Observed numbers of cases
The cases were all children under 15 years of age diagnosed with leukaemia between 1st January 1990 and 31st December 2001 and living around nuclear sites at the time of diagnosis. The cases were provided by the French National Registry of Childhood Leukemia and Lymphoma, which registered 5,330 cases of acute leukaemia from 1990 to 2001 for the mainland France (
Clavel et al, 2004). Eighty-one percent of all the cases of acute leukaemia were acute lymphoblastic leukaemia (ALL) and 17% were acute myeloid leukaemia (AML). The incidence of acute leukaemia among children aged less than 15 years varies markedly with age (
Clavel et al, 2004). In particular, ALL incidence shows a typical peak at age 2 years for girls and 3 for boys. Therefore, the analyses were also performed by age group (0–4, 5–9 and 10–14 years).
The Chooz and Civaux nuclear power plants were connected to the grid in 1997 and 1999 respectively. Therefore, the childhood leukaemia cases around those two sites were taken into account from 1997 for Chooz and 1999 for Civaux.
Expected numbers of cases
Age- and gender-specific population counts by “commune” (the smallest French administrative division), were derived from the national censuses of March 1990 and March 1999 provided by the National Institute for Statistics and Economic Studies (INSEE). A “
département” is an administrative geographic unit including 383 “communes” on average. The annual number of births by gender for each “
commune” and the annual age- and gender-specific population estimates for each
“département” were available for each year from 1990 to 2001 (INSEE). They were used to obtain age- and gender-specific population estimates from 1991 to 1998 and for years 2000 and 2001 for each
“commune”. The number of person-years for a given year and a given
“commune” were subsequently calculated using those estimates. National age- and gender-specific incidence rates for childhood leukaemia in France (1990–2001), based on the National Registry data, were used as reference rates to derive annual expected numbers of cases for each age-group and
“commune” near nuclear sites. The annual expected numbers of cases were used to estimate standardized incidence ratios (
SIR) defined as the ratio of the observed over the expected number of cases. The exact 95% confidence intervals (95%CI) for these ratios were given using a Poisson distribution.
Exposure assessment
The spatial distribution of the exposure of the population around French nuclear installations due to gaseous radioactive discharge has been assessed by the Institute for Radiation Protection and Nuclear Safety (IRSN) using radionuclide discharge data, local climate data, and a mathematical model of nuclide transfers in the environment (
Morin and Backe, 2002,
2003). Four types of installations were selected: nuclear power plants (NPP), nuclear fuel cycle plants (one fuel production facility, one site with a fuel conversion facility and a fuel enrichment facility), a nuclear fuel reprocessing plant, and two research centres. All the 19 French NPPs have been taken into account. They include 2 to 6 reactors ranging from 900 electric megaWatts (MWe) to 1450 MWe. For plants of other types (nuclear fuel cycle plants, nuclear fuel reprocessing plant, research centres), only a few sites were selected to test the feasibility of the approach and its consistency with the available data. The selected nuclear fuel cycle plants include the nuclear fuel processing plant at Romans-sur-Isère and the fuel conversion plant and the fuel enrichment plant at Pierrelatte. Two nuclear research centres at Saclay and Cadarache were selected together with the La Hague nuclear fuel reprocessing plant. Previous dose calculation work for the population living near the La Hague reprocessing plant has been carried out in France (
GRNC, 1999;
Laurier et al, 2000;
Rommens et al, 2000). Only the average dose delivered to the Beaumont-La Hague
“canton” (a geographic unit which, in this case, includes ten
“communes”) was determined. No zoning was carried out to differentiate the
“communes”.For the NPPs, the average annual discharge levels and discharge compositions for recent years were determined for each of the two types of NPP, namely the 900 MWe and 1300 MWe NPPs. The discharges from the 1450 MWe Chooz and Civaux NPPs were assumed similar to the discharge from the 1300 MWe NPPs. The typical composition of NPP gaseous discharges was taken into account. The composition includes the following nuclides: tritium, carbon 14, argon 41, krypton 85, xenon 133, xenon 135, iodine 131, iodine 133, cobalt 58, cobalt 60, caesium 134 and caesium 137. Carbon 14 was not measured in the gaseous effluent of the French NPPs until very recently. Therefore, no discharge data on carbon 14 were available, and the levels were assumed equivalent to the limit specified in recent discharge permits for this radionuclide. The average discharge levels and compositions for a period of several years (3- to 5-year period, depending on the data available) were assessed for the other plants. Those levels and compositions were assumed to be representative of discharges for recent years (after 1995) and have been used as input data in the dose estimations. Local climate data on wind speed and direction, vertical stability of the atmosphere and rain frequency were used for each site when available. Data were collected from the documents submitted by operators in discharge authorization applications. For a few nuclear plants, data on vertical stability and on rain frequency were not available and the national average data were used.
The FOCON96 1.0 (Rommens et al, 1999) code was used to calculate the doses from routine discharge into the atmosphere. This code includes a model of gas and aerosol dispersion in the atmosphere. The model is based on a Gaussian model with the modeling of vertical and horizontal standard deviation developed by Doury (1976). The code also includes models of dry and wet deposition of aerosols on soil, grass and vegetable leaves. Wet deposition is based on a model of plume scavenging by the rain. The code takes into account root absorption of nuclides by vegetables and grass. Contamination of meat and milk is also modeled in the code. The default values of the transfer parameters proposed in the FOCON 96 code have been used. The main pathways have been taken into account: inhalation, ingestion (vegetables, meat and milk), external exposure from the plume (in the atmosphere or in the water) and deposits (aerosol deposition on the ground and sediment deposition on beaches or river banks). Protection by buildings was not taken into account in the assessment of external exposure. National average food consumption rates have been taken into account (INSEE, 1991). Only estimates of the local part of food production were taken into account (from national average data). It was assumed that 100% of the year was spent in the “commune”. The dose coefficient to red bone marrow (RBM) calculated by the International Commission of Radiological Protection (2002) was used for internal pathways and the US Federal Guidance dose coefficient for external exposure (Eckerman and Ryman, 1993).
RBM doses from gaseous discharge were estimated around each site on a polar grid (252 assessment points) around the stack with the FOCON 96 code. The doses were interpolated on a 250-m square mesh using a SPLINE method (G3GRID procedure of the SAS© software). The RBM dose for each “commune” is the average of the four nearest mesh points around the town-hall of the “commune”.
Areas under study
The studied area was defined as all
“communes” located in 40 km squares centered on 24 French nuclear installations. There are 36,354
“communes” in France with an average population per
“commune” of 1,609 people. A total of 2,107 French
“communes” were included in the study. A strict partition of the areas under study had to be maintained in order to ensure the statistical independence of the observations. Therefore, when study areas around two sites overlapped (this occurred for 68
“communes”), the
“communes” were assigned to the site for which the estimated RBM dose was the highest. The total RBM dose obtained by adding the RBM dose estimates for the two sites was then assigned to the
“commune” considered. Since the Tricastin NPP and Pierrelatte plant are very close to each other, they were considered as a single site throughout the study. This explains the reference to 23 sites (18 NPPs) rather than the original 24 sites (19 NPPs). All of the 2,107
“communes” were subsequently divided into five zones defined on the basis of the estimated dose. Each of the lowest two categories of estimated dose included approximately a tertile of the expected number of cases in order to obtain stable incidence estimates in each category. The corresponding arithmetic means were 0.021 μSv/y and 0.057 μSv/y, respectively. In order to cover the full range of variation of the estimated dose, the third category was then divided into three categories using a logarithmic scale. The corresponding arithmetic means were 0.141 μSv/y, 0.553 μSv/y and 2.13 μSv/y, respectively. Each of the five categories was constructed as aggregations of the
“communes” whose estimated dose was within the limits of the category.
Statistical analysis
The present study investigated for the existence of an increase in the
SIR of childhood leukaemia with increasing estimated radiation dose due to gaseous discharge from nuclear sites. The following four tests were used: Fisher’s chi-square test, the likelihood ratio test, a linear risk score test, and Stone’s Poisson maximum test. Fisher’s chi-square test and the likelihood ratio test based on the Poisson regression models were used to examine the heterogeneity of the five predefined categories of estimated dose. The linear risk score test and Stone’s Poisson maximum test explicitly investigate for an increase in
SIR with increasing estimated dose. The linear risk score test used was adapted from those used by
Bithell et al (1994) and discussed by
Bithell (1995): for the test, each case is scored on the basis on the estimated dose of the
“commune” under consideration. The Stone’s Poisson maximum test is based on the maximum value of the
SIR as
“communes” ranked by increasing estimated dose are aggregated into a region of greater size (
Stone, 1988). The latter two tests were applied to the five predefined dose-based categories and to the estimated dose considered quantitatively for each
“commune”. The powers of the linear risk score tests and Stone’s Poisson maximum test to detect a given risk pattern have been discussed by several authors (
Bithell, 1995 -
White-Koning et al, 2004 -
COMARE 2005).
For all four tests, both an external and an internal reference were used. In the case of tests using an external reference, rejecting the null hypothesis (i.e. a uniform SIR of 1 irrespective of the estimated dose) might evidence the existence of a trend in the relative risk with the estimated dose, or might be due to an excess risk in the overall study area compared to the whole of France, which was considered as the external reference in the study. Using an internal reference enables only the distribution of cases within the study area to be considered and ignores the difference between the overall observed and expected numbers of cases around a given site.
The 5% critical values and the p-values of the four test statistics were estimated using simulation methods with R-software. Under the null hypothesis of a uniform SIR of 1 irrespective of the estimated dose, the null distributions of the four tests were determined from 10,000 simulations. The simulations were based on a Poisson distribution with expected values equal to the expected numbers of cases in the case of an external reference, or on a multinomial distribution with expected values proportional to the expected numbers of cases in the case of an internal reference. The 5% critical values and the p-values were then derived from the null distributions.
The analyses were performed for the 23 sites, for all cases (0–14 years) and for the complete period (1990–2001), and then separately, by age group (0–4, 5–9 and 10–14 years), period (1990–1995–1996–2001), and leukaemia type (ALL, AML). The 18 NPPs were analyzed as a separate subgroup because of their common characteristics. The possible heterogeneity of the 23 sites led to a more detailed study of each site individually. Bonferroni’s method was used in order to correct for multiple testing.