Stability of Wertheimer–Leeper wire codes as a measure of exposure to residential magnetic fields over a 9- to 11-year interval

Article metrics


The Wertheimer–Leeper (W–L) wire code is a construct used as a surrogate indicator of residential exposure to electromagnetic fields. However, little is known about how changes in electrical distribution systems affect wire code assignment. The wire code was determined for 150 homes in the Seattle, WA, area twice, 9–11 years apart. For each home, the authors evaluated whether the electrical configuration around the home and the wire code changed between the two time points. The effect of wire code misclassification on observable odds ratios was evaluated, given hypothetical true control distributions and two different dose–response curves. There was an electrical configuration change for 77 (51.3%) homes, which resulted in a wire code change for 29 (19.3%) homes. Eight (5.3%) other homes had a wire code change due to mapping errors or methodological inconsistencies. Misclassification masked the shape of a threshold (nonlinear) dose–response curve and changed the slope of a linear dose–response curve. Although the wire code detected less than half of electrical configuration changes, misclassification of exposure over time may change odds ratios and mask possible dose–response relationships.


The Wertheimer–Leeper (W–L) wire code (Wertheimer and Leeper, 1979) is an exposure metric that has been used in many studies of magnetic fields and adult or childhood cancers, producing mixed results for both (Hardell et al., 1995; Ahlbom et al., 2000; Greenland et al., 2000). This metric assigns exposure by inferring the magnetic field strength inside a residence from a visual inspection of all external electrical distribution structures around the home (Wertheimer and Leeper, 1979). The W–L wire code was initially envisioned as a relatively objective measure of long-term exposure that did not require personal recall or contact with study participants. However, case–control studies using a wire code based on the current electrical configuration may not accurately reflect the wire code that would have been assigned during etiologically more relevant time periods in the past (Savitz et al., 1989, 1993; Hardell et al., 1995).

Of primary concern is that changes over time in the electrical distribution system around a residence could lead to misclassification of exposure across wire code categories. In one formal evaluation of wire code stability, Dovan et al. (1993) mapped 81 residences in the Denver, CO, area 5 years apart. They reported that the wire code assignment differed for 10% of residences, primarily due to human mapping errors. Only one difference was due to an electrical configuration change, suggesting that electrical configurations and wire codes were stable over a 5-year span in a large metropolitan area.

Little is known about configuration changes in electrical distribution systems in other geographic areas over a longer time interval and their effect on wire code assignment. This is relevant for interpreting studies of adult cancers with long latency periods and for case–control studies of childhood cancer that ascertain cases retrospectively. The present study evaluates the extent of electrical configuration changes over a 9- to 11-year period and estimates the amount of exposure misclassification resulting from using current wire code configurations to reflect past exposure. Additionally, an attempt was made to determine how this degree of wire code misclassification might impact the odds ratio estimates for disease.


Selection of Homes and Data Collection

The residences in this study were drawn from a population-based case–control study of adult nonlymphocytic leukemia and magnetic field exposure conducted in the Seattle metropolitan area between 1985 and 1986, denoted hereafter as the “original study” (Severson et al., 1988). During the study, 114 cases and 133 controls were interviewed and an attempt was made to map each home occupied by the subject within 15 years prior to the reference date (diagnosis date for cases, interview date for controls). This consisted of drawing a scaled map of all electrical equipment within 140 ft of study residences, while blinded to case–control status. For each electrical structure (e.g., substations, transformers, distribution lines), the distance to the closest part of the home was measured and other relevant information, such as the wire thickness or line voltage, was obtained.

Between 1995 and 1996, a subset of the residences from the original study was mapped a second time, denoted hereafter as the wire code stability (WCS) study. The remapping occurred during the data collection period of a separate case–control study of breast cancer and magnetic field exposure (Breast Cancer Study) in the Seattle area (Davis et al., 2002). For the WCS study, we obtained the address of the home for which the subject reported residing the longest, 3–10 years before the reference date. Of 247 potential homes, complete address information was available and original maps were drawn for 182 residences.

When attempting to remap these 182 homes for the WCS study, technicians were blinded to the case–control status, the wire code assigned by the previous study and the algorithm used to assign the wire code. Scale maps of the wire configuration were drawn using a protocol that followed the procedures in the original study. Despite this, information about mapping secondary distribution lines in the original study could not be located, and therefore, the WCS study employed the protocol used in the concomitant Breast Cancer Study (Davis et al., 2002). In addition, the WCS study was only able to obtain an estimate of the wire thickness of distribution lines; thus, exact wire gauges determined in the original study were collapsed into matching categories.

Since the abstraction protocol for the original maps was unavailable, they were reabstracted using the WCS study protocol to ensure consistency between the two time points. The abstraction process obtained information necessary for assigning the W–L wire code, including the type, distance and characteristics of distributions lines that passed the residence. To ensure quality control, 10% of the maps for both studies was abstracted independently by two technicians.

Data Analysis

For this analysis, the wire code determined in the original study is the exposure classification that would have been found during the historical time period. The wire code determined in the WCS study is the observable exposure classification during study implementation. A difference between the two is defined as an exposure misclassification.

Assessing Electrical Configuration Changes

To better understand electrical configuration stability over time, we assessed, for each residence, whether and how the electrical configuration changed between the two studies. Since the abstracted data did not take into account the spatial orientation of the wires, each map pair was visually inspected. We assessed changes in the thickness or phase of primary distribution lines; the addition, removal, movement or extension of distribution lines; and the replacement of banked secondary distribution lines. We identified the specific configuration difference(s) that led to an electrical configuration change.

For some map pairs, it was unclear whether the apparent movement of a distribution line was due to an actual electrical configuration change or a mapping error. Therefore, a cut point was established to differentiate between the two using data from an interrater reliability study conducted as part of the Breast Cancer Study (Davis et al., 2002). Three technicians measured 31 unique electrical lines from 18 homes. Using the percent difference of each measurement from the mean measured distance for that line, the pooled estimate of the error variance (0.00195) and its standard deviation (0.044) were determined from a one-way ANOVA. A cutoff of two times the standard deviation (0.088) was chosen. Thus, if the percent difference between the distance found in the WCS study and the mean distance of the two studies was less than 8.8%, we concluded that the apparent movement was due to a mapping error, provided that no other evidence indicated an electrical configuration change.

Assessing Wire Code Misclassification

For both time points, one of four standard W–L wire codes was assigned (Wertheimer and Leeper, 1979): very low current configuration (VLCC), ordinary low current configuration (OLCC), ordinary high current configuration (OHCC) and very high current configuration (VHCC). Residences for which a W–L wire code difference existed were further examined to determine why the wire code changed: a mapping error, a methodological difference between original and WCS mapping protocols, or a change in the electrical configuration.

To determine the effect of misclassification on risk estimates, we generated a misclassification matrix relating the wire code distributions between the two studies (Armstrong et al., 1992). Each matrix cell corresponds to the proportion of individuals with wire code j from the original study who were classified as wire code i in the WCS study, where i and j are VLCC, OLCC, OHCC or VHCC. Given hypothetical, true case and control exposure distributions, this matrix can be used to determine observable case and control exposure distributions and corresponding odds ratios. Two misclassification matrices were generated. One included all 150 eligible residences and one excluded eight homes in which the wire code misclassification was due to a mapping error, identified using the distance cutoff of 8.8% described above (four homes), or due to the difference in mapping protocols between the studies (four homes). Results from the two matrices were similar. We present data excluding the eight homes mentioned above, as we were primarily interested in the misclassification effect due to electrical configuration changes.

Several parameters had to be defined to generate hypothetical, true case exposure distributions (see Appendix): the number of cases and controls, the control exposure distribution and the odds ratios for each wire code category, with VLCC as the reference. We applied a 1:1 ratio between cases and controls, using 500 cases. We examined three distinct control wire code distributions based on the literature of adult cancer and magnetic field exposure (Wertheimer and Leeper, 1982; Severson et al., 1988; Wrensch et al., 1999; Davis et al., 2002).

Four forms of a dose–response relationship were examined: (1) linear — a monotonically increasing risk of disease with increasing exposure category; (2) plateau — all exposure categories, beyond the lowest, have a similar excess disease risk; (3) threshold — has little or no excess risk of disease until a certain exposure category is reached; and (4) linear/plateau — a monotonically increasing risk of disease that plateaus at a certain exposure category. Data for plateau and linear/plateau dose–response curves are not presented as the results are similar to that for the linear and threshold dose–response curves, respectively. Similar patterns were found using several sets of odds ratios, so we only present one set of odds ratios.

We determined the observable distribution of cases and controls by multiplying the respective true distributions by the misclassification matrix, assuming nondifferential misclassification. A logistic regression model was used to determine the observable odds ratio point estimates for the OLCC, OHCC and VHCC categories, using VLCC as the reference.


Technicians were unable to map 24 of 182 eligible residences from the original study. Most (n=20) were classified as VLCC or OLCC in the original study. Reasons homes could not be mapped included: the home no longer existed (7 homes), had long driveways with “no trespassing” signs (4 homes), could not be located (12 homes) or unknown reason (1 home). Eight residences were excluded because the original study map had missing data needed to determine the W–L wire code: wire thickness or size (three homes) or information about whether an electrical line was active (five homes). An additional eight homes were excluded because the wire code change was not due to an electrical configuration change (see Methods section).

Of the remaining 142 homes, an electrical configuration change was found for 77 (54.2%) homes (Table 1). Changes included the addition, removal, extension or movement of a distribution line (59 homes); changes in the thickness or phase of a distribution line (7 homes); replacement of 4-kV banked systems (8 homes); and modifications to the residence structure that changed its distance to the electrical system (3 homes). Although the majority of changes were due to the addition, removal, extension or movement of a distribution line, we found no clear pattern that explained the distribution of the electrical configuration changes.

Table 1 Type of electrical configuration change between the original and WCS study, stratified by the magnitude and direction of the wire code change

Twenty-nine (37.7%) of the 77 homes with an electrical configuration change also had a wire code change between the original and WCS study. Most wire code changes were due to the addition, removal, extension or movement of a distribution line. Nineteen (65%) of the 29 homes with a wire code change had an increase in wire code category from the original to the WCS study. The majority moved into adjacent wire code categories. Five of the six that changed into a nonadjacent category had an increase in wire code classification over time.

About one-third of the homes originally classified as VHCC were misclassified as OHCC in the WCS study (Table 2). Over 25% of homes classified as OLCC in the original study was misclassified in the WCS study, with about equal proportions misclassified to higher or lower wire code categories. A moderate proportion (17.3%) of homes originally classified as VLCC was misclassified in the WCS study.

Table 2 Proportion of homes classified in each wire code category in the WCS study by the original classification

Misclassification masked the true shape of the threshold dose–response curve (Figure 1). For the true control distributions examined, the observable threshold dose–response curves were flattened compared to the true curve. This was because the observable odds ratio for the OHCC category moved away from the null. For the linear dose–response curve, the observable slope was different than the true slope; however, the linear trend was maintained. The slopes of observable linear dose–response curves decreased by 11.4–31.4%, compared to the true slope of 0.35.

Figure 1

True and observable odds ratios for three true control distributions by wire code exposure, with VLCC as the reference category, for a threshold and linear dose–response curves. (——) True odds ratio; (–--) 35% VLCC, 25% OLCC, 25% OHCC, 15% VHCC; (—□--) 15% VLCC, 55% OLCC, 25% OHCC, 5% VHCC; (—×--) 65% VLCC, 15% OLCC, 15% OHCC, 5% VHCC.


The purpose of this study was to determine the extent of changes in residential electrical configurations and the resulting effect on W–L wire codes, over a 9- to 11-year period. We examined how this misclassification may impact the observable wire code distributions and odds ratios given several true control distributions and dose–response relationships. Substantial electrical configuration changes occurred between the original and WCS studies. While we cannot pinpoint why such changes occurred, the Seattle metropolitan area was growing rapidly between the original and WCS studies. Although fewer than 40% of electrical configuration changes resulted in a wire code change, this misclassification impacted the observable odds ratios and dose–response trends.

The W–L wire code appears to be somewhat insensitive to electrical configuration changes. One possible explanation may be that since the wire code assignment is based on the electrical line conferring the highest presumed magnetic field exposure inside the house, a change in another line would not cause a change in the assigned wire code. Alternatively, if the electrical line determining the wire code moved between — but did not cross — the cutoff distances for wire code categories, no wire code change would occur. Thus, although the wire code appears to remain stable over time, its insensitivity to electrical configuration change suggests that it may not be measuring ambient magnetic fields inside the home well. This is of particular concern, since the wire code only accounts for 15–20% of the variance of magnetic fields in the home (Savitz et al., 1989, 1993; Ahlbom et al., 2000). However, no information exists about how magnetic fields inside the home are affected by electrical configuration changes around the residence.

We found that a substantial proportion of homes originally classified as OHCC and VHCC was misclassified in the WCS study; this is similar to the findings of Dovan et al. (1993). We also found that, of the four wire code changes due to a mapping error in our study, three were originally classified as OHCC. This suggests that homes with a high wire code may be more susceptible to misclassification, perhaps because the electrical configurations are complex and thus more measurements are required. Another study examined potential misclassification due to technician or mapping errors over a short time period (1 month or 1 year), reporting a high agreement between the two study mappers (Tarone et al., 1998). This indicates that mapper error may be a small source of error; our results support this conclusion as the wire code misclassification for only four homes was due to a mapping mistake.

Exposure misclassification changed the slope of a linear dose–response curve and masked a threshold (nonlinear) dose response. Statistical simulation studies have found that under extreme misclassification with three exposure levels, dose–response relationships can be inverted or distorted when observed odds ratios shift away from unity in intermediate exposure categories (Dosemeci et al., 1990; Birkett, 1992; Correa-Villasenor et al., 1995). This occurred for the observable OHCC risk estimate in the threshold dose–response curve, slightly distorting the true shape. The results of this analysis have two important implications. First, they suggest that it may not be possible to observe the true dose–response relationships in studies using wire codes because misclassification can either mask nonlinear or change linear relationships. Second, risk estimates associated with intermediate wire code categories may not necessarily be conservative, even if assuming nondifferential exposure misclassification, which usually attenuates the observable odds ratios.

The present study has several limitations. First, many homes not mapped in the WCS study were classified in the lowest exposure categories in the original study. Most of these homes could not be mapped because they no longer existed or could not be located, suggesting that such homes may have been located in areas experiencing many changes in the physical layout of streets and buildings as well as in electrical configurations. If so, we may be underestimating the amount of misclassification in these lowest wire code categories. Second, the visual inspection of the map pairs to assess electrical configuration change was subjective and sometimes was based on knowledge about patterns of electrical equipment used in the Seattle area. Although every effort was made to be conservative and obtain a consensus opinion between two coders, it is possible that the changes were either over- or underestimated. Third, the cutoff for differentiating between mapping errors and actual electrical configuration change was based on an interrater reliability study using technicians from the WCS study, which likely underestimates intermapper error between the original and WCS technicians. Thus, the cutoff we used to differentiate mapper error and true electrical configuration change may be too small, leading to a possible overestimation of electrical configuration change. The determination of electrical configuration change depended on this cutoff for 11 residences (14.3% of homes found to have an electrical configuration change). Fourth, the amount of misclassification of wire codes may not be generalizable to rural areas or other time periods.

Despite these limitations, this study has several strengths. First, we used similar mapping and abstraction protocols for both time points. Second, we were able to look at changes in wire code and electrical configuration in a different geographic region and over a longer time period than previous studies. This interval is likely to be more relevant for cancers with long latency periods or retrospective study designs.

Interpreting studies of cancer and wire code has been difficult because studies of both adult and childhoood cancers are heterogeneous (Wertheimer and Leeper, 1982; Severson et al., 1988; Wrensch et al., 1999; Davis et al., 2002; Greenland et al., 2000). Our results may explain some of these inconsistencies, particularly for adult cancers since they tend to have long latency periods. Error due to exposure misclassification may explain null findings since nondifferential misclassification generally attenuates the observed odds ratio. Further, variability may also be introduced because nondifferential misclassification sometimes can lead to nonconservative odds ratios in the intermediate wire code categories, particularly if misclassification in the reference category is low, but is moderate to high in the upper levels (Correa-Villasenor et al., 1995). Our study and that of Dovan et al. (1993) found more misclassification in the higher exposure categories, suggesting that the assumption of conservative risk estimates is likely to be incorrect.

The possible insensitivity of the wire code scheme to electrical configuration changes suggests that this exposure metric may not accurately represent magnetic field levels inside the home, which is the underlying exposure of interest. This would imply that using the wire code introduces further misclassification with respect to the true magnetic field exposure. Understanding how electrical configuration changes affect measured magnetic fields in the home would further elucidate the appropriateness of the wire code metric.

The degree of exposure misclassification estimated in this study can be used to determine whether previously published studies had sufficient power to observe an association. Further, information about the misclassification in wire codes over time, in combination with data about how wire code exposure levels correlate with magnetic fields in the home, may allow the calculation of adjusted risk estimates associated with magnetic fields. However, given the potential problems of using the wire code metric, the development of improved methods for better measuring historical magnetic field exposure will likely advance knowledge in this area more than further studies using the wire code.

Further reading