Radon gas (radon-222) is a natural radioactive gas that emanates from soils and can concentrate inside houses. Radon and its α-particle-emitting decay products (denoted collectively here as “radon”) are the primary sources of the population’s natural exposure to ionizing radiation.1 As most of an inhaled dose of radon is deposited in the airways, radon exposure increases the risk for lung cancer.2,3 A minor part of an inhaled dose can also be distributed to other organs, including the red bone marrow4–6 and the brain.6 Exposure to high doses of ionizing radiation is one of the few well-established risk factors for childhood cancer, but the picture is less clear for lower doses.7 The hypothesis of an association between radon exposure and childhood cancer is based on several geographic correlation studies showing associations between average regional radon levels and incidence or mortality rates.8–15 Although these results were not consistent, most of the associations were for leukemias.
The results of correlation studies must, however, be interpreted with caution. Domestic concentrations of radon were measured in a few case-control studies of childhood leukemias (discussed below), but those studies gave inconsistent results.16–20 A recent case-control study showed an association between radon exposure and leukemia among miners,21 but a meta-analysis of 11 cohort studies showed no convincing association.22 A UK case-control study showed no association between measured domestic radon and adult leukemia.23
We have used a large population-based case-control data set from Denmark to test the hypothesis that domestic radon exposure increases the risk for childhood leukemia, central nervous system (CNS) tumor, and malignant lymphoma. We assessed individual radon exposure using a newly developed prediction model for indoor radon concentrations, which we have applied to the complete residential history of the children.24
Eligible cases were leukemia, tumor of the CNS, or malignant lymphoma diagnosed in children in Denmark when they were under the age of 15 years, during the period 1 January 1968 to 31 December 1994. The incident cases were identified from the Danish Cancer Registry,25 which also provided the unique personal identification number that includes sex and date of birth. Children born outside Denmark were excluded.
Two control children matched by sex and birthday (within 1 year) were chosen for each case of leukemia, 3 for each case of CNS tumor, and 5 for each malignant lymphoma. The controls were randomly drawn from the files of the Danish Central Population Registry and likewise identified by personal identification number. Eligible children were those born in Denmark and living in Denmark with no previous diagnosis of cancer at the age at diagnosis of the corresponding case.
The parents of case and control children were identified by record linkage with the files of the Danish Central Population Registry, and the residential history of each family was ascertained for the “childhood period,” ie, from birth to the age at diagnosis (cases) or the age at diagnosis of the corresponding case (controls). Addresses were provided jointly by the Central Population Registry and the 275 local population registries of Denmark and identified according to municipality, town, postal code, street, building number, and floor; the dates of moving in and out were noted.
The addresses were linked with the Cross-reference Register to obtain geographical coordinates (denoted in the following as “geocodes”) and with the Building and Dwelling Register to obtain construction data for each building. Local authorities were contacted to collect these data for addresses with no automatic match in the registers. The geocodes were used to determine the type of soil at each address from a detailed high-resolution digital map of soils in Denmark.26
We constructed a model for predicting radon concentrations in Danish dwellings on the basis of register data.24 In brief, the model was developed by regression analysis of 3116 radon measurements conducted over a period of 1 year with CR-39 track detectors in the living room of Danish dwellings. Geographical region, soil type, and a number of house characteristics (such as type of house, floor, basement, and building materials) were used as predictive factors; the model had an r2 of 0.45 when tested against independently collected data. We used the prediction model, with seasonal correction based on typical monthly variations in indoor radon levels, to calculate domestic radon concentrations for the addresses in the present study. More details on model development, prediction uncertainty, and errors of the measurements are provided elsewhere.24 For use in the risk analyses, we calculated cumulated radon exposure by cumulating the product of exposure level and time for each address occupied during childhood. For addresses with no calculated radon levels (due to failed geocoding or missing input data for the model), the time-weighted average concentration for the part of the childhood period with known radon concentration was imputed as a best estimate.
The children were divided into 3 exposure groups with cutoff points at the 50th and the 90th percentiles of the distribution of cumulated radon exposure between case and control children combined. Rate ratios (RRs) and 95% confidence intervals (CIs) for childhood cancer were calculated for the intermediate and high exposure groups relative to the low exposure group from conditional logistic regression models for individually matched datasets with the Phreg procedure of SAS (version 8.2; SAS Institute, Cary, NC). RRs were also estimated by increments of 103 Bq/m3-years from a linear dose-response analysis. Possible associations were analyzed for all included childhood cancers combined, for the main diagnostic groups (leukemias, CNS tumors, and malignant lymphomas), and for morphologic subtypes of these cancers. We performed sensitivity analyses to gain further insight into the relation between exposure and cancer risk by (1) an analysis of the shape of the dose-response relation by a graphical presentation of RRs for 16 exposure groups over the whole exposure range and by quadratic-logistic spline models,27 (2) analyses restricted to children with predicted radon concentrations at all addresses and with a precision in geocodes of all addresses better than 200 m, (3) analyses restricted to children who had lived in a single-family house throughout childhood (radon concentrations in Danish single-family houses are generally higher and more variable than concentrations in apartments)28 and (4) analyses with adjustment for the following potential confounding factors: birth order of the child, mother’s age, traffic density, and electromagnetic fields from nearby high-voltage facilities. These factors were selected among suspected risk factors for childhood cancer because they might also be associated with radon levels due to an uneven distribution by geography or housing characteristics. All 4 factors had been collected for use in previous studies29,30 and were available for 82% of the children in the present study.
We investigated the association between radon level and the risk for ALL in different age groups, as children below the age of 2 years might comprise a subgroup with a distinct etiology.31 Further, we investigated the association between radon level and the risk for ALL stratified in 4 subperiods of year of diagnosis defined as quartiles of ALL cases. The proportion of ALL cases attributable to cumulated radon exposure given the observed association was calculated by standard methods.32
We identified 2400 cases and sampled 6697 controls (Table 1). Radon concentrations were predicted for 18,899 (94%) of the 20,077 residences in which the 9097 children had lived, at all relevant addresses for 90% of both cases and controls, at part of the relevant addresses for 8% of cases and 9% of controls, and for none of the relevant addresses for 2% of cases and 1% of controls. On average, both cases and controls lived at 2.2 addresses during childhood (2.0 and 2.0 addresses for leukemia cases and controls, 2.2 and 2.3 for CNS tumor cases and controls, and 2.5 and 2.5 addresses for cases of malignant lymphomas and their controls). The predicted radon concentrations at the 18,899 addresses ranged from 4 to 254 (mean, 48) Bq/m3, with 10th, 50th, and 90th percentiles of the distribution equal to 10, 41, and 102 Bq/m3, respectively. The geocodes of at least one of the addresses of these children were less precise than 200 m for only 33 cases (2%) and 117 controls (1%). Of the 8976 children for whom radon exposure was estimated, 5230 (58%) had lived in a single-family house throughout childhood.
Table 2 shows that the RR for all childhood cancers combined was close to 1.0 for the 2 higher exposure groups when compared with the reference group. The results showed a tendency towards an elevated risk for leukemia with higher exposure, whereas no association was seen for CNS tumors or malignant lymphomas, most of the RRs estimates being slightly lower than 1.0. None of the morphologic subtypes of CNS tumors (ependymoma, astrocytoma, medulloblastoma) or malignant lymphomas (Hodgkin disease, non-Hodgkin lymphoma) was associated with residential radon concentrations (data not shown).
Table 3 shows an association between predicted cumulated radon exposure and the risk for ALL, with a 21% higher rate in the intermediate exposure group and a 63% higher rate in the highest exposure group, corresponding to a 56% increase in the ALL rate per 103 Bq/m3-years. No statistically significant association was found for acute nonlymphoblastic leukemia or for “other leukemia.” While the rate ratios for acute nonlymphoblastic leukemia were below 1.0, they were slightly increased for the group of “other leukemia”; some of the cases in the latter group were probably acute lymphoblastic leukemia classified as “leukemia not otherwise specified” in the Danish Cancer Registry.
Use of a linear term in the logistic regression model for dose-response analysis gave a satisfactory model fit. Application of a quadratic-logistic spline model suggested that the log-linear model was adequate (data not shown). Figure 1 shows the log-linear fit of the dose-response relationship together with rate ratios of 16 exposure categories, compared with the reference group of ≤50 Bq/m3-years.
Age-stratified trend analyses for ALL gave rate ratios of 2.22 (95% CI = 0.80–6.18) per 103 Bq/m3-years for children aged 2 to 4, 1.67 (0.85–3.30) per 103 Bq/m3-years for those aged 5 to 9, and 1.35 (0.79–2.32) per 103 Bq/m3-years for those aged 10 to 14 years. Thus, the results showed increased rate ratios in all 3 age groups, with overlapping confidence intervals. The rate ratios for children under the age of 2 years had very wide confidence intervals because of small numbers and low cumulative radon exposure. Nevertheless, the lowest rate ratios (using radon exposure both as a categorical and a continuous variable) were observed for children aged 0 to 1 year, and the log-linear model suggested no increase in risk with increasing exposure (data not shown).
Table 4 shows that adjustment for mother’s age, birth order, traffic density at the residence, and electromagnetic fields from high-voltage facilities slightly increased the rate ratio for ALL for the highest exposure group and in the trend analysis. When the data were restricted to those for children for whom radon concentrations had been predicted for all addresses during childhood and with better precision in the geocodes of all addresses than 200 m, all the rate ratios increased slightly. When the analysis was based only on children who had lived in a single-family house throughout childhood, the rate ratios for the intermediate and high exposure groups increased to 1.41 and 2.09, respectively, corresponding to a rate ratio of 2.44 for ALL per 103 Bq/m3-years.
Stratification of the ALL findings by year of diagnosis showed an association in all four 6 to 7 year periods, with a tendency towards a stronger association for the most recent period. Using the identical exposure group definition as in Tables 2 to 4, the RRs for highest exposure group versus the lowest exposure group were 1.5 (CI = 0.6–3.7), 1.6 (0.6–4.1), 1.4 (0.6–3.2), and 2.0 (0.8–4.9) for the periods 1967–1974, 1975–1981, 1982–1987, and 1988–1994, respectively (data not shown).
The present study showed an association between cumulated domestic radon exposure and the risk for childhood ALL, but not for any other leukemia, CNS tumors, or malignant lymphomas. The dose-response analysis suggested a 56% increase in ALL rate per 103 Bq/m3-years. The association with ALL persisted in sensitivity analyses and after adjustment for potential confounders. If we assume that the observed risk association in the categorical analyses of ALL is causal, and assuming a population exposure corresponding to that of the controls of the ALL cases, an estimated 9% of all cases of ALL in Denmark would be attributable to radon exposure.
We took advantage of complete nationwide case ascertainment in a high-quality cancer register, an ideal frame for control sampling from a nationwide population register, complete residential histories, and a model-based exposure assessment method with no bias due to nonparticipation and with an assessment of exposure independent of case or control status.
We used a mathematical model to predict radon concentrations in the dwellings occupied by the children.24 In previous case-control studies, radon was measured in present or former residences of study participants.16–20 The measurement method (in contrast to our prediction model) captures actual daily ventilation of the rooms and, therefore, nondifferential misclassification of exposure inevitably occurred in the present study due to imprecise radon predictions. Radon levels predicted recently by the model with the same input data as used in the present study were compared with 758 radon measurements performed between 1985 and 2001 in homes other than those of the present study, and homes other than those used to develop the model.24 Table 5 shows that misclassification occurred as could be expected for predictions, with a geometric standard deviation of about 2, which means that 32% of predictions are expected to be discrepant from the true concentration by more than a factor of two.24 Nevertheless, 80% of the low exposures and 60% of the high exposures were correctly identified by the prediction. Furthermore, the mean, the 25th, 50th, 75th, and 95th percentiles of the distributions of the measured radon concentrations all increased monotonously through 5 groups of predicted concentrations (Fig. 2), indicating some capacity of the model to correctly rank classes of radon concentrations in homes.
Moreover, the comparison between predicted and measured concentrations showed that the misclassification (or error) was on the whole independent of the calculated concentration,24 as might be expected when exposure is predicted from a model.33 Thus, the data corresponded to a Berkson error model;34 trend estimates derived from exposure data with a Berkson error can be expected to be unbiased.35,36 Further errors occur when (1) estimates for radon concentration in the living room are used as a proxy for personal exposure and, eventually, for dose to the red bone marrow and (2) when recently predicted (or measured), radon concentrations are used as proxies for exposure that occurred during previous decades. Thus, several uncertainties could also introduce a classic error structure (in addition to Berkson error), which can introduce bias in the exposure-response relationship. In the present study, we anticipate that these errors have affected cases and controls randomly; hence, the expected direction of bias is probably an underestimation of any association.
The model was validated against measurement data from 1985 to 2001 but was in fact used for addresses occupied as early as the 1950s. It is difficult to quantify errors associated with this historical extrapolation of the model. Historical changes in diagnostic methods and disease classification may have introduced uncertainty into the end point variable (eg, methods for diagnosis of subtypes of leukemia have improved over the enrollment period). Nevertheless, the association between radon levels and childhood ALL was apparent in each of 4 historical time intervals, with only a slightly stronger risk association in the most recent period (1988–1994). This might be due to either more precise radon model estimates or more precise end point definition.
The most important strength of our modeling approach is that the register-based input data for the model allowed us to assess lifetime exposure for virtually the entire study population without introduction of selection bias due to nonparticipation (a serious concern in studies based on measurements). The response rates were usually below 50% in those studies.16,17,20 The few nonparticipants in our study are due to problems in obtaining or geocoding addresses, thus affecting cases and controls randomly. Second, the modeling approach enabled us to assess exposure at nearly 19,000 addresses, which is hardly realistic in a measurement study. Nevertheless, we had to give up calculating the radon concentration at one or more of the addresses for 10% of the children because it was not possible to geocode the address or to collect all the needed input data for the model. In principle, the residential movement pattern, and thereby the radon concentration, might be different for these 10% of the children, leaving the remaining 90% with a slightly different radon level than that of the whole study population. However, since the percentage is equal for cases and controls and since this 10% “nonparticipation” depends on objective factors unrelated to the child, the family or the disease, we have no reason to believe that it would affect radon concentrations differently for cases and controls. Although our strategy for imputation in case of missing exposure data may have produced a bias towards underestimation of a true risk association,37 the observed association between domestic radon and ALL persisted in analyses including only the 90% of the children without imputed radon concentrations (Table 4). The latter exclusion strategy for avoiding gaps in the exposure history is not expected to introduce bias into the risk estimation.37
The complete residential history of each child during childhood enabled us to estimate cumulative radon exposure, whereas average radon concentration during the measurement period was used in previous studies. In a study in the United States, radon measurements in homes inhabited for more than 70% of the childhood period were available for 54% cases and 34% of controls; the odds ratio for ALL was 1.44 (95% CI = 0.9–2.3) for >148 Bq/m3 compared with <37 Bq/m3.16 For a 5-year exposure period, this would correspond to cumulated exposures of >0.74 compared with <0.19 × 103 Bq/m3-years. This corresponds well to the rate ratio for ALL of 1.63 (95% CI = 1.05–2.53) in the present study (Table 3). The result of the matched analysis in the US study (rate ratio = 1.02; 95% CI = 0.5–2.0)16 is also in agreement with our result owing to wide overlap of the CIs. A small German study17 of children with acute leukemia (primarily ALL) showed an odds ratio of 1.30 (95% CI = 0.32–5.33) for an exposure contrast of >70 versus <70 Bq/m3, also in agreement with our findings. The odds ratios of 5 to 7 for the higher exposure categories in a small Egyptian study19 are far higher than those in the present and other studies.
In a large study in the United Kingdom, radon was measured in homes at the time of diagnosis for children with cancer (50% of eligible cases and 31% of eligible controls).20 The study showed a lower risk in association with higher radon levels for all types of childhood cancer, including ALL, with an odds ratio of 0.77 (95% CI = 0.61–0.99) for exposures >30 Bq/m3 compared with those <8 Bq/m3. For a 5-year-old child, this contrast in average exposure would correspond to a cumulated exposure of >0.15 versus <0.04 × 103 Bq/m3-years. The authors of the UK study suggested that the inverse risk association might be related to socioeconomic and household differences between cases and controls due to nonparticipation.20 Due to the lower exposures in the UK than in Denmark and the methodologic differences between the UK and the Danish studies (particularly the balancing between detailed up-to-date exposure assessment [UK] versus complete coverage of the study population [Denmark]), it is difficult to directly compare the findings for ALL, although the results of the 2 studies appear to give different messages.
The lack of evidence for an association between radon levels and acute nonlymphoblastic leukemia is in accordance with a case-control study in Canada and the United States, which showed no overall association between measured domestic radon concentration and acute myeloid leukemia in children.18
Several criteria can be used to evaluate whether a statistical association is likely to be causal. The association between radon exposure and ALL in the present study was relatively weak; however, a strong association was not expected because the doses of radiation from indoor radon are relatively low. The observed association persisted and was slightly stronger (1) after adjustment for potential confounders, (2) among children with more precise exposure assessment, and (3) in the subgroup of children who had lived in a single-family house throughout childhood (which implies higher exposure levels and higher exposure contrasts), indicating internal consistency of the study. Moreover, the association showed a linear dose-response pattern, which has also been observed for radiation-induced adult ALL38 and an association was present for children of all ages except very young children, in whom environmental postnatal exposures are less likely to cause cancer because of the short exposure period. Disregarding the very young children (below the age of 2 years), the results showed a pattern of higher risk for ALL in association with radon for younger children (2.22, 1.67, 1.35 per 103 Bq/m3-year for ages 2–4, 5–9, and 10–14 years, respectively). This pattern might be interpreted either as higher sensitivity among younger children or as the importance of absolute radon concentration beyond the contribution to cumulative exposure, since radon concentration would have to be 4 times higher for a 3-year-old child than for a 12-year-old child to obtain the same cumulative exposure. However, chance might also explain the pattern.
Red bone marrow receives lower doses of radiation from inhaled radon than other tissues,1,39 but previous dosimetric calculations suggest that radon might still be responsible for a proportion of childhood leukemias.6 Thus, as a rough estimate it has been suggested that about 6% of fatal childhood leukemias might be attributed to radon in the United Kingdom.6 This number is similar to our estimate of 9% of ALL cases attributable to radon in Denmark. The population-weighted average annual radon concentration in Denmark is 59 Bq/m3,28 which is more than twice that measured in the United Kingdom and the Netherlands, somewhat higher than the mean concentrations measured in Canada and the United States, and about half the mean concentrations in Finland and Sweden.40
There are alternatives to a causal interpretation. The finding for ALL in the present study could be due to chance or to confounding. Socioeconomic status has previously been linked to the risk for childhood leukemia and may also be linked to the type of the residence, which may affect indoor radon levels. We have previously shown that individual social class of the family is not associated with childhood leukemia in Denmark, whereas the risk for leukemia in childhood was higher in areas of low socioeconomic status.41 We would expect a higher proportion of families living in apartments (with relatively low radon levels) in such areas, which might confound the risk estimates towards an inverse association between radon and risk for ALL. Thus, this cannot explain the ALL finding of the present study. Lifestyle factors such as smoking and diet may differ between families living in apartments and single-family houses, and might have confounded the overall results. However, no such factor has been linked to the risk for childhood cancers,7 and the risk association between radon and ALL in the present study persisted in analyses restricted to children who had lived only in single-family houses during childhood.
The specificity of the association between radon and ALL indicates that a general bias is not a likely explanation for our results. However, a biologic explanation for why an association should be restricted to ALL among children is not obvious. Radiation also causes other types of leukemia among adults, and a recent German study showed associations between residential proximity to nuclear power plants and both ALL and acute nonlymphoblastic leukemia in children, although the result was not significant for the latter.42 A possible different effect of low-dose radiation on different subtypes of childhood leukemia deserves attention in future studies.
1. Laurier D, Valenty M, Tirmarche M. Radon exposure and the risk of leukemia: a review of epidemiological studies. Health Phys. 2001;81:272–288.
2. Krewsky D, Lubin JH, Zielinski JM, et al. Residential radon and risk of lung cancer. Epidemiology. 2005;16:137–145.
3. Darby S, Hill D, Auvinen A, et al. Radon in homes and risk of lung cancer: collaborative analysis of individual data from 13 European case-control studies. BMJ. 2005;330:223–226.
4. Richardson RB, Eatough JP, Henshaw DL. Dose to red bone marrow from natural radon and thoron exposure. Br J Radiol. 1991;64:608–624.
5. Allen JE, Henshaw DL, Keitch PA, et al. Fat cells in red bone marrow of human rib: their size and spatial distribution with respect to the radon-derived dose to the haemopoietic tissue. Int J Radiat Biol. 1995;68:669–678.
6. Kendall GM, Smith TJ. Doses from radon and its decay products to children. J Radiol Prot. 2005;25:241–256.
7. Little J. Epidemiology of Childhood Cancer. Lyon: IARC; 1999.
8. Lucie NP. Radon exposure and leukaemia. Lancet. 1989;2:99–100.
9. Henshaw DL, Eatough JP, Richardson RB. Radon as a causative factor in induction of myeloid leukaemia and other cancers. Lancet. 1990;335:1008–1012.
10. Alexander FE, McKinney PA, Cartwright RA. Radon and leukeamia. Lancet. 1990;335:1336–1337.
11. Collman GW, Loomis DP, Sandler DP. Childhood cancer mortality and radon concentration in drinking water in North Carolina. Br J Cancer. 1991;63:626–629.
12. Thorne R, Foreman NK, Mott MG. Radon in Devon and Cornwall and paediatric malignancies. Eur J Cancer. 1996;32A:282–285.
13. Gilman EA, Knox EG. Geographical distribution of birth places of children with cancer in the UK. Br J Cancer. 1998;77:842–849.
14. Kohli S, Noorlind BH, Lofman O. Childhood leukaemia in areas with different radon levels: a spatial and temporal analysis using GIS. J Epidemiol Community Health. 2000;54:822–826.
15. Evrard AS, Hemon D, Billon S, et al. Ecological association between indoor radon concentration and childhood leukaemia incidence in France, 1990–1998. Eur J Cancer Prev. 2005;14:147–157.
16. Lubin JH, Linet MS, Boice JD, et al. Case-control study of childhood acute lymphoblastic leukemia and residential radon exposure. J Natl Cancer Inst. 1998;90:294–300.
17. Kaletsch U, Kaatsch P, Meinert R, et al. Childhood cancer and residential radon exposure—results of a population-based case-control study in Lower Saxony (Germany). Radiat Environ Biophys. 1999;38:211–215.
18. Steinbuch M, Weinberg CR, Buckley JD, et al. Indoor residential radon exposure and risk of childhood acute myeloid leukaemia. Br J Cancer. 1999;81:900–906.
19. Maged AF, Mokhtar GM, El-Tobgui MM, et al. Domestic radon concentration and childhood cancer study in Cairo, Egypt. Environ Carcino Ecotox Revs. 2000;C18:153–170.
20. UK Childhood Cancer Study Investigators. The United Kingdom Childhood Cancer Study of exposure to domestic sources of ionising radiation. 1: radon gas. Br J Cancer. 2002;86:1721–1726.
21. Rericha V, Kulich M, Rericha R, et al. Incidence of leukemia, lymphoma, and multiple myeloma in Czech uranium miners: a case-cohort study. Environ Health Perspect. 2006;114:818–822.
22. Darby SC, Whitley E, Howe GR, et al. Radon and cancers other than lung cancer in underground miners: a collaborative analysis of 11 studies. J Natl Cancer Inst. 1995;87:378–384.
23. Law GR, Kane EV, Roman E, et al. Residential radon exposure and adult acute leukaemia. Lancet. 2000;355:1888.
24. Andersen CE, Raaschou-Nielsen O, Andersen HP, et al. Prediction of 222Rn in Danish dwellings using geology and house construction information from central databases. Radiat Prot Dosimetry. 2007;123:83–94.
25. Storm HH, Michelsen EV, Clemmensen IH, et al. The Danish Cancer Registry–history, content, quality and use. Dan Med Bull. 1997;44:535–539.
26. Geological Survey of Denmark and Greenland. Digital Soil Map of Denmark, 1:200. 000. Copenhagen: Geological Survey of Denmark and Greenland; 1999.
27. Greenland S. Dose-response and trend analysis in epidemiology: alternatives to categorical analysis. Epidemiology. 1995;6:356–365.
28. Andersen CE, Ulbak K, Damkjær A, et al. Radon in Danish dwellings. Copenhagen: National Institute of Radiation Hygiene; 2001.
29. Raaschou-Nielsen O, Hertel O, Thomsen BL, et al. Air pollution from traffic at the residence of children with cancer. Am J Epidemiol. 2001;153:433–443.
30. Olsen JH, Nielsen A, Schulgen G. Residence near high voltage facilities and risk of cancer in children. BMJ. 1993;307:891–895.
31. Spector LG, Davies SM, Robison LL, et al. Birth characteristics, maternal reproductive history, and the risk of infant leukemia: a report from the Children’s Oncology Group. Cancer Epidemiol Biomarkers Prev. 2007;16:128–134.
32. Olsen JH. Avoidable cancers in the Nordic countries. Aims and background. APMIS Suppl. 1997;76:1–8.
33. Clayton D. Models for the longitudinal analysis of cohort and case-control studies with inaccurately measured exposure. In: Dwyer JH, Feinleib M, Lippert P, et al., eds. Statistical Model for Longitudinal Studies of Health. Oxford: Oxford University Press; 1992.
34. Carrol RJ. Measurement error in epidemiological studies. In: Armitage P, Colton T, eds. Encyclopedia of Biostatistics. Vol. 3. Chichester, England: John Wiley & Sons, Ltd; 1998:2491–2519.
35. Prentice RL. Covariate measurement errors and parameter estimation in a failure time regression model. Biometrika. 1982;69:331–342.
36. Steenland K, Deddens JA, Zhao S. Biases in estimating the effect of cumulative exposure in log-linear models when estimated exposure levels are assigned. Scand J Work Environ Health. 2000;26:37–43.
37. Weinberg CR, Moledor ES, Umbach DM, et al. Imputation for exposure histories with gaps, under an excess relative risk model. Epidemiology. 1996;7:490–497.
38. United Nations Scientific Commitee on the Effects of Atomic Radiation. Unscear 2000 Report Vol. II: Effects. New York, NY: United Nations; 2006.
39. Kendall GM, Smith TJ. Doses to organs and tissues from radon and its decay products. J Radiol Prot. 2002;22:389–406.
40. United Nations Scientific Commitee on the Effects of Atomic Radiation. Unscear 2000 Report Vol. I: Sources. New York, NY: United Nations; 2006.
41. Raaschou-Nielsen O, Obel J, Dalton S, et al. Socioeconomic status and risk of childhood leukaemia in Denmark. Scand J Public Health. 2004;32:279–286.
42. Kaatsch P, Spix C, Schulze-Rath R, et al. Leukaemia in young children living in the vicinity of German nuclear power plants. Int J Cancer. 2008;122:721–726.
© 2008 Lippincott Williams & Wilkins, Inc.