Childhood cancer in relation to distance from high voltage power lines in England and Wales: a case-control study
BMJ 2005; 330 doi: https://doi.org/10.1136/bmj.330.7503.1290 (Published 02 June 2005) Cite this as: BMJ 2005;330:1290
All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
Draper et al’s (1) observation that: “ Compared with those who lived
more than 600 m from a transmission line at birth, children who lived
within 200 m had a relative risk of leukaemia of 1.69 (95% confidence
interval 1.13 to 2.53); those born between 200 and 600 m had a relative
risk of 1.23 (1.02 to 1.49)” prompted a “rapid response” from Roman et al
(2) which concluded: “Their positive result over 100m may [therefore] be
explained not by an excess of cases but by a deficit of controls.”
Draper et al matched cases and controls within the registration
districts where the cases’ births were registered. There are some 400 of
them and they vary greatly in size, age standardised rates (ASR) for
childhood cancers and fractions of their populations exposed to
transmission line fields. The Committee on Medical Aspects of Radiation in
the Environment (COMARE) (3) produced a report on the geographic incidence
of childhood cancer over the years 1969 – 1993, which covers most of the
period of the Draper study and Jeffers (4) suggested that the geographic
volatility revealed by COMARE might explain why the relative risks which
were generated using the “leukaemia” controls disappeared when the “total
cancer” controls were used instead (1, 5).
It will be shown here how the volatilities in registration district
populations, incidence rates and exposed fractions can, in combination,
give rise to the underestimate of the exposed fraction of controls
suggested by Roman.
Total population =N
Population of j’th registration district = N(j)
Average age standardised incidence rate =R
Total number of cases / controls =NR
Average exposed fraction in population = A
ASR in j’th district, R(j) = R (1 + r(j))
Exposed fraction in j’th district, A(j) = A (1 + a(j))
Number of exposed controls = Sum N(j) R(j) A(j)
Giving an estimated exposed fraction
= Exposed controls / Total number of controls
= (A/N) Sum N(j) (1+r(j))(1+a(j))
=A + (A/N)Sum N(j) r(j) a(j)
= average exposed fraction (A) + offset
It should be noted that, if the ASR were constant over the districts,
the a(j), their range and the offset would all be zero, making the
estimated exposed fraction equal to the population average, A.
In England and Wales, the COMARE report shows an average ASR of 37.7
for leukaemia of and 113 (per million children per year) for total
cancers. The extremes of the two geographic ASR distributions are similar
to the extent that both diseases have their minima in West Glamorgan and
their maxima in Buckinghamshire. For leukaemia, the minimum is 29.1 and
the maximum is 48.3, giving a minimum “r” of -0.228, a maximum of 0.281
and a range (leukaemia) of 0.509. For the total cancers, the extremes are
132.2 and 94.1, resulting in a minimum value for “r” of -0.167, a maximum
of 0.170 and a range (total cancers) of 0.337. In districts which are not
crossed by transmission lines, A(j) = 0 and a(j) = -1, as a consequence
the range is very much larger than that for r. The values of “a” depend
on geography and are independent of the type of control but because new
houses would have been constructed during the course of the study, both A
and a(j) may change with time and the summations above apply to averages
over the duration of the study. The ASRs derived from the COMARE report
are averaged over the period 1969 – 1993.
Draper et al show 39 out of 9700 “leukaemia” controls within 200 m of
a transmission line, making the estimated exposed fraction 0.402%. For
total cancers, 151 out of 29081 were within a similar distance, making the
estimated exposed fraction 0.519%. One thus has two “estimated exposed
fraction,range” pairs: (0.402%, 0.509) for the leukaemia controls and
(0.519%, 0.337) for the total cancer controls. Extrapolating to zero
range gives a third pair (0.75%, 0). It was shown above that the estimated
exposed fraction is equal to the population average when the range is zero
and this estimate of 0.75% is larger than both Draper’s measured exposed
fraction of 64/9700 (= 0.66%) for the leukaemia cases and the 0.402%
estimated for the controls. An excess risk is no longer predicted.
When the calculation is repeated for all of the controls within 600
m, one obtains a value of 3.32% for population average exposed fraction
which is equal to that for the cases. An excess risk is, again, no longer
predicted.
Whilst recognising that the analysis here relies on drastic
simplifying assumptions and that Draper et al leave open the possibility
that the variations in estimated risk may be due to chance; the analysis
provides some support for Roman et al’s contention that the elevated risk
ratios reported by Draper et al are the consequence of underestimates of
the exposed fraction of controls rather than an excess of cases. It is
suggested that the effects of geographic volatility merit further
consideration.
References
1. Draper, G., Vincent, T., and Swanson, J.; Childhood cancer in
relation to distance from high voltage power lines in England and Wales:
a case control study, BMJ 2005; 330: 1290.
2. Roman, E, Day, N, Eden, T, McKinney, P and Simpson, J.; Childhood
Cancer and distance from high-voltage power lines – what do the data mean?
bmj.com, 5 Jul 2005
3. Committee on Medical Aspects of Radiation in the Environment; The
distribution of childhood cancers in Great Britain 1969 – 1993. (2006)
http://www.comare.org.uk/comare_docs.htm
4. Jeffers, D; Geography and Controls, bmj.com, 19 March 2008.
5. Kheifets L, Feychting M and Schuz J; Control selection in the
Study of Childhood cancer in relation to distance from high voltage power
lines in England and Wales. bmj.com, 28 Jun 2005.
Competing interests:
None declared
Competing interests: No competing interests
Draper et al’s result 1 that: “Compared with those who lived >
600 m from a line at birth, children who lived between 200 and 600 m had a
relative risk of leukaemia of 1.69 (95% confidence interval 1.13 to 2.53);
those born between 200 and 600 m had a relative risk of 1.23 (1.02 to
1.49)”, coupled with their conclusion that: “If the association is
causal, about 1% of childhood leukaemia in England and Wales would be
attributable to these lines”, has made their paper an important document
in the political debate regarding the case for precautionary policies to
limit EMF exposure 2, 3.
The authors qualify their data with the comment: “The results are
highly significant but could nevertheless be due to chance - for example,
if the leukaemia controls are not sufficiently representative of the
relevant population. Some support for this explanation can be derived
from the different distance distributions observed for the leukaemia and
non-leukaemia controls in table 1. Comparison of the leukaemia cases with
the latter still suggests that there is an increased risk for leukaemia
but it is much lower than that found using the matched controls. We
emphasise, however, that the use of matched controls is the most
appropriate approach.
The impact of control selection was discussed in the “Rapid
Responses” from Hepworth et al, Kheifets et al, Roman et al and the
original authors. Hepworth et al concluded their letter with the comment:
“The findings of this study are of interest in that they point towards
geographical correlates of risk for childhood leukaemia but do not support
the hypothesis that electromagnetic fields have a causal role.” The
possible influence of geographical variation will be considered below.
Draper et al matched cases and controls within registration districts
and a map of England and Wales showing their boundaries can de downloaded
from the “National Statistics” web site
(http://www.statistics.gov.uk/geography/reg_districts.asp).
There are currently some 400 districts and the map shows that there are
vast differences in size. Draper et al’s study ran from 1962 – 1995 and,
because large numbers of registration districts were abolished in the
local government reorganisation of 1974, there were many more of them at
the beginning of the study than at the end.
The authors were considering the possible impact, on childhood
cancer, of the 7000 km of transmission lines which make up the National
Grid in England and Wales. Cases were considered to be “exposed” up to a
distance of 600 m either side of the lines, giving a total “exposed” area
of 8400 km2 which is comparable to that of North Yorkshire, which is the
largest English county (administrative area 8038 km2). There is clearly
scope for considerable geographic variation within such a large area.
A map of the National Grid can be downloaded from chapter 6 of the
National Grid Company’s Seven Year Statement
(http://www.nationalgrid.com/uk/ ). It is apparent that many districts
are not crossed by its transmission lines and, in Draper et al’s terms,
the populations of these districts will be unexposed. (Although a lot of
districts are not crossed by transmission lines they are still likely to
host many lower voltage distribution lines.)
The authors took one control per case, matched for sex and date of
birth within the registration district.
The total number of cases and controls = S1
= Sum over time and registration districts of (district population *
Age-standardised rate of disease incidence)
The number of exposed controls = S2
= Sum over time and districts of (population * ASR * fraction of
population which is exposed)
The fraction of the controls which is exposed is then S2 / S1 and, if
the ASR’s were uniform over the districts, then the calculated exposed
fraction would be equal to that for the total population
In its eleventh report, the “Committee on Medical Aspects of
Radiation in the Environment” (COMARE) 4 provides data on “The
distribution of childhood cancers in Great Britain 1969 – 1993.” Over
this period, which is comparable to that of the Draper et al study, there
were significant geographical differences in the regional incidence of
leukaemia and other cancers. The calculated exposed fractions are,
therefore, not likely to be equal to that for the total population and, as
Draper et al found, the distance distributions for the leukaemia and total
cancer controls will be different. To quote from section 3.6 of the
report:
“In the counties of England and Wales the rates for all cancers
combined vary between 132.2 per million (Buckinghamshire) and 94.1 (West
Glamorgan), a ratio of 1.40. For leukaemia the rates varied between 48.3
(Buckinghamshire) and 29.1 (West Glamorgan), a ratio of 1.66”
It is of interest to note that, at 1.66, the Buckinghamshire / West
Glamorgan ratio for leukaemia is almost equal to the excess risk of 1.69
reported by Draper et al.
The COMARE data show that variation in the leukaemia ASR’s, at 1.66
is greater than that for all cancers combined (1.40). A calculation of
the exposed fraction using the “all cancers” controls is thus likely to be
closer to the value for the total population than a calculation based on
the leukaemia controls. It is therefore suggested that geographic
variation may provide an explanation for the differences between the risk
ratios calculated using the leukaemia and total cancer controls and a
reason for favouring the use of total cancer controls.
References
1 Draper, G., Vincent, T., and Swanson, J.; Childhood cancer in
relation to distance from high voltage power lines in England and Wales: a
case control study., BMJ 2005; 330: 1290
2 Stakeholder advisory group on ELF EMFs (SAGE). Precautionary approaches
to ELF EMF. Supporting document S4 (2007)
http://www.rkpartnership.co.uk/sage/
3 Cross Party Inquiry into childhood leukaemia and Extremely Low
frequency electric and magnetic fields (ELF EMF); (2007);
http://ePolitix.com - CPIELFEMF
4 Committee on Medical Aspects of Radiation in the Environment; The
distribution of childhood cancers in Great Britain 1969 – 1993. (2006)
http://www.comare.org.uk/comare_docs.htm
Competing interests:
None declared
Competing interests: No competing interests
Results of the study by Draper et al have been considered mainly in
terms of electric and magnetic field influences. The `indirect` ion
hypothesis of Fews et al (1) relating to health hazards of electrically
enhanced airborne particulate pollutant deposition was briefly discussed,
but the possible relevance of `direct` biological activity (2) by
powerline associated air ionization has not been considered.
In general, the extent of corona activity at overhead powerlines may
have been underestimated, but has become apparent recently following the
introduction of the daytime ultraviolet corona detection camera. As
illustrated on the "Daycor" website (3), corona activity is particularly
associated with support insulators and other overhead line junction
equipment.
Groom & Chalmers (4) reported that negative space charge, near a
132 kV powerline, was liberated at line support insulators and could be
detected principally near transmission towers. Unlike magnetic field
effects therefore, any health impacts attributable to airborne
electroactivity would be expected to show some non-random distribution
along the axis of an electricity utilization line corridor, with a
tendency for `health impact` sites to be clustered near support towers or
poles. As the Draper et al study obtained grid references for all
relevant high voltage line pylons in England & Wales, the study data
could provide the basis for evaluating the relative importance of
`airborne electroactivity` and `magnetic field` interpretations of health
impacts. A linear non-random distribution of adverse health impacts would
support the view that some airborne products of electrical corona
discharge may be biologically active.
1. Fews AP, Henshaw DL, Wilding RD, Keitch PA, 1999. Corona ions from
powerlines and increased exposure to pollutant aerosols. Int. J. Radiat.
Biol. 75 1523-1531.
2. Sidaway GH, 2008. Environmental and social impacts of electricity
utilization : broadening the debate. The Environmentalist, in press : DOI
10.1007/s10669-007-9160-2
3. <http://www.daycor.com/Technology/Corona-phenomenon.html>
4. Groom KN, Chalmers JA, 1967. Negative charges from high-tension
power cables in fog. J.Atmos. Terr. Phys. 29 613-615.
Competing interests:
None declared
Competing interests: No competing interests
Editor – In the paper by Draper and colleagues [1], the authors state
in their conclusions that “there is no obvious source of bias in the
choice of cases or controls”. Although we agree with the authors in their
statement that “registration for childhood cancer is nearly complete”, we
are concerned that the cases included in the analysis group may be prone
to censoring since the inclusion criteria was based on cases both born and
diagnosed in England and Wales between 1962-1995.
Cases extracted from the National Register of Childhood Tumours
(NRCT) [2] were diagnosed between 1962 and 1995 aged under fifteen years
and the analysis was restricted to those born during the same period. The
aim of the study was to investigate whether there was a relationship
between distance of the home address at birth and high voltage power lines
for the subsequent development of childhood cancer. This approach will
have excluded a proportion of individuals diagnosed with cancer under the
age of 15 years but who were born after 1980 and diagnosed beyond 1995. A
complete birth cohort from 1962-1995 would require ascertainment of cases
from 1962-2010.
In addition, there were a number of missing birth addresses (n=1700,
5%) and we would be interested to know whether the proportion differed
across the study time period. Our experience would suggest a higher
proportion of cases with missing birth address records near the beginning
of the study period. Birth address postcodes were not routinely available
during the 1960s possibly reducing the completeness of the cohort during
the early period.
The main aim of the study was to examine risk around the time of
birth and therefore we believe that a more appropriate analysis would have
been one where the inclusion criteria are based on a complete birth
cohort. It would be interesting for the authors to re-analyse their data
by restricting cases to those diagnosed from 1962 onwards and born between
1962 and 1980 to minimise any potential effect from censoring.
1. Draper G, Vincent T, Kroll ME, Swanson J. Childhood cancer in
relation to distance from high voltage power lines in England and Wales: a
case-control study. BMJ 2005;330:1290-2.
2. Stiller CA, Allen MB, Bayne AM, et al. United Kingdom: National
Registry of Childhood Tumours, England and Wales, 1981-1990. In
International Incidence of Childhood Cancer: Volume 2. eds Parkin DM,
Kramárová E, Draper GJ et al. pp.365-367. Lyon: IARC Scientific
Publications No 144, 1998.
Competing interests:
None declared
Competing interests: No competing interests
Why isn't there a discussion about a possible correlation between
incidence of leukemia and r*B, that is the product between the distance
from the power line and the strength of the magnetic field?
The electric field induced by the time-varying magnetic field is one
possible cause of the health effects, and as far as I can see this induced
field is proportional not to the magnetic field itself, but to the above
product.
Competing interests:
None declared
Competing interests: No competing interests
Roman and colleagues say that we used distance as a proxy for
magnetic field exposure; this is correct only in a rather weak sense of
the word “proxy”. They go on to say that we “acknowledge [that] this is a
crude estimate [of power-frequency magnetic field exposure]”; we said
nothing like this. The distance analyses are similar to those used by the
writers of the letter in their capacity as authors of the United Kingdom
Childhood Cancer Study (UKCCS) paper [1]. We shall be presenting our
analysis of calculated magnetic fields in a subsequent paper, and we
regard the distance analysis in this paper as a separate analysis in its
own right.
We drew attention in our paper to the possibility that the leukaemia
controls are, by chance, unrepresentative.
In discussing this point, however, Roman et al make invalid
comparisons in the graph accompanying their letter. Their two sets of
comparison data refer to addresses in the 1990s. Our study extends over a
much longer period (1962-1995), during which there were increases in the
numbers of lines and of houses situated close to lines. The numbers quoted
in our paper relate to the whole of this period. Their finding that there
is a smaller proportion of addresses close to the line when comparing the
average over the whole of this period with data for the 1990s is
unsurprising given the time trend in the number of houses near lines.
Our unpublished data show that when data relating to more closely
comparable periods are used we actually have, for most of the distances
considered, higher proportions of leukaemia controls living near lines
than are found for the two comparison groups, not lower as they suggest:
see attached graph. (We have taken the values for UKCCS controls from
table 1 of [1].) Even when the periods are comparable, the distributions
of birth addresses (our data) and diagnosis addresses and all homes (their
comparison data) are not necessarily expected to be the same .
To summarise: we suggested ourselves that the distribution of our
leukaemia controls means that chance has to be more seriously considered
as an explanation for our results, but the evidence for this comes from
internal comparisons within our data and not from suggested comparisons to
other data.
1 UK Childhood Cancer Study Investigators. Childhood cancer and
residential proximity to power lines. Br J Cancer 2000; 83:1573-80.
Competing interests:
GJD, TJV and MEK: no conflict of interest. JS is employed by National Grid Transco and worked on this project with their permission
Competing interests: No competing interests
The authors estimate an increased attribution, if the association
were causal, of about 5 cases of childhood leukaemia per year in England
and Wales, among some 400,000 children with birth address within 600
metres of National Grid powerlines. About half of those five cases would
be within 200 metres.
The NRPB [1] estimated that about two attributable cases per year in
the UK would be associated with time-weighted average (TWA) magnetic
fields (MF) above 0.4mT (and none below), of which about half a case would
be attributable to exposures from powerlines. MF in excess of 0.4mT from
powerlines would probably all occur well within 200 metres.
The population of the UK is about 13% more than that of England and
Wales. National Grid powerlines include all those at 275 and 400 kV but
exclude almost all those at 132 kV and lower voltages. The Draper study
has a more restricted definition, in terms of both geography and exposure
sources, than the key MF studies. If the only relevant cause were MF above
0.4mT, the estimate from the Draper study of 2.5 attributable cases per
year would at first sight be surprising, compared with only 0.5 from MF
studies, even though both estimates are imprecise.
Does this resurrect the "wire code paradox", said to be resolved in
[2], in another form? A question arises as to whether inappropriate
metrics in the MF studies tend to suppress any association compared with
proximity to powerlines.
MF exposure metrics in the constituent studies in both [2] and [3]
are generally arithmetic averages with respect to time over 24 hours or
more during the year preceding diagnosis. Some are weighted to track the
individual case or control exposure over time.
One quirk in [3] is the decision "to use geometric means from all
studies, because they are less affected by outliers". For positive numbers
not all the same, the geometric mean G will always be less than the
arithmetic mean A. That will mean G is less affected by high outliers but
more affected by low ones (and critically affected by a zero!). This might
be one deficiency in exposure metric for this seminal pooled MF analysis.
Another possible deficiency, having regard to melatonin hypotheses,
might be the dilution of night-time exposure by 24-hour averaging.
Although MF from powerlines in the UK are lower at night than in daytime,
they may be the dominant night-time source. Analysis of pooled German
studies [4] found an OR of 4.28 (1.25-14.7) for night-time exposures above
0.4mT. That gives an attributable RR of 3.28, compared with the
attributable RR of 1 from 24-hour TWA, which lies behind the estimate of
half a case per year near powerlines in the UK.
The above figures from [1], [3] and [4] could be broadly reconciled
as follows, taking proximity to powerlines as a proxy for night-time
exposure. Take the normal (non-attributable) EMF-associated cases to be,
like the NRPB conjectured attributed cases, 0.5 per year from powerlines
sources and 1.5 from non-powerlines sources. Then take the excess
(attributable) cases to be 1.5 from powerlines and 0.5 from non-powerline
sources. This preserves the overall RR of 2 while allowing the night-time
(powerline) RR to be about 4 and incidentally implying a daytime RR of
about 1.3. As the data are so imprecise, such a reconciliation is not to
be taken prescriptively; it merely indicates a possible broad
compatibility.
As well as dilution of night-time exposure above 0.4mT, might there
be dilution above 0.2mT, where UK data suggest there are far more
children? A joint EEA/WHO review [5] notes "If regression dilution were
concealing a relative risk of 1.5 for children exposed to between 0.2 and
0.4mT, then the annual number of attributable cases might be six or
seven". Metric dilution might also contribute to the concealment of such
an association for night time exposure; Schutz [4] found OR = 3.21 (1.33-
7.80) above 0.2mT.
For the avoidance of doubt, these comments aim to raise questions,
not to infer conclusions. The questions open possibilities for the
findings of increased hypothetically attributable cases within 200 metres
to be reconciled in terms of exposure metric deficiencies in MF studies.
Other questions, e.g. about the distributions of controls, may point the
other way. I don’t want to exaggerate a marginal consideration, and
acknowledge the perspective in the editorial comparing 5 attributable
cases with 500 others. But given public (and scientific) interest, might
it yet be worth re-analysing MF studies, such as UKCCS, for night-time
exposure?
[1] Board of NRPB Response Statement R3/2001.
[2] Greenland et al, Epidemiology 11(6), 624-634, 2000.
[3] Ahlbom et al, British Journal of Cancer 83(5), 692-698, 2000.
[4] Schutz et al, Int J Cancer 91, 728-735, 2001.
[5] Tamburlini et al, EEA Environmental Issue report No. 29, 2002.
Competing interests:
None declared
Competing interests: No competing interests
Prof. Roman makes an interesting point but the distribution of
housing density with distances from power lines, which was produced
recently for the UKCCS, should not be applied over the timescale of the
Draper et al study.
The study considered data for the period 1962-95 and there have been
profound social, environmental and economic changes during this time.
The 400-kV electical network was developed during the 1960's and,
initially, much of it ran through industrial landscapes. Most of the heavy
manufacturing industry which these lines were designed to serve is long
gone and, as in London Docklands, for example, urban development has taken
its place. This has resulted in the construction of housing near to
existing over head power lines and, as a consequence the number of houses
near lines increases with time.
Mention is also made in some of the reponses to the effect of power
line corona on particulate pollution. The start of the study falls
between the Clean Air Act of 1956 and its extension in 1968. The early
stages of the period covered by the study would have seen a marked fall in
general pollution levels.
Competing interests:
None declared
Competing interests: No competing interests
In their reply to responses received, the authors say:
"O’Carroll’s calculations are based on the assumption that the results at
0-200 metres are due to magnetic fields while those at 200-600 metres are
due to chance. We do not think it justifiable to make an arbitrary
division of our results into two bands."
I made no such assumption. I simply estimated how much of the
associated excess incidence found by the authors would be in the range 0-
200 metres. On that point I concluded "So the stronger statistical
findings in the range 0 - 200 metres alone support about half the
increased attribution".
I made no such arbitrary division. The authors made the division in
presenting their results. I made comments on the statistical nature of
their results so divided. I made only tentative conclusions about possible
reasons for their results and I made no assumptions about causation. Far
from using arbitrary banding or incurring statistical effects of so doing,
I cautioned against taking the results from 200 to 600 metres out of
context to support or deny an effect.
Finally, I was careful to distinguish between association and cause,
contrary to the general allegation which the authors make against several
respondents.
Competing interests:
None declared
Competing interests: No competing interests
Re: Childhood cancer in relation to distance from high voltage power lines in England and Wales: a case-control study
Dear Editor
I was a Consultant at The United Leeds Hospital Trust from 1975 at Cookridge Hospital with special responsibility for paediatric oncology. I noticed a cluster of children with leukaemia, lymphoma and other cancers in the town of Stanley near Wakefield in the late 70s. I approached the newly appointed Epidemiologist Ray Cartwright and on a site visited noticed a profusion of overhead power lines.
We worked with Gerald Draper and I have continued to take an interest in this topic. I tried to contact Gerald by email but the domain wasn’t recognised. I helped in the development of the Children’s Cancer Register
I would appreciate your help in directing me to the current situation regarding this work.
In particular as an honorary member of the the CCLG I am following the Late Effects of Childhood Cancer.
Thank you.
Yours sincerely
Sheila Cartwright BSc MBBS FRCR
Competing interests: No competing interests