Document rp5dqzZ5Jqj55jrq0ZrOxRY7G

Journal of Epideniiologv and Cornnumiry Health 1997;Sl:15 1-159 Hazard proximities of childhood cancers in Great Britain from 1953-80 151 E G Knox, E A Gilman Abstract birth and death, the proximity effect was -Study objectives Firstly, to examine re- limited to the birth addresses. -lationships between the birth and death Conclusions Childhood cancers are geo- addresses of children dying from leuk- graphically associatedwith two main types aemia and cancer in Great Britain, and the of industrial atmospheric effluent namely: sites of potential environmental hazards; (1) petroleum derived volatiles and (2) kiln and secondly to measure relative case and furnace smoke and gases, and efflu- densities close to, and at increasing dis- ents from internal combustion engines. -tances from, different hazard types. Design Home address postcodes (PCs) (JEpidemwl Community Health 1997;51:151-159) and their map coordinates were identified at birth and at death in children who died Our previous studies showed that childhood from leukaemia or cancer. Potentiallyhaz- leukaemias and cancers occurred in small geo- ardous industrial addresses and PCs were graphical clusters.' * Among 9411 childhood listed from business and other directories, registrations of leukaemia and lymphoma in and map coordinates obtained from the Great Britain between 1966 and 1983, for Central Postcode Directory or else located whom we knew the residential postcode (PC) directly on Ordnance Survey (OS) maps. coordinates, there were 264 case pairs/triplets Railway lines and motorways were digit- who were separated by less than 150m; and in ised from OS maps. Numbers of deaths England and Wales there were 520 registration (and births) at successiveradial distances pairs who shared a common census enu- fromthese hazards were counted and com- meration district (ED). Among 22 458 child- pared with expected numbers. The latter hood leukaemia and cancer deaths in Great were based on a count of all PCs at similar Britain between 1953 and 1980, 503 pairs distances. Relative case density ratios at shared a comxhon PC. All these numbers were successive distances &om the hazards about twice the expected values. The excesses were obtained from observed and expected were statistically significant, and they were not numbers, aggregated over similar sites. explained by national variations of ED and PC This was repeated for different hazard sizes (numbers of households and of postal types and results were tested for evidence delivery points). Studies of house numbering of systematic centrihgal case density gra- revealed a further excess of pairs which bridged -dients. the boundaries of adjacent PCs. Participants and setting AU 22 458 chil- The close address clusters were about 300 m dren dying from leukaemia or cancer aged in diameter, but there was also a surrounding 0-15 years, in England, Wales, and Scot- zone of moderately enhanced risk, tapering -land, between 1953 and 1980. towards normal over a radius of 2-3km. PC Main results Relative excesses of leuk- clustering was stronger among birth addresses aemias and of solid cancers were found than among death addresses. It was not limited near the following: (1) oilrefineries, major to the leukaemias and it involved all childhood oil storage installations, railside oil dis- cancers. Some of the close pairs were con- tribution terminals and factories making cordantly affected siblings living at the same bitumen products; (2) motor car factories, address, but their exclusion did not alter the coach builders, and car body repairers; (3) overall conclusion. The scale of the geo- Cottage, Great Comberton, Worcestershire WRlO 3DU E G Knox major users of petroleum products including manufacturers of solvents, paint sprayers, fibreglass fabricators, paint and varnish makers, plastics and detergent manufacturers, and galvanisers; (4) users graphical clusters was commensurate with that of previously demonstrated space-time clusters among the 1966-83 leukaemia registrations, and also among the births and onsets of the fatal 1953-80 l e ~ k a e m i a s . ~ ~ ~ Department of Public Health and Epidemiology, Medical School, University of of kilns and furnacesincludingsteelworks, power stations, galvanizers, cement makers, brickworks, crematoria and aluminium, zinc, and irodsteel foundries; (5) If the geographic clustering is genuine and not a demographic artefact, as is now clear, then it must reflect the existence of localised environmental hazards. Many of them must B-gham, Bvrmngham B15 ZTT E A Gilman airfields, railways, motorways and har- have persisted over periods of years, although bours. The findings for leukaemiasand for the space-time interactions show that some solid cancers were indistinguishable. The could have been transient or intermittent. They hazard proximities of birth addresses were act most powerfully within a few hundred ~ for publication stronger than for death addresses. For metres, but some of them reach out to a few 1996 children who had moved house between kilometres. I 152 A previous study showed that the map locations of the 9411 registrations were closer to a variety of potentially hazardous sites than were a randomly chosen set of P C S .T~he 264 sub-150 m pairs among them showed an even stronger relationship. The apparent hazards included oil refineries, oil storage and distribution depots, railway lines, and other industrial sites. Effective ranges extended as far as 5km from the sources. This suggested a hazard related to large scale usages of fossil fuels, especially petroleum, operating through leakage or evaporationor combustion -perhaps all three. The matching secular trends of childhood leukaemia and of petroleum usage over many decades accords with this hypothesis. However, there were several problems of interpretation. First, the proximity investigations were undertaken initially to test the reality of the then uncertain clustering, and any causal speculations involved a second use of the same data. This introduced an unsatisfactory circularity. Second, no child population denominators were available for geographical analyses conducted on this small scale. The use of randomly selected PCs as indicators of the geographical distribution of those at risk entailed several demographic uncertainties. The most important was the possibility that numbers of children in PCs close to industrial sitesmight show a systematicpositive bias. This caveat was also elaborated by commentators.' Third, there was a question of environmental confounding; hazards of one kind might be related geographicallyto hazards of another kind, with a consequent risk of false attribution. This applied especially to railways. Finally, the close (300 m) spacing of the core clusters did not match the wider (c 5 km) radii of the main hazard proximity effects. These problems demanded additional study. Like the previous one, this reinvestigation lacks a true measure of child exposure on the requisite fine scale. No such data are generally available. The fundamental problems of interpretation are therefore the same. However, it is based on a fresh data set - larger and more comprehensive, and including addresses and dates of birth as well as of onset and death. It records solid cancers as well as leukaemias. But most crucially it starts from the now well supported premise that persistent local hazards must indeed exist. This no longer needs to be demonstrated. The present task is simply to discover what they are. Methods The cancer data set used here is the second of the two noted above, and it has been described elsewhere.689It records all 22 458 deaths from leukaemias and other cancers in children aged 0-15 years in England, Wales, and Scotland between 1953and 1980.After excluding deaths in the Northern Isles of Scotland, and a few recording errors, we were left with a working file of 22448 death addresses. The birth addresses of many were also available. Although these data are limited to fatal cases, the great majority of the children affected in these years died within the time limits of the investigation; over 75% within five years of diagnosis.'o The existence and locations of putative environmentalhazardswere determined in several ways. Point hazards such as oil refineries, gasworks, airtields, and crematoria were identified through direct searches of Ordnance Survey (OS) maps and street atlases. Many factories were found in classified business directories." Others such as motor car works, nuclear installations, benzene refineries, and TV transmitters were taken from specialised lists. The lists and directories supplied addresses and PCs, which were then translated to map referencesvia the CentralPostcode Directoy. Where ``business" PCs could not be traced in our "residential" directory we used the alphabetically nearest residential PC (last character) as the index position. Business addresses listed as "Company House" were avoided in favour of actual factories. Some locations were reconfirmed or amended through map inspections or site visits. Detected errors were typically between 100 and 300 m. Series of particular hazard types usually consisted of 30-50 locations taken from the index in alphabetic order. For a few uncertain results, extensions of these lists enabled us to reject a suspect finding. Linear cartographic features - railways and motorways - were digitised from OS maps using a digitising tablet. This registered strings of point coordinates and allowed the labelling of their beginnings, ends, breaks, and branch points. Tablet coordinates were translated to map positions against the key punched positions of the map sheet corners. The railways were digitised in 11separate regions, with London excluded because of its density and complexity. Motorways outside the recently completed M25 London Orbital (which was itself omitted) were digitised as a single set. For the point hazards, numbers of cases and numbers of PCs were counted in successively greater circles around each source. Crude densities of cases per 1000PCs within each annulus were calculated from the aggregated numbers, and many close proximity concentrations were found. However, we feared that higher case densities in some regions of the country may have coincided with geographical concentrations of certain industries. The calculations were therefore repeated using a form of indirect regional standardisation. This was based upon direct measurements of case densities in 56 urban and non-urban areas, as described later. Region specific expected numbers of cases, within successive annuli, were then calculated. The expected number was the product of the regional case density per PC and the local number of PCs within the annulus. The observed and expected results were summed to generate a standardised density ratio (SDR) for each successive circle around each hazard type. The method is set out in an appendix. For linear map features (rail, road) we found and measured the shortest case to hazard distance. Each address was examined against every node in the net and perpendiculars were k nazarcrpmxrmates of ch-d cancers 153 (x)Table 1 Residential postcode densities (per km2) at di#eerent distancesjivm index points &ranee em) Index (No) ( M . 3 -0.6 -1.0 -2.0 -3.0 - 4.0 - 5 . 0 -7.0 - 10.0 Random PCs Random cases Refineries' Airfields Oil storest Chemical works+ Crematoria Car making Safe sites? (500) (700) (21) (192) (149) (83) (64) (193) (195) 94.05 98.12 0.51 2.43 25.80 37.67 27.63 62.21 105.21 70.42 75.68 2.25 3.74 33.75 33.76 50.53 58.98 88.48 60.96 66.31 5.04 6.62 36.80 34.91 59.65 63.03 77.99 47.83 52.66 9.56 8.20 32.38 35.47 53.75 56.09 67.06 39.81 46.04 15.06 9.72 26.14 32.01 49.60 47.92 57.10 35.99 42.48 14.11 10.13 22.85 28.68 42.81 42.18 51.35 33.21 38.94 16.35 10.19 19.86 27.10 37.33 36.38 47.72 30.28 34.79 15.82 10.32 16.52 24.84 32.06 31.17 41.65 26.67 30.37 13.84 9.37 15.44 17.92 27.76 27.67 42.90 'Including 6 benzene refineries, t Combined major oil storage locations, and railside (mainly) oil distribution terminals. $ Combined "petrochemical" and other "indusmal chemicals"works plus other factories noted to have storage tank facilities. 1See text. Total land area of England, Scotland and Wales is 229 870 h2T.he overall density of the 1 372 359 residential postcodes is 5.97/km2. dropped to inter-node segments. Cancer ad- dresses were allocated to parallel bands on either side of the line rather than to annuli. A similar allocation of all PCs was impractical here, so numbers in different distance bands were estimated from the locations of 10000 randomly selected PCs. The counted numbers were adjusted according to the sampling fraction, and this was used in tum to calculate a regionally standardised expected number, as above. In the absence of population based denominators for individual PCs, alternative small group analyses were considered. A reclassification of addresses by census ED could have supplied an alternative set of coordinates, and could have provided local estimates of child years at risk. Unfortunately, appropriate census ED data were not available for the greater part of the study period. This choice would also have discarded the superior geographical resolution of the PCs. There are approximately 11 PCs per ED in England and Wales, and other studies using census based child denominators found in practice that EDs had to be aggregated into even larger blocks. These studies also failed to show clustering at these relatively coarse 1e~els.I~This approach was unlikely to meet the needs of a hazard proximity study pursuing very close range effects: and which was founded upon an initial demonstration of clustering within much smaller areas. The matched case-control format of our main data source suggested another method for displaying comparable proximity patterns among non-cancer children. Unfortunately, control addresses had not been postcoded, matched controls were available only for about two thirds of cases, and they had been geographically matched according to place of residence at the time that the "case" had died. They were not thus really suitable for this task. Over matching in this respect was likely to obscure any true proximity effects.In pragmatic terms the extra work was beyond current resources and it seemed unlikely to attract them. As with our previous proximity study, the crucial demographic question was whether ap- parent case concentrations might result entirely from systematic concentrations of PCs with large child populations around the types of hazard examined. Again, no appropriate data were available within the distance ranges considered. Regional standardisation eliminates only one component of the problem and less resolved population groupings (eg EDs) could offer no advantages. This focal demographic artefact was entirely hypothetical, and unsupported by external evidence,but we felt that it should be considered seriously. Fortunately, we found ways of examining the issue within our own data. Results SHORT RANGE DEMOGRAPHIC EFFECTS Crude case densities per 1000 PCs were meas- ured in different parts of the counuy: 56 rectangular areas covering major cities and inner cities, and residual surroundmg areas. They were greater in population-densethan in sparse areas. The mean overall density was 16.36 deaths per 1000residential PCs, but individual values descended from Covenuy (29.98), Scunthorpe (27.78), Doncaster (24.30), Glasgow (24.01), Liverpool (23.27), Hull (23.15), Birmingham (23.07), central London (22.41) ... to ... Southampton (14.32), Gloucester (13.49), Brighton (12.06), Exeter (11.16), and Bournemouth (9.34); and the residual nonurban zones of northern England, Wales, and southwest England (10.28, 9.60, 7.79). Although correlations with high population density have been demonstrated el~ewhere,'~ these regional variations do not accurately follow known mortality patterns. They must stem in part from the different ways in which PC boundaries are delineated. PCs in populationdense indusmal zones probably have more delivery points (DPs) per PC, and possibly more children per DP. Such excesses would create a false overall association between industrial hazards and cancer deaths per 1000 PCs. This necessitated the development of the stratification and standardisation procedures outlined above. Unfortunately, broad area standardisations provide no protection against the falsifylng effects of high DP PCs in the immediate vicinities of industrial premises. There is no available measure of industrially related demographic distributions on this scale. However, we tested the proposition indirectly by calculating densities of PCs per km2around each of the hazard types examined. The results are sum- 134 . 1.30 1.20 - 0 Mail order x Furniture A Beer .gu) 1.10 - E .c>g 1.00 - 0 -0 U P) 2 0.90 - E m -0 C 2 0.80- v) 0.70 - '0.60 I' I /I I I II II -0.3 -0.6 -1 -2 -3 -4 -5 -7 -10 Radial distance bands (km) Figure 1 Indusmallwsidential competition. r1.70 t1.60 + *Oil refineries Oil storage sites Bitumen products 1 Oil terminals x Benzene refining A Chemical works 1.50 - A '' 0.70 -0.3 -0.6 I II I -1 -2 -3 -4 -5 Radial distance bands (kml II -7 -10 idential PCs nearby. Oil refineries, nuclear power stations, and airfields, for example, vir- tually exclude human habitation Within a km or more of their centres. Other hazards with evidence of local industrial-residential competition included chemical works, oil distribution terminals, crematoria, and harbours. However, not all hazard types exhibited these residential exclusions, notably the less se- gregated installations labelled as "safe sites". Factories closely integrated with residential areas might show very local competition for space within individual PCs. This could result in artificially low densities of cancers per 1000 PCs. This was confirmed directly. Figure 1 shows SDRs surrounding the premises of 36 paper manufacturers, 21 soap manufacturers, 35 cotton spinners and weavers, 34 brewers, 31 mail order firms, and 38 furniture makers - the 195 locations aggregated as the "safe sites" of table 1. In these instances we know that the low short range case densitiesper 1000 PCs did not stem from low PC densities per km'. Together with their later regression towards the mean regional values, these findings most probably indicate short range domesticindustrial competition for space within PCs adjacent to or containing a factory. The general regional relationship between low PC densities per km2and low case densities per 1000 PCs, the latter probably mediated through an excess of low DP PCs, also accords with these short range findings; but there was no way in which this could be specifically confirmed within the available data. Formal exclusion of a demographic artefact capable of mimicking source related case concentrations . was not possible. However, this artefact remains entirely hypothetical, without the support of prior evidence; and despite these searches of our own data, no positive evidence in its favour was found. Indeed, the balance of evidence indicates an artefact capable only of reducing factory-adjacent cancer density estimates. Figure 2 Oil refining and storage. marised in table 1. PC densities are compared around: 0 A random sample of PCs; 0 A random sample of cancer deaths; and 0 Potential hazards of several different types. Local PC densities around random PCs and around cancer cases were much greater than the overall national value ( 5.97 PCs/km2)based upon total land area. This is because the sample selection process is biased in favour of densely inhabited regions. At increasing distances, the high densities tapered towards the national value as this selection effect became diluted. The much lower values around some of the hazard sites, sometimes increasing at greater distances, probably indicate a dissociation between homes and factories. Many large installations are situated in rural areas, or at coastal or estuarine sites, or grouped together in industrial estates so that there are few res- PETROLEUM BASED INDUSTRIES Figure 2 shows the pattern of SDRs in separate bands around 15 major oil refineries, 17 major oil storage areas, and 168 lesser oil distribution terminals and storage sites - mostly rail to road or rail to factory transfer sidings or factory depots. Short range SDRs were irregular through small numbers; but for the refineries and major storage sites there was a clear SDR excess between 1 and 5 km (see table 2 for the aggregated results). Figure 2 also illustrates results for 45 manufacturers of industrial chemicals, excluding the 10 classified as petrochemical factories. Petrochemical factories showed a non-significant excess and the others showed none at all; nor did the six major UK benzene refineries, plants which at this time produced about a million tonnes per year. Critics of our previous paper thought that our results supported a link between benzene and leukaemia,67but we found no direct evidence on this, then or now. Hazard proximities of childhood cancers 155 Ta6le 2 Proximities of death addresses to point hazards Hazard type of Distance No of No SDR* 1.` sites e m ) cases expected P Oil refinenes Oil farms Oil distribution terminals BiNmen products Detergent makers Plastics formers Car makers Coach builders Solvent makers Spray painting Galvanizers Paint makers Rubber manufacturers Varnish makers Fibreglass fabrication Adhesives makers Powder coating Steel works Aluminium casting Zinc casting Car batteries Power stations Cement works Crematoria Crematoria Brickworks Airfields Rail yards Harbours 15 17 168 22 33 28 31 32 32 33 69 28 97 13 37 29 26 44 24 43 14 69 12 64 64 27 192 106 82 1.0-5.0 1.0-5.0 1.0-5.0 0.0-5.0 0.0-3.0 0.0-2.0 0.0-5.0 0.0-5.0 0.0-3.0 0.0-5.0 0.0-5.0 0.0-5.0 0.0-5.0 0.0-2.0 0.0-3.0 0.0-3.0 0.0-5.0 0.0-5.0 0.0-2.0 0.0-3.0 0.0-3.0 0.0-3.0 0.0-3.0 0.0-1.0 0.0-3.0 0.0-2.0 0.0-3.0 0.0-1.0 0.0-2.0 329 735 31765642 1200 2203813 2025 2094291 6580 35569759 337 1739381 784 1385249 1523 1405399 209 2136072 142 847069 688 259.9 621.5 3411.0 1512.8 1039.3 1274899..22 1840.1 1788456..72 5922.1 35426773..72 286.3 1628081..12 682.4 1371309.s1 1389.9 836831..04 158.8 1281340..96 117.6 734632..65 585.6 1.27 1.18 1.10 1.23 l1..1144 1.10 l1..11 17 1.11 1l ..0066 1.18 1l ..1l 6l 1.15 1l ..0154 1.10 11..1286 1.32 11..1481 1.21 1.13 1.17 18.34 20.74 3145..1018 24.85 342s.970 18.48 2223..1468 73.09 1129..6858 8.97 1174..5073 15.11 46..2601 12.74 2285..1387 15.86 3651..6700 5.08 235..9577 17.91 CO.001 co.001 <O.Ool CO.001 co.001 <C0O.0.0501 co.001 <<O0..0O0o1l co.001 <COO.O.0101 c0.01 C<OO..O00o1l co.001 <cO0..0051 co.001 <COO..O00o1l co.001 <<O0..0O0o1l c0.02 <CO0..O02ol <o.ooi * SDR (standardised density ratio) in this table accumulates individual bands shown in figures across the ranges given in column 3. Most sources gave significant results at several boundaries - exemplified here by the crematoria. For most hazards the mble displays SDRs at 2.0, 3.0, or 5.0 km, usually selecting the maximum O E (observedexpccted) ratio. Significance levels were sometimes greater at other boundaries. showed pronounced excesses at ranges beyond the very shortest. Paint manufacture (28 sites) - -had a less powerful effect. These three, together with car manufacture and coach building, com- prise the 193 sites whose adjacent PC densities are described in table 1. Brake manufacture (13 sites) and tyre making (23 sites) showed no effects. Other forms of rubber manufacture showed a small relative excess within 5 km (1.06) but none at shorter ranges (0-3km). Among other solvent based processes ex- amined, there were excesses in the vicinities of varnish makers, fibreglassfabricators, adhesives makers, and factoriesundertaking powder coat- ing of metals. Despite the use of solvent based metal cleaning, 71 electroplating factories ex- hibited no apparent effect. Factories (22 sites) making halogenated hydrocarbons - chlor- inated and fluorinated - likewise had no ap- parent effect, but 32 other solvent manu- facturers showed highly significant cumulative excesses (p<O.O01) up to 5 k m . After some short were range irreg-ularities, foundfor 22 significant effects Of bitumen based products, 33 detergent factories, and 28 plastics factories. .Car making +Paint making *Spray painting OCoach building xSolvent making AGalvanizing 0.70 IIIIIII -0.3 -0.6 -1 -2 -3 -4 -5 -7 -10 Radial distance bands (km) Figure 3 Automobile manufacture. AUTOMOBILE INDUSTRY It has been claimed l5 that a greater quantity of volatile organic compounds is discharged when painting a car than is emitted from its exhaust during the whole of its road career. The 31 car factories examined (fig 3) showed a clear excess of leukaemias and cancers, which was greatest at 0.3-0.6km, but extended up to 5km. Coach building and body repair firms (32 sites) had a less powerful but significant effect over the same range. Non-car factories pursuing related trades, and using similar organic materials, were also examined. Among them were 69 galvanizing plants - which use solvents for metal cleaning - and 33 spray painting contractors. They METAL. CASTING AND REFINING The associations with car manufacture and with galvanizingsuggest alternativenon-solvent exposures: namelyto metal casting, metal forming and welding. Examination revealed (fig 4 and table 2) short range risks associated with aluminium, zinc, and iron and steel casting. Steelworks showed a moderate effect. Firms offering contract welding construction showed nothing, but the addresses are probably remote from their on-site work locations. Casting and refining of lead (22 sites) showed no excesses, although car battery factories exhibited a strong effect. Lead casting is a relatively low temperature process and the battery making effect may be related to the manufacture of battery casings, plastics forming or solvent usage, rather than lead processing. "Hot process" associations were not limited to metal forming. After some short range irregularities, powerful medium range excesses were evidentfor power stations, crematoria and cement works. Cement rail terminals had no effect. Brickworks showed a local effect but potteries (48 sites) and gasworks (98 sites) did not. The contrast between brick and pottery manufacture may be an effect of the relative scales of production, and the absence of an apparent hazard from gasworks may be explained in terms of local population exclusions or perhaps in terms of their history. In 1965/6 20 million tonnes of coal and oil were used in the production of gas, but by 197011 this was down to 8 million; and by 1976 to 0.2 million, as gas manufacture finally vanished.I6 Even before 1970, and still in the era of gas manu- facture, many local gasworks were closed or reduced to simple storage sites following the introduction of high pressure coal and oil gasification processes. These new processes, and the import of liquefied natural gas, led to the ? r1- 1.70 1.60 -1.50 .-ln 0 c 1.40 - E .-> cln 1.30 - C ;P) 1.20 - .-0ln1 E 1.10 - m D Cm 1.00 - 6 0.90 - \\ Brickworks +Zinc casting * Steel works 0 Power stations x Cement works A Crematoria ' ' `0.80 - ' 0.70 -0.3 -0.6 III I -1 -2 -3 -4 -5 Radial distance bands (km) II -7 -10 Figure 4 Kilns and furnaces. 1.50 .-ln 0 CI 1.40 E `cjj 1.30 C A .Railways +Rail yards *Motorways x Harbours A Airfields Radways The railway network was examined in the same way. The pooled observed and expected n u bers across all 11 rail regions showed a strong and significant excess within 2 (or 3 or 4) km of a rail line and a significant deficit beyond 4km. The distance distributions differed in differentregions but there were significant SDR excesses in at least one sub-4 km band in 9 of the 11 regions. The main exception was in industrial Lancashire and north Cheshire, which showed no difference. Here, a very high proportion of control PCs were within 2 km of a railway and thus afforded no effective basis for comparing near and far distances. Railway goods yards showed a significant short range SDR of 1.13. Harbours, most of them serviced by railway lines, and heavily laden with other industrial sources, showed an SDR of 1.17 within 2.0 km. Distance related SDRs within 4 km of railways and motorways were readjusted using pooled data beyond 4.0km as the reference level, and these adjusted relative risks are shown (on an extended scale) in figure 5. As with many industrial sites, as well as with harbours and airfields (also in fig 5), the railway risk was lower at 0-0.3 km than at rather greater ranges, probably from residential-industrial competition for space in railway adjacent PCs. For the motorways this was less pronounced. The rail and motorway analyses were repeated for the 9411 leukaemiaregistrations, with the same results. 1.20 5 1.10 m 0 g 1.00 t-tj 0.90 -0.3 -0.6 Linearfeatures 7 > IIIII -1 -2 -3 -4 -5 Radial distance bands (km) 10 I -7 20 I -10 OTHER INDUSTRIES A wide range of industrial locations gave negative results. In addition to those already mentioned they included agricultural fertilizer rail terminals, MOD rail sidings, T V transmitters, cake and biscuit bakers, dry battery makers, magnetic tape makers, nuclear plants, PVC compound manufacturers, and makers of wood preservatives. Figure 5 Transport facilities. construction of high pressure long distance pipelines and to centralised production. TRANSPORT ROUTES Motorways Distances were measured between all leukaemidcancer bearing PCs and the nearest motorway (outside and excluding the M25) and a frequency distribution compiled. This was repeated for 10000 randomly selected PCs and the result used to calculate regionally standardised expected values. SDRs were calculated for an extended set of parallel distance bands up to and beyond 20 km.There was a significant excess of cases within 4.0 km of a motorway, and a significant deficit beyond 4.0km. The calculation was reDeated without the use of local standardisation, with the same result. AIRBORNE DRIFT Easting and northing biases of cases, relative to expected cases, were measured in all analyses. Nothing was found. For some, the calculations were repeated with specific drifts added to the coordinates of the hazards, to represent the effects of the prevailing wind. The results were compared with those of zero and counter prevailing wind drifts. No evidence of any drift was found and centrifugal risk gradients generally diminished with increasing imposed biases, in any direction. In the case of motorways, for example, a "dummy" linear hazard displaced 8km N and 8km E from the actual network anulled the cancer proximity pattern. BIRTH ADDRESS OR DEATH ADDRESS? This question was tested first against a composite list of those "hot process" sites whose individual effects had already been demonstrated. It included steelworks (44 sites), power stations (69 sites), cement works (12 Hazard pmrimities of childhood cancers 157 sites), crematoria (64 sites), aluminium Casting densities to childhood leukaemia risks, but works (24 sites), and zinc casting works (43 within much broader zones. Neither the relative sites). Proximities were measured against the distance methods developed for investigating following: particular nuclear site^'^-^^ nor the assembly of 0 The full list of death addresses, 0 The subset known not to have changed their PCs between birth and death, 0 The birth addresses of those who moved, and e The death addresses of those who moved. The full death address list showed an observedexpected ( O E ) ratio of 1.15 within 3 km. Non-movers showed a ratio of 1.19. The enumerated civil populations into, for example, 10km squares or 25 km radius circles,'7z2met the technical needs of the present search. The continuing demographic difficulties of a PC based method are clear, and a totally unambiguous demonstration of proximity clustering has stillnot been possible, but the present investigation stands on less error prone foundations than did the previous one.6 First, our recent analyses have effectively dispelled cav- birth addresses of movers also showed 1.19 eats about the reality of short range case clus- and the death addresses of movers a ratio of 1.15. All these excesses were highly significant (p<O.OOl), suggesting that while early ex- posures were more important than later ones, they could also be effective at later ages. This corresponds with our earlierfindings in relation to PC sharing pairs.2 However, detailed examination revealed that many of those who had changed their PCs had moved only a short distance. The mean distance was 35.6 km, but the distribution was highly skewed and 49.9 % of this group had moved less than 5.0 km and 65.1 % less than 1 O . O k m . The examination was therefore repeated tering and the existence of geographically localised hazards is not now in doubt. Proximity studies are no longer concerned with this issue and can be directed solely at asking what those hazards might be. Second, some of the hazard associationsmatch those previouslypostulated. The prior hypothesis of a petroleum associated leukaemia risk seems to be confirmed in a very large and independent set of disease data, and against a wider range of environmental sources. Third, the new analysis bases its estimates of geographical population dismbutions upon much larger numbers of PCs. The main remaining source of ambiguity is among 3213 children who had moved more a hypothetical systematic excess of children at than lOkm, and where both addresses were risk in PCs close to suspect sources. A local accurately coded. Within 3.0 km of the hazard excess of high DP PCs or of high child DPs closest to either address, the birth places could create a false apparent excess of cancers. showed an O E a ratio of 1.09 (572E26.9). Broad areal density standardisation (as used) The ratio for the death addresses around the provides no protection against hidden, small same sources was only 0.80 (420L526.9).Each scale demographic structuring of this kind. Nor was significant, as was the difference between could the use of available population data, the numerators (set on a common de- such as EDs or electoral wards, be expected to nominator). Their contrary directions reflect resolve this problem. However, our own data the imposed separation, through selection, of the two addresses; if one address is close to a particular hazard then the other is necessarily some distance away. Similar results were obtained for other hazard types and for different revealed low rather than high areal PC densities adjacent to many putative hazards. There were clear signs of competitive exclusion of residential PCs from the vicinities of very large plants: and of houses from those residential migration distances. Among children who had PCs which were very closeto factories.Notably, moved more than 5.0km and where at least low case densities were found within very short one address was within 10 km of a galvanising plant, we found 186 births but only 83 deaths within 3 km of the nearest source. We conclude that early exposure is the more important. Cancer initiations in migrating chil- distances of non-volatile organic compound, non-combustion sites. Local demographicvariations of these kinds could readily mask a genuine short range hazard related risk or create false appearances of a low risk; but they could dren may indeed be limited to exposures occurring within a short period before or after birth. The hazard proximities of the death addresses, as well as of birth addresses, in the not easily create a false excess. The major risk mimicking potential artefact therefore remains entirelyhypothetical.There was never any prior evidence to show that it existed and our own total data set could reflect only their strong indirect evidence now suggests that it does not. geographical associations. Further reassurance against false positive demographic artefacts appeared in analyses of cases who had migrated between birth and Discussion death. The highly asymmetric ratios of 572/ The logistical and methodological difficulties 420 birth/death addresses within 3km of a of conducting intimate proximity analyses on "hot process" source among children who had so large a scale are reflected in the paucity of migrated lOkm or more and a 186/83 birth/ comparable reported studies. Apart from sev- death ratio within 3 km of galvanizing plants eral proximity studies of childhood cancers and among children who had migrated 5.0km or leukaemias around nuclear in~tallations,"-~~more are both independent of these demo- we found little which could be related to the graphic contingencies. These results also sug- findings described here. Several large scale gest a powerful generalmethod for investigating geographical-ecological investigations have re- age sensitive exposures where there are no lated socioeconomic factors and population demographic data available at all. We shall describe its properties and develop its usage in a later study. Effective apparent exposures were concentrated around two main kinds of installation namely: (a) producers, refiners, distributors, and industrial users of petroleum fuels and volatile petroleum products, and (b) manufacturing processes using high temperature furnaces, kilns, and combustion chambers. Some sources, notably internal combustion engines and oil fired furnaces, met both criteria. These specific associations and their absence around other sources were too coherent to suggest a demographic artefact. The few inconsistencies were probably due to different scales of production or restricted periods of operation. There was littleto suggest systematic confounding from adjacent sitings of different processes. With few exceptions (eg pottery manufacture) the main factory based hazards were distributed independently across many regions. The rail and motorway associations were relatively ambiguous and could more readily be attributed to adjacent industrial hazards, but even here the most obvious direct common factor would be the use of petroleum fuel, especially diesel. Diesel is relatively nonvolatile and it is difficult to envisage a spillage based or access based mechanism affecting both sources, so if fuel combustion is the common factor the most eligible materials must be blow-by fumes from crankcases and substances emitted from exhausts. However, combustion fumes could not ac- count for all the findings and there was no indication of any other class of effluent capable of explaining the full range of effects. None of the carcinogenic industrial species listed in the IARC monographs would easily serve as a common determinant.23The specific role of benzene, much discussedin the literature and in the media, was not supported by an examination of the major benzene manufacturing sites. Its reputed leukaemia toxicity may have led to special containment measures, but it would be surprising if none had escaped. Rather than one specific effector, we must therefore consider the possibility that there might be several - from among the aromatics, other volatile organic substances, arsenic, metal fumes, soot and smoke compounds, radioactive elements, carbon monoxide, oxides of nitrogen and sulphur, and others. Alternatively, chemical species from different hazard types are known to combine in the atmosphere to form secondary pollutant^.^^ The suspect primary substances are oxides of nitrogen and volatile organic substances, discharged singly or in combination. The secondarypollutants include ozone, aldehydes, ketones, nitrogenated organics, nitrates, hydrogen peroxide, ionised oxygen, and many others. Alternative modes of access to such substances include the following: 0 Diffusing gases and volatiles reaching the children or their pregnant mothers directly, 0 Mediation through parental/occupational germ-cell injury, 0 OccupationaVdomestic Contamination via clothing or through other means. The last is feasible only for a limited range of sources (eg car factories or paint spraying contractors) and it is unlikely that sufficient numbers of parents were employed in oil storage installations and crematoria to explain the excess cancers in their vicinities. This is also against an exclusively parental occupational germ-cell or fetal exposure mechanism. For a comprehensive single explanation we must favour direct exposure of pregnant women or young children to airborne substances diffusing into the surrounding environment. The apparent absence of tumour type specificities for different sources possibly reflects an intimate mix of several potent substances with different tissue specificities or a single common effector capable of reaching many different tissues and invoking a polyvalent response, as in the case of prenatal medical irradiation.' Al- ternatively, it might reflect only the limited geographical resolution of our analyses. The O B ratios are all modest and hazard adjacent type specific excesses could easily be buried among cases unrelated to the source in question. Finally, we have to ask whether the moderatelyincreaseddensitiesof cases around these hazards also explain the previously reported high local concentrations represented by same PC and same ED pairs." The first reached out as far as 3 km and sometimes 5 km, while the close pairs were detected within diameters of about 300m. They do not necessarily stem from a common mechanism, although two considerations suggest that they might. First, identified sources must vary in the toxicities or containment of their emissions, and case densities surrounding the more dangerous could be nearer to those within the small clusters. Second, populations within high exposure zones are themselves so patchy that any resulting cancers are necessarily concentrated within small sub-areas. We can not at present resolve these issues, and patterns of exposure could be far more intricate than our present results have indicated. Combustion based and volatile organic substances must be emitted by many less obvious sources at lower elevations, diffusing over shorter ranges, and leading to more concentrated effects than those surrounding major installations. Candidates include domestic and commercial heating systems, oil storage bunkers, oil delivery spillages, garage blocks, small workshops, bus stations, school or hospital chimneys, municipal incinerators, petrol stations, and many others. There are also many close clustered exposures to non-effluent hazards, each capable of reaching cellular targets and damaging the genetic material, or immunological processes, or the mechanism of cell division. They include domestic radon and gamma radiation, medical radiation, ambient electromagnetic fields, infections of mother or child and prenatal drug exposures. All of them have been implicated although none offers a comprehensive and exclusive explanation.9 22 25-30 To solve these questions we need Hazard proximities of childhood cancers 159 an extended and better differentiated record of exposures, and more discriminating modes of examination. Appendix CALCULATION OF STANDARDISED DENSITY RATIOS (SDR) Consider a set of h hazards of similar type, H. The area surrounding each hazard consistsof concenmc annuli, A,, containing 0, observed cases and Phr PCs. Each hazard is located in one of 56 rectangular zones Z,, in which k cases and n PCs give a zonal case density D,=kJnz and the expected number of cases in each of the A,,, annuli is Summing observed and expected values for all hazards of a given type gives the SDR for each successive annulus around a hazard of type H as The functions of regional standardisation are to calculate local expected numbers, weighted according to corresponding regional case densities; and to calculate ratios (SDR) between the aggregated observed and expected numbers. For restricted subsets of cases, such as specific tumom, or children who moved house, the regional case densities are first reduced in proportion with the size of the subset. We thank Dr A M Stewart for access to the data of the Oxford Survey of Childhood Cancers. The postcoding of the OSCC records was part of a research programme supported by the Medical Research Council and by the Three Mile Island Public Health Fund (USA). EGK's expenses during the current work were defrayed through a Leverhulme Emeritus Research Fellowship. 1 Knox EG, Gilman EA. Leukaemia dusters in Great Brit- ain.2. Geographical concentrations. 3 Epidemiol Com- muniy Healrh. 1992;46:573-76. 2 Knox EG, Gilman EA. Spatial clustering of childhood cancers in Great Britain. J Epidemwl Communify Health. 1996;50313-19. 3 Gilman EA, Knox EG. Temporal-spatial distribution of childhood leukaemias and non-Hodgkin lymphomas in Great Britain. In: Draper GJ ed. The geographical epidemw1o.w of childhood leukaemia and non-Ho&kin lymphomas in Great Brirain 196683. Studies on medical and population subjects No53. London: HMSO, 1991. 4 Knox EG, Gilman EA. Leukaemia clusters in Great Britain. 1. Space-timeinteractions.3Epidemiol Community Health. 1992;46:566-72. 5 Gilman EA, b o x EG. Childhood cancers: space-time dis- tribution in Britain. 3Epidemiol Communiry Health. 1995; 49158-63. 6 Knox EG. Leukaemia clusters in childhood: geographical analysis in Britain. J Epidemiol Communiry Health. 1994; 48:369-76. 7 Bithell JF, Draper GJ. Apparent association between ben- zene and childhood leukaemia:doubts concerninga report by Knox.J Epidemwl CommuniryHealrh. 1995;49:437-39. 8 Stewart AM, Webb J, Hewin D. A survey of childhood malignancies. BMJ 1958;1:1495-508. 9 Knox EG, Stewart AM, Kneale GW, Gilman EA. Prenatal irradiation and childhood cancer. 3 Radio1 Rot 198737: 177-89. 10 Stiller CA. Population based survival rates for childhood cancer in Britain. BMJ 1995;309:1612-26. 11 Kompass: CBI-UK. London: Reed Information Services, 1993. 12 Kelb'r.London: Reed Information Services, 1994. 13 Alexander FE. Investigations of localised spatial clustering, and extra-Poisson variation. In: Draper GJ ed. The geographical epidemiologyof childhood leukaemia and wn-Hodgkin lymphomas in &at Britain 1966-83. Studieson medical and population subjects No53. London: HMSO, 1991. 14 Muirhead C R Childhood leukaemiain metropolitanregions in the United States: a possible relation to population density? Cancer Causes and Conml. 1995;6383-8. 15 Gibbs WW. An ounce of prevention. Scienrific American. November 1994;271:84. -16 Williams TI. A historv of the British PBS industrv. Oxford: Oxford University Ress, 1981. 17 Cook-MozaEari PJ, Ashwood FL, Vicent T,Forman D, Alderson M. Cancer incihncc in the vicini@ of nudear i n r d a t i a c , England and W& 1959-80. Studies on medical and population subjects No51. HMSO. London. 1987. 18 C o o k - M o d & PJ, Darby SC, Doll R, et 01. Geographical variation in mortality from leukaemia and other cancers in England and Wales in relation to proximity to nuclear installations, 1967-78. Br 3 Cancer 1989;59476-85. 19 Bithell JF, Stone RA. O n statistical methods for analysing the geographical distribution of cancer cases near nuclear installations. 3Epidnnwl Communiy Health. 1989;43:79- 85. 20 Bithell JF,Dutton SJ.Draper GJ, Neary NM. Distribution of childhood leukaemias and non-Hodgkin Lymphomas near nuclear installations in England and Wales. BMJ 1994;309:50 1-5. 21 Bithell JF. Statistical methods for analysing point-source exposures. In:Elliott P, CuzickJ, English D, Stem R eds. Geographical and environmental epidemiology: methods for small-area studies. oxford Oxford University Press, 1992; 221-30. 22 Knox EG, Stewart AM, Gilman EA, Kneale GW. Back- ground radiation and childhood cancers. J R&l R o r 1988;8:+18. 23 International Agency for Research on Cancer. WHO mono- graphs on the evaluation of cam'nogeniic risks w humans. In p a m d a r , Vol60. Some induspial chemicals. 1994. Lyon: IARC. 24 Kelly TJ, Mukund R, Spicer CW, Pollack AJ.,Con- centrations and transformations of hazardous an pollutants. E n v i m Sci Tcchnol 1994;28:37&87. 25 Greaves MF.Speculations on the cause of childhood acute lymphoblastic leukaemia. Leukaemia 1988;2 1 2 t 5 . 26 Kiden LJ. Evidence for an infective cause of chlldhood leukaemia: comparison of a Scottish new town with nuclear reprocessing sites in Britain. Lancer 1988;Z1323-27. 27 Kinlen LJ, Clarke K, Hudson C. Evidence from population mixing in British new towns 1946-85 of an infecnve basis for childhood leukaemia. Lancer 1990;336577-82. 28 Kinlen LJ. Epidemiological evidence for an infective basis in childhood leukaemia. B r J Cancer 1995;71:1-5. 29 Kinlen LJ, John SM. Wartime evacuation and mortality from childhood leukaemia in England and Wales in 1945-9. BMJ 1994;309:1197-202 30 Coleman M, Beral V. A review of epidemiological studies of the healtheffectsof livingnear or working with electricity generation equipment. Int J Epidemwl 1988;17:1-13. I