Previous studies have shown (1) that childhood cancers and leukaemias in Great Britain exhibit geographical clustering of birth places: (2) they occur at increased densities around industrial sites with large scale combustion processes or using volatile organic compounds (VOCs), or which incinerate waste: (3) among children who moved house between birth and death the first addresses were closer to these hazards than were the later ones and migrations were more often directed away from a nearby hazard than towards one.^1–7 The increased effectiveness of early exposure, combined with the known effects of obstetric radiation exposures,⁸ suggests that these diseases are often initiated prenatally.

The use of hazard proximity birth-death comparisons was initially dictated by the absence of a suitable set of non-cancer controls in the study from which the case material was extracted. These data were designed to measure the effects of obstetric radiation by comparing cases with paired non-cancer controls⁸; but the controls had been geographically matched with the cases and could not then be used to make geographical comparisons. The migration based method gave coherent results but was open to the possible objection that the migration patterns in the cancer children may have reflected a general population movement, perhaps related to area demolitions and subsequent rehousing. The recent publication of independent and comprehensive national pollution data now affords an opportunity to retest the atmospheric birth hazard hypothesis, and to identify specific chemical hazards. This is the objective of this study.

The UK National Atmospheric Emissions Inventory (NAEI) has recently published through its web site, detailed geographical displays of emissions of many different chemical species for 2001. These maps⁹ were downloaded and individual pixels—resolved at 412.8 metres per pixel—were translated to grid references (see appendix, available on line http://www.jech.com/supplemental). Emission levels are expressed “‘per square km per year” on a seven point colour coded scale, using units that vary from grams (dioxins), through kilograms (chromium, nickel), to tonnes (sulphur dioxide, PM10 particles). Some were measured directly and others by ascertaining activities with known emission characteristics. Lower scale values are indicated as broad map zones, but the highest levels are shown as small clusters of red pixels or by individual pixels, often appearing to represent individual sources. The maps are readily available for inspection.

Except for the red pixel hotspots, the main scale divisions were too broad for effective comparisons of birth and death addresses. NAEI also points out that because of atmospheric diffusion the emission estimates do not directly represent the air we breathe. However, as in earlier studies, it is possible to compare birth and death addresses by measuring hotspot distances. The “case centred” method, used again here, examines the surroundings of each address in turn to identify the nearest of the hotspots. The selection is entirely objective and the resulting comparative distance measurements are available in very large numbers.

METHODS

The case material was extracted from a file of all 22 458 deaths from leukaemia or other cancer occurring before the 16th birthday in Great Britain between 1953 and 1980. They were classified into 11 main groups (lymphatic, myeloid, monocytic and unclassified/other leukaemias: lymphomas, nephroblastoma, CNS tumours, neuroblastoma, bone cancers, other solid cancers, and fatal “benign” tumours). Home addresses at death were always recorded and where parents were subsequently interviewed, the birth address was also obtained and the postcode later identified. Map references were extracted from the Central Postcode Directory. Most were probably accurate to within 0.1 km of the actual address although larger PCs may have incurred errors up to about 1.0 km. This dataset has been described in detail elsewhere.^3,4 To ameliorate the effects of the interval between the survey dates and the 2001 date of the NAEI maps, this analysis is based upon those children who died in the second half of the study period, 1966–1980. Orkney and Shetland are excluded. Each leukaemia type and cancer type was analysed individually, together with the grouped reticuloendothelial cancers, the grouped “solid” cancers, and all cancers together.

The UK emissions maps were downloaded from the NAEI web site⁹ into a bitmap file (*.bmp) and unwanted areas were edited out. Relations between pixel northings/eastings and grid coordinates were solved against small islands and promontaries (see appendix). The maps are orientated true (grid) north-south with identical northing and easting scales of 412.8 metres per pixel. Subsequently measured distances from addresses to hazardous emission locations are accurate to within about 500 metres. Mean nearest hotpoint distances (truncated to a maximum 50.0 km), and means of individual distance differences, were tested against standard errors: and outward/inward ratios through χ² tests. All results subsequently quoted were extremely significant and p values are not individually set out.

Estimates of the relative risks of being born close to, or distant from, an emission hazard were first calculated among children who had moved more than 1.0 km between birth and death, and who had one address within 1.0 km of the nearest and the other address more than 1.0 km away from the nearest hazard point. Viewed against a presumed overall short term cross boundary equilibrium of migrations in the general population (see later), outward/inward ratios substantially greater than 1.0 were taken to represent the selective cancer-initiating effects of the birth locations themselves. These relative risks were then used to calculate the proportions of sub-1.0 km cancer births attributable to each hazard; and then, through reference to the total number of sub-1 km births, to estimate the proportion of all cancers that might be so caused.

RESULTS

Large scale combustion emissions

Table 1 gives the migration patterns in relation to the major fossil fuel emissions, within the above subset. Carbon dioxide (expressed as tonnes of carbon) represents the entire sum of fossil fuel use including coal, oil, and gas. The outward/inward ratio of 2.05 indicates a movement of cancer migrants away from zones of high fuel consumption and, presumably, high population densities and industrial concentrations. Sulphur dioxide emissions, which arise mainly from coal burning, failed to mirror this net outward movement suggesting that the main correlates of the differential migration must be oil and/or gas combustion. PM10, nitrogen oxides, and carbon monoxide are all preferentially associated with oil burning, particularly in internal combustion engines, and their high outward/inward ratios indicate that transport emissions contribute a major part of the overall carbon related effect.

Viewed against a presumed overall migration equilibrium, these ratios show the relative cancer risks among inner zone compared with outer zone births. The findings were mirrored in significant excesses of mean death distances over birth distances. The mean within-pair differences also differed from zero.

Volatile organic compounds

Non-methane volatile organic compounds (NMVOC) are discharged in quantities comparable with materials listed in the last section. In roughly equal proportions they reflect (a) solvent use, (b) engine exhaust fuel residues and fuel evaporation, and (c) other industrial/refinery processes. They include benzene, benz(a)pyrene, and 1,3-butadiene, each of which is discharged in tonnage or subtonnage quantities; also dioxins, whose upper range is lower bounded by only 0.1 gram/km²/year. Benzene is a suspected leukaemogen while benz(a)pyrene, 1,3-butadiene, and dioxins are known carcinogens. Benzene is a major component of motor fuel while atmospheric 1,3-butadiene is largely derived from petrol and diesel engine exhausts. Table 2 shows the results of migration analyses.

The very high outward/inward ratio (3.78) within 1.0 km of the 1,3-butadiene hotspots strongly suggests a specific association. Despite the fogging effects of the limited map resolutions, crossings of a 0.7 km boundary returned a ratio of 4.66 (1273/273); while 0.5 km and 0.3 km gave 7.65 (819/107) and 8.39 (411/49). The butadiene analyses were repeated with biases attached to all the hotpoint locations. When they were “moved” by 1.0 km south and west, the 7.65 ratio (above) was reduced to 2.50 and a similar north easterly bias reduced it to 2.34. These findings indicate a ground level source with distance dilution and a very local carcinogenic action. They also confirm the precision of the map location data for 1,3-butadiene.

Dioxins also showed increasing ratios at shorter distances; from 2.21 at 1.0 km to 3.15 at 0.7 km and 3.24 at 0.5 km. Carbon monoxide showed a qualitatively similar, but much weaker, effect.

Metals and non-metals

Table 3 summarises results for copper, zinc, beryllium, lead, calcium, cadmium, and mercury, and table 4 for nickel, manganese, tin, vanadium, arsenic, selenium, and chromium. None showed birth-death asymmetries comparable with those shown in tables 1 and 2. Lead and cadmium showed small birth proximity excesses but discharges of these metals are closely associated with the fuel hungry processes represented in tables 1 and 2 and the effect is almost certainly indirect. Despite the near eradication of leaded petrol, the linear distributions of lead on the NAEI maps show that it is still related to transport. Calcium discharges also showed a small emigration excess but are associated with lime and cement manufacture, processes using large amounts of fossil fuel. Mercury, another metal with a modestly asymmetric migration pattern, is associated with waste incineration and hence with dioxins. All of these relations are probably indirect, with the high points for each substance acting as proxy for other emissions.

Several substances showed small reverse effects; an excess of death place proximities. Arsenic, nickel, and vanadium recorded inward/outward ratios of 1.391, 1.597, and 1.718 respectively. Arsenic is associated with industrial coal burning, vanadium with industrial fuel oil uses, and nickel with both, and we might have expected to see parallels rather than the inverse. NAEI data again suggest an indirect origin. All three emissions have been greatly reduced in recent years and their recent topographical distributions all show major conurbations with “hollow emission centres”, presumably arising from changed fuel uses in city centres. The probable explanation is through an inverse statistical relation with high traffic densities.

Risk assessments

Tables 1 to 4 optimised birth-death contrasts by excluding children who had not moved very far and those who had moved entirely within high exposure or entirely within low exposure zones. However, this prior selection barred assessment of an “attributable risk” (AR) in the total population, an assessment demanding inclusion of children who did not cross any critical boundaries or who did not move house at all. The most strongly interacting emission types were therefore reanalysed with such children included and the results are presented in table 5. The outward/inward ratios in table 5 are less dramatic than those in tables 1 and 2 because they incorporate many less discriminatory records, but they are still highly significant .

Table 5 uses relative risks (RR) from tables 1 and 2 to calculate attributable proportions (RR−1)/RR for the inner zones and applies them to all those born inside 1.0 km—that is, rows 2 to 4 of the table. Absolute attributable numbers are then derived and expressed as a proportion of all cancer children, giving an overall attributable risk (AR).

It is difficult to compare results for different emissions, whether in terms of RR or AR, because of the varying ways in which different hotspot cut off points were defined. Less extreme cut off values will generate a less extreme RR, but the concomitant increase in numbers of hazard points may then increase estimates of AR. In Table 5 the hotspots for dioxins and for 1,3-butadiene were comparatively infrequent and gave the highest RRs; while the abundant hotspots for NMVOC and nitrogen oxides gave lower RR values, yet “accounted for” larger proportions of child cancers. The several AR estimates can not be simply added together as each emission type acts to some extent as proxy for others but it is clear that these migration derived RR and AR estimates have important implications. They show that most childhood cancers are probably initiated by close perinatal encounters with one or more of these high emissions sources. The low atmospheric concentrations of these substances, and a timing analogy with the effects of fetal irradiation⁸ suggest that the mother may act as an accumulating filter and pass carcinogens across the placenta. However, effective direct exposures in early infancy, or through breast milk, or even preconceptually, cannot be excluded.

The migration phenomena described in tables 1 to 5 refer to the sum of all leukaemias and cancers but they were equally evident in each tumour class separately and in children dying at different ages. The mean age at death among migrants (97.4 months) was greater than among non-migrants (70.8 months) but this was simply because a later death allowed more time to become a migrant. Both 1,3-butadiene and dioxins are believed to disturb sex determination so the data were examined for sex ratio differences at shorter and longer distances from their emission hotspots. No evidence of any gradient was found.

DISCUSSION

The time interval between the cancer records and the emissions data must be considered. NAEI describes many recent changes in fuel uses and in manufacturing, power production, and waste disposal practices. However, most of them were improvements and while previous high emissions may have diminished, most of those still existing in 2001 would be at least as active when these children were born. These changes underline the special advantages of basing analyses upon hotspot distances rather than emission levels.

The basic method, and all the results, depend upon the premise of a migration equilibrium between hotspot related inner and outer zones in the general child population. The differential migrations among the cancer children, an approximate 0.1% subset of the total, can then be seen as a consequence of early risk variations in different places. In the early postwar years, however, much substandard and bomb damaged city centre housing was demolished and replaced, some of it by out of town building, and families with young children may have been motivated on amenity grounds to move away from emission sources such as oil storage facilities, gasworks, factories, chimneys, railways, and road traffic; at least where work and family economics and municipal housing waiting lists permitted. We have to ask whether such selective population movements could have accounted for the findings in the cancer subset.

A first consideration is the necessary scale of any explanatory demographic movements. For example, the 1,3-butadiene findings in table 5 would require a general emigration rate of 49.6% (2227 of 4490) among children born within 1.0 km over a period corresponding with the mean age of these cancer deaths (7.4 years). After allowing for incoming replacements the net outward flow is still 36.1% (2227–604)/4490; sufficient over the full survey period to have depopulated all near contact zones. Personal preferences in the face of competitive domestic economics are an integral part of migration dynamics but there are no grounds for supposing that children incubating cancer should be any different in these respects from other children.

A second consideration is the universality of the findings around the high risk sites, irrespective of the varying sizes of their inside populations and outside populations. A third is the absence of any evidence of major outward demographic movements around other industrially related emissions such as zinc, copper, tin, nickel, chromium, arsenic, and sulphur dioxide. A fourth is the specificity of the significant emission types; which compels a specific interpretation. All of them are associated with waste incineration or with oil combustion or evaporation and several are known carcinogens. Some showed an increasing relative risk at shortening ranges and a sharp decrease when their given locations were marginally displaced; and all of them correspond with previously identified types of hazardous industrial installation.

It is always difficult to say whether any particular result points to a specific carcinogenic effect, or whether it is acting as proxy for other geographically associated substances. For example, carbon monoxide is a powerful indicator of transport activity, itself a generator of many other substances, so we need not conclude that it must itself be an active substance. (Although it may be.) Benzene is also difficult to assess because the original leukaemogenic suspicion might itself spring from indirect associations. Nitrogen oxides, with their complex trains of secondary atmospheric reactions raise similar issues, as do many other substances, including 1,3-butadiene.^10,11 However, the present associations with 1,3-butadiene, dioxins, and benz(a)pyrene are sufficiently specific to conclude that they are probably among the truly active agents.

A literature search returned no similar studies and no similar findings. Systematic attempts to link the birthplaces of cancer children to specific exposures have been largely limited to ionising and electromagnetic radiations. There are some cancer studies around incinerators in the general population, but not specifically related to children or to birthplaces. There are many reports of the carcinogenic effects of 1,3-butadiene and dioxins in rats and mice¹²; and of human occupational exposures to 1,3-butadiene in the synthetic rubber industry. Some of the human reports have decided that 1,3-butadiene and dioxins are significant human carcinogens,^13,14 while others give more cautious “probable” or “possible” verdicts^15–17 and yet others report doubtful or negative effects.^18–21 Two mass exposure episodes to carcinogens so far remain uninformative including the major escape of dioxins at Seveso and a possible relation between an alleged leukaemia epidemic in Basra and the Gulf war oil well fires in Kuwait and southern Iraq. These uncertainties have inhibited any consistent public health policy. For example, some countries are taking serious efforts to control emissions of 1,3-butadiene,²² while others have decided to await more information.²³ The massive evidence on this substance, reported here, should allow no further delays.

The main additional policy implications are for a major redirection of the research effort relating to childhood cancer. This should now concentrate on determining the exact timings of these chemically determined initiations—whether in early infancy or prenatally, or even preconceptually; and a search for appropriate engineering and social solutions.