GIS Model of Power Lines Used to Study EMF and Childhood Leukemia

What's New in GIS in Public Health
http://www.atsdr.cdc.gov/GIS/conference98/news/jan2000.html
January 2000

by Joseph Bowman, National Institute for Occupational Safety and Health

A GIS model of household magnetic fields from power lines has shown a significant association with childhood leukemia where exposure measurements showed none (1,2). This study by a collaboration between NIOSH and the University of Southern California (USC) sheds new light on the difficult question of whether electric and magnetic fields (EMF) from AC electricity cause cancer (3). This GIS wiring model can be used in other epidemiologic studies where residential EMF is a possible risk factor, such as the Long Island Breast Cancer Project (see Public Health GIS News and Information, November 1999).

The suspected link between EMF and cancer has been a scientific mystery and the subject of contentious debate for nearly two decades. Part of the problem has been exposure assessment. The original report by Wertheimer and Leeper (4) assessed exposures with a simple "wire code" that crudely predicted residential magnetic fields from observable configurations of the electrical lines around the subject's homes. When better funded studies took household measurements, childhood leukemia risks were more strongly associated with wire codes than with the EMF measurements. This "wire code paradox" led to a decade-long debate over the meaning of the observed associations. Were they artifacts due to study flaws or undetected leukemogens correlated with electric lines? Or was household EMF actually a carcinogen, but measurements failed to show an association due to exposure assessment errors and selection biases that wire codes somehow avoided?

To approach this wire code paradox in a new way, the NIOSH/USC collaboration proposed a wiring configuration model that might predict long-term magnetic field exposures better than either wire codes or measurements. The model was based on the formulas of electromagnetism where the unknown parameters were determined by a regression against household measurements. The hypothesis was that magnetic field exposures predicted by the model should be more strongly associated with leukemia than either measurements or wire codes. Data to test this hypothesis was available from an EMF-leukemia study in Los Angeles County which used both wire codes and 24-hour measurements of the magnetic fields in the child's bedroom (5).

The key to this EMF exposure model was a GIS procedure for analyzing the wiring around the homes where the subjects had lived since conception. The USC team had originally used GIS in the late 1980s to determine wire codes objectively. Field technicians had sketched maps of all electrical lines around the subject's homes and measured relevant distances (see figure). From electric utilities in the area, they obtained scaled maps of the neighborhood electric lines which had detailed information on voltage, wire thickness and other parameters relevant to magnetic field production.

These wiring and geographic data were entered into an early GIS program called Facility Mapping System with AutoCAD (Facility Mapping Systems, Mills Valley, CA). FMSAC then produced scale maps with the attribute data attached to the lines, poles, transformers, and the subject's residence. Since the electric utilities had not yet started using GIS for their wire maps, the maps had to be traced on a computerized tablet, and other data entered by hand. A standardized coding manual and rigorous quality control measures were crucial to successful data entry. In the end, FMSAC could extract a database for each residence that contained all the data needed for both wire coding and the more sophisticated exposure model.

The exposure model was based on the Law of Biot and Savart, the equation for the magnetic field at a distance from a wire carrying an electric current (1). From electrical engineering principles, we derived a reasonable formula for the magnetic fields in the child's bedroom. All the necessary parameters in this formula were provided by our GIS data except for the current in the lines. The unknown currents were assumed to be linear functions of relevant wire configuration parameters. with coefficients to be determined by regression against 24-hr average magnetic fields measured in the child's bedroom. To calculate these empirical current functions, Prof. Duncan Thomas (USC) developed a stepwise non-linear regression procedure.

For the regression, the GIS databases were merged with measurement data on the homes where we had obtained access. The study design attempted EMF measurements only at the residence where the subject lived the longest, and access was denied at many such residences. So wiring data was available was available for 709 residences, but measurements were taken in only 315 homes. (This was a potential source of selection bias in the original study.) The model's predictions produced a bootstrap correlation of 0.40 with the measured fields, an improvement on the 0.27 correlation obtained with the wire code.

 The risk analysis from the case-control data was then repeated with the predicted magnetic fields (2). Although the measured fields had no association with childhood leukemia (p for trend = 0.88), the risks were significant for the highest predicted magnetic fields (OR = 2.00, 95% CI = 1.03-3.89), and a significant dose-response was seen (p for trend = 0.02). When exposures were determined by an empirical Bayes combination of predictions and measurements, the odds ratio (OR = 2.19, 95% CI = 1.12-4.31) and the trend showed somewhat greater significance (p=0.007).

These findings support the hypothesis that magnetic fields from electrical lines are causally related to childhood leukemia, but that this association has been inconsistent among epidemiologic studies due to different types of exposure assessment error. This result bolsters the conclusions of a recent U.S. risk assessment which found EMF to be a possible carcinogen (3).

The GIS wire configuration model appears to assess the leukemia risks from a child's long-term residential magnetic field exposures better than the 24-hr measurements. One reason is that the model can be assess exposures with more subjects and more previous residences because the maps do not require access to homes. This increases the study's power and reduces the potential for selection bias.

This wiring model should also be better for retrospective exposure assessments since electric lines in residential neighborhoods seldom change. In contrast, EMF measurements are strongly influenced by short-term fluctuations in electrical usage which create errors in assessing long-term average exposures. The regression over many residences tends to average out such fluctuations. Where measurement data is available, the empirical Bayes estimator combines the advantages of measurements and modeling.

Our exposure model can be applied to other studies with far less work than the original study. In the decade since the USC team collected its data, GIS technology has completely taken over mapping at electric utilities. Instead of laboriously visiting residences and tracing maps, a wiring configuration model can be developed today by merging GIS databases. The only active data collection needed is magnetic field measurements in homes. In order to apply our model to new electrical service areas, regression against local measurements is desirable because the parameters depend on the engineering details of the electrical distribution system.

The Long Island Breast Cancer Study Project (LIBCSP) appears to be a perfect setting for the wiring model. EMF has been implicated as a breast cancer risk factor by animal toxicology, cellular studies and some preliminary epidemiology (3). LIBCSP is therefore measuring magnetic field exposures and mapping adjacent electric lines in a subset of 1200 subjects. In addition, NCI has just contracted for a $4.8 million GIS system which will incorporate data on many possible environmental risk factors. According to the GIS contractor's Internet site (6), their data sources include high-voltage powerlines (from the US Geologic Survey) and electric distribution lines (from the Long Island Light Co). When the GIS system is complete in a couple years, all the data needed to apply the wiring model to the LIBCSP should be readily available.

A study with the GIS exposure model would be especially important if the present breast cancer study finds no association with EMF measurements, raising the specter of the wire code paradox. In that case, the wiring configuration model could clarify the assessment of EMF's breast cancer risks. [Contact: Dr. Bowman, Radiation Section, Division of Biomedical and Behavioral Sciences, Cincinnati, OH at voice (513) 533-8143 or email jdb0]

References

1.Bowman JD, Thomas DC, Jiang L, Liang F, Peters JM. Residential magnetic fields predicted from wiring configurations: I. Exposure model. Bioelectromagnetics, 20:399-413, 1999.

2. Thomas DC, Bowman JD, Jiang L, Liang F, Peters JM. Residential magnetic fields predicted from wiring configurations: II. Relationships to childhood leukemia. Bioelectromagnetics, 20:414-422, 1999.

3.National Institute of Environmental Health Sciences. NIEHS Report on Health Effects from Exposure to Power-line Frequency Electric and Magnetic Fields. NIH Report No. 99-4493, 1999. (Also available at http://www.niehs.nih.gov /emfrapid/.)

4.Wertheimer N, Leeper E. Electrical wiring configurations and childhood cancer. Am J Epidemiol. 109:273-284, 1979.

5.London SJ, Thomas DC, Bowman JD, Sobel E, Cheng TC, Peters JM. Exposure to residential electric and magnetic fields and risk of childhood leukemia. Am J Epidemiol 134: 923- 937, 1991.

6.The Geographical Information System for the Long Island Breast Cancer Study Project. Internet address: http://www. healthgis- li.com.