smoke on cardiorespiratory mortality.

Methods Generalised linear Poisson regression models were used to investigate whether deprivation index (as measured by the Carstairs deprivation index) modified the acute effect of black smoke on mortality in two largest Scottish cities (Glasgow and Edinburgh) between January 1981 and December 2001. Lag periods of up to 1 month were assumed for the effects of black smoke.

Results Deprivation index significantly modified the effect of black smoke on mortality, with black smoke effects generally increasing as level of deprivation increased. The interaction coefficient from a parametric model assuming a linear interaction between black smoke (μg/m−3) and deprivation in their effect on mortality—equivalent to a test of ‘linear trend’ across Carstairs categories—was significant for all mortality outcomes. In a model where black smoke effects were estimated independently for each deprivation category, the estimated increase in respiratory mortality over the ensuing 1-month period associated with a 10 μg/m3 increase in the mean black smoke concentration was 8.0% (95% CI 5.1 to 10.9) for subjects residing in the ‘most’ deprived category (Carstairs category 7) compared to 3.7% (95% CI −0.7 to 8.4) for subjects residing in the ‘least’ deprived category (Carstairs category 1).

Conclusions The results suggest a stronger effect of black smoke on mortality among people living in more deprived areas. The effect was greatest for respiratory mortality, although significant trends were also seen for other groups. If corroborated, these findings could have important public health implications.

  • Cardiovascular
  • respiratory
  • air pollution
  • environment
View Full Text

Statistics from Altmetric.com

Introduction

The link between social deprivation and ill health has been clearly shown in various studies, with populations living in deprived areas exhibiting levels of mortality substantially in excess of those in affluent areas.1 2 In the first decade (1981–1991) of our study period, these differentials increased in Scotland as inequalities in health widened (eg, the difference in rates for ischaemic heart disease between affluent and deprived groups increased over those 10 years).3 4 However, the relative contribution of environmental and other factors to these differentials in health and mortality has yet to be unravelled. One such factor is particulate air pollution. A link between exposure to particulate air pollution and increased mortality has been consistently shown.5 6 However, whether the impact of air pollution on mortality is greater for individuals experiencing greater socioeconomic deprivation is less clear. A recent study investigating the impact of the London Congestion Charge concluded that the health benefits associated with a reduction in air pollutant levels would be greater in deprived compared to more affluent areas.7 In contrast, a study in Vancouver, Canada8 found some indication of an increased risk of mortality at higher levels of socioeconomic deprivation (as defined by mean family income levels of the enumeration district of usual residence) for some pollutants (eg, nitrogen dioxide) but not particulate pollutants.

Our previous work (a time series analysis investigating the effect of black smoke on mortality), which did not address social deprivation, observed black smoke pollution to have a significant effect on mortality.9 A 10 μg/m3 increase in daily black smoke was associated with a 1.7% increase in all cause mortality, and increases of 0.4%, 5.4% and 2.1% in cardiovascular, respiratory and ‘other cause’ mortality, respectively, over the ensuing 30-day period. To investigate whether these effects were greater for people experiencing greater socioeconomic deprivation, classification of the deaths by deprivation categories (based on area of residence) has been undertaken. The measure of deprivation chosen for this study was the deprivation score of Carstairs and Morris.10 This is an ecological measure of socioeconomic deprivation/affluence, calculated from four census-derived variables (unemployment, overcrowding, non-car ownership and low social class).

The aim of this study therefore was to investigate (using a time series analysis) whether deprivation index modified the acute effect of black smoke on mortality in two Scottish cities.

Methods

Health data

The Information and Services Division (ISD) of the National Health Service (NHS) National Services Scotland supplied mortality data for the period January 1981 to December 2001 in the two Scottish cities of Edinburgh and Glasgow (data from Aberdeen was also initially considered, but later excluded due to postcode mapping issues resulting from extensive postcode changes during 1995).11 In Edinburgh and Glasgow, only those postcode sectors within 10 km of a black smoke monitoring site were considered for inclusion in the study. The outcomes considered in this study were deaths from all non-accidental causes, deaths from cardiovascular causes (International Statistical Classification of Diseases and Related Health Problems, ninth edition (ICD-9) codes 410–414, 426–429, 434–440) and deaths from respiratory causes (ICD-9 codes 480–487 and 490–496), essentially including cardiac and cerebral ischaemia (but excluding cerebral haemorrhage), chronic obstructive pulmonary disease, asthma and pneumonia.12 Mortality from ‘other’ causes (non-cardiorespiratory causes excluding accidental causes) was also considered.

Socioeconomic data

The mortality data supplied by ISD included geographical information in the form of the postcode sector of usual residence for each of the deaths. The Carstairs deprivation score (1–7, with 1 being the least deprived) for each postcode sector was obtained and used to ascribe a deprivation score to each deceased person. Carstairs scores are an unweighted combination of four census (1991 census in this instance) variables, namely, unemployment (unemployed men aged 16 and over as percentage of all men), overcrowding (persons in households with one or more persons per room as proportion of all residents in households), non-car ownership (residents in households with no car as proportion of all residents in households) and low social class (residents in households with an economically active head of household in social class IV or V as a proportion of all residents in households).10

Meteorological and air pollutant data

The UK Meteorological Office supplied hourly meteorological data for each of the three cities. These data were used to calculate the daytime mean temperature for each day, taken as the average of the 7.00–23.00 hourly values. Our previous work13 had observed a non-linear relationship between temperature and mortality (mortality increased as temperature decreased but this increase was steeper at temperatures below 11°C). We also observed the effect of temperature on mortality to persist for up to 1 month. Thus, to model temperature, we created two linear temperature variables, representing temperature above and below 11°C. These variables were defined in terms of usual daytime mean temperature, t, as follows:

High=t-11 if t≥11°C Low=t-11 if t<11°C0 otherwise0 otherwise

The daily ‘high’ and ‘low’ temperature values were then averaged over different time periods (0, 1–6, 7–12, 13–18, 19–24 and 25–30 days prior to the death) to provide the lagged ‘high’ and ‘low’ temperature variables.

For Edinburgh, daily mean black smoke measurements were obtained from a centrally located site within the city. This site was deemed most suitable for the study for its location and for the fact that it was operational during the majority (1981–1997) of the study period. To impute for missing data at this site and to extend the dataset to December 2001, a multiple imputation (ordinary least squares) method available within the software package SOLAS V.2.0 was used.14 This method uses the predictive information in a user specified set of covariates to impute the missing values. For Edinburgh, the covariates were the aforementioned black smoke site plus data from two additional black smoke sites, one of which was operational during the first part of the study period (1981–1991) with the second site operational for the latter part of the study period (1997–2001). In order to represent the uncertainty about which values to impute, several values (5 in this study) are imputed for each missing value, thus resulting in five different black smoke series to use as input variables for the later analyses.

For the Glasgow conurbation, daily black smoke measurements from seven separate sites were available for the period of interest (1981–2001). Thus, given the size of the conurbation coupled with the availability of data, the Glasgow population was divided into seven separate groups based on associating postcode sectors with the nearest monitoring site. As for Edinburgh, a multiple imputation method was used to replace missing black smoke data at the Glasgow sites.14 In this instance, the covariates used to predict the missing values were all aforementioned seven Glasgow sites, resulting in five different possible black smoke series for each of the seven sites. Prior to imputation, approximately 14% of the total black smoke data (Edinburgh and Glasgow combined) was missing. After imputation, this was reduced to less than 1%.

Statistical analyses

The study used a time series analysis with the daily number of deaths—either total non-accidental or the subtotal in the cause subgroups described previously—in each deprivation category as the outcome variable. All analyses were carried out in Splus (V.2000) using generalised linear (Poisson) models (GLMs) with natural cubic splines to capture seasonal and other long-term effects. The convergence tolerances of the GLM function were set to a low value of 10−9 with a limit of 1000 iterations.15 Data for the eight areas were analysed separately and the resulting estimates combined using inverse variance weighting.16

In previous work9 we found evidence for black smoke effects on mortality at lags up to 1 month. In view of this, a single variable representing the mean black smoke concentration over the same and preceding 30 days (ie, averaged over 0–30 days) was included in the regression model. The regression models for each area also included weekday (indicator variables); season (a smoothed function of date with 7 degrees of freedom per year); the 12 temperature variables described above; Carstairs category (indicator variables for each—except 1—of the 7 categories) and terms representing the interaction between black smoke and deprivation. Two approaches were taken: a parametric approach with the interaction term being a product of the black smoke variable and Carstairs score treated as a continuous variable (taking a value 1–7) (model 1), and secondly a non-parametric approach where interactions between the black smoke variable and each Carstairs category (except one) were fitted giving six interaction terms in total (model 2). Model 1 had the form:ln(deaths)=termsrepresentingcalendartime,dayofweek,temperature+i=1,i57αiIi+βBS+ηCBSwhere BS is black smoke averaged over 0–30 days, Ii is an indicator variable for Carstairs category i, i=1,4,6,7, C is Carstairs score and {α123467 ,β, η} are parameters estimated from the data.

Model 2 had the form:ln(deaths)=termsrepresentingcalendartime,dayofweek,temperature+=i=1,i57αiIi+βBS+i=1,i57ηiIiBSThe overdispersion parameter (estimated from the GLM models) was close to 1 suggesting little additional variation beyond Poisson variation, and as such a simple Poisson model was assumed.

Results

According to the 1991 Census, there were approximately 1.4 million people in the area covered by the study, of which 38% (table 1) was assigned to the two most deprived deprivation categories (categories 6 and 7). Table 1 shows means of daily deaths, temperature and, black smoke and deprivation categories (1991 Census based) for each of the eight areas. Daily mean black smoke concentrations ranged from a low of 10.8 μg/m3 in Dalmarnock to 18.8 μg/m3 in the most centrally located Glasgow black smoke site, City Chambers.

Table 1

Daily mean mortality, black smoke concentration and daytime mean temperature 1981–2001. Population (1991 census) by area (black smoke site) and Carstairs category

Table 2 shows results obtained by combining information from the parametric interaction models for each area; these models assumed a linear interaction between black smoke and deprivation as measured by the Carstairs deprivation categories. The interaction coefficients for the four outcomes are shown in the second row of the table; a test of whether the interaction was not equal to zero—equivalent to a test of ‘linear trend’ across Carstairs categories—was significant for all four outcomes. To illustrate the implication of these coefficients, the implied effects of black smoke at each of the Carstairs deprivation categories were calculated from the model. For example, a 10 μg/m3 increase in daily black smoke was associated with an increase in all cause mortality over the next 30 days of 2.9% (2.0, 3.9) in the most deprived areas (Carstairs category 7) and a change of −1.1 (−2.2, 0.1) in the least deprived areas (Carstairs category 1). To illustrate the general consistency (in terms of direction and size) of the observed interaction between black smoke and deprivation across the study area, the interaction coefficients for each of the eight geographical areas are provided in table 3.

Table 2

Estimated percentage increase (and 95% CIs) in mortality over the ensuing 1-month period associated with a 10 μg/m3 increase in the mean black smoke concentration on any given day, at each Carstairs category (assuming linear interaction between black smoke and deprivation)

Table 3

Estimates and 95% confidence limits for interaction coefficients (assuming linear interaction between black smoke and deprivation) for each area and for each outcome

Figure 1 shows the estimated percentage increase (and 95% CIs) in mortality over the ensuing 1-month period associated with a 10 μg/m3 increase in the mean black smoke concentration on any given day, at each Carstairs category. The results shown here are for all areas combined and are based on the non-parametric models, which estimate black smoke effects for each deprivation category independently. The estimated increase in respiratory mortality over the ensuing 1-month period associated with a 10 μg/m3 increase in the mean black smoke concentration was 8.0% (95% CI 5.1 to 10.9) for residents of the ‘most’ deprived areas (Carstairs category 7) compared to 3.7% (95% CI −0.7 to 8.4) for residents of the ‘least’ deprived areas (Carstairs category 1). These results are very similar to those implied by the parametric models. For cardiovascular and other cause mortality, respectively, the same increase in black smoke was associated with a 1.0% (95% CI −0.6 to 2.6) and a 3.4% (95% CI 1.8 to 5.1) increase at deprivation category 7 and a −3.8% (95% CI −6.2 to −1.8) and −0.6% (95% CI −3.1 to 2.1) change at deprivation category 1.

Figure 1

Estimated percentage increase (and 95% CIs) in mortality over the ensuing 1-month period associated with a 10 μg/m3 increase in the mean black smoke concentration on any given day, at each Carstairs category.

Discussion

This study presents evidence that the acute effect of black smoke on mortality is modified by deprivation index, with people living in the more deprived areas typically having a larger risk of pollutant-related mortality than those living in less deprived areas. The interaction effect between air pollution and deprivation index was greatest for respiratory mortality although significant trends were also seen for other groups.

Previous studies, all conducted outside the UK examining the possible variation of pollutant-related mortality and morbidity with deprivation index, have been somewhat inconclusive. Forastiere et al (2007) observed that the pollutant effect on number of hospital admissions prior to death was greater for people of lower social class compared to more affluent residents.17 A study in Sao Paulo18 observed the PM10 (particles of 10 μm or less) effect on mortality to be slightly greater (but not significantly so) in wealthier areas compared to the poorer areas. A later study in Sao Paulo19 which investigated the effect modification of PM10 on respiratory mortality in older subjects by various indicators of socioeconomic deprivation, found the effect to be strongly negatively correlated with percentage of people with college education and family income. A study in Hamilton, Canada,20 which investigated the associations between air pollution, deprivation and mortality within five geographical zones, observed that two socioeconomic indicators (low educational attainment and high manufacturing) significantly and positively modify the effect of SO2 and PM (as measured by the coefficient of haze, COH) on mortality. A study in Cook County, Illinois21 which used a case crossover design to investigate possible modifiers of the PM10 effect on hospital admissions for older subjects for heart and lung disease, also found that a number of indicators of socioeconomic deprivation (household income, percentage of adults with higher education and percentage of adults who spoke a language other than English in the home) did not modify the pollutant effect.

The studies mentioned so far (as well as the present study) have all assessed socioeconomic deprivation at group levels (eg, postcode sector, zones, or enumeration districts). The issue of (pollutant) effect modification by socioeconomic deprivation has also been addressed in a number of studies, in which information on individual level socioeconomic factors has been available. The Health Effects Institute reanalysis of the American Cancer Society study and the Harvard Six Cities study indicated that in both (cohort) studies, the effect of fine particles did appear to vary with educational level; the association between an increase in fine particles and mortality tended to be higher for individuals without a high school education compared to those educated to high school level or beyond.22 Individual level socioeconomic data was also used in a recent time series study in Hong Kong, which requested the person reporting a death to fill in a questionnaire on behalf of the deceased.23 The results of the study suggested that the effect of PM10 on mortality was greater in groups living in rented compared to private housing and in blue collar workers compared to white collar workers. However, they found no evidence of effect modification by their third socioeconomic indicator, education attainment.

Laurent et al,24 who reviewed the literature (published prior to May 2006), similarly concluded that it was not possible to definitively conclude that the effect of air pollution on mortality was modified by socioeconomic deprivation. However, they did conclude that while those studies that measured socioeconomic deprivation at city or country level tended to find no association, those studies which used finer geographic resolutions (like our own) found mixed results and those studies using individual level socioeconomic characteristics (for example the cohort studies) tended to find a positive association.

Populations residing in deprived areas may be at an increased risk of pollutant-related mortality and morbidity for a number of reasons. Firstly, they may reside and/or work in subareas where they would be exposed to higher levels of air pollution and hence exposure estimates of black smoke and other pollutants from measurement sites may be underestimated for people living in highest deprived areas. For example, manual workers in polluted heavy industries such as steel making or shipbuilding might be more likely to live in socially deprived districts. Secondly, people living in more deprived areas may potentially experience greater health problems due to smoking habits, material deprivation, poor diet and stress and therefore may be more susceptible to the effects of air pollution. Moreover, other microenvironmental factors such as poor heating in the home may contribute. Finally, poorer people may effectively have less access to healthcare even though the UK NHS is designed to afford equitable healthcare to all. The combination of increased exposure and increased susceptibility could mean these groups are likely to be a higher risk group for air pollution effects.25

An unexpected finding of this study was the apparent suggestion that particulate pollution had a ‘protective’ effect on cardiovascular mortality for those subjects residing in the least deprived areas (ie, those areas with a Carstairs deprivation score of 1). Furthermore, this effect was seen in results from the parametric and non-parametric models (if it was just the former one could argue that it was a result of forcing the interaction to be linear). The reason behind this finding is unclear but it seems highly unlikely that this is a ‘true’ relationship. Of further interest, in models looking only at the main effect of black smoke on mortality (ie, not looking at any interactions) a significant association between black smoke and cardiovascular mortality was not observed. This study also observed the effect of black smoke on ‘other cause’ mortality to be significantly modified by deprivation index. There are several plausible explanations for this finding. This group contained all subjects that did not have a specified cardiac or respiratory code recorded as the primary (underlying) cause of death. However, the ‘mode’ or secondary cause of death in many diseases, regardless of the primary underlying pathology, is often cardiac, vascular or respiratory and it is possible that air pollution contributes to this final ‘mode’ of death. A second explanation is that a true association exists between air pollution and causes of death other than cardiorespiratory causes. In this study the group ‘other’ causes includes deaths from lung cancer; an outcome which has been previously linked to air pollution.26 Goodman et al27 also observed an association between black smoke and ‘other’ causes of death (defined as all cause deaths minus cardiorespiratory deaths).

Although the results of our study have suggested that the effect of black smoke on mortality is modified by deprivation index it is not possible to discern which specific socioeconomic variables are responsible for this effect. The ecological measure of socioeconomic deprivation/affluence used in this study was the Carstairs deprivation index which is calculated from four (1991) census-derived variables (unemployment, overcrowding, non-car ownership and low social class), all of which have been shown to be correlated with each other28 and may be correlated to further, unmeasured variables. Carstairs scores based on 1991 census data were deemed most suitable for the present study, as 1991 was the midpoint of the study period (1981–2001). Although the deprivation experienced in any given postcode sector might have changed during the 20-year study period, significant changes in deprivation categories between censuses are unusual.29 A final point to note is that the present study was an ecological study and therefore does not permit valid inferences to the individual level; the plausibility of any generalisation to individuals is dependent in part on the social homogeneity of the areas studied; and some postcode sectors may be highly heterogeneous.19

The results from the non-parametric models did not suggest a strictly linear relationship between deprivation and the effect of air pollution on health. This may in part be due to confounding by area; much of the data for affluent and deprived people is derived from different areas. One possible approach to better address potential confounding by area would have been to use a combined model for all areas (with areas included as dummy variables). However, the computational power required to analyse a combined model was beyond the capability of the available resources. In addition, a combined model would have assumed a common time smooth across all areas and while we might assume this to be true for seasonal effects it is unlikely to be true for infectious disease epidemics or other ‘unknown’ factors.

The only pollutant considered in this study was black smoke (data on black smoke for our study area was extensive in spatial and temporal terms whereas data on other pollutants was much more limited) and it is possible that some of the observed effects could be due to confounding by other, unmeasured pollutants. However, studies have consistently pointed to particulate matter as being the most important pollutant in terms of health effects.5 In line with recent findings,9 27 this study also considered pollutant concentrations lagged up to 1 month prior to the health event. A further point to note is that this study (in common with many other time series studies investigating the effect of air pollution on health) estimated exposure to black smoke for the study population by applying black smoke measurements from a single fixed monitoring site in each case (ie, the nearest monitoring site to the postcode of the subjects home address). However, an individual's personal exposure to air pollution will depend on a number of factors including the time spent in different microenvironments, for example, indoors or at work, and the activities undertaken while there.30 As such, the approach undertaken in the present study is likely to have led to some degree of exposure misclassification.

Similarly, it is possible that the observed effects were due to inadequate adjustment for confounders. However, this study used a standard time series methodology with natural cubic splines to capture seasonal and other long-term effects. In previous analyses, a sensitivity analysis on this population suggested that increasing the number of degrees freedom per year to control for season and trend only decreased the air pollution estimates slightly.9 In addition, there has been some debate as to the correct level of adjustment for temperature in air pollutant time series studies. A number of studies have observed the effect of cold temperature on mortality to persist at lag periods in excess of 2 weeks13 31 suggesting that the inclusion in the regression models of temperature lagged by only a few days (which has tended to be the norm for air pollutant studies) may be insufficient and lead to an overestimation of the air pollutant effect. In response, this study adopted a stringent approach in the control for temperature and included temperature measurements lagged up to 1 month in the regression models. As such, it is unlikely that the observed pollutant effects were due to inadequate adjustment for temperature. It is also possible that some or all of the observed (respiratory) effects were due to influenza epidemics or pollen (neither of which this study formally adjusted for). However, our seasonal model should have been adequate to reduce the potential confounding effect of these two variables.32

In conclusion, we have found evidence that deprivation index in two cities in the UK modifies the effect of black smoke on mortality. Social factors should therefore be considered alongside other factors which may interact with exposure to air pollution in increasing the risk of adverse health outcomes.33 A public health approach which appears to be driven in a reductionistic manner by air quality standards34 in isolation should be amplified by parallel consideration of other factors such as temperature9 and sociodemographic factors as shown in this paper. If corroborated, these findings could have important public health implications.

What this paper adds

  • The link between social deprivation and ill health has been clearly shown in various studies, with populations living in deprived areas exhibiting levels of mortality substantially in excess of those in affluent areas.

  • However, the relative contribution of environmental (eg, air pollutants) and other factors to these differentials in health and mortality has yet to be unravelled.

  • The effect of black smoke on mortality appears to be stronger among people living in more deprived areas.

  • The effect was greatest for respiratory mortality although significant trends were also seen for other groups.

  • A public health approach to consider the impact of particulate pollution on health should take into consideration potential interactions between pollution and social deprivation.

Acknowledgments

This work was supported by a research grant, number G9900747, from the Medical Research Council to RMA and colleagues. The views expressed in this publication are those of the authors and not necessarily those of the Medical Research Council.

References

View Abstract

Footnotes

  • Funding Medical Research Council, London, UK.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

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.