Broadcast Transmitters and Risk of Childhood Cancer
Broadcast Transmitters and Risk of Childhood Cancer
The study was based on data from the Swiss Childhood Cancer Registry (SCCR) and the Swiss National Cohort. The SCCR includes cancer patients aged less than 21 years at diagnosis. For patients under age 16 years at diagnosis, at least 95% of incident cases are registered. The Swiss National Cohort is a database containing data on all Swiss buildings, households, and persons. It is based on probabilistic record linkages of census data sets from 1990 and 2000 with each other and with national birth, mortality, and emigration data sets. Participation in the Swiss census is compulsory, and coverage for the 2000 census was estimated to be 98.6%.
We considered all cancer diagnoses made in Switzerland classified according to the International Classification of Childhood Cancer, Third Edition (ICCC-3), with a special focus on leukemia (ICCC-3 code I), acute lymphoblastic leukemia (ICCC-3 code I.a), and CNS tumors (ICCC-3 code III), including benign tumors.
Time-to-event Analysis. We used 2 strategies to analyze the data: a time-to-event analysis and an incidence density cohort analysis. For the time-to-event analysis, we included children who were under age 16 years and living in Switzerland on the date of the 2000 census (December 5, 2000). Time at risk started on the date of the census and lasted until the date of diagnosis, death, emigration, the child's 16th birthday, or December 31, 2008, whichever occurred first. Incident cancer cases in the Swiss National Cohort were identified by means of a probabilistic linkage with the SCCR using information on date of birth, sex, place of residence, place of birth, and parents' birthdates if available. The resulting data set contained the diagnosis date of cancer cases and information on potential confounders for all study participants: sex, birth order (within each household), socioeconomic status of the parents (highest education, socioprofessional category), and geospatial data for place of residence on the census date.
Incidence Density Cohort Analysis. For the incidence density cohort analysis, no linkage between SCCR and Swiss National Cohort data was necessary. We included in this cohort all SCCR-registered patients diagnosed between January 1985 and December 2008 and residing in Switzerland at the time of diagnosis. For a Poisson regression analysis, person-years at risk accrued during a census year (1990, 2000) were calculated for each cell of a cross-tabulation between exposure categories, sex, and 1-year age strata. Cell-specific person-years for noncensus years were then estimated by inter-/extrapolation from corresponding values in the census years, with adjustments for national population-level changes by sex and age, which were known for all years. Details on this procedure are provided by Spycher et al..
For this study, we focused on broadcast transmitters emitting medium-wave (0.5–1.6 MHz), short-wave (6–22 MHz), very high frequency (VHF; 174–230 MHz), and ultra-high frequency (UHF; 470–862 MHz) EMFs, which includes analogous television transmitters (VHF and UHF bands), terrestrial digital audio broadcast transmitters (VHF band), and digital terrestrial video broadcast transmitters (UHF band). All models considered antenna height, transmission duration, the horizontal and vertical directions of the emissions, and local topography. We included all VHF and UHF transmitters in Switzerland with an output power of more than 100 kW (11 transmitters), as well as transmitters with an output power between 10 kW and 100 kW if more than 30,000 persons lived within a 5-km radius (11 transmitters). Population density was considered as a selection criterion because transmitters in a highly populated area may cause relevant exposure, whereas remote transmitters (mainly in the alpine region) were not expected to be relevant for population exposure. RF-EMF levels from these transmitters were modeled by the Federal Office of Communications for an area with a radius of 10 km around each transmitter for the years 1990 and 2000. For the modeling, the Institut für Rundfunktechnik 2-dimensional (IRT_2d) model was applied using CHIRplus_BC software from LS Telcom (Lichtenau, Germany).
RF-EMF exposure levels from all Swiss short- and medium-wave radio transmitters with an output power greater than 1 kW (9 transmitters) were modeled on the basis of the Fresnel Deygout method using ICS-Telecom software from ATDI (Paris, France). For these transmitters, modeling was carried out within a radius of 20 km for the years 1993 and 1997. For overlapping modeled areas, the exposure levels of all transmitters were summed.
In the time-to-event analysis, RF-EMF exposure to radio and television transmitters at baseline was assessed for each study participant at the place of residence using the modeled RF-EMFs from 2000 and 1997, respectively. In the incidence density cohort analysis, place of residency on the date of diagnosis was used for exposure assignment. For children diagnosed before 1995, exposure assessment was based on the models for 1990 and 1993. Thereafter, RF-EMF exposure was assessed using the modeled RF-EMFs from 2000 and 1997, respectively.
Geospatial data on potential confounders were extracted from digital maps using ArcGIS (ESRI, New York, New York), based on the place of residence. Data on background γ radiation were available from the Swiss radiation maps with a grid cell resolution of 2 km. Digital maps with power lines with a resolution of 1:25,000 were provided by the Federal Inspectorate for Heavy Current Installations. We extracted distances to the traffic network in 2000 from digital maps on the traffic network with a resolution of 1:25,000 (VECTOR25-maps), published by the Federal Office of Topography. Data on distances to the nearest orchards, vineyards, and golf courses, for the estimation of exposure to agricultural pesticides, were obtained from the Swiss land-use statistics (Arealstatistik Schweiz) for the year 1997, published by the Swiss Federal Statistical Office, with a grid cell resolution of 100 m. We geocoded the location of the pediatric cancer centers manually. Data on ambient benzene, particulate matter with an aerodynamic diameter less than 10 μm, and nitrogen dioxide exposure were available from a digital map with a grid cell resolution of 100 m (benzene: 400 m), published by the Swiss Agency for the Environment, Forests and Landscape. Residential radon exposure was estimated from a nationwide radon prediction model.
For the time-to-event and incidence density cohort analyses, the same RF-EMF exposure categories were used with a priori chosen cutpoints at 0.05 V/m and 0.2 V/m to differentiate between low, medium, and high exposure. All study participants living in an area not covered by the modeling were included in the reference category. A cutpoint of 0.05 V/m for the reference category was chosen because this value is unlikely to be exceeded due to broadcasting outside the modeling area. A cutpoint of 0.2 V/m for high exposure corresponds roughly to the first quartile of the study population being exposed to RF-EMF levels of more than 0.05 V/m. For short- and medium-wave transmitters, the exposure variable was dichotomized at 0.05 V/m because of the lower levels.
In addition to categorical exposure classification, we also carried out linear exposure-response modeling in the time-to-event analysis using exposure to the broadcast transmitters as a continuous predictor and expressing the hazard ratio per 0.1-V/m increase in exposure. For these analyses, exposure levels outside the modeled area were set to 0.001 V/m.
For the time-to-event analysis, Cox proportional hazards regression models were applied using age as the underlying time scale. Period effects were considered by splitting the follow-up time into two 4-year blocks. The basic models always included adjustment for sex. Furthermore, we decided a priori to adjust for exposure to the potential leukemia risk factors benzene, natural background ionizing γ radiation, distance to the nearest high-voltage power line, and degree of urbanization. We tested the relevance of additional potential confounding factors in the time-to-event analysis by including one confounder at a time in the model and applying a change-in-estimation criterion of 10%. We also conducted a sensitivity time-to-event analysis that excluded all children not living in an area covered by the exposure modeling (i.e., >10 km or >20 km from any transmitter).
For the incidence density analysis, we conducted a Poisson regression analysis that adjusted for sex, age, and calendar year. Separate analyses were conducted for the period up to 1995 and the period after 1995. Results of the incidence density analysis for leukemia were also stratified by age, using 1 and 6 years of age as cutpoints.
Methods
Study Population
The study was based on data from the Swiss Childhood Cancer Registry (SCCR) and the Swiss National Cohort. The SCCR includes cancer patients aged less than 21 years at diagnosis. For patients under age 16 years at diagnosis, at least 95% of incident cases are registered. The Swiss National Cohort is a database containing data on all Swiss buildings, households, and persons. It is based on probabilistic record linkages of census data sets from 1990 and 2000 with each other and with national birth, mortality, and emigration data sets. Participation in the Swiss census is compulsory, and coverage for the 2000 census was estimated to be 98.6%.
We considered all cancer diagnoses made in Switzerland classified according to the International Classification of Childhood Cancer, Third Edition (ICCC-3), with a special focus on leukemia (ICCC-3 code I), acute lymphoblastic leukemia (ICCC-3 code I.a), and CNS tumors (ICCC-3 code III), including benign tumors.
Time-to-event Analysis. We used 2 strategies to analyze the data: a time-to-event analysis and an incidence density cohort analysis. For the time-to-event analysis, we included children who were under age 16 years and living in Switzerland on the date of the 2000 census (December 5, 2000). Time at risk started on the date of the census and lasted until the date of diagnosis, death, emigration, the child's 16th birthday, or December 31, 2008, whichever occurred first. Incident cancer cases in the Swiss National Cohort were identified by means of a probabilistic linkage with the SCCR using information on date of birth, sex, place of residence, place of birth, and parents' birthdates if available. The resulting data set contained the diagnosis date of cancer cases and information on potential confounders for all study participants: sex, birth order (within each household), socioeconomic status of the parents (highest education, socioprofessional category), and geospatial data for place of residence on the census date.
Incidence Density Cohort Analysis. For the incidence density cohort analysis, no linkage between SCCR and Swiss National Cohort data was necessary. We included in this cohort all SCCR-registered patients diagnosed between January 1985 and December 2008 and residing in Switzerland at the time of diagnosis. For a Poisson regression analysis, person-years at risk accrued during a census year (1990, 2000) were calculated for each cell of a cross-tabulation between exposure categories, sex, and 1-year age strata. Cell-specific person-years for noncensus years were then estimated by inter-/extrapolation from corresponding values in the census years, with adjustments for national population-level changes by sex and age, which were known for all years. Details on this procedure are provided by Spycher et al..
Exposure Assessment
For this study, we focused on broadcast transmitters emitting medium-wave (0.5–1.6 MHz), short-wave (6–22 MHz), very high frequency (VHF; 174–230 MHz), and ultra-high frequency (UHF; 470–862 MHz) EMFs, which includes analogous television transmitters (VHF and UHF bands), terrestrial digital audio broadcast transmitters (VHF band), and digital terrestrial video broadcast transmitters (UHF band). All models considered antenna height, transmission duration, the horizontal and vertical directions of the emissions, and local topography. We included all VHF and UHF transmitters in Switzerland with an output power of more than 100 kW (11 transmitters), as well as transmitters with an output power between 10 kW and 100 kW if more than 30,000 persons lived within a 5-km radius (11 transmitters). Population density was considered as a selection criterion because transmitters in a highly populated area may cause relevant exposure, whereas remote transmitters (mainly in the alpine region) were not expected to be relevant for population exposure. RF-EMF levels from these transmitters were modeled by the Federal Office of Communications for an area with a radius of 10 km around each transmitter for the years 1990 and 2000. For the modeling, the Institut für Rundfunktechnik 2-dimensional (IRT_2d) model was applied using CHIRplus_BC software from LS Telcom (Lichtenau, Germany).
RF-EMF exposure levels from all Swiss short- and medium-wave radio transmitters with an output power greater than 1 kW (9 transmitters) were modeled on the basis of the Fresnel Deygout method using ICS-Telecom software from ATDI (Paris, France). For these transmitters, modeling was carried out within a radius of 20 km for the years 1993 and 1997. For overlapping modeled areas, the exposure levels of all transmitters were summed.
In the time-to-event analysis, RF-EMF exposure to radio and television transmitters at baseline was assessed for each study participant at the place of residence using the modeled RF-EMFs from 2000 and 1997, respectively. In the incidence density cohort analysis, place of residency on the date of diagnosis was used for exposure assignment. For children diagnosed before 1995, exposure assessment was based on the models for 1990 and 1993. Thereafter, RF-EMF exposure was assessed using the modeled RF-EMFs from 2000 and 1997, respectively.
Geospatial data on potential confounders were extracted from digital maps using ArcGIS (ESRI, New York, New York), based on the place of residence. Data on background γ radiation were available from the Swiss radiation maps with a grid cell resolution of 2 km. Digital maps with power lines with a resolution of 1:25,000 were provided by the Federal Inspectorate for Heavy Current Installations. We extracted distances to the traffic network in 2000 from digital maps on the traffic network with a resolution of 1:25,000 (VECTOR25-maps), published by the Federal Office of Topography. Data on distances to the nearest orchards, vineyards, and golf courses, for the estimation of exposure to agricultural pesticides, were obtained from the Swiss land-use statistics (Arealstatistik Schweiz) for the year 1997, published by the Swiss Federal Statistical Office, with a grid cell resolution of 100 m. We geocoded the location of the pediatric cancer centers manually. Data on ambient benzene, particulate matter with an aerodynamic diameter less than 10 μm, and nitrogen dioxide exposure were available from a digital map with a grid cell resolution of 100 m (benzene: 400 m), published by the Swiss Agency for the Environment, Forests and Landscape. Residential radon exposure was estimated from a nationwide radon prediction model.
Statistical Analysis
For the time-to-event and incidence density cohort analyses, the same RF-EMF exposure categories were used with a priori chosen cutpoints at 0.05 V/m and 0.2 V/m to differentiate between low, medium, and high exposure. All study participants living in an area not covered by the modeling were included in the reference category. A cutpoint of 0.05 V/m for the reference category was chosen because this value is unlikely to be exceeded due to broadcasting outside the modeling area. A cutpoint of 0.2 V/m for high exposure corresponds roughly to the first quartile of the study population being exposed to RF-EMF levels of more than 0.05 V/m. For short- and medium-wave transmitters, the exposure variable was dichotomized at 0.05 V/m because of the lower levels.
In addition to categorical exposure classification, we also carried out linear exposure-response modeling in the time-to-event analysis using exposure to the broadcast transmitters as a continuous predictor and expressing the hazard ratio per 0.1-V/m increase in exposure. For these analyses, exposure levels outside the modeled area were set to 0.001 V/m.
For the time-to-event analysis, Cox proportional hazards regression models were applied using age as the underlying time scale. Period effects were considered by splitting the follow-up time into two 4-year blocks. The basic models always included adjustment for sex. Furthermore, we decided a priori to adjust for exposure to the potential leukemia risk factors benzene, natural background ionizing γ radiation, distance to the nearest high-voltage power line, and degree of urbanization. We tested the relevance of additional potential confounding factors in the time-to-event analysis by including one confounder at a time in the model and applying a change-in-estimation criterion of 10%. We also conducted a sensitivity time-to-event analysis that excluded all children not living in an area covered by the exposure modeling (i.e., >10 km or >20 km from any transmitter).
For the incidence density analysis, we conducted a Poisson regression analysis that adjusted for sex, age, and calendar year. Separate analyses were conducted for the period up to 1995 and the period after 1995. Results of the incidence density analysis for leukemia were also stratified by age, using 1 and 6 years of age as cutpoints.
