REVOLT opposing unnecessary, excessive and intrusive powerline development

opposing unnecessary, excessive and intrusive powerline development

REVOLT Newsletter 190

Revolt news 12/06/2005


Revolt did not issue a press release at the time of publication of the Draper study on 4 June. Instead this considered and detailed response is issued after scrutiny of the full original paper, and with apologies for the detail. Summative conclusions are given at the end of the response.

***** *****

Comments on Draper et al, BMJ vol. 330, 1290 - 1293, 4 June 2005.


These comments, circulated with Revolt email news, aim for accuracy and balance, not for campaigning goals. Revolt's position statement, and commendations by public inquiry inspectors, reflect that approach. Any commentators may be influenced by their background. These comments are primarily from the viewpoint of a chartered mathematician after scrutiny of the full original paper. After more detailed comments, my conclusions are summarised at the end.

The Draper study, reported in BMJ last week, was heralded by some campaigners as a major study likely to change the case for precaution, or even for a ban, on exposures from powerlines. Because of the long delay in reaching publication, concerns were expressed that the results were being withheld. While it is a major study in scale, some of the statistics are weak, particularly in the newer findings at greater distances from powerlines, and are not independently replicated. They therefore do not change the case for precaution as radically as some campaigners suggest.

On the other hand some respondents have seized upon those results at greater distances (up to 600 metres), where magnetic fields may be negligible, to disparage any hypothesis for a magnetic field effect. Such disparaging claims do not stand up to mathematical scrutiny and such weak results cannot be relied upon in this way. Notwithstanding possibly negligible fields at 600 m, such claims also overlook the possibility of associated exposure in time spent closer to the line, for example at nursery or school. Further, while hastily relying on weak statistical results, such claims dismiss the stronger statistical association with childhood leukaemia established for flux densities above 0.4 microTesla, which are consistent with and broadly reinforced by the Draper study findings in the 0 - 200 m distance range.

How results may be diluted and potential effects understated

Some features limit the ability of the study to reach more significant conclusions and may tend to dilute results and lead to understatement, rather than better identification, of any underlying causal factors.

The exposure metric in this study is proximity of birth address to power lines. The many studies and meta-analyses finding generally positive associations have used different metrics, with field level represented by peak, spot, 24-hour average and time-weighted average field values up to three months or a year before diagnosis, as well as proximity to electrical sources. If there is an effect or set of effects, then using the most appropriate metrics should give sharper results.

Uncertainty in type of field metric is compounded by uncertainty in the timing of exposure. Genetic evidence suggests, in many cases, a two- stage causal process of in-utero genetic damage followed by conversion in childhood to the disease. Address at birth may be correlated with exposure in utero, and perhaps to a lesser extent with exposure in childhood. Uncertainties in these correlations (which are not addressed in the paper) would again tend to dilute causal evidence.

It would be better in general to study the relation of exposure in utero with incidence of genetic damage (identifiable by blood tests), and, separately, the relation of exposure prior to diagnosis in children with genetic damage to the incidence of the disease, though this approach was not available within the scope of the present study.

Whereas other studies and meta-analyses have provided suggestive evidence of a causal link to childhood leukaemia from EMF exposure shortly prior to diagnosis, this study provides a statistical association with EMF exposure at the birth address, and therefore an indicator towards a possible causal link with pre-natal EMF exposure. This much is largely (but not entirely) new, but given the uncertainties with the metric and timing, such an indication is also uncertain.

Statistical features

The results declare a significant (P < .01) trend in risk with inverse distance. However, Table 1 shows some odd features in the data. Firstly, the relative risk (RR) is more like a step function with distance (suggesting a possible exposure threshold) than an inverse power relation. The authors did not test for that. Secondly, within the 200 - 600 metre range there is a strange counter-trend; as results in this range are barely statistically significant (CI 1.02 - 1.49) this suggests chance variations.

Within the 200 - 600 metre range the largest RR (1.36) is found at the outer band of 500 - 599 metres. The weakness of association in this range, together with the counter-trend, again suggests chance rather than there being a local peak (or indeed any) effect beyond 500 metres. This is not to deny the possibility of such an effect, but simply to note that these results do not particularly support it, given all the uncertainties of the metrics and statistics.

Therefore the paper does not, in my view, justify a conclusion about effects up to 600 metres. At the same time, the study does not rule out such a possibility. Professor Henshaw's response is to be noted in this connection: that charged particles arising from corona discharge from powerlines have been found at greater distances and can increase deposition of pollutants in the lungs, although the size and impact of any consequent effect on leukaemia incidence has not been clearly established.

The authors fairly say: "If the association is causal, about 1% of childhood leukaemia in England and Wales would be attributable to these lines, though this estimate has considerable statistical uncertainty". That is what leads to about 5 attributable cases per year, some ten times higher than suggested in previous studies (notably UKCCS). The 1% of cases seems to be derived from a slightly underestimated relative risk of about 1.25 (of which 0.25 is attributable) on the 4% of children living within 600 metres.

Discounting the contribution from the most uncertain range of 200 - 600 metres would leave a population of about 1.3 % of children (it may be a bit lower as population density may be less very close to powerlines) within 200 metres with a relative risk of about 1.7 of which 0.7 is attributable. That still amounts to about 0.9 % of cases, which would only reduce the estimated attributable outcome to 4.5 cases per year, or, within the broad approximation and uncertainty of these figures, still around 5 cases per year. So the stronger statistical findings in the range 0 - 200 metres alone still support the increased attribution.

Looking further at Table 1, there are large variations in the distributions of the three sets of controls, even with large numbers in each set. The controls need not be representative of population density, except for population of controls matching the particular disease cases, and therefore the three sets may properly differ in distribution due to different matching characteristics. However, the paper does not identify any such differences in matching characteristics between the three sets of cases; indeed all diseases are matched for sex, approximate date of birth and birth registration district, none of which would seem likely to lead to large differences in the distributions.

Controls in this study are not matched for age (e.g. at diagnosis) as they might be in studies of pre-diagnosis exposure. The calculated relative risks are for lifetime incidence of the disease. But there is no obvious reason why that should have a large effect on distributions of controls.

The controls within 200 metres of power lines, compared as numbers per 10,000 controls, were respectively 40, 68 and 52 for the three disease sets. The relatively low number of controls (40) for leukaemia enhances the relative risk found, but even if the average control density of 53 per 10,000 were used, there would still be a relative risk of about 1.3, so maldistribution of controls would not wholly explain the finding. The authors say: "Comparison of the leukaemia cases with the non-leukaemia controls still suggests that there is an increased risk for leukaemia but it is much lower than that found using the matched controls. However the use of the matched controls is the most appropriate approach."

These differences in distributions of controls may therefore indicate the scale of chance or methodological differences, showing that they are comparable with the observed trends of association. Indeed, a spurious "protection" effect of low relative risks is observed for CNS/brain tumours within 200 metres of the powerlines. All this underlines the uncertainty of the results, albeit some of them are statistically significant.

The authors are to be commended on stating plainly "there is an association between ...", whereas some past papers and some responses have been unwilling to accept an association (albeit plain by mathematical definition). As with the 0.4 microTesla doubling of risk, the association is definite, but the question of cause remains with some uncertainty.

Where the authors say "Thus our results do not seem to be compatible with the existing data", it should be clear from the context that this applies only to the range 200 - 600 metres. Such incompatibility underlines the statistical weakness in that range, but should not be seen as undermining previous firm associations at higher exposures.

Mechanism and causation

The authors also say "There is no accepted biological mechanism to explain the epidemiological results; indeed the relation may be due to chance or confounding". True enough, although much hangs on the definition of "accepted", as some scientists would accept some identified mechanisms as plausible for a causal hypothesis. It would have been fairer to say also that (1) there is a body of scientific evidence to support potential biological mechanisms; (2) while these particular results contain many uncertainties, taken together with the previously established associations, they are mutually reinforcing in underpinning the case for a probably understated causal role of EMFs, while emphasising the need for more focused studies.


1. The study is important in that it is on a large scale and deals with proximity of birth address to powerlines. In contrast, other key studies, which lie behind the established statistical association of childhood leukaemia with magnetic flux density exposure above 0.4 microTesla, refer mainly to exposures in childhood prior to diagnosis. The extent that this study might represent exposure to EMF in utero or in childhood prior to diagnosis is unclear.

2. The study finds statistically significant results of two kinds. First there are stronger results for birth addresses within 200 metres of a power line. Second there are weaker results in the range 200 - 600 metres, which on close inspection show statistical quirks and are likely to be spurious. The authors recognise the statistical weakness but do not point out the quirks.

3. The results within 200 metres are consistent with and broadly reinforce the established data doubling the risk of childhood leukaemia for children exposed prior to diagnosis above 0.4 microTesla. However, they suggest the number of attributable cases from high voltage power lines in the UK would be about 5 per year, some ten times more than previous estimates. This may be a reflection of a greater effect of pre- natal exposure compared with pre-diagnosis exposure, and/or a better estimate of numbers exposed, but this is not clear. There are uncertainties in both estimates.

4. The results in the range 200 - 600 metres are likely to be spurious. They should not be relied on to support or deny an effect at 600 metres nor to claim that the greater population up to that distance is at risk. Further the argument that these results are incompatible with magnetic field levels should not be relied on to dismiss magnetic field hypotheses nor to counter the established statistical association with magnetic flux densities.

5. The uncertainty between pre-natal and pre-diagnosis exposure, noting the evidence for in utero genetic damage being a common precursor for childhood leukaemia, tends to dilute results and understate any potential underlying causation. Better focused studies on the two stages would be helpful and could be much more robust statistically.

Post script

Press releases last week from several interested organisations (government advisory, research labs, academic, cancer charities, professional) contrasted in their responses to this study. Light heartedly, I found myself marking them against the following criteria: * accuracy and avoidance of error or spin; * relation to context and other studies and avoidance of deviation; * overall balance and perspective; * clarity and style.

Pleasingly the top mark was won by the HPA radiation division, the former NRPB. I have criticised them in the past for schoolboy howlers of scientific error and for hasty and unbalanced knee-jerk responses, as well as for spin. This response was careful and well balanced, and put well in context. Only once did it refer to "the possible association" when the association is a matter of statistical fact; Draper et al called their association just that, not a "possible association". But that minor shortcoming was offset by many good points.

The very short IEE press release received the lowest score. It seemed to zoom in on the 600 metre results without appreciating their statistical weakness and quirks, and without recognising the difference in reliability between the results within 200 metres and those up to 600. Then it concluded that "the study points away from magnetic fields from power lines as a cause of childhood leukaemia", which seems to place undue reliance on the spurious results while ignoring the stronger statistics within 200 metres and the still stronger statistics of association with field levels.

Mike O'Carroll 12.6.05




Custom Search

Search the web

Custom Search