Results
For the time-to-event analysis, 1,332,944 children aged ≤15 years on the census date were identified in the Swiss National Cohort database. Of these, 45,590 children with an unclear place of residence were excluded from the analysis (Figure 1). In total, 1,287,354 children with 7,627,646 person-years accumulated during the study period were considered for the analysis. We identified 1,127 cancer cases in the SCCR that were diagnosed between the 2000 census date and 2008 (Figure 1). Of these, 997 could be linked to the Swiss National Cohort database (283 leukemia cases and 258 CNS tumor cases).
(Enlarge Image)
Figure 1.
Linkage of the database of the Swiss Childhood Cancer Registry to that of the Swiss National Cohort (SNC) for a study of radio-frequency electromagnetic fields and childhood cancer, Switzerland, 2000–2008. ALL, acute lymphoblastic leukemia; CNS, central nervous system.
Figure 2 shows the total field levels by distance to the closest transmitter for all residences in the modeled study area. The Spearman correlation between total field levels and distance to the closest transmitter was −0.462 (95% CI: −0.464, −0.460). Eleven percent of all children were exposed to a predicted RF-EMF between 0.05 V/m and 0.2 V/m, and 4% were exposed above 0.2 V/m. From the whole study sample, 51% lived within the modeled area. Arithmetic mean exposure for this sample within the modeled area was 0.14 V/m, with a median value of 0.02 V/m, a 90th percentile value of 0.16 V/m, and a maximum value of 9.77 V/m. Mean exposure was higher in urban areas (0.17 V/m) than in suburban (0.14 V/m) and rural (0.08 V/m) areas.
(Enlarge Image)
Figure 2.
Modeled strengths of radio-frequency electromagnetic fields according to distance from children's households to the nearest broadcast transmitter within the modeled areas, Switzerland, 1997–2000.
Results from the time-to-event analysis are shown in Table 1. Compared with the group of children exposed to a predicted RF-EMF below 0.05 V/m, hazard ratios for the highest exposure category (≥0.2 V/m) were 1.03 (95% CI: 0.74, 1.43) for all cancers, 0.55 (95% CI: 0.26, 1.19) for leukemia, 0.62 (95% CI: 0.27, 1.43) for acute lymphoblastic leukemia, and 1.68 (95% CI: 0.98, 2.91) for CNS tumors when considering all transmitters (Table 1). The linear exposure-response analyses provided a result pattern similar to that of the categorical analyses, although the positive correlation with CNS tumors reached statistical significance for all types of transmitters. The linear analyses indicated that none of the additional potential confounding factors materially altered the hazard ratios (Figure 3). Restricting the analysis to children who were living within the modeled exposure area had virtually no impact on the results (data not shown).
(Enlarge Image)
Figure 3.
Impact of various confounding factors on the hazard ratio for childhood cancer per 0.1-V/m increase in exposure to radio-frequency electromagnetic fields in time-to-event analysis, Switzerland, 2000–2008. Potential confounding factors were added to the full model one at a time. The basic model adjusted for age and sex. The full model additionally adjusted for environmental γ radiation, benzene exposure, distance to the nearest high-voltage power line, and degree of urbanization. The relevance of additional potential confounding factors was tested by including one confounder at a time in the model. Linear adjustment was used for birth order (within each household), radiation, benzene, particulate matter with an aerodynamic diameter less than 10 μm, and nitrogen dioxide. Categorical adjustment was used for distance to highways (<40 m, 40–<100 m, 100–400 m, or >400 m), distance to class 1 roads (<20 m, 20–<50 m, 50–200 m, or >200 m), distance to high-voltage railway lines (<50 m, 50–<200 m, 200–600 m, or >600 m), exposure to agricultural pesticides (<50 m, 50–<100 m, 100–200 m, or >200 m to orchards; <100 m, 100–<250 m, 250–500 m, or >500 m to vineyards; <750 m, 750–<1,500 m, 1,500–3,000 m, or >3,000 m to golf courses), distance to the nearest pediatric medical center (<5 km, 5–<15 km, 15–30 km, or >30 km), domestic radon exposure (50th (77.7 Bq/m and 90th (139.9 Bq/m percentiles), parental socioeconomic status (low, medium, high, or no information), and parental job position (low, medium, high, unemployed/retired/housewife/volunteer work, or no information). Bars, 95% confidence intervals (CIs).
The incidence density cohort analysis accumulated 30.2 million person-years at risk and comprised 4,246 cancer cases, including 971 cases from the time-to-event analyses with geocoded addresses at the time of diagnosis. Results for the whole study period and for period-stratified analyses are shown in Table 2. Again leukemia tended to be negatively associated with predicted RF-EMFs. There was no indication of association between CNS tumor risk and predicted RF-EMF exposure from all transmitters. However, analyses restricted to the period up to 1995 yielded borderline-significant increased incidence rate ratios for all cancers in the high exposure category (incidence rate ratio = 1.23, 95% CI: 0.98, 1.55). For the period after 1995, the corresponding incidence rate ratio was significantly decreased (incidence rate ratio = 0.69, 95% CI: 0.54, 0.87). Stratifying the analyses for leukemia into different age groups that might represent different etiologies did not indicate effect modification by age (Table 3).
The results were similar when we restricted the analyses to VHF and UHF transmitters (see Web Tables 1 and 2, available at http://aje.oxfordjournals.org/). For short- and medium-wave transmitters, hazard ratios in the time-to-event analysis tended to be somewhat higher but not statistically significant, based on few cases, and without indications of a linear exposure-response association.