Discussion

Areas of the UK that had high domestic consumption of coal in 1951/1952 now have raised mortality from a wide variety of causes, including cardiovascular and respiratory diseases and certain cancers. The correlations are strong, statistically significant and independent of all available variables that might be considered major confounders, whether assessed in 1951 or currently, including social class, level of education, overcrowding and unemployment.

Very few comparable data exist in the literature, particularly with such a long follow-up period. Our findings accord with an increasing body of evidence that early life exposure to air pollution has detrimental long-term health effects.6 15 23 In particular, two recent studies demonstrated important findings. In the first, children exposed to the great London Smog of 1952 were found to have greater risk of asthma in adulthood, compared with unexposed children.24 In the second, which was based on the ONS longitudinal study, air pollution concentrations at the site of residence, assessed every 10 years from 1971, were associated with increased total, respiratory and cardiovascular mortality.25 Our conclusions also accord with historical studies of coal use and mortality in Britain, which estimated that each SD increase in coal use raised mortality by 5%–15% in infants and 5% in adults.26

Our study necessarily depended on published crude estimates of coal consumption as an index of domestic air pollution. We contend that it is very reasonable to assume that the quantity of coal burnt relates to the amount of pollution emitted, although local factors such as climate and wind strength would also have a major influence. The consumption measures in our study were shown to correlate with subsequent measurements of pollution in the UK in the early 1960s,13 when coal was still a major fuel and source of pollution. This approach is also supported by a number of studies carried out over the last 60 years, showing correlations with diverse health outcomes including reduced early growth,3 chronic bronchitis and lung cancer.13–15 While the consumption data come from a short time-period between 1951 and 1952, they are likely to reflect smoke emission over many years. The UK coal production was fairly constant during the late 1940s, declined a little during the 1950s and only fell dramatically during the 1960s and subsequent decades.12 In the postwar years, economic necessity drove the use of low-grade, bituminous domestic coal while better-quality ‘hard’ coals were exported. A typical, inefficient domestic grate burning low-grade coal would have produced lots of smoke, rich in a wide variety of potentially toxic compounds, including heavy metals, sulfur and complex mixtures of aliphatic and aromatic hydrocarbons.27 Confirmation of this can be found in the lung tissue from autopsies of people exposed to the 1952 London smog, which was shown to contain both ultrafine carbonaceous and metal PM.28 Children would have been exposed both in the home and in their local environment, although it is not clear from our data which of these would have been the most important source. However, our study suggests that it was domestic fuel consumption that had the greatest long-term adverse health effects, with industrial pollution or pollution from power stations or other sources causing relatively smaller effects (table 2). The probable explanation for this is that most industrial and power station pollution was vented through tall chimneys, decreasing ground-level pollution in approximate proportion to the inverse square of chimney height.29

Current levels of pollution in the 342 local areas were based on ambient levels of fine PM_2.5 because large cohort studies have shown that this measure correlates best with long-term health effects of pollution.30–32 The effects of PM_2.5 on current mortality (table 3) in the present study are consistent with previously published estimates,20 but importantly the mortality correlations with coal consumption persisted after adjustment for this measure of current pollution.

Our data do not allow determination of the age at which children were most vulnerable to the effects of pollution, although a cohort-based analysis (figure 3) suggests that exposures around the time of birth or infancy are likely to have been most important, especially for all-cause and respiratory mortality. We can also only speculate on the mechanism by which air pollution arising from coal consumption, predominantly in a domestic setting, might increase the risk of mortality. There may have been direct mutagenic effects, especially in relation to the increased cancer risk. Another possibility is that the pollutants triggered mechanisms that selected alternative developmental pathways in the young, perhaps through epigenetic modification of gene expression.33

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

The strength of the geographical correlations between domestic consumption and current mortality according to successive birth cohorts. For all-cause and respiratory mortality, the strongest associations were observed among those born in 1952/1953 suggesting that exposure around the time of birth had the greatest effect on mortality. For cardiovascular and cancer mortality, the peak associations were more blunted, suggesting that exposure over a wider age range was associated with subsequent mortality.

Elevated relative risks were observed for respiratory diseases, especially COPD, asthma and pneumonia (table 4), which is consistent with past evidence that black smoke exposure is a risk factor for these conditions.14 15 23 The strongest association we found, however, was with tuberculosis, which is less supported by existing literature. A recent systematic review and meta-analysis concluded that the evidence for such an association was inconclusive, but it was reliant on case-control and cross-sectional studies.34 The findings in the present study raise the possibility that early exposure to air pollution could be a critical factor in determining susceptibility to tuberculosis, although other factors such as population density cannot be completely excluded. Cardiovascular disease was also found to be strongly associated with domestic fuel consumption, which accords with earlier UK findings,15 23 the Harvard Six Cities study31 and a multicentre study of European cohorts.35 The strength of the association was similar with ischaemic heart disease and stroke. The association with rheumatic heart disease was strong and confirms the findings of a previous study based on a subset of large Country Boroughs in England and Wales, and further supports the suggestion that an increase in susceptibility to infectious diseases is linked with exposure to air pollution in infancy or early childhood.36 An additional finding was a strong association with hypertension, which is supported by a meta-analysis that showed both short-term and long-term exposure to some air pollutants increases the risk of hypertension.37 The data on cancers (table 5) show wide variation in the level of associated risk. They show that the strongest associations are observed with epithelial cancers of the respiratory system (lip, oral cavity and pharynx; larynx; trachea and lung) and the upper gastrointestinal system (stomach and liver). The major reproductive, urinary and haematological cancers were not positively associated with domestic pollution with the exception of cervical cancer, for which there is evidence that cigarette smoking is an important cofactor.38 Malignant melanoma had a lower relative risk in the areas with high consumption possibly because of reduced sunlight exposure. Some cancers showed small reductions in relative risk, for example, brain cancer, and although the reasons are not clear, this may be due to residual confounding, which is evident because of the very large numbers of deaths and high sensitivity of this analysis.

There are a number of potential limitations to our study. Because neighbouring geographical areas tend to have similar levels of coal consumption, spatial autocorrelation may lead to CIs that are narrower than they should be. However, an analysis evaluating the influence of spatial autocorrelation (described in detail in online supplementary appendix 2) suggest that this is not a serious problem in the coal consumption data. Calculations incorporating the autocorrelation term into regression analyses increase the very narrow CIs, as shown in table 2, by 7.1%.

An inevitable issue with evaluating the association of early exposures with causes of late-life mortality is the long latency between, in this case, the measurements of coal consumption and the outcomes. It could be argued that internal migration might have weakened the associations that we observed as selective migration has been proposed as an explanation for geographical variations in mortality. This is supported by some studies39 but not others.40 41 Migration is complex as, while migrants are generally healthier, those moving short distances tend to have higher mortality than those moving long distances.42 43 Although we do not have migration data, most migration during the time period of this study occurred over short distances, with only a small minority migrating significant distances.44 A more recent study of migration based on the 2001 census, while documenting the complex age and social class influences on migration, suggested that the declining industrial areas of South Wales, Yorkshire, Greater Manchester and Lancashire and the North-East of England, which were the areas that had high coal consumption, had in-migration and out-migration levels in all age groups that were far below the National average.45 In addition, our analysis allows for the factors that tend to drive migration such as education and socioeconomic status.46 Finally, it is difficult to see how selective migration could explain the specific links between coal consumption and some conditions and not others (table 5).

Any undetected bias in our observational study would be most likely due to confounding from an unknown source. However, a sensitivity analyses suggested that such a variable would have to correlate very strongly (r>0.5) with both domestic air pollution and respiratory disease, for example, to confound the relationships we have observed. There are no obvious candidates for such a confounder that were not addressed by our analysis, apart from air pollutants from sources other than the combustion of coal and solid fuels, measurements of which are unavailable for the early 1950s. As many of these would have been geographically localised and specific to certain industries, it is difficult to see how they could be confounders, especially as our analyses incorporate adjustment for socioeconomic factors (eg, occupation/urbanisation) that would reflect differential exposure to these substances. In addition, we have examined the effects of the major confounding variables that were available in the 1951 and 2001 census and that are likely to be related to both residence in an area of high pollution and to the various outcomes that we have studied. The 1951 census contained several additional measures of household amenities, which were also examined in regression analyses. We found that these did not explain any further variance in our analyses beyond that due to the confounding variables that we used in the final analyses. Adjusting the risk estimates associated with domestic consumption for social class, density, unemployment and education reduced their magnitude, but they still remained strongly significant (table 3 and figure 2). However, it could be argued that some of these variables are so closely related to the measures of pollution that the regression-based adjustment has resulted in conservative risk estimates. Unsurprisingly, population density was highly correlated with domestic fuel use because fuel was rationed on a per household basis. In our analyses based on a published technique,47 we were able to show that pollution had a more marked effect than population density but the close relationship between these variables prevented us from using this in the adjusted regression models and, as a result, we cannot confidently exclude the influence of early population density. However, we would also argue that, in this context, density (population per unit area) may be a less important confounder than overcrowding (proportion of households with >1 person per room). There is considerable observational evidence linking overcrowding with physical health, especially in adults48 and in a review of the environmental factors most associated with health outcomes, Evans and Kantrowitz concluded that areal indices of density were less important than overcrowding in determining the health outcomes associated with high population density.49 Finally, our data showing linear relationships between fuel usage and mortality (figure 2) are unlikely to be driven by the influence of density on pathogen transmission where saturation effects would be expected.50 Although we did not have data on the prevalence of tobacco smoking, we do not think that this is likely to have seriously confounded our results. In the early 1950s, smoking was a social norm (in 1951 78% of men used tobacco), smoking was associated with low socioeconomic status,51 which was included in our analyses and standardisation of the data for lung cancer rates as a proxy for smoking did not explain the correlations we observed.