Evaluating the long-term consequences of air pollution in early life: geographical correlations between coal consumption in 1951/1952 and current mortality in England and Wales

Objective To evaluate associations between early life air pollution and subsequent mortality. Design Geographical study. Setting Local government districts within England and Wales. Exposure Routinely collected geographical data on the use of coal and related solid fuels in 1951–1952 were used as an index of air pollution. Main outcome measures We evaluated the relationship between these data and both all-cause and disease-specific mortality among men and women aged 35–74 years in local government districts between 1993 and 2012. Results Domestic (household) coal consumption had the most powerful associations with mortality. There were strong correlations between domestic coal use and all-cause mortality (relative risk per SD increase in fuel use 1.124, 95% CI 1.123 to 1.126), and respiratory (1.238, 95% CI 1.234 to 1.242), cardiovascular (1.138, 95% CI 1.136 to 1.140) and cancer mortality (1.073, 95% CI 1.071 to 1.075). These effects persisted after adjustment for socioeconomic indicators in 1951, current socioeconomic indicators and current pollution levels. Conclusion Coal was the major cause of pollution in the UK until the Clean Air Act of 1956 led to a rapid decline in consumption. These data suggest that coal-based pollution, experienced over 60 years ago in early life, affects human health now by increasing mortality from a wide variety of diseases.

•Use of national mortality data with virtually complete ascertainment and a large number of deaths (> 3.5 million).
•Coal consumption data provides an integral of air pollution over a 12 month period.
•The analysis also allows for current pollution levels •The study depends on the use of geographical correlations together with census-derived measures of socio-economic status; individual-level data are not available.

INTRODUCTION
Air pollution has detrimental effects on health at all ages but little is known about the long-term impact of early life exposure to pollutants. 1 This is despite accumulating evidence that the young are especially vulnerable. Developing tissues and organs are particularly susceptible due to their high cell proliferation rates in prenatal life. Postnatally, infants and young children also have increased vulnerability due to immature detoxification systems, high infection rates and patterns of behaviour that increase their pollutant exposure. 1 Pregnant mothers exposed to air pollution are more likely to give birth prematurely and have offspring with intrauterine growth retardation and poor growth in infancy. 2 3 Such outcomes are linked with long-term adverse health effects, including increased risks of respiratory and cardiovascular disease. 4 In common with many other countries, the UK had high levels of air pollution in the past. 5 Most of this was due to coal use in industry and as a domestic fuel, which gave rise to the "great stinking fogs" that previously characterised London and other major cities. 5 The London smog of 1952 killed an estimated 4,000 people 6 and led to the first UK Clean Air Act being passed in 1956. Following the 1952 smog, a national survey of smoke and SO 2 concentrations was set up during the 1960s to monitor the progress of pollution control following the Act. From around 1960, there was a progressive reduction in black smoke and SO 2 emissions, with nitrogen oxides, ozone and microparticulates from transport becoming the dominant forms of air pollution. 7 Little is known about the long-term health effects of the combustion of the vast quantities of coal burnt which still exceeded 200 million tons per year in England and Wales at the middle of the 20 th century. 8 Studies published in the 1950s and 1960s suggested that there were strong cross-sectional correlations between the amounts of coal used in domestic fires and childhood respiratory disease 9 or other major causes of mortality in adults 10   Historical Geographical Information System (GIS). 13 14 As the LFO areas either exactly corresponded to LGDs or were defined as aggregates of them, constructing digital boundary data for LFO areas was straightforward. However, modern LGDs are not necessarily simple aggregates of the LFO areas, so the first stage of the redistricting procedure was to re-allocate the fuel consumption data to the 12,504 Civil Parishes of 1951. Digital boundaries already existed for the parishes, and the re-allocation was based on what proportion each parish's total population, as listed by the 1951 census, was of the overall population of the containing LFO area. The fuel data were then re-allocated to the modern areas by digitally overlaying the two sets of boundaries: whatever proportion of a 1951 parish's area fell within a modern district, that proportion of the parish's fuel consumption was assigned to the district.
Historically, air pollution was worse in predominantly industrial towns with a high population density; these towns also tended to have large populations of factory workers who left school early and worked in low status occupations. 10 These are important potential confounders in a study of air pollution and disease, as poverty and overcrowding are known to be associated with many of the major causes of mortality. We used a series of variables extracted from the published reports of the 1951 census: male unemployment; indices of social class, which were based on a weighted average of the proportions of each of five social classes within each area; overcrowding, assessed as the proportion of households with more than one person per room; density, the number of people per acre; and educational achievement, the proportion of employed men in each area who left school before the age of 15 years. We also allowed for the impact of current levels of air pollution which were obtained from published data reporting the modelled annual average concentrations of fine particulate matter (PM 2.5 ) in each of the 342 LGDs. 15

Statistical methods
We studied deaths in people aged between 35 and 74 years, as this was the approximate age range of the generation who were infants or young children in 1951-52, and excluded older age groups where the cause of death was less likely to have been ascertained accurately. 16 We calculated the number of deaths expected for each five-year age and sex group within the LGDs by multiplying the number of people in each group by the national age-and sex-specific death rates in 1993-2012. We expressed the number of observed deaths in each area as a percentage of the expected deaths for that group, i.e. standardised mortality ratios (SMRs). We used Poisson regression to model SMRs expressing the effect of fuels and other variables on mortality in terms of the hazard ratio (or relative risk) per SD change in the explanatory variable. This weights the analyses appropriately to allow for variation in population size across LGDs. A Fisher-Yates transform was used to standardise pollution and socioeconomic variables.
This yielded variables with zero means, unit standard deviations (SDs) and symmetrical distributions, allowing direct comparison of the strength of associations with these variables.

Patient involvement
No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for design or implementation of the study. No patients were asked to advise on interpretation or writing up of results. The analysis was based on geographical death rates and no individuals were identifiable. The mean current PM 2.5 concentrations in the 342 areas was 9.48 (SD 1.67) µg/m 3 and ranged from 5.5 in Gwynedd (Wales) to 14.9 µg/m 3 in the City of Westminster, London. Table 2 shows the relationship between the various sources of fuel burnt in 1951/2 (excluding smokeless fuels) and subsequent all-cause mortality or mortality due to respiratory, cardiovascular disease and cancer in the 342 areas. The influence of socioeconomic indicators (social class, education, crowding, density and unemployment) are also shown together with the levels of current pollution (PM 2.5 ). In univariate analyses, all variables were significantly associated with both all-cause and cause-specific mortality rates although the effects of domestic and industrial consumption tended to be most strongly related to the mortality outcomes. In multivariate analyses, however, domestic consumption was strongly and significantly associated with both all-cause and cause-specific mortality rates, an effect which was markedly stronger than any other variable included in the regression analysis. There was a very high correlation (r=0.97) between domestic consumption and population density, resulting in unacceptable collinearity in regression analyses that contained both variables. This occurred because both had the same denominator, the area of the LGD. To allow the influence of both of these factors to be modelled statistically, we created two further variables: the sum of and the difference between domestic use and density. These variables were uncorrelated (r=0.0). These were entered into the regression models in Table 2, which shows both the unadjusted coefficients and the coefficients adjusted for all other variables ( Table 2 lower panel). The positive regression coefficients for the sum variable show that the joint effects of domestic use and density on all four causes of mortality were powerful.
However, associations with the difference variable suggest that the influence of domestic use was stronger than that of density. Smokeless fuels represented a small proportion of the total fuels used and contributed little to the mortality risks. The effects of the potential confounders were tested singly and in combination. There was no evidence of non-linear relationships. Figure 2 shows the associations between quartiles of domestic fuel consumption and all-cause mortality or the major mortality categories. These associations were progressive and remained strong and statistically significant for each cause group after adjustment for social class, education, crowding, unemployment and current PM 2.5 . They were present in both genders and were similar during the ICD 9 and ICD 10 periods. Figure 3 shows 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 1952/3 suggesting that exposure at 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. Tables 3 & 4 show the relationship between domestic fuel consumption and specific causes of mortality.
As before, the data are shown before and after adjustment for the four socioeconomic indicators and current PM 2.5 levels. The respiratory conditions responsible for most deaths, chronic obstructive pulmonary disease and pneumonia, were strongly and consistently associated while associations with asthma mortality were somewhat weaker. However, the largest relative risks were with tuberculosis.
While all four major causes of cardiovascular mortality were significantly associated with domestic consumption, the strongest relationship was with rheumatic heart disease.
The data on associations of domestic fuel use with mortality from specific cancers were striking (Table   4). There were strong relationships with epithelial cancers; lip, oral cavity and pharynx (adjusted hazard ratio, 1.283); laryngeal cancers (1.323) and lung (1.219). The upper gastrointestinal cancers; oesophagus (1.131), stomach (1.166) and liver (1.174), were more strongly associated with domestic pollution than the lower gastrointestinal cancers. There were weak associations with bladder and kidney cancers. In contrast, the major reproductive and haematological cancers showed no increased risk, with the exception of cervical cancer (1.229). Melanoma and brain cancer were associated with a reduced relative risk.

DISCUSSION
Areas of the UK which had high domestic consumption of coal and related non-smokeless solid fuels in 1951/52 currently have raised mortality from a wide variety of different causes including cardiovascular or respiratory disease and certain cancers. The correlations were strong, statistically significant and independent of the available variables that might be considered likely major confounders such as social class, level of education, overcrowding, unemployment and density. They were also independent of current levels of air pollution as indicated by current concentrations of microparticulate pollution (PM 2.5) .
Very few comparable data exist in the literature, particularly with such a long follow-up period.
However, our findings accord with a body of evidence that early life exposure to air pollution has detrimental long-term health effects. 11 17 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 to unexposed children. 18 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. 19 Our conclusions also accord with historical studies of coal use and mortality in Britain which estimate that each SD increase in coal use raised mortality by 5-15% in infants and 5% in adults. 20 Our study depended on coal consumption as an index of domestic air pollution as it is not unreasonable to assume that the quantity of coal burnt relates to the amount of pollution emitted. The consumption measures in our study were shown to correlate strongly with subsequent measurements of pollution in the UK, 9 and this approach is also validated by a number of studies carried out over the last 60 years showing correlations with diverse health outcomes including reduced early growth, 2 3 chronic bronchitis and lung cancer. [9][10][11] In the post-war years, economic necessity drove the use of low-grade, bituminous domestic coal while better-quality "hard" coals were usually 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, sulphur and complex mixtures of aliphatic and aromatic hydrocarbons. 21 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 particulate matter. 22 Children would have been exposed both in the home and in their local environment although it is not clear from our data which would have been the most important source. However, our study does suggest that it was domestic exposure that had the greatest long-term adverse health effects, with industrial pollution or pollution from power stations or other sources causing relatively smaller effects ( Current levels of pollution in the 342 local areas were based on ambient levels of fine particulate matter (PM 2.5 ); the results of large cohort studies suggest that these levels are most closely associated with long-term health effects on pollution. [24][25][26] . Although the effects on current mortality (Table 2) are consistent with published estimates 15 in the inclusive models with early coal consumption, there was no separate effect observed on mortality.
Our data do not allow determination of the age at which children were most vulnerable to the effects of pollution although the cohort-based analysis ( Figure 3) suggests that exposures at around the time of birth or infancy are likely to be most important especially for all-cause and respiratory mortality.
Although the consumption data come from a short time-period between 1951 and 1952, they are likely to reflect smoke emission over many years. UK coal production was fairly constant during the late 1940s, declined a little during the 1950s and only fell dramatically after 1960. 8 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. 27 Elevated relative risks were observed for respiratory diseases, especially COPD, asthma and pneumonia, which is consistent with past evidence that black smoke exposure is a risk factor for these conditions. 10 11 17 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. 28 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. Cardiovascular disease was also found to be strongly associated with domestic fuel consumption which accords with earlier UK findings 11 17 the Harvard Six Cities study 25 and a multicentre study of European Cohorts. 29 The strength of the association was similar with ischaemic heart disease and stroke. The association with rheumatic heart disease was strong and confirms a previous study based on a subset of large Country Boroughs in England and Wales. 30 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. 31 The data on cancers ( and liver). The major reproductive 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 co-factor. 32 Malignant melanoma had a lower relative risk in the areas with high consumption possibly because of reduced sunlight exposure, while the reason for reduced relative risk observed with brain cancer is not clear, but may be due to residual confounding.
There are a number of potential limitations to our study. An inevitable consequence of evaluating early effects on major causes of late life mortality is the long latency between the measurements of coal consumption and the mortality data. It could be argued that internal migration might have weakened the associations that we observed as selective migration has been proposed as an explanation of geographical variations in mortality which is supported by some studies 33 but not others. 34 35 Migration is complex as while migrants are generally healthier, those moving short distances tend to have higher mortality than those moving long distances. 36  were the areas that had high coal consumption, had very much below average in-migration and outmigration for all age groups. 39 In addition, our analysis allows for the factors which tend to drive migration such as education and socioeconomic status. 40 Finally, it is difficult to see how selective migration could explain the specific links between coal consumption and some conditions and not others (Table 4). Any undetected bias in our observational study would be most likely to be 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. We have examined the effects of the major confounding variables that were available in the 1951 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 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 for social class, density, unemployment and education reduced their magnitude marginally but they but still remained strongly significant (Table 2 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, 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. Areal indices of density may be less important than overcrowding in determining the health outcomes associated with high population density. 41 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 as firstly, smoking is associated with low socioeconomic status 42 which was included in our analyses and secondly standardisation of the data for lung cancer rates, as a proxy for smoking, did not explain the correlations we observed.

Footnotes
Contributors: DIWP discovered the dataset on coal consumption in 1951/2 and together with CO conceived the study. HS and PA advised on the use of the geographical data and carried out the redistricting analysis. DIWP, CO and AJ were responsible for the statistical analysis. The manuscript was written by DIWP, CO, AJ, HS and SH. All authors participated in a critical revision of the text and approved the final manuscript. DIWP is responsible for the overall content as the corresponding author.

Funding:
The work was supported by the Medical Research Council.
Competing interests: None declared.
Ethics approval: As the study was based on routinely collected mortality data no ethics approval was necessary.

Objective
The short-term effects of air pollution are well-documented, but less is known about whether early life pollutant exposure has long-term adverse effects.

Design
Geographical study

Setting
Local authority areas in England and Wales

Exposures
Routinely collected data on the use of coal and related solid fuels in 1951-2 were used as an index of air pollution from domestic and industrial sources in different geographical areas of England and Wales.

Main outcome measures
We evaluated the relationship between these data and both all-cause and disease-specific mortality in the geographically equivalent 342 current local authority areas in 1993-2012.

Results
Of all the sources of solid fuel, domestic usage had the most powerful effect on mortality. There were strong relationships between domestic fuel use and all-cause mortality (relative risk per SD increase in

Conclusion
Coal was the major cause of pollution in the UK until the Clean Air Act of 1956 led to a rapid decline in consumption. Although based on geographical correlations, these data raise the possibility that coalbased pollution, experienced over 60 years ago in early life, affects human health now by increasing mortality from a wide variety of diseases. •Use of national mortality data with virtually complete ascertainment and a large number of deaths (> 3.5 million).
•Coal consumption data provide an integral of air pollution over a 12 month period.
•The analysis also allows for current pollution levels •The study depends on the use of geographical correlations together with census-derived measures of socio-economic status; individual-level data are not available. fires and the consequent exposure of mothers and young children to pollutants is widespread in many low-and middle-income countries. [2][3][4] There is accumulating evidence that the young are especially vulnerable. Developing tissues and organs are particularly susceptible due to their high cell proliferation rates in prenatal life. Postnatally, infants and young children also have increased vulnerability due to immature detoxification systems, high infection rates and patterns of behaviour that increase their pollutant exposure. 2 Pregnant mothers exposed to air pollution are more likely to give birth prematurely and have offspring with intrauterine growth retardation and poor growth in infancy. 5 6 Such outcomes are linked with long-term adverse health effects, including increased risks of respiratory and cardiovascular disease. 7 In common with many other countries, the UK had high levels of air pollution in the past. 8 Most of this was due to coal use in industry and as a domestic fuel, which gave rise to the "great stinking fogs" that previously characterised London and other major cities. 8    Historical Geographical Information System (GIS). 16 17 As the LFO areas either exactly corresponded to LGDs or were defined as aggregates of them, constructing digital boundary data for LFO areas was straightforward. However, modern LGDs are not necessarily simple aggregates of the LFO areas, so the first stage of the redistricting procedure was to re-allocate the fuel consumption data to the 12,504 Civil Parishes of 1951. Digital boundaries already existed for the parishes, and the re-allocation was based on what proportion each parish's total population, as listed by the 1951 census, was of the overall population of the containing LFO area. The fuel data were then re-allocated to the modern areas by digitally overlaying the two sets of boundaries: whatever proportion of a 1951 parish's area fell within a modern district, that proportion of the parish's fuel consumption was assigned to the district. Historically, air pollution was worse in predominantly industrial towns with a high population density; these towns also tended to have large populations of factory workers who left school early and worked in low status occupations. 13 These are important potential confounders in a study of air pollution and disease, as poverty and overcrowding are known to be associated with many of the major causes of mortality. To control for these we used a

Statistical methods
We studied deaths in people aged between 35 and 74 years, as this was the approximate age range of the generation who were in their first decade of life in 1951-52, and excluded older age groups where the cause of death was less likely to have been ascertained accurately. 20 We calculated the number of deaths expected for each five-year age and sex group within the LGDs by multiplying the number of people in each group by the national age-and sex-specific death rates in 1993-2012. We expressed the number of observed deaths in each area as a percentage of the expected deaths for that group, i.e. as standardised mortality ratios (SMRs). We used Poisson regression to model SMRs, expressing the effect of fuels and other variables on mortality in terms of the hazard ratio (or relative risk) per SD change in the explanatory variable. This weights the analyses appropriately to allow for variation in population size across LGDs. A Fisher-Yates transform was used to standardise pollution and socioeconomic variables.
This yielded variables with zero means, unit standard deviations (SDs) and symmetrical distributions, allowing direct comparison of the strength of associations with these variables.  The mean current PM 2.5 concentrations in the 342 areas was 9.48 (SD 1.67) µg/m 3 and ranged from 5.5 in Gwynedd (Wales) to 14.9 µg/m 3 in the City of Westminster, London. Table 2 shows the relationship between the various sources of fuel burnt in 1951/2 (excluding smokeless fuels) and subsequent all-cause mortality or mortality due to respiratory, cardiovascular disease and cancer in the 342 areas. The influence of socioeconomic indicators (social class, education, crowding, density and unemployment) are also shown. In univariate analyses, all variables were significantly associated with both all-cause and cause-specific mortality rates although the effects of domestic and industrial consumption tended to be most strongly related to the mortality outcomes. Multivariate analyses were carried out to evaluate the relative effects of fuel usage and the major confounding variables derived from the 1951 census. This analysis shows that domestic consumption was strongly and significantly associated with both all-cause and cause-specific mortality rates, an effect which was markedly stronger than the other sources of fuel. There was a very high correlation (r=0.97) between domestic consumption and population density, resulting in unacceptable collinearity in regression analyses that contained both variables. This occurred because both had the same denominator, the area of the LGD. To allow the influence of both of these factors to be modelled statistically, we created two further variables: the sum of and the difference between domestic use and density. These variables were uncorrelated (r=0.0). These were entered into the regression models in Table 2, which shows both the unadjusted coefficients and the coefficients adjusted for all other variables ( suggest that the influence of domestic use was stronger than that of density.  Table 3 shows the relationship between current socioeconomic indicators or microparticulate air pollution and all-cause or cause-specific mortality. The lower section of Smokeless fuels represented a small proportion of the total fuels used and contributed little to the mortality risks. The effects of the potential confounders were tested singly and in combination. There was no evidence of non-linear relationships. Figure 2 shows the associations between quartiles of domestic fuel consumption and all-cause mortality or the major mortality categories. These associations were progressive and remained strong and statistically significant for each cause group after adjustment for social class, education, crowding, unemployment in 1951, and for current socioeconomic indicators and PM 2.5 concentrations. They were present in both genders and were similar during the ICD 9 and ICD 10 periods. Figure 3 shows 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 1952/3 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. Tables 4 & 5 show the relationship between domestic fuel consumption and specific causes of mortality.
As before, the data are shown before and after adjustment for past and current socioeconomic indicators and current PM 2.5 levels. The respiratory conditions responsible for most deaths, chronic obstructive pulmonary disease and pneumonia, were strongly and consistently associated with domestic fuel consumption while associations with asthma mortality were somewhat weaker. However, the largest relative risks were with tuberculosis. While all four major causes of cardiovascular mortality were significantly associated with domestic consumption, the strongest relationship was with rheumatic heart disease.
The data on associations of domestic fuel use with mortality from specific cancers were striking ( cancers showed no increase in risk, with the exception of cervical cancer (1.131). Some cancers, for example melanoma and brain, were associated with a somewhat reduced relative risk.

DISCUSSION
Areas of the UK that had high domestic consumption of coal and related non-smokeless solid fuels in 1951/52 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, such as social class, level of education, overcrowding, unemployment and density assessed at birth and currently. They were also independent of current levels of air pollution as indicated by concentrations of microparticulate pollution (PM 2.5 ).
Very few comparable data exist in the literature, particularly with such a long follow-up period.
However, our findings accord with a body of evidence that early life exposure to air pollution has detrimental long-term health effects. 14 21 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 to unexposed children. 22 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. 23 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. 24 Our study necessarily depended on coal consumption as an index of domestic air pollution. We contend that is very reasonable to assume that the quantity of coal burnt relates to the amount of pollution emitted. The consumption measures in our study were shown to correlate with subsequent measurements of pollution in the UK, 12 and this approach is also validated by a number of studies carried out over the last 60 years, showing correlations with diverse health outcomes including reduced early growth, 5 chronic bronchitis and lung cancer. [12][13][14] In the post-war years, economic necessity drove the use of lowgrade, 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, sulphur and complex mixtures of aliphatic and aromatic hydrocarbons. 25 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 particulate matter. 26 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 pollution was vented through tall chimneys, decreasing ground-level pollution in approximate proportion to the inverse square of chimney height. 27 Current levels of pollution in the 342 local areas were based on ambient levels of fine particulate matter (PM 2.5 ) because large cohort studies have shown that this measure correlates best with long-term health effects of pollution. [28][29][30] The effects of PM 2.5 on current mortality (Table 2) in the present study are consistent with previously published estimates. 19 Our data do not allow determination of the age at which children were most vulnerable to the effects of pollution although the 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.
Although the consumption data come from a short time-period between 1951 and 1952, they can be considered a reasonable reflection of smoke emission over many years. UK coal production was fairly constant during the late 1940s, declined a little during the 1950s and only fell dramatically after 1960. 11 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. 31 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. 13 14 21 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. 32 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, 14 21 the Harvard Six Cities study, 29 and a multicentre study of European Cohorts. 33 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. 34 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. 35 The data on cancers ( 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. 36 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. 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 studies 37 but not others. 38 39 Migration is complex as, while migrants are generally healthier, those moving short distances tend to have higher mortality than those moving long distances. 40  North-East of England, which were the areas that had high coal consumption, had in-migration and outmigration levels in all age-groups that were far below the National average. 43 In addition, our analysis allows for the factors that tend to drive migration such as education and socioeconomic status. 44 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 to be 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. 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 (  13 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, 45 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 may be a less important confounder than overcrowding. There is considerable observational evidence linking overcrowding with physical health, especially in adults 46 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. 47 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 as firstly, smoking is associated with low socioeconomic status, 48 which was included in our analyses and secondly, standardisation of the data for lung cancer rates, as a proxy for smoking, did not explain the correlations we observed.

Conclusions
Although this is an ecological study based on geographical correlations, the results raise the possibility that domestic air pollution, experienced over 60 years ago by young children, affects human health now, by increasing mortality from a wide variety of diseases. The effect of early life exposure was much stronger than that of current pollution assessed by concentrations of microparticulates. Importantly, the findings raise the possibility that the health effects of air pollution have been underestimated as most studies do not have data on early life exposure, where vulnerability may be greatest, and even the effects of the lower levels of pollution experienced by current Western populations appears to have been underappreciated. Indeed, a recent mortality study based on the Medicare population suggests that a significant disease burden is associated with pollutant levels that are below current statutory standards. 49 The evidence that air pollution has been having a major detrimental impact on health for decades and

Footnotes
Contributors: DIWP discovered the dataset on coal consumption in 1951/2 and together with CO conceived the study. HS and PA advised on the use of the geographical data and carried out the redistricting analysis. DIWP, CO and AJ were responsible for the statistical analysis. The manuscript was written by DIWP, CO, AJ, HS and SH. All authors participated in a critical revision of the text and approved the final manuscript. DIWP is responsible for the overall content as the corresponding author.

Funding:
The work was supported by the Medical Research Council.
Competing interests: None declared.
Ethics approval: As the study was based on routinely collected mortality data no ethics approval was necessary.

Objective
To evaluate the impact of early life air pollution on subsequent mortality.

Design
Geographical study

Exposure
Routinely collected geographical data on the use of coal and related solid fuels in 1951-2 were used as an index of air pollution.

Main outcome measures
We evaluated the relationship between these data and both all-cause and disease-specific mortality among men and women aged 35-74 in Local Government Districts between 1993 and 2012.

Results
Domestic (household) coal consumption had the most powerful effect on mortality. There were strong correlations between domestic coal use and all-cause mortality (relative risk per SD increase in fuel use

Conclusion
Coal was the major cause of pollution in the UK until the Clean Air Act of 1956 led to a rapid decline in consumption. These data suggest that coal-based pollution, experienced over 60 years ago in early life, affects human health now by increasing mortality from a wide variety of diseases. •Use of national mortality data with virtually complete ascertainment and a large number of deaths (> 3.5 million).
•Coal consumption data provide an integral of air pollution over a 12 month period.
•The analysis also allows for current pollution levels •The study depends on the use of geographical correlations together with census-derived measures of socio-economic status; individual-level data are not available.  In common with many other countries, the UK had high levels of air pollution in the past. 9 Most of this was due to coal use in industry and as a domestic fuel, which gave rise to the "great stinking fogs" that previously characterised London and other major cities. 9 The London smog of 1952 killed an estimated 12,000 people 10    Historical Geographical Information System (GIS). 17 18 As the LFO areas either exactly corresponded to LGDs or were defined as aggregates of them, constructing digital boundary data for LFO areas was  Historically, air pollution was worse in predominantly industrial towns with a high population density; these towns also tended to have large populations of factory workers who left school early and worked in low status occupations. 14  LGDs. 20

Statistical methods
We studied deaths in people aged between 35 and 74 years, as this was the approximate age range of the generation who were in their first decade of life in 1951-52, and excluded older age groups where the cause of death was less likely to have been ascertained accurately. 21 We calculated the number of deaths expected for each five-year age and sex group within the LGDs by multiplying the number of people in each group by the national age-and sex-specific death rates in 1993-2012. We expressed the number of observed deaths in each area as a percentage of the expected deaths for that group, i.e. as standardised mortality ratios (SMRs). We used Poisson regression to model SMRs, expressing the effect of fuels and other variables on mortality in terms of the hazard ratio (or relative risk) per SD change in the explanatory variable. This weights the analyses appropriately to allow for variation in population size across LGDs. A Fisher-Yates transform was used to standardise pollution and socioeconomic variables.
This yielded variables with zero means, unit standard deviations (SDs) and symmetrical distributions, allowing direct comparison of the strength of associations with these variables.   were significantly associated with both all-cause and cause-specific mortality rates although the effects of domestic and industrial consumption tended to be most strongly related to the mortality outcomes.
Multivariate analyses were carried out to evaluate the relative effects of fuel usage and the major confounding variables derived from the 1951 census. This analysis shows that domestic consumption was strongly and significantly associated with both all-cause and cause-specific mortality rates, an effect which was markedly stronger than the other sources of fuel. Smokeless fuels represented less than 15% of the total fuels used (Table 1) and in separate regression analyses had no statistically significant independent effect on mortality rates. (Data not shown) There was a very high correlation (r=0.97) between domestic consumption and population density, resulting in unacceptable collinearity in regression analyses that contained both variables. This occurred because both had the same denominator, the area of the LGD. To allow the influence of both of these factors to be modelled statistically, we created two further variables: the sum of and the difference between domestic use and density. These variables were uncorrelated (r=0.0). These were entered into the regression models in Table 2, which shows both the unadjusted coefficients and the coefficients adjusted for all other variables (  Table 3 shows the relationship between current socioeconomic indicators or microparticulate air pollution and all-cause or cause-specific mortality. The lower section of Table 3 shows that the effect of domestic coal usage is attenuated by successive adjustments for socioeconomic indicators from the 1951 census, for indicators from the 2001 census together with particulate exposure and for both of these sets of factors. However, in all cases the association with domestic coal consumption remained strong and statistically significant.
The effects of the potential confounders were tested singly and in combination. There was no evidence of non-linear relationships. Figure 2 shows the associations between quartiles of domestic fuel consumption and all-cause mortality or the major mortality categories. These associations were progressive and remained strong and statistically significant for each cause group after adjustment for social class, education, crowding, unemployment in 1951, and for current socioeconomic indicators and PM 2.5 concentrations. They were present in both genders and were similar during the ICD 9 and ICD 10 periods. Figure 3 shows 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 1952/3 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. Tables 4 & 5 show the relationship between domestic fuel consumption and specific causes of mortality.
As before, the data are shown before and after adjustment for past and current socioeconomic indicators and current PM 2.5 levels. The respiratory conditions responsible for most deaths, chronic obstructive pulmonary disease and pneumonia, were strongly and consistently associated with domestic fuel consumption while associations with asthma mortality were somewhat weaker. However, the largest relative risks were with tuberculosis. While all four major causes of cardiovascular mortality were significantly associated with domestic consumption, the strongest relationship was with rheumatic heart disease.
The data on associations of domestic fuel use with mortality from specific cancers were striking ( cancers showed no increase in risk, with the exception of cervical cancer (1.131). Some cancers, for example melanoma and brain, were associated with a somewhat reduced relative risk.

DISCUSSION
Areas of the UK that had high domestic consumption of coal and related non-smokeless solid fuels in 1951/52 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 22 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 to unexposed children. 23 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. 24 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. 25 Our study necessarily depended on published crude estimates of coal consumption as an index of domestic air pollution. We contend that 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. While there is no information about exposure to other pollutants these are likely to have been of minor importance compared with coal combustion. 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][14][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. 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 post-war 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, sulphur and complex mixtures of aliphatic and aromatic hydrocarbons. 26 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 particulate matter. 27 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 groundlevel pollution in approximate proportion to the inverse square of chimney height. 28 Current levels of pollution in the 342 local areas were based on ambient levels of fine particulate matter (PM 2.5 ) because large cohort studies have shown that this measure correlates best with long-term health effects of pollution. [29][30][31] 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 the 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. 32 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 22 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. 33 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 22 the Harvard Six Cities study, 30 and a multicentre study of European Cohorts. 34 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. 35 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. 36 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. 37 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. 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 studies 38 but not others. 39 40 Migration is complex as, while migrants are generally healthier, those moving short distances tend to have higher mortality than those moving long distances. 41 42 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. 43 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 outmigration levels in all age-groups that were far below the National average. 44 In addition, our analysis allows for the factors that tend to drive migration such as education and socioeconomic status. 45 Finally, it is difficult to see how selective migration could explain the specific links between coal consumption and some conditions and not others ( 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, 46 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 adults 47 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. 48 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. 49 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 as firstly, smoking is associated with low socioeconomic status, 50 which was included in our analyses and secondly, standardisation of the data for lung cancer rates, as a proxy for smoking, did not explain the correlations we observed.

Conclusions
Although this is an ecological study based on geographical correlations, the results raise the possibility that domestic air pollution, experienced over 60 years ago by young children, affects human health now, by increasing mortality from a wide variety of diseases. The evidence that air pollution has been having a major detrimental impact on health for decades and continues to do so underlines the importance of routine environmental tracking and surveillance systems in detecting and avoiding the harmful health effects of environmental pollution. The data presented in this analysis also have implications for the long-term health of the populations of countries that still depend on large amounts of coal for their domestic markets. This includes newly industrialized countries such as India or China, where coal is a major energy source, and resource-poor countries, where indoor air pollution from cook stoves results in heavy exposure of women and young children to pollution. 7

Footnotes
Contributors: DIWP discovered the dataset on coal consumption in 1951/2 and together with CO conceived the study. HS and PA advised on the use of the geographical data and carried out the redistricting analysis. DIWP, CO and AJ were responsible for the statistical analysis. The manuscript was written by DIWP, CO, AJ, HS and SH. All authors participated in a critical revision of the text and approved the final manuscript. DIWP is responsible for the overall content as the corresponding author.

Funding:
The work was supported by the Medical Research Council.
Competing interests: None declared.

Objective
To evaluate associations between early life air pollution and subsequent mortality.

Design
Geographical study

Exposure
Routinely collected geographical data on the use of coal and related solid fuels in 1951-2 were used as an index of air pollution.

Main outcome measures
We evaluated the relationship between these data and both all-cause and disease-specific mortality among men and women aged 35-74 in Local Government Districts between 1993 and 2012.

Conclusion
Coal was the major cause of pollution in the UK until the Clean Air Act of 1956 led to a rapid decline in consumption. These data suggest that coal-based pollution, experienced over 60 years ago in early life, affects human health now by increasing mortality from a wide variety of diseases. •Use of national mortality data with virtually complete ascertainment and a large number of deaths (> 3.5 million).
•Coal consumption data provide an integral of air pollution over a 12 month period.
•The analysis also allows for current pollution levels •The study depends on the use of geographical correlations together with census-derived measures of socio-economic status; individual-level data are not available.  In common with many other countries, the UK had high levels of air pollution in the past. 9 Most of this was due to coal use in industry and as a domestic fuel, which gave rise to the "great stinking fogs" that previously characterised London and other major cities. 9 The London smog of 1952 killed an estimated 12,000 people 10    Historical Geographical Information System (GIS). 17 18 As the LFO areas either exactly corresponded to LGDs or were defined as aggregates of them, constructing digital boundary data for LFO areas was  Historically, air pollution was worse in predominantly industrial towns with a high population density; these towns also tended to have large populations of factory workers who left school early and worked in low status occupations. 14  LGDs. 20

Statistical methods
We studied deaths in people aged between 35 and 74 years, as this was the approximate age range of the generation who were in their first decade of life in 1951-52, and excluded older age groups where the cause of death was less likely to have been ascertained accurately. 21 We calculated the number of deaths expected for each five-year age and sex group within the LGDs by multiplying the number of people in each group by the national age-and sex-specific death rates in 1993-2012. We expressed the number of observed deaths in each area as a percentage of the expected deaths for that group, i.e. as standardised mortality ratios (SMRs). We used Poisson regression to model SMRs, expressing the effect of fuels and other variables on mortality in terms of the hazard ratio (or relative risk) per SD change in the explanatory variable. This weights the analyses appropriately to allow for variation in population size across LGDs. A Fisher-Yates transform 22 was used to standardise pollution and socioeconomic variables. This yielded variables with zero means, unit standard deviations (SDs) and symmetrical distributions, allowing direct comparison of the strength of associations with these variables.

Patient and Public Involvement
No patients or public were involved in this study.  Figure 1   were significantly associated with both all-cause and cause-specific mortality rates although the effects of domestic and industrial consumption tended to be most strongly related to the mortality outcomes.
Multivariate analyses were carried out to evaluate the relative effects of fuel usage and the major confounding variables derived from the 1951 census. This analysis shows that domestic consumption was strongly and significantly associated with both all-cause and cause-specific mortality rates, an effect which was markedly stronger than the other sources of fuel. Smokeless fuels represented less than 15% of the total fuels used (Table 1) and in separate regression analyses had no statistically significant independent effect on mortality rates. (Data not shown) There was a very high correlation (r=0.97) between domestic consumption and population density, resulting in unacceptable collinearity in regression analyses that contained both variables. This occurred because both had the same denominator, the area of the LGD. To allow the influence of both of these factors to be modelled statistically, we created two further variables: the sum of and the difference between domestic use and density. These variables were uncorrelated (r=0.0). These were entered into the regression models in Table 2, which shows both the unadjusted coefficients and the coefficients adjusted for all other variables (  Table 3 shows the relationship between current socioeconomic indicators or microparticulate air pollution and all-cause or cause-specific mortality. The lower section of Table 3 shows that the effect of domestic coal usage is attenuated by successive adjustments for socioeconomic indicators from the 1951 census, for indicators from the 2001 census together with particulate exposure and for both of these sets of factors. However, in all cases the association with domestic coal consumption remained strong and statistically significant.
The effects of the potential confounders were tested singly and in combination. There was no evidence of non-linear relationships. Figure 2 shows the associations between quartiles of domestic fuel consumption and all-cause mortality or the major mortality categories. These associations were progressive and remained strong and statistically significant for each cause group after adjustment for social class, education, crowding, unemployment in 1951, and for current socioeconomic indicators and PM 2.5 concentrations. They were present in both genders and were similar during the ICD 9 and ICD 10 periods. Figure 3 shows 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 1952/3 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. Tables 4 & 5 show the relationship between domestic fuel consumption and specific causes of mortality.
As before, the data are shown before and after adjustment for past and current socioeconomic indicators and current PM 2.5 levels. The respiratory conditions responsible for most deaths, chronic obstructive pulmonary disease and pneumonia, were strongly and consistently associated with domestic fuel consumption while associations with asthma mortality were somewhat weaker. However, the largest relative risks were with tuberculosis. While all four major causes of cardiovascular mortality were significantly associated with domestic consumption, the strongest relationship was with rheumatic heart disease.
The data on associations of domestic fuel use with mortality from specific cancers were striking ( cancers showed no increase in risk, with the exception of cervical cancer (1.131). Some cancers, for example melanoma and brain, were associated with a somewhat reduced relative risk.

DISCUSSION
Areas of the UK that had high domestic consumption of coal and related non-smokeless solid fuels in 1951/52 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 to 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 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][14][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. 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 post-war years, economic necessity drove the use of lowgrade, 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, sulphur 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  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y 10 particulate matter. 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 particulate matter (PM 2.5 ) because large cohort studies have shown that this measure correlates best with long-term health effects of pollution. [30][31][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 the 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 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 study, 31 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  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   11 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 auto-correlation may lead to lead to confidence intervals that are narrower than they should be. However, an analysis evaluating the influence of spatial autocorrelation (described in detail in 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 confidence intervals, for example 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 studies 39 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 to be due to confounding from an unknown source. However, a sensitivity analyses suggested that such a variable would have to correlate  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   12 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 "Air Toxics" from sources other than the combustion of coal and other 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 adults 48 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 is 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.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Conclusions
Although this is an ecological study based on geographical correlations, the results raise the possibility that domestic air pollution, experienced over 60 years ago by young children, affects human health now, by increasing mortality from a wide variety of diseases. The effect of early life exposure was much stronger than that of current pollution assessed by concentrations of microparticulates. Importantly, the findings raise the possibility that the health effects of air pollution have been underestimated as most studies do not have data on early life exposure, where vulnerability may be greatest, and even the effects of the lower levels of pollution experienced by current Western populations appears to have been underappreciated. Indeed, a recent mortality study based on the Medicare population suggests that a significant disease burden is associated with pollutant levels that are below current statutory standards. 52 The evidence that air pollution has been having a major detrimental impact on health for decades and continues to do so underlines the importance of routine environmental tracking and surveillance systems in detecting and avoiding the harmful health effects of environmental pollution. The data presented in this analysis also have implications for the long-term health of the populations of countries that still depend on large amounts of coal for their domestic markets. This includes newly industrialized countries such as India or China, where coal is a major energy source, and resource-poor countries, where indoor air pollution from cook stoves results in heavy exposure of women and young children to pollution. 7

Footnotes
Contributors: DIWP discovered the dataset on coal consumption in 1951/2 and together with CO conceived the study. HS and PA advised on the use of the geographical data and carried out the redistricting analysis. DIWP, CO and AJ were responsible for the statistical analysis. The manuscript was written by DIWP, CO, AJ, HS and SH. All authors participated in a critical revision of the text and approved the final manuscript. DIWP is responsible for the overall content as the corresponding author.

Funding:
The work was supported by the Medical Research Council.
Competing interests: None declared.
As described in the methods section, the coal consumption data (x) for each area (x1,x2áåå š342) used in Table 2 were transformed using a Fisher-Yates normal transformation. Tabulating the mean values of the product xixj in bands derived according to the value of the inter-area distance dij confirms that the correlation falls rapidly in this way.

Study design 4
Present key elements of study design early in the paper 5-6 Setting 5 Describe the setting, locations, and relevant dates, including periods of recruitment, exposure, follow-up, and data collection 5-6 Participants 6 (a) Cohort study-Give the eligibility criteria, and the sources and methods of selection of participants. Describe methods of follow-up Case-control study-Give the eligibility criteria, and the sources and methods of case ascertainment and control selection. Give the rationale for the choice of cases and controls Cross-sectional study-Give the eligibility criteria, and the sources and methods of selection of participants 5-6 (b) Cohort study-For matched studies, give matching criteria and number of exposed and unexposed Give the source of funding and the role of the funders for the present study and, if applicable, for the original study on which the present article is based 25 *Give information separately for cases and controls in case-control studies and, if applicable, for exposed and unexposed groups in cohort and cross-sectional studies.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47