Society for Radiological Protection J. Radiol. Prot. 34 (2014) 873–889
Journal of Radiological Protection doi:10.1088/0952-4746/34/4/873
Childhood cancer and exposure to corona ions from power lines: an epidemiological test J Swanson1, K J Bunch2, T J Vincent2 and M F G Murphy2 1
National Grid, 1–3 Strand, London WC2N 5EH, UK Formerly Childhood Cancer Research Group, University of Oxford, New Richards Building, Old Road Campus, Headington, Oxford OX3 7LG, UK 2
E-mail: [email protected] Received 12 June 2014, revised 12 September 2014 Accepted for publication 1 October 2014 Published 30 October 2014 Abstract
We previously reported an association between childhood leukaemia in Britain and proximity of the child’s address at birth to high-voltage power lines that declines from the 1960s to the 2000s. We test here whether a ‘corona-ion hypothesis’ could explain these results. This hypothesis proposes that corona ions, atmospheric ions produced by power lines and blown away from them by the wind, increase the retention of airborne pollutants in the airways when breathed in and hence cause disease. We develop an improved model for calculating exposure to corona ions, using data on winds from meteorological stations and considering the whole length of power line within 600 m of each subject’s address. Corona-ion exposure is highly correlated with proximity to power lines, and hence the results parallel the elevations in leukaemia risk seen with distance analyses. But our model explains the observed pattern of leukaemia rates around power lines less well than straightforward distance measurements, and ecological considerations also argue against the hypothesis. This does not disprove the corona-ion hypothesis as the explanation for our previous results, but nor does it provide support for it, or, by extension, any other hypothesis dependent on wind direction. Keywords: epidemiology, power line, corona, childhood cancer (Some figures may appear in colour only in the online journal)
0952-4746/14/040873+17$33.00 © 2014 IOP Publishing Ltd Printed in the UK
873
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
1. Introduction This paper reports a test to investigate whether a previously reported excess of childhood leukaemia close to high-voltage overhead power lines in the UK fits the predictions of a ‘coronaion hypothesis’. In 1979, Wertheimer and Leeper [1] reported an association between overhead electrical wiring outside homes and the incidence of childhood cancer. They suggested this could be caused by the magnetic fields produced by the electrical wiring. Since then, a further thirty or so epidemiological studies of childhood leukaemia, and numerous more of other health outcomes, have been reported, investigating either proximity to power lines, or directly the magnetic field produced by such power lines and other sources. On the basis of fairly consistent associations between magnetic fields and childhood leukaemia in epidemiological studies, but generally negative laboratory evidence, power-frequency magnetic fields were classified in 2001 as ‘possibly carcinogenic’ by IARC [2]. One source of magnetic fields within the home is high-voltage power lines outside the home. Excesses of childhood leukaemia close to power lines have been observed and have been taken as part of the evidence for a causal role of magnetic fields; e.g.in the seminal 2000 pooled analysis [3], the risk from studies looking just at power lines was in fact marginally higher than that from studies measuring fields from all sources. However, Henshaw et al [4–7] have suggested additional mechanisms by which a high-voltage power line might affect health. In particular, they point out [5] that power lines produce atmospheric ions, called corona ions, which blow away from the power line and which might combine with existing airborne pollutants so as to increase any health effects of those pollutants. We previously reported [8] an epidemiological study of childhood cancer and proximity to power lines in England and Wales. This found an association with childhood leukaemia, but one that extended too far from the power line to be explicable in terms of magnetic fields. We subsequently reported [9] that this excess declined over the decades from the 1960s to the 2000s in Britain. In this paper, we report a new analysis of our previous data to investigate whether our findings could be explained by the originally proposed ‘corona ion’ mechanism [5]. Since our 2005 paper raised the issue of risks at larger distances from power lines than previously investigated, other studies investigating similar distances have either reported [10, 11] or are in progress [12], and so investigating possible mechanisms for such effects, of which this paper is one example, has wider relevance to understanding the possible risks from magnetic fields. If an alternative mechanism for risks from power lines were established, such that magnetic fields did not need to be invoked to explain such risks, this would weaken the overall evidence for risks from magnetic fields. 2. Background 2.1. The proposed mechanism
High-voltage overhead power lines produce high electric fields on the surface of the conductors. Power lines are usually designed so that this field is below the ionisation threshold of air. However, where the electric field is enhanced by surface irregularities, it can be high enough to cause ionisation. Such irregularities are produced by water droplets (the crackling noise characteristic of high-voltage power lines in wet weather is produced by this ionisation) and also by damage to the conductor (e.g. a broken strand protruding) and accretions to the conductor (e.g. chaff from harvesting fields below the power line). Breakdown over the surface of the insulators can also produce ions. 874
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
This process of ionisation of the air at the surface of the conductors is known as ‘corona’. Full-blown corona, sufficient to be audible, is confined to wet weather or, in dry weather, to a minority of lines with specific design characteristics or under specific conditions. However, virtually all high-voltage power lines do produce some corona ions even in dry weather. The great majority of corona ions recombine in the vicinity of the conductors. A minority escape and are blown away by the wind. These attach either to other airborne particles or, ultimately, to the ground. They can be measured either by charged-particle detectors or by their effect on the static electric field. Their concentration is usually greatest up to a few hundred metres from the line, but they have sometimes been measured several kilometres away. The proposed mechanism for health effects is that when these corona ions attach themselves to existing airborne pollutants (specific suggestions include radon daughter products and combustion products from traffic or smoking), they increase the charge on those pollutant particles and hence increase the probability of the particles being retained in the lungs or airways when breathed in, thereby increasing the ‘dose’ of those pollutants to the body and increasing any health effects those pollutants have. Therefore, individuals exposed to those ions—broadly, people living downwind of high-voltage power lines—would be at increased risk of any health effects produced by pollutants. One specific suggestion is that some pollutants are a causal factor for childhood leukaemia, and therefore those living downwind of power lines should be at higher risk of childhood leukaemia. Several aspects of this proposed mechanism have been questioned. In particular, there is debate as to whether the physical effects in question—the transfer of charge from corona ions to airborne pollutants, and the consequent increased retention of those pollutants in the airways—are large enough to have significant health effects [13–16], and whether airborne pollution is associated with childhood leukaemia [17, 18]. In this paper, we do not address the plausibility of the proposed mechanism; we simply describe the results of an epidemiological test of the hypothesis. 2.2. Previous epidemiological investigations of the proposed mechanism
Preece noted that in the south-west of England, prevailing winds are from the south-west. He therefore devised a test based on quadrants: he approximated that all homes in the north-east quadrant relative to the closest point of the line could be regarded as downwind of the line and therefore exposed to corona ions, and all homes in the south-west quadrant as upwind and therefore not exposed. This quadrant model has two main problems. First, it assumes that the wind blows uniformly from directions between 180° and 270° but never between 0° and 90°. Secondly, it considers only the closest point of the line to the address of interest, ignoring the rest of the line. Preece applied this test to data on adult cancers, and reported in the media (‘Costing the Earth’, BBC Radio 4, 21 September 2000) that all cancers combined, and specifically lung and mouth cancer, were more common in the ‘exposed’ quadrant than the ‘non-exposed’. The United Kingdom Childhood Cancer Study published an analysis of cancer risk with proximity to power lines and included an analysis of the presence of a 275/400 kV line within 400 m with the stated purpose of testing this hypothesis. They did not include upwind/downwind differences so their analysis is more similar to our previous analyses of distance than to the present paper; they found a non-statistically-significant elevation for acute lymphocytic leukaemia, relative risk (RR) 1.42, 95% confidence interval (CI) 0.85–2.37, consistent with our previous results. We are aware of no other epidemiological tests of this proposed mechanism, and none at all in relation to childhood cancer, other than our own previous preliminary test [8] discussed below. 875
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
0-199 m Bunch et al 2014 PM 2.5 PM10 atmospheric radionuclides
relative risk and 95% CI
5
0.8
4
0.6
3
0.4
2
0.2
1
0
0 1960
1970
1980
1990
2000
normalised concentration/dose
1
6
2010
by decade
Figure 1. Relative risks and 95% confidence intervals for residential distance at birth
0–199 m from power lines compared to >1 000 m from [9] (left axis) compared to UK concentrations of two different sizes of particulate matter, normalised to 1970 = 1 [21] and annual effective dose in the Northern hemisphere from radionuclides produced in atmospheric testing, normalised to 1963 = 1 [22] (right axis).
2.3. Previous findings from this study
In our most recent analysis [9], we expanded our original analysis [8], so that we now potentially consider 53 515 cases of childhood cancer plus matched controls in England, Wales and Scotland, born and diagnosed from 1962 to 2008, recorded in the National Registry of Childhood Tumours. We investigated the possibility of a relation between incidence of childhood cancer and the distance from the residence at birth to the nearest high-voltage overhead power line of 132, 275 or 400 kV. For CNS/brain tumours and for diagnoses other than leukaemia and CNS/brain, our results did not suggest any association. However, for leukaemia, we find an increase in risk that was highest (RR 4.50 for 0–200 m compared to >1 000 m) in the 1960s and has declined over the decades such that it no longer appears to be present (RR 0.71) in the most recent decade, the 2000s. For 275/400 kV lines, the elevated risk, when present, extended to 600 m; for 132 kV lines, perhaps only to 200 m. Magnetic fields at the level implicated by the previous epidemiological studies are found only within about 100 m of the lines [19], and it is almost impossible to see how the whole of this elevation in risk, at distances up to 600 m or even only 200 m, could be produced by magnetic fields. Other explanations are possible, including chance or unrepresentative controls; we discuss these further elsewhere [20]. However, an obvious candidate explanation is the corona-ion hypothesis. The predictions of that hypothesis match our findings in predicting elevated risks over a similar range (a few hundred metres), and the concentrations of at least some airborne pollutants have reduced over the period concerned in a way that matches our decline in risk. We show these temporal trends in figure 1 for two particular examples, concentrations of particulate matter, and radionuclides from atmospheric nuclear testing. These coincidences of prediction do not, of course, prove that the corona-ion hypothesis is the explanation of our results; there will have been other relevant factors, such as the origin and size distribution of airborne pollution, that have also changed; and as we discuss further below there has been no corresponding overall decline in leukaemia rates. Nonetheless, we consider these coincidences, taken with the public concern originating from the original hypothesis and 876
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
Table 1. Results from the simpler ‘quadrant’ model using data in [8].
Case/control ratios for subjects within 600 m
Leukaemia Central nervous system tumours Other tumours
Exposed (NE quadrant) 1.03 1.13 0.90
Non-exposed (SW quadrant) 1.28 1.37 1.18
Neutral (NW and SE quadrants) 1.44 0.93 0.90
the media reports of associations with adult cancers, striking enough to mandate the further investigation presented here. We reported a simple test of the corona-ion hypothesis in our original paper [8], using the ‘quadrant’ model described above, and concluded ‘we have no evidence to support this hypothesis’. The quantitative findings, which we did not present previously, are given here in table 1. Because of the limitations in the quadrant model, we do not regard this test as particularly influential. 3. Methods Here we use a much more detailed model to calculate exposure to corona ions for each subject in our study. It models the exposure as the product of four terms: a source-strength term, modelling the different propensity of different power lines to produce ions; a distance-frompower-line term, modelling how the concentration of corona ions varies with distance; a windrose term, modelling the amount of time the wind blows in each direction, and a wind-speed term. There are many uncertainties in the proposed mechanism and therefore uncertainties in how best to model its effects. We have made assumptions and simplifications where necessary, described below. We recognise that details of our model could be questioned, and we explore its robustness through sensitivity analyses reported below, but we believe the model is valid for distinguishing addresses with high exposure from addresses with low exposure, and it is certainly an improvement on the quadrant model. 3.1. Overview of method
Details of case and control selection and of derivation of grid references were given in our main results paper [9]. For each subject we know the grid reference of the address at birth. The present analysis is confined to subjects born in England and Wales, as we did not have wind data for Scotland. For each address, we identify power lines having any of their length within 600 m of that address and that were in existence in the year of interest. For consistency with our previous analysis, this analysis includes all 275/400 kV power lines plus 132 kV power lines in those areas of the country where we had data on year of construction for at least 80% of these lines. We create a series of point sources at 10 m intervals along such lines and, for each point, we multiply together the source-strength term for that power line, the distance term for the distance from that point to the subject address, the proportion of time the wind was blowing along the bearing from that point to the subject address, and the average wind speed along that bearing. The resulting exposure (in arbitrary units) for each point is summed over all the relevant points to give the total corona exposure at the address. 877
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
weighting (arbitrary units)
1
upwind
downwind 0
50
400
600
distance from power line / m
Figure 2. The distance weighting term for the variation of corona ions from a power line.
This model addresses several of the weaknesses of the quadrant model. First, it considers all the length of the power line that is within 600 m of the address rather than just the closest point. Secondly, it uses actual wind directions based on meteorological records rather than prevailing winds, and includes all directions a wind blows from over the year, appropriately weighted. We now describe the calculation of each of the four terms in turn. 3.2. Distance term
The ions produced by a power line can sometimes be detected kilometres away, but measurements by ourselves and others (e.g. [5]) lead us to model the observed average variation of the concentration of ions with distance from the power line, when the wind is transverse to the line, as shown in figure 2. We know that the distance variation is also affected by other factors, such as atmospheric conditions, but we are unable to model these. The distance term required for each individual point, in order for them to sum (at distances that are large when compared to the separation of the individual points) to this overall variation, is this overall variation multiplied by one-over-distance (we applied the one-over-distance multiplier beyond 50 m). The reduction in this term below 50 m reflects the fact that ions are produced on the conductors above ground and do not normally reach ground until they have travelled some distance. 3.3. Source-strength term
The production of corona ions depends on the electric field on the surface of the conductors. This in turn depends on the design of line, the voltage, and the number and size of conductors. Most power lines are built to one of a small number of standard designs, and we approximate all power lines as being in one of five categories as shown in table 2. For 275 kV and 400 kV lines, this is a good approximation, as alternative designs (particularly alternative conductors) proliferated only in the 1990s, towards the end of the period of this study. For 132 kV lines there is a greater variety. In particular, some 132 kV lines have twin conductor bundles and some single, which affects the electric field considerably, but we did not have access to these data, so chose to represent all 132 kV lines with an average of the values for twin and single bundles. We model the corona-ion production as proportional to the conductor surface electric field. We know this is not correct, in that corona is not proportional to the field but is produced only when a threshold is exceeded, and this usually happens in places where the field is increased 878
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
Table 2. Electric fields on the surface of conductors, used as ‘source strength’ weighting factors, for the five categories of power line.
Conductor Bundle Quad (4 conductors in square) Twin (2 conductors spaced horizontally) Mixed twin and single
29 mm diameter spaced 0.305 m 29 mm diameter spaced 0.305 m Average of twin (2 different designs) and single (3 designs)
Voltage (kV)
Electric field (kV m−1)
400 275 400 275 132
1 076 698 1 778 1 222 1 140
by surface perturbations, rather than by the unperturbed field which we calculate. However, we have no way of modelling this level of detail and we have no way of distinguishing, for example, lines in good repair from lines in bad repair. This method probably orders the different categories of line correctly even if the precise weightings are approximate; 400 kV lines with twin conductor bundles are indeed usually observed to be the greatest producers of corona, and 275 kV lines with quad bundles are rarely observed to produce any significant corona. The electric field varies round the circumference of a conductor; we used the maximum value on each bundle, averaged over all the bundles on that power line. 3.4. Wind-rose term
We obtained wind data from eight meteorological stations, in most cases from 1987 to 2002. We averaged over all available years and used the result for all subjects regardless of birth year. The locations of the meteorological stations are shown in figure 3. These had been selected by National Grid for other, operational, reasons and were not within our choice. For each subject we used the data from the nearest station. The distance from subject to meteorological station varied from 2 km to 170 km with a median of 37 km. The data give the number of hours for which the wind blows at each combination of angle (in ten degree sectors) and speed (grouped in 3 m s−1 intervals). We show the results for one station, Yeovilton, in figure 4. The wind-rose term for each point is the proportion of time the wind blew in the direction corresponding to the bearing of the address from that point. 3.5. Wind-speed term
Using the same wind data as for the wind-rose term, we calculated the average wind speed in the direction from each point to the address. 3.6. Alternative calculations of exposure
Each of the terms in our model involve assumptions. We have made our best assumptions, but even with recent advances [23, 24], knowledge of the behaviour of corona ions is not yet sufficient for us to be confident in these. In addition to the analysis based on the primary corona model, calculated by the product of the four terms described in sections 3.2–3.5 above, we therefore examined the sensitivity of our results to the assumptions by constructing four alternative secondary models where, in turn, each of the terms is omitted (in the case of the distance term, it is not completely omitted but replaced by a simpler term comprising just the one-over-distance variation and a truncation at 600 m, that is, omitting the corona-ion-specific distance variation of figure 2). 879
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
Figure 3. Locations of meteorological stations providing wind data.
3.7. Statistical analysis
We analyse the data by conditional logistic regression, using standard methods for an individually matched case-control study, as in our previous papers. For our primary calculation of exposure and for each of our four subsequent sensitivity analyses, we group the subjects into five categories: the reference category, consisting of all addresses that are more than 600 m from the nearest power line, and which have zero exposure in all models; and four categories defined by quartiles of the calculated exposure for addresses within 600 m of the line. For each exposure calculation, we calculated the quartile boundaries once, based on the totality of the subjects, then applied these boundaries consistently for all analyses of that variable. (In [9], we used a different reference category, >1 000 m, because we were testing for risks from 600 m to 1 000 m; having found that there were no elevated risks here, we now revert to the >600 m reference category from [8].) 4. Results This paper is based on the same series of case and control subjects as our previous study, except that three addresses were excluded because of missing data. There were 7 347 subjects with an address within 600 m of a relevant power line. For each address and for lengths of line within 600 m we carried out the calculations described above for points at 10 m intervals. We 880
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
2
0
1
270
90
0
180
Figure 4. Wind Rose for Yeovilton 1987–2002. Radial axis is the fraction of time the
wind blew from each direction, in 10° intervals, normalised to a uniform distribution of directions.
identified a total of 716 013 such points, with between 1 and 930 points per address, a mean of 97 each. Table 3 shows, by decade of diagnosis and in aggregate, the results for our primary exposure calculation. For leukaemia only, table 4 compares the results for the primary and the four secondary exposure calculations, again by decade of diagnosis and in aggregate. Table 5 presents selected relative risks, for the primary exposure calculation and for leukaemia only, for subjects whose closest line is 275/400 kV compared to 132 kV (in each case the exposure calculation for that subject still includes all nearby lines). 5. Discussion The model for calculating corona-ion exposure we have used here is undoubtedly an improvement on the previous quadrant model, but it still has considerable limitations. These include: • use of wind data from just eight meteorological stations, applied, in extreme cases, to subjects over a hundred km distant, with no allowance for the effect on the wind direction of local topography or proximity to coasts; • modelling of the source strength of each line only in terms of the voltage and conductor bundle, without taking into account the other factors which, whilst not yet identified or understood, must affect corona production; and • applying one representative and simplified distance variation to all points. Nonetheless, we regard our model, despite its limitations, as the best that is likely to be achievable on present understanding and data, and we consider, by taking account of key factors such as voltage and conductor bundle, and wind direction, it provides a valid discrimination between high exposures and low exposures. Table 4 shows that none of the risks from 881
Q4 Q3 Q2 Q1 Q0 Trend Q4 Q3 Q2 Q1 Q0 Trend Q4 Q3 Q2 Q1 Q0 Trend
1962–1969
882
1980–1989
1970–1979
Exposure
Period
52 43 53 54 3 084
44 52 32 35 3 100
14 13 6 9 972
Cases
39 47 34 43 3 129
30 42 41 35 3 115
2 10 4 9 989
Controls 1.84–106.5 0.51–2.95 0.42–5.32 0.38–2.66
0.98–1.16 0.87–2.11 0.58–1.41 1.11–2.85 0.83–1.93 0.99–1.17
14.00 1.22 1.50 1.00 1.00 1.32 1.41 1.29 0.76 1.10 1.00 1.07 1.35 0.90 1.78 1.26 1.08
1.08–1.61 0.86–2.30 0.83–2.00 0.46–1.24 0.68–1.82
95% CI
RR
Leukaemia
44 36 33 31 2 160
18 22 28 25 2 021
4 4 4 6 551
Cases
44 33 31 33 2 176
25 15 30 21 2 023
8 3 3 4 551
Controls 0.50 3.00 0.67 2.00 1.00 0.93 0.67 1.75 0.96 1.10 1.00 0.99 0.97 1.12 1.14 0.95 1.00 1.01
RR
0.93–1.11
0.88–1.11 0.61–1.54 0.67–1.89 0.64–2.02 0.54–1.67
0.73–1.19 0.35–1.25 0.86–3.56 0.56–1.65 0.60–2.02
0.15–1.66 0.31–28.84 0.11–3.99 0.50–8.00
95% CI
CNS/brain tumours
60 70 55 70 4 235
41 39 45 52 3 548
11 14 13 12 1 402
Cases
68 66 76 61 4235
46 33 43 50 3554
13 14 14 15 1396
Controls
0.75 0.92 1.09 0.79 1.00 0.96 0.82 1.16 1.03 1.18 1.00 1.00 0.88 1.15 0.68 1.12 1.00 0.97
RR
Other diagnoses
0.91–1.04
0.93–1.09 0.61–1.28 0.80–1.64 0.47–0.98 0.76–1.64
0.83–1.12 0.53–1.27 0.72–1.88 0.66–1.60 0.77–1.82
0.32–1.78 0.40–2.08 0.48–2.47 0.36–1.73
95% CI
Table 3. Relative risks and 95% confidence intervals for primary exposure calculation, three diagnostic groups, by decade of diagnosis and overall. In this and subsequent tables: ‘trend’ is the RR for each quartile increase in exposure. Q0 is the reference category (zero exposure) and Q1–Q4 the four successive quartiles of non-zero exposure. Bold indicates p 1 000 m) and corona-
ion analysis (trends across exposure categories). Because there are four quartiles of exposure, the trend RRs are plotted on an axis expanded by a factor of 4, to allow a very approximate visual comparison.
across the whole study area, they could constitute a potential confounding factor that we are not able to adjust for. Further, there are many sources of ions other than corona, such as combustion products, from transport, industry and gas cooking, waterfalls, etc, which will lead to misclassification; and there will be geographical variations of any airborne pollutants with which the corona ions interact. Table 3 does not suggest any pattern of risk with calculated exposure for brain tumours or for ‘other’ cancers. Further discussion here therefore focuses on leukaemia. Because the model assigns zero exposure outside 600 m and non-zero exposure inside 600 m, and because we already know that there are raised risks for leukaemia within 600 m in the earlier decades, we expect leukaemia risk to be associated with calculated exposure. The key question is: does the calculated exposure predict risk better than distance on its own? Table 6 compares selected estimates of risk from the two analyses. Figure 5 presents one of these comparisons graphically. Other comparisons between the two sets of results are possible and we present table 6 and figure 5 as illustrative rather than definitive. As expected, the corona-ion analysis does produce elevated risks in the earlier decades. But, with one exception, there is no clear trend with increasing exposure, and the risks are less than the equivalents for the distance analysis. The exception is the risk for the top quartile in the 1960s; we draw attention to this, but are inclined to regard it as a chance result. Further, of the four terms that make up the primary exposure calculation, the one that is most distinctive of the hypothesis is the wind-rose term; whatever uncertainty there may be about the individual terms, wind direction, even when assessed imperfectly as here, is clearly fundamental to the specific hypothesis being tested. From the sensitivity analysis presented in table 4, we see that removing the wind-rose term does not result in a consistent reduction in risk as would be expected. Thus, we conclude that the simple corona-ion hypothesis that we test here explains the observed pattern of leukaemia rates around power lines less well than straightforward distance. Our findings therefore do not support this corona-ion hypothesis as the explanation of these results. In view of the many uncertainties and approximations, however, this analysis is not a definitive rebuttal of this hypothesis, and, as discussed above, would not be relevant 887
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
to any development of the hypothesis which no longer predicted stronger effects downwind compared to upwind, or in which the effects were confounded by other factors. Further, our test relates only to childhood cancers, not to other diseases. Further arguments against the hypothesis come from the time trends illustrated in figure 1. If the reduction in risk around power lines were because the power lines still produce corona ions, but the numbers of airborne leukaemogens involved have reduced, we would expect that the reduction in these leukaemogens across the country as a whole, the majority of which is not close to power lines, would have led to a corresponding reduction in overall leukaemia rates. Recorded leukaemia rates have in fact increased. At least some of this is attributable to improved diagnosis [25], but there is still a discrepancy, which could be explained only if some other risk factor had increased in proportion to compensate, a concept that is not impossible but is inelegant. This paper has considered just the corona hypothesis and found little evidence in its favour. As discussed in our previous papers and the subsequent correspondence, there are many other possible explanations for the observed association between childhood leukaemia and proximity to high-voltage overhead power lines. Acknowledgments We thank Professors Denis Henshaw and Alan Preece for contributing to the original development of our corona-ion model, Alasdair Philips for drawing our attention to the fall in atmospheric radionuclides, and Messrs Henshaw and Philips for helpful comments on this paper; however, none of them are thereby indicating agreement with our methods or conclusions. This study was undertaken as part of a project funded initially by the United Kingdom Department of Health Radiation Protection Programme and subsequently by the charity Children with Cancer UK. The Childhood Cancer Research Group also received funding from the Department of Health and the Scottish Ministers. The Childhood Cancer Research Group closed down completely in April 2014. The views expressed here are those of the authors and not necessarily those of the Department of Health and the Scottish Ministers. JS is employed by National Grid who provided staff time but no other funding. A written contract exists between the Childhood Cancer Research Group and National Grid specifying that the Childhood Cancer Research Group has complete control over the conduct, interpretation, and publication of this study. This paper has not been approved by anyone in National Grid other than JS in his capacity as author and does not necessarily represent National Grid’s views. References [1] Wertheimer N and Leeper E 1979 Electrical wiring configurations and childhood cancer Am. J. Epidemiol. 109 273–84 [2] IARC 2002 Non-Ionizing Radiation Part 1: Static and Extremely Low-Frequency (ELF) Electric and Magnetic Fields (Geneva: World Health Organization) p 1 [3] Ahlbom A et al 2000 A pooled analysis of magnetic fields and childhood leukaemia Br. J. Cancer 83 692–8 [4] Henshaw D L, Ross A N, Fews A P and Preece A W 1996 Enhanced deposition of radon daughter nuclei in the vicinity of power frequency electromagnetics fields Int. J. Radiat. Biol. 69 25–38 [5] Fews A P, Henshaw D L, Wilding R J and Keitch P A 1999 Corona ions from powerlines and increased exposure to pollutant aerosols Int. J. Radiat. Biol. 75 1523–31 [6] Fews A P, Henshaw D L, Keitch P A, Close J J and Wilding R J 1999 Increased exposure to pollutant aerosols under high voltage powerlines Int. J. Radiat. Biol. 75 1505–21 888
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
[7] Henshaw D L, Ward J P and Matthews J C 2008 Can disturbances in the atmospheric electric field created by powerline corona ions disrupt melatonin production in the pineal gland? J. Pineal Res. 45 341–50 [8] Draper G, Vincent T, Kroll M E and Swanson J 2005 Childhood cancer in relation to distance from high voltage power lines in England and Wales: a case-control study BMJ 330 1290 [9] Bunch K J, Keegan T J, Swanson J, Vincent T J and Murphy M F 2014 Residential distance at birth from overhead high-voltage powerlines: childhood cancer risk in Britain 1962–2008 Br. J. Cancer 110 1402–8 (Electronic) [10] Sermage-Faure C et al 2013 Childhood leukaemia close to high-voltage power lines—the Geocap study Br. J. Cancer 108 1899–906 [11] Pedersen C et al 2014 Distance from residence to power line and risk of childhood leukemia: a population-based case-control study in Denmark Cancer Causes Control CCC 25 171–7 [12] Kheifets L et al 2013 Epidemiologic study of residential proximity to transmission lines and childhood cancer in California: description of design, epidemiologic methods and study population J. Expo. Sci. Environ. Epidemiol (doi:10.1038/jes.2013.48). [13] Swanson J and Jeffers D E 1999 Possible mechanisms by which electric fields from power lines might affect airborne particles harmful to health J. Radiol. Prot. 19 213–29 [14] Jeffers D 2001 A note on the charging of aerosols by overhead line corona Radiat. Prot. Dosim. 95 181–3 [15] Advisory Group on Non-Ionising Radiation 2004 Particle Deposition in the Vicinity of Power Lines and Possible Effects on Health (Chilton: National Radiological Protection Board) p 1 [16] Jeffers D 2007 Modelling and analyses do not support the hypothesis that charging by power-line corona increases lung deposition of airborne particles Radiat. Prot. Dosim. 123 257–61 [17] Raaschou-Nielsen O and Reynolds P 2006 Air pollution and childhood cancer: a review of the epidemiological literature Int. J. Cancer 118 2920–9 (Print) [18] Pyatt D and Hays S 2010 A review of the potential association between childhood leukemia and benzene Chem. Biol. Interact. 184 151–64 (Electronic) [19] Kroll M E, Swanson J, Vincent T J and Draper G J 2010 Childhood cancer and magnetic fields from high-voltage power lines in England and Wales: a case-control study Br. J. Cancer 103 1122–7 (Electronic) [20] Swanson J, Vincent T, Kroll M and Draper G 2006 Power-frequency electric and magnetic fields in the light of Draper et al Ann. N. Y. Acad. Sci. 1076 318–30 [21] Murrells T P et al 2010 UK Emissions of Air Pollutants 1970–2008 (Didcot: AEA Technology plc) Report No.: AEAT/ENV/R/3036 [22] UNSCEAR 2000 Sources and Effects of Ionizing Radiation (Vienna: United Nations Scientific Committee on the Effects of Atomic Radiation) [23] Matthews J C 2012 Diurnal variations of atmospheric potential gradient disruption near to high voltage power lines J. Atmos. Sol.-Terr. Phys. 77 235–40 [24] Matthews J C 2012 The effect of weather on corona ion emission from AC high voltage power lines Atmos. Res. 113 68–79 [25] Kroll M E, Carpenter L M, Murphy M F G and Stiller C A 2012 Effects of changes in diagnosis and registration on time trends in recorded childhood cancer incidence in Great Britain Br. J. Cancer 107 1159–62
889
Copyright of Journal of Radiological Protection is the property of IOP Publishing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Journal of Radiological Protection doi:10.1088/0952-4746/34/4/873
Childhood cancer and exposure to corona ions from power lines: an epidemiological test J Swanson1, K J Bunch2, T J Vincent2 and M F G Murphy2 1
National Grid, 1–3 Strand, London WC2N 5EH, UK Formerly Childhood Cancer Research Group, University of Oxford, New Richards Building, Old Road Campus, Headington, Oxford OX3 7LG, UK 2
E-mail: [email protected] Received 12 June 2014, revised 12 September 2014 Accepted for publication 1 October 2014 Published 30 October 2014 Abstract
We previously reported an association between childhood leukaemia in Britain and proximity of the child’s address at birth to high-voltage power lines that declines from the 1960s to the 2000s. We test here whether a ‘corona-ion hypothesis’ could explain these results. This hypothesis proposes that corona ions, atmospheric ions produced by power lines and blown away from them by the wind, increase the retention of airborne pollutants in the airways when breathed in and hence cause disease. We develop an improved model for calculating exposure to corona ions, using data on winds from meteorological stations and considering the whole length of power line within 600 m of each subject’s address. Corona-ion exposure is highly correlated with proximity to power lines, and hence the results parallel the elevations in leukaemia risk seen with distance analyses. But our model explains the observed pattern of leukaemia rates around power lines less well than straightforward distance measurements, and ecological considerations also argue against the hypothesis. This does not disprove the corona-ion hypothesis as the explanation for our previous results, but nor does it provide support for it, or, by extension, any other hypothesis dependent on wind direction. Keywords: epidemiology, power line, corona, childhood cancer (Some figures may appear in colour only in the online journal)
0952-4746/14/040873+17$33.00 © 2014 IOP Publishing Ltd Printed in the UK
873
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
1. Introduction This paper reports a test to investigate whether a previously reported excess of childhood leukaemia close to high-voltage overhead power lines in the UK fits the predictions of a ‘coronaion hypothesis’. In 1979, Wertheimer and Leeper [1] reported an association between overhead electrical wiring outside homes and the incidence of childhood cancer. They suggested this could be caused by the magnetic fields produced by the electrical wiring. Since then, a further thirty or so epidemiological studies of childhood leukaemia, and numerous more of other health outcomes, have been reported, investigating either proximity to power lines, or directly the magnetic field produced by such power lines and other sources. On the basis of fairly consistent associations between magnetic fields and childhood leukaemia in epidemiological studies, but generally negative laboratory evidence, power-frequency magnetic fields were classified in 2001 as ‘possibly carcinogenic’ by IARC [2]. One source of magnetic fields within the home is high-voltage power lines outside the home. Excesses of childhood leukaemia close to power lines have been observed and have been taken as part of the evidence for a causal role of magnetic fields; e.g.in the seminal 2000 pooled analysis [3], the risk from studies looking just at power lines was in fact marginally higher than that from studies measuring fields from all sources. However, Henshaw et al [4–7] have suggested additional mechanisms by which a high-voltage power line might affect health. In particular, they point out [5] that power lines produce atmospheric ions, called corona ions, which blow away from the power line and which might combine with existing airborne pollutants so as to increase any health effects of those pollutants. We previously reported [8] an epidemiological study of childhood cancer and proximity to power lines in England and Wales. This found an association with childhood leukaemia, but one that extended too far from the power line to be explicable in terms of magnetic fields. We subsequently reported [9] that this excess declined over the decades from the 1960s to the 2000s in Britain. In this paper, we report a new analysis of our previous data to investigate whether our findings could be explained by the originally proposed ‘corona ion’ mechanism [5]. Since our 2005 paper raised the issue of risks at larger distances from power lines than previously investigated, other studies investigating similar distances have either reported [10, 11] or are in progress [12], and so investigating possible mechanisms for such effects, of which this paper is one example, has wider relevance to understanding the possible risks from magnetic fields. If an alternative mechanism for risks from power lines were established, such that magnetic fields did not need to be invoked to explain such risks, this would weaken the overall evidence for risks from magnetic fields. 2. Background 2.1. The proposed mechanism
High-voltage overhead power lines produce high electric fields on the surface of the conductors. Power lines are usually designed so that this field is below the ionisation threshold of air. However, where the electric field is enhanced by surface irregularities, it can be high enough to cause ionisation. Such irregularities are produced by water droplets (the crackling noise characteristic of high-voltage power lines in wet weather is produced by this ionisation) and also by damage to the conductor (e.g. a broken strand protruding) and accretions to the conductor (e.g. chaff from harvesting fields below the power line). Breakdown over the surface of the insulators can also produce ions. 874
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
This process of ionisation of the air at the surface of the conductors is known as ‘corona’. Full-blown corona, sufficient to be audible, is confined to wet weather or, in dry weather, to a minority of lines with specific design characteristics or under specific conditions. However, virtually all high-voltage power lines do produce some corona ions even in dry weather. The great majority of corona ions recombine in the vicinity of the conductors. A minority escape and are blown away by the wind. These attach either to other airborne particles or, ultimately, to the ground. They can be measured either by charged-particle detectors or by their effect on the static electric field. Their concentration is usually greatest up to a few hundred metres from the line, but they have sometimes been measured several kilometres away. The proposed mechanism for health effects is that when these corona ions attach themselves to existing airborne pollutants (specific suggestions include radon daughter products and combustion products from traffic or smoking), they increase the charge on those pollutant particles and hence increase the probability of the particles being retained in the lungs or airways when breathed in, thereby increasing the ‘dose’ of those pollutants to the body and increasing any health effects those pollutants have. Therefore, individuals exposed to those ions—broadly, people living downwind of high-voltage power lines—would be at increased risk of any health effects produced by pollutants. One specific suggestion is that some pollutants are a causal factor for childhood leukaemia, and therefore those living downwind of power lines should be at higher risk of childhood leukaemia. Several aspects of this proposed mechanism have been questioned. In particular, there is debate as to whether the physical effects in question—the transfer of charge from corona ions to airborne pollutants, and the consequent increased retention of those pollutants in the airways—are large enough to have significant health effects [13–16], and whether airborne pollution is associated with childhood leukaemia [17, 18]. In this paper, we do not address the plausibility of the proposed mechanism; we simply describe the results of an epidemiological test of the hypothesis. 2.2. Previous epidemiological investigations of the proposed mechanism
Preece noted that in the south-west of England, prevailing winds are from the south-west. He therefore devised a test based on quadrants: he approximated that all homes in the north-east quadrant relative to the closest point of the line could be regarded as downwind of the line and therefore exposed to corona ions, and all homes in the south-west quadrant as upwind and therefore not exposed. This quadrant model has two main problems. First, it assumes that the wind blows uniformly from directions between 180° and 270° but never between 0° and 90°. Secondly, it considers only the closest point of the line to the address of interest, ignoring the rest of the line. Preece applied this test to data on adult cancers, and reported in the media (‘Costing the Earth’, BBC Radio 4, 21 September 2000) that all cancers combined, and specifically lung and mouth cancer, were more common in the ‘exposed’ quadrant than the ‘non-exposed’. The United Kingdom Childhood Cancer Study published an analysis of cancer risk with proximity to power lines and included an analysis of the presence of a 275/400 kV line within 400 m with the stated purpose of testing this hypothesis. They did not include upwind/downwind differences so their analysis is more similar to our previous analyses of distance than to the present paper; they found a non-statistically-significant elevation for acute lymphocytic leukaemia, relative risk (RR) 1.42, 95% confidence interval (CI) 0.85–2.37, consistent with our previous results. We are aware of no other epidemiological tests of this proposed mechanism, and none at all in relation to childhood cancer, other than our own previous preliminary test [8] discussed below. 875
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
0-199 m Bunch et al 2014 PM 2.5 PM10 atmospheric radionuclides
relative risk and 95% CI
5
0.8
4
0.6
3
0.4
2
0.2
1
0
0 1960
1970
1980
1990
2000
normalised concentration/dose
1
6
2010
by decade
Figure 1. Relative risks and 95% confidence intervals for residential distance at birth
0–199 m from power lines compared to >1 000 m from [9] (left axis) compared to UK concentrations of two different sizes of particulate matter, normalised to 1970 = 1 [21] and annual effective dose in the Northern hemisphere from radionuclides produced in atmospheric testing, normalised to 1963 = 1 [22] (right axis).
2.3. Previous findings from this study
In our most recent analysis [9], we expanded our original analysis [8], so that we now potentially consider 53 515 cases of childhood cancer plus matched controls in England, Wales and Scotland, born and diagnosed from 1962 to 2008, recorded in the National Registry of Childhood Tumours. We investigated the possibility of a relation between incidence of childhood cancer and the distance from the residence at birth to the nearest high-voltage overhead power line of 132, 275 or 400 kV. For CNS/brain tumours and for diagnoses other than leukaemia and CNS/brain, our results did not suggest any association. However, for leukaemia, we find an increase in risk that was highest (RR 4.50 for 0–200 m compared to >1 000 m) in the 1960s and has declined over the decades such that it no longer appears to be present (RR 0.71) in the most recent decade, the 2000s. For 275/400 kV lines, the elevated risk, when present, extended to 600 m; for 132 kV lines, perhaps only to 200 m. Magnetic fields at the level implicated by the previous epidemiological studies are found only within about 100 m of the lines [19], and it is almost impossible to see how the whole of this elevation in risk, at distances up to 600 m or even only 200 m, could be produced by magnetic fields. Other explanations are possible, including chance or unrepresentative controls; we discuss these further elsewhere [20]. However, an obvious candidate explanation is the corona-ion hypothesis. The predictions of that hypothesis match our findings in predicting elevated risks over a similar range (a few hundred metres), and the concentrations of at least some airborne pollutants have reduced over the period concerned in a way that matches our decline in risk. We show these temporal trends in figure 1 for two particular examples, concentrations of particulate matter, and radionuclides from atmospheric nuclear testing. These coincidences of prediction do not, of course, prove that the corona-ion hypothesis is the explanation of our results; there will have been other relevant factors, such as the origin and size distribution of airborne pollution, that have also changed; and as we discuss further below there has been no corresponding overall decline in leukaemia rates. Nonetheless, we consider these coincidences, taken with the public concern originating from the original hypothesis and 876
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
Table 1. Results from the simpler ‘quadrant’ model using data in [8].
Case/control ratios for subjects within 600 m
Leukaemia Central nervous system tumours Other tumours
Exposed (NE quadrant) 1.03 1.13 0.90
Non-exposed (SW quadrant) 1.28 1.37 1.18
Neutral (NW and SE quadrants) 1.44 0.93 0.90
the media reports of associations with adult cancers, striking enough to mandate the further investigation presented here. We reported a simple test of the corona-ion hypothesis in our original paper [8], using the ‘quadrant’ model described above, and concluded ‘we have no evidence to support this hypothesis’. The quantitative findings, which we did not present previously, are given here in table 1. Because of the limitations in the quadrant model, we do not regard this test as particularly influential. 3. Methods Here we use a much more detailed model to calculate exposure to corona ions for each subject in our study. It models the exposure as the product of four terms: a source-strength term, modelling the different propensity of different power lines to produce ions; a distance-frompower-line term, modelling how the concentration of corona ions varies with distance; a windrose term, modelling the amount of time the wind blows in each direction, and a wind-speed term. There are many uncertainties in the proposed mechanism and therefore uncertainties in how best to model its effects. We have made assumptions and simplifications where necessary, described below. We recognise that details of our model could be questioned, and we explore its robustness through sensitivity analyses reported below, but we believe the model is valid for distinguishing addresses with high exposure from addresses with low exposure, and it is certainly an improvement on the quadrant model. 3.1. Overview of method
Details of case and control selection and of derivation of grid references were given in our main results paper [9]. For each subject we know the grid reference of the address at birth. The present analysis is confined to subjects born in England and Wales, as we did not have wind data for Scotland. For each address, we identify power lines having any of their length within 600 m of that address and that were in existence in the year of interest. For consistency with our previous analysis, this analysis includes all 275/400 kV power lines plus 132 kV power lines in those areas of the country where we had data on year of construction for at least 80% of these lines. We create a series of point sources at 10 m intervals along such lines and, for each point, we multiply together the source-strength term for that power line, the distance term for the distance from that point to the subject address, the proportion of time the wind was blowing along the bearing from that point to the subject address, and the average wind speed along that bearing. The resulting exposure (in arbitrary units) for each point is summed over all the relevant points to give the total corona exposure at the address. 877
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
weighting (arbitrary units)
1
upwind
downwind 0
50
400
600
distance from power line / m
Figure 2. The distance weighting term for the variation of corona ions from a power line.
This model addresses several of the weaknesses of the quadrant model. First, it considers all the length of the power line that is within 600 m of the address rather than just the closest point. Secondly, it uses actual wind directions based on meteorological records rather than prevailing winds, and includes all directions a wind blows from over the year, appropriately weighted. We now describe the calculation of each of the four terms in turn. 3.2. Distance term
The ions produced by a power line can sometimes be detected kilometres away, but measurements by ourselves and others (e.g. [5]) lead us to model the observed average variation of the concentration of ions with distance from the power line, when the wind is transverse to the line, as shown in figure 2. We know that the distance variation is also affected by other factors, such as atmospheric conditions, but we are unable to model these. The distance term required for each individual point, in order for them to sum (at distances that are large when compared to the separation of the individual points) to this overall variation, is this overall variation multiplied by one-over-distance (we applied the one-over-distance multiplier beyond 50 m). The reduction in this term below 50 m reflects the fact that ions are produced on the conductors above ground and do not normally reach ground until they have travelled some distance. 3.3. Source-strength term
The production of corona ions depends on the electric field on the surface of the conductors. This in turn depends on the design of line, the voltage, and the number and size of conductors. Most power lines are built to one of a small number of standard designs, and we approximate all power lines as being in one of five categories as shown in table 2. For 275 kV and 400 kV lines, this is a good approximation, as alternative designs (particularly alternative conductors) proliferated only in the 1990s, towards the end of the period of this study. For 132 kV lines there is a greater variety. In particular, some 132 kV lines have twin conductor bundles and some single, which affects the electric field considerably, but we did not have access to these data, so chose to represent all 132 kV lines with an average of the values for twin and single bundles. We model the corona-ion production as proportional to the conductor surface electric field. We know this is not correct, in that corona is not proportional to the field but is produced only when a threshold is exceeded, and this usually happens in places where the field is increased 878
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
Table 2. Electric fields on the surface of conductors, used as ‘source strength’ weighting factors, for the five categories of power line.
Conductor Bundle Quad (4 conductors in square) Twin (2 conductors spaced horizontally) Mixed twin and single
29 mm diameter spaced 0.305 m 29 mm diameter spaced 0.305 m Average of twin (2 different designs) and single (3 designs)
Voltage (kV)
Electric field (kV m−1)
400 275 400 275 132
1 076 698 1 778 1 222 1 140
by surface perturbations, rather than by the unperturbed field which we calculate. However, we have no way of modelling this level of detail and we have no way of distinguishing, for example, lines in good repair from lines in bad repair. This method probably orders the different categories of line correctly even if the precise weightings are approximate; 400 kV lines with twin conductor bundles are indeed usually observed to be the greatest producers of corona, and 275 kV lines with quad bundles are rarely observed to produce any significant corona. The electric field varies round the circumference of a conductor; we used the maximum value on each bundle, averaged over all the bundles on that power line. 3.4. Wind-rose term
We obtained wind data from eight meteorological stations, in most cases from 1987 to 2002. We averaged over all available years and used the result for all subjects regardless of birth year. The locations of the meteorological stations are shown in figure 3. These had been selected by National Grid for other, operational, reasons and were not within our choice. For each subject we used the data from the nearest station. The distance from subject to meteorological station varied from 2 km to 170 km with a median of 37 km. The data give the number of hours for which the wind blows at each combination of angle (in ten degree sectors) and speed (grouped in 3 m s−1 intervals). We show the results for one station, Yeovilton, in figure 4. The wind-rose term for each point is the proportion of time the wind blew in the direction corresponding to the bearing of the address from that point. 3.5. Wind-speed term
Using the same wind data as for the wind-rose term, we calculated the average wind speed in the direction from each point to the address. 3.6. Alternative calculations of exposure
Each of the terms in our model involve assumptions. We have made our best assumptions, but even with recent advances [23, 24], knowledge of the behaviour of corona ions is not yet sufficient for us to be confident in these. In addition to the analysis based on the primary corona model, calculated by the product of the four terms described in sections 3.2–3.5 above, we therefore examined the sensitivity of our results to the assumptions by constructing four alternative secondary models where, in turn, each of the terms is omitted (in the case of the distance term, it is not completely omitted but replaced by a simpler term comprising just the one-over-distance variation and a truncation at 600 m, that is, omitting the corona-ion-specific distance variation of figure 2). 879
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
Figure 3. Locations of meteorological stations providing wind data.
3.7. Statistical analysis
We analyse the data by conditional logistic regression, using standard methods for an individually matched case-control study, as in our previous papers. For our primary calculation of exposure and for each of our four subsequent sensitivity analyses, we group the subjects into five categories: the reference category, consisting of all addresses that are more than 600 m from the nearest power line, and which have zero exposure in all models; and four categories defined by quartiles of the calculated exposure for addresses within 600 m of the line. For each exposure calculation, we calculated the quartile boundaries once, based on the totality of the subjects, then applied these boundaries consistently for all analyses of that variable. (In [9], we used a different reference category, >1 000 m, because we were testing for risks from 600 m to 1 000 m; having found that there were no elevated risks here, we now revert to the >600 m reference category from [8].) 4. Results This paper is based on the same series of case and control subjects as our previous study, except that three addresses were excluded because of missing data. There were 7 347 subjects with an address within 600 m of a relevant power line. For each address and for lengths of line within 600 m we carried out the calculations described above for points at 10 m intervals. We 880
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
2
0
1
270
90
0
180
Figure 4. Wind Rose for Yeovilton 1987–2002. Radial axis is the fraction of time the
wind blew from each direction, in 10° intervals, normalised to a uniform distribution of directions.
identified a total of 716 013 such points, with between 1 and 930 points per address, a mean of 97 each. Table 3 shows, by decade of diagnosis and in aggregate, the results for our primary exposure calculation. For leukaemia only, table 4 compares the results for the primary and the four secondary exposure calculations, again by decade of diagnosis and in aggregate. Table 5 presents selected relative risks, for the primary exposure calculation and for leukaemia only, for subjects whose closest line is 275/400 kV compared to 132 kV (in each case the exposure calculation for that subject still includes all nearby lines). 5. Discussion The model for calculating corona-ion exposure we have used here is undoubtedly an improvement on the previous quadrant model, but it still has considerable limitations. These include: • use of wind data from just eight meteorological stations, applied, in extreme cases, to subjects over a hundred km distant, with no allowance for the effect on the wind direction of local topography or proximity to coasts; • modelling of the source strength of each line only in terms of the voltage and conductor bundle, without taking into account the other factors which, whilst not yet identified or understood, must affect corona production; and • applying one representative and simplified distance variation to all points. Nonetheless, we regard our model, despite its limitations, as the best that is likely to be achievable on present understanding and data, and we consider, by taking account of key factors such as voltage and conductor bundle, and wind direction, it provides a valid discrimination between high exposures and low exposures. Table 4 shows that none of the risks from 881
Q4 Q3 Q2 Q1 Q0 Trend Q4 Q3 Q2 Q1 Q0 Trend Q4 Q3 Q2 Q1 Q0 Trend
1962–1969
882
1980–1989
1970–1979
Exposure
Period
52 43 53 54 3 084
44 52 32 35 3 100
14 13 6 9 972
Cases
39 47 34 43 3 129
30 42 41 35 3 115
2 10 4 9 989
Controls 1.84–106.5 0.51–2.95 0.42–5.32 0.38–2.66
0.98–1.16 0.87–2.11 0.58–1.41 1.11–2.85 0.83–1.93 0.99–1.17
14.00 1.22 1.50 1.00 1.00 1.32 1.41 1.29 0.76 1.10 1.00 1.07 1.35 0.90 1.78 1.26 1.08
1.08–1.61 0.86–2.30 0.83–2.00 0.46–1.24 0.68–1.82
95% CI
RR
Leukaemia
44 36 33 31 2 160
18 22 28 25 2 021
4 4 4 6 551
Cases
44 33 31 33 2 176
25 15 30 21 2 023
8 3 3 4 551
Controls 0.50 3.00 0.67 2.00 1.00 0.93 0.67 1.75 0.96 1.10 1.00 0.99 0.97 1.12 1.14 0.95 1.00 1.01
RR
0.93–1.11
0.88–1.11 0.61–1.54 0.67–1.89 0.64–2.02 0.54–1.67
0.73–1.19 0.35–1.25 0.86–3.56 0.56–1.65 0.60–2.02
0.15–1.66 0.31–28.84 0.11–3.99 0.50–8.00
95% CI
CNS/brain tumours
60 70 55 70 4 235
41 39 45 52 3 548
11 14 13 12 1 402
Cases
68 66 76 61 4235
46 33 43 50 3554
13 14 14 15 1396
Controls
0.75 0.92 1.09 0.79 1.00 0.96 0.82 1.16 1.03 1.18 1.00 1.00 0.88 1.15 0.68 1.12 1.00 0.97
RR
Other diagnoses
0.91–1.04
0.93–1.09 0.61–1.28 0.80–1.64 0.47–0.98 0.76–1.64
0.83–1.12 0.53–1.27 0.72–1.88 0.66–1.60 0.77–1.82
0.32–1.78 0.40–2.08 0.48–2.47 0.36–1.73
95% CI
Table 3. Relative risks and 95% confidence intervals for primary exposure calculation, three diagnostic groups, by decade of diagnosis and overall. In this and subsequent tables: ‘trend’ is the RR for each quartile increase in exposure. Q0 is the reference category (zero exposure) and Q1–Q4 the four successive quartiles of non-zero exposure. Bold indicates p 1 000 m) and corona-
ion analysis (trends across exposure categories). Because there are four quartiles of exposure, the trend RRs are plotted on an axis expanded by a factor of 4, to allow a very approximate visual comparison.
across the whole study area, they could constitute a potential confounding factor that we are not able to adjust for. Further, there are many sources of ions other than corona, such as combustion products, from transport, industry and gas cooking, waterfalls, etc, which will lead to misclassification; and there will be geographical variations of any airborne pollutants with which the corona ions interact. Table 3 does not suggest any pattern of risk with calculated exposure for brain tumours or for ‘other’ cancers. Further discussion here therefore focuses on leukaemia. Because the model assigns zero exposure outside 600 m and non-zero exposure inside 600 m, and because we already know that there are raised risks for leukaemia within 600 m in the earlier decades, we expect leukaemia risk to be associated with calculated exposure. The key question is: does the calculated exposure predict risk better than distance on its own? Table 6 compares selected estimates of risk from the two analyses. Figure 5 presents one of these comparisons graphically. Other comparisons between the two sets of results are possible and we present table 6 and figure 5 as illustrative rather than definitive. As expected, the corona-ion analysis does produce elevated risks in the earlier decades. But, with one exception, there is no clear trend with increasing exposure, and the risks are less than the equivalents for the distance analysis. The exception is the risk for the top quartile in the 1960s; we draw attention to this, but are inclined to regard it as a chance result. Further, of the four terms that make up the primary exposure calculation, the one that is most distinctive of the hypothesis is the wind-rose term; whatever uncertainty there may be about the individual terms, wind direction, even when assessed imperfectly as here, is clearly fundamental to the specific hypothesis being tested. From the sensitivity analysis presented in table 4, we see that removing the wind-rose term does not result in a consistent reduction in risk as would be expected. Thus, we conclude that the simple corona-ion hypothesis that we test here explains the observed pattern of leukaemia rates around power lines less well than straightforward distance. Our findings therefore do not support this corona-ion hypothesis as the explanation of these results. In view of the many uncertainties and approximations, however, this analysis is not a definitive rebuttal of this hypothesis, and, as discussed above, would not be relevant 887
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
to any development of the hypothesis which no longer predicted stronger effects downwind compared to upwind, or in which the effects were confounded by other factors. Further, our test relates only to childhood cancers, not to other diseases. Further arguments against the hypothesis come from the time trends illustrated in figure 1. If the reduction in risk around power lines were because the power lines still produce corona ions, but the numbers of airborne leukaemogens involved have reduced, we would expect that the reduction in these leukaemogens across the country as a whole, the majority of which is not close to power lines, would have led to a corresponding reduction in overall leukaemia rates. Recorded leukaemia rates have in fact increased. At least some of this is attributable to improved diagnosis [25], but there is still a discrepancy, which could be explained only if some other risk factor had increased in proportion to compensate, a concept that is not impossible but is inelegant. This paper has considered just the corona hypothesis and found little evidence in its favour. As discussed in our previous papers and the subsequent correspondence, there are many other possible explanations for the observed association between childhood leukaemia and proximity to high-voltage overhead power lines. Acknowledgments We thank Professors Denis Henshaw and Alan Preece for contributing to the original development of our corona-ion model, Alasdair Philips for drawing our attention to the fall in atmospheric radionuclides, and Messrs Henshaw and Philips for helpful comments on this paper; however, none of them are thereby indicating agreement with our methods or conclusions. This study was undertaken as part of a project funded initially by the United Kingdom Department of Health Radiation Protection Programme and subsequently by the charity Children with Cancer UK. The Childhood Cancer Research Group also received funding from the Department of Health and the Scottish Ministers. The Childhood Cancer Research Group closed down completely in April 2014. The views expressed here are those of the authors and not necessarily those of the Department of Health and the Scottish Ministers. JS is employed by National Grid who provided staff time but no other funding. A written contract exists between the Childhood Cancer Research Group and National Grid specifying that the Childhood Cancer Research Group has complete control over the conduct, interpretation, and publication of this study. This paper has not been approved by anyone in National Grid other than JS in his capacity as author and does not necessarily represent National Grid’s views. References [1] Wertheimer N and Leeper E 1979 Electrical wiring configurations and childhood cancer Am. J. Epidemiol. 109 273–84 [2] IARC 2002 Non-Ionizing Radiation Part 1: Static and Extremely Low-Frequency (ELF) Electric and Magnetic Fields (Geneva: World Health Organization) p 1 [3] Ahlbom A et al 2000 A pooled analysis of magnetic fields and childhood leukaemia Br. J. Cancer 83 692–8 [4] Henshaw D L, Ross A N, Fews A P and Preece A W 1996 Enhanced deposition of radon daughter nuclei in the vicinity of power frequency electromagnetics fields Int. J. Radiat. Biol. 69 25–38 [5] Fews A P, Henshaw D L, Wilding R J and Keitch P A 1999 Corona ions from powerlines and increased exposure to pollutant aerosols Int. J. Radiat. Biol. 75 1523–31 [6] Fews A P, Henshaw D L, Keitch P A, Close J J and Wilding R J 1999 Increased exposure to pollutant aerosols under high voltage powerlines Int. J. Radiat. Biol. 75 1505–21 888
J Swanson et al
J. Radiol. Prot. 34 (2014) 873
[7] Henshaw D L, Ward J P and Matthews J C 2008 Can disturbances in the atmospheric electric field created by powerline corona ions disrupt melatonin production in the pineal gland? J. Pineal Res. 45 341–50 [8] Draper G, Vincent T, Kroll M E and Swanson J 2005 Childhood cancer in relation to distance from high voltage power lines in England and Wales: a case-control study BMJ 330 1290 [9] Bunch K J, Keegan T J, Swanson J, Vincent T J and Murphy M F 2014 Residential distance at birth from overhead high-voltage powerlines: childhood cancer risk in Britain 1962–2008 Br. J. Cancer 110 1402–8 (Electronic) [10] Sermage-Faure C et al 2013 Childhood leukaemia close to high-voltage power lines—the Geocap study Br. J. Cancer 108 1899–906 [11] Pedersen C et al 2014 Distance from residence to power line and risk of childhood leukemia: a population-based case-control study in Denmark Cancer Causes Control CCC 25 171–7 [12] Kheifets L et al 2013 Epidemiologic study of residential proximity to transmission lines and childhood cancer in California: description of design, epidemiologic methods and study population J. Expo. Sci. Environ. Epidemiol (doi:10.1038/jes.2013.48). [13] Swanson J and Jeffers D E 1999 Possible mechanisms by which electric fields from power lines might affect airborne particles harmful to health J. Radiol. Prot. 19 213–29 [14] Jeffers D 2001 A note on the charging of aerosols by overhead line corona Radiat. Prot. Dosim. 95 181–3 [15] Advisory Group on Non-Ionising Radiation 2004 Particle Deposition in the Vicinity of Power Lines and Possible Effects on Health (Chilton: National Radiological Protection Board) p 1 [16] Jeffers D 2007 Modelling and analyses do not support the hypothesis that charging by power-line corona increases lung deposition of airborne particles Radiat. Prot. Dosim. 123 257–61 [17] Raaschou-Nielsen O and Reynolds P 2006 Air pollution and childhood cancer: a review of the epidemiological literature Int. J. Cancer 118 2920–9 (Print) [18] Pyatt D and Hays S 2010 A review of the potential association between childhood leukemia and benzene Chem. Biol. Interact. 184 151–64 (Electronic) [19] Kroll M E, Swanson J, Vincent T J and Draper G J 2010 Childhood cancer and magnetic fields from high-voltage power lines in England and Wales: a case-control study Br. J. Cancer 103 1122–7 (Electronic) [20] Swanson J, Vincent T, Kroll M and Draper G 2006 Power-frequency electric and magnetic fields in the light of Draper et al Ann. N. Y. Acad. Sci. 1076 318–30 [21] Murrells T P et al 2010 UK Emissions of Air Pollutants 1970–2008 (Didcot: AEA Technology plc) Report No.: AEAT/ENV/R/3036 [22] UNSCEAR 2000 Sources and Effects of Ionizing Radiation (Vienna: United Nations Scientific Committee on the Effects of Atomic Radiation) [23] Matthews J C 2012 Diurnal variations of atmospheric potential gradient disruption near to high voltage power lines J. Atmos. Sol.-Terr. Phys. 77 235–40 [24] Matthews J C 2012 The effect of weather on corona ion emission from AC high voltage power lines Atmos. Res. 113 68–79 [25] Kroll M E, Carpenter L M, Murphy M F G and Stiller C A 2012 Effects of changes in diagnosis and registration on time trends in recorded childhood cancer incidence in Great Britain Br. J. Cancer 107 1159–62
889
Copyright of Journal of Radiological Protection is the property of IOP Publishing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
