Associations between Childhood Cancer and Ionizing Radiation: Results of a Population-based Case-Control Study in Germany

Exposure to ionizing radiation is one of the few established risk factors for childhood cancers. Ionizing radiation is a harmful exposure because it causes DNA changes and may, thus, act as an initiator in carcinogenesis. The larger the dose, the more cells will be affected, and inadequate DNA-repair may result in mutations, which may change the behavior of cell growth and reproduction. The effects of radiation exposure in utero are particularly well documented (1, 2), whereas effects of preconception exposures are disputed. Radiation exposure of the mother before conceiving the child has not frequently been examined, and, to our knowledge, there is only one study that suggests a positive association with childhood leukemia (3). Preconception exposure of the father, on the other hand, has been investigated more frequently, the reason probably being a study by Gardner et al. (4). In that study, it was observed that fathers of children with leukemia who were employed at the Sellafield nuclear reprocessing installation were more likely to have received radiation doses of more than 100 mSv than fathers of control children. However, the Gardner hypothesis that radiation-induced DNA damage of paternal germ cells can cause childhood cancer has been criticized by others, mainly because of its inconsistency with cancer rates in the offspring of Japanese atomic bomb survivors (5, 6). Hence, the potential role of paternal preconception exposure is considered to be controversial.

Besides preconception exposures and exposure during pregnancy, postnatal diagnostic X-ray examinations in infancy are also a potential risk factor for childhood cancer. This paper presents results with respect to pre- and postnatal exposures to ionizing radiation from a questionnaire-based case-control study carried out by the GCCR.² The study comprises 2358 cases (1184 leukemias, 234 NHLs, and 940 solid tumors), and 2588 controls.

The case-control study presented in this report consists of two parts: a NW-part and a NI-part (a part restricted to geographic areas around nuclear installations and selected control regions), using the same procedures for recruitment of cases and selection of controls as well as the identical questionnaires and interview techniques.

The NW- part of our case-control study comprised the following diagnostic groups: acute leukemia, NHL, and a group of solid tumors. The group of solid tumors (central nervous system tumors, neuroblastomas, nephroblastomas, bone tumors, and soft tissue sarcomas) was recruited not only as an additional study group but also to enable the investigation of possible recall bias effects in the leukemia group with respect to other potential risk factors besides ionizing radiation. Cases were identified from the GCCR, which has an estimated rate of ascertainment of more than 95% (7). Cases were eligible if one of the diseases mentioned above was diagnosed in a child younger than 15 years of age between October 1992 and September 1994 and if the child lived in West Germany (i.e., not in the former German Democratic Republic) at the date of diagnosis.

The second part of our case-control study was embedded in an incidence study investigating childhood malignancies in the vicinity of German nuclear installations. The study population consists of cases of childhood acute leukemia or NHL diagnosed between January 1980 and September 1994, age less than 15 years, born after July 1, 1975, and living in a nuclear-installation area (i.e., at most, 15 km away from a nuclear installation) or a matched control region at the date of diagnosis. Details on the choice of the control regions were described previously (8).

For each case, one control matched for sex, date of birth within 1 year, and district of residence (the smallest administrative unit in Germany) was selected from complete files of local offices for the registration of residents. These files are an excellent sampling frame for population-based studies, because they permit the sampling of individuals including children. The registration office of the district where the case lived at the date of diagnosis was asked for a list of four addresses of children of the same sex and with a date of birth as close as possible to that of the corresponding case. We randomly chose one control from this list, avoiding the case who might have been sampled as a potential control. If a selected family did not participate in the study, we contacted another family from the remaining names. This procedure of control selection was repeated until a selected family consented to participate or until no more potential controls were left.

Structured questionnaires developed by the United States Children’s Cancer Group (9) served as a model to develop our own questionnaire, which is shorter than the original Children’s Cancer Group questionnaire. Questionnaires were mailed by the physician responsible for the cancer treatment (cases) or by the study center at the GCCR (controls) and were to be returned to the study center. In addition, telephone interviews were performed to obtain further information on rare exposures and to complete missing information from the questionnaire. In nearly all of the instances, both of the parents were interviewed. The data were coded by trained personnel, and, in the case of missing or implausible information, the parents were phoned a second time to confirm or correct the given information. We performed double data entry to ensure a high data quality standard. After correction of discrepancies between the two data files, automatic plausibility checks were performed using purpose-written SAS programs. In case of implausible data, the parents were contacted again via phone to solve the problem. If the participating family had no phone (cases 5.0%, controls 4.8%; see Table 1), we relied solely on the information from the questionnaire.

In the mailed questionnaire, all of the parents’ occupations and their respective durations were recorded, and there was an additional question specifically referring to occupations in a nuclear installation. If this question was answered positively, we asked the parents in the subsequent telephone interview whether they had been working under dosimetric surveillance and requested information on cumulative preconception radiation exposure. In the questionnaire, parents also had to specify whether they had been occupationally exposed to radiation in the year preceding pregnancy and/or during pregnancy. Another question referred to X-ray examinations in the 2 years preceding the birth of the child. The mother was additionally asked for X-ray examinations during pregnancy. Furthermore, we gathered information on diagnostic X-ray examinations (site and date) of the child and asked whether the child had been treated in a neonatal intensive care unit.

At the end of the telephone interview, parents were asked for their education level and monthly net income. A high SES was defined by at least one parent having a university degree or a monthly net family income of more than DM 6000 (≈ US$2800). If parents could not be interviewed or were not willing to provide date on education and/or income (cases, 13.2%; controls, 13.5%), we inspected the questionnaires and interview forms for other information related to SES (e.g., current profession) and performed a “best guess” classification of SES on the basis of plausible considerations. Such a procedure is sometimes called “expert rating” also.

Odds ratios and 95% CIs were computed by conditional logistic regression models for matched sets using the PHREG procedure of SAS 6.12. Analyses for 1:1 matched pairs were performed separately for leukemias, NHLs, and the group of solid tumors. Because the NHL group was relatively small and not all of the participating subjects had a matched correspondent, analyses were repeated using a post hoc m:n frequency matching (10). By “m:n matched analysis,” we mean a conditional logistic regression model for frequency-matched data that is stratified for sex, age (groups of 1 year), year of birth, and residence within a nuclear-installation area and, furthermore, adjusted for degree of urbanization (rural, mixed, or urban). These analyses were also carried out using the SAS procedure PROC PHREG. The results of the m:n matched analyses were generally similar to those of the 1:1 matched analyses but yielded somewhat more stable estimates and smaller CIs because they included subjects not having a 1:1 matched partner also and compared each diagnostic group with all of the 2588 available controls. Therefore, in this paper, we preferably present results of the m:n matched analyses. The subgroup of leukemia cases diagnosed within the first 18 months of life was compared with all of the controls for whom the corresponding reference date was within the first 18 months of life using m:n matched analyses.

Logistic regression models were always adjusted for SES. Additional analyses with adjustment for parental education and family income instead of SES were also performed routinely. However, because information on income and/or education was missing for 13% of all of the subjects, these additional analyses will not always be presented in this paper but will be mentioned when the adjustment changed the results noticeably.

From the GCCR, 2989 cases were identified as eligible for at least one of the two parts of the study. Eighty-three (2.8%) of the case families were not contacted, mainly for psychological reasons. Of the remaining 2906 case families, 2454 (84.4%) returned a questionnaire (Table 1). A total number of 3886 control families was contacted to obtain 2815 (72.4%) responders. Hence, there were 254 nonparticipating cases and 1071 nonparticipating controls. The major reasons for nonparticipation were refusals (cases, 85.8%; controls, 74.1%), losses to follow-up (cases, 5.5%; controls, 17.1%) and insufficient knowledge of the German language (cases, 2.7%; controls, 3.5%). The response in the NW-part was somewhat better than in the NI-part (Table 1), which might be explained by the more retrospective nature of the NI-part, including cases diagnosed from 1980 onwards. After inspection of the questionnaires, 254 subjects (96 cases and 158 controls) were excluded from the study population because violations of the inclusion criteria were found. Furthermore, available questionnaires of 69 controls were not processed because the corresponding cases did not participate. Hence, data on 2358 cases (1184 leukemias, 234 NHLs, 399 central nervous system tumors, 160 neuroblastomas, 147 nephroblastomas, 97 bone tumors, and 137 soft tissue sarcomas) and 2588 controls were analyzed. Of the 1184 leukemia cases, 104 were diagnosed within the first 18 months of life and were contrasted with 333 controls having a comparable age at reference date. A 1:1 matched control was recruited for 1010 leukemias (NW-part, 663; NI-part, 449), 199 NHLs (NW-part, 147; NI-part, 74), and 798 solid tumors (NW-part only).

Demographic characteristics of cases and controls are shown in Table 2. As expected, cases are similar to controls on the matching factors of sex and age, as well as on the degree of urbanization that depends on the matching factor of district of residence. Parents of controls generally had a better education and a higher income than parents of case children, which is reflected in a higher proportion of families having a high SES among controls. In Table 3, the relationship between potential risk factors and parental education and family income is shown. All of the risk factors are positively associated with either parental education or family income or both.

A preconception occupational exposure to ionizing radiation was reported by 196 fathers (4.1%) and 226 mothers (4.6%). Whereas the vast majority of these mothers worked in medical professions, there was a variety of different professions among fathers. Most ORs associated with these exposures are slightly above unity (Table 4). Table 5 shows that, in the subgroup of leukemia cases who were diagnosed before their 19^th month of life, ORs were more pronounced (OR = 2.74, 95% CI: 1.01–7.44 for paternal exposure; OR = 2.34, 95% CI: 0.91–6.02 for maternal exposure). The same is true for maternal occupational exposure during pregnancy (OR = 3.30, 95% CI: 0.82–13.3), which was also significantly more prevalent among NHL cases (OR = 3.87, 95% CI: 1.54–9.75). When an adjustment for family income and parental education instead of the two-level SES variable was performed, most OR estimates were somewhat more pronounced. In the subgroup of children who were younger than 19 months, ORs increased to 3.08 (95% CI: 1.07–8.83), and 3.16 (95% CI: 1.18–8.44) for paternal and maternal occupational exposure in the year before pregnancy, respectively, and to 5.94 (95% CI: 1.14–31.0) for maternal occupational exposure during pregnancy.

Mothers of 16 children (6 leukemias, 3 solid tumors, and 7 controls) had worked in a nuclear installation but only 3 mothers (1 of a child with a solid tumor, 2 of controls) had worked under dosimetric surveillance. Among the fathers, 104 had worked in a nuclear installation. Of these, 64 reported having worked under dosimetric surveillance; 35 had not been monitored; and, for 5 fathers, no further information was available. Only 47 of the 64 fathers who had been monitored had worked in the same job before the conception of their child. Preconception paternal occupation with dosimetric monitoring was not a statistically significantly elevated risk factor for any of the diagnostic groups (OR = 1.80, 95% CI: 0.71–4.58 for leukemia; OR = 0.49, 95% CI: 0.09–2.74 for NHL; OR = 1.04, 95% CI: 0.30–3.62 for solid tumors). Again, when analyses were adjusted for family income and parental educational level, the estimated OR for leukemias was somewhat higher (OR = 2.20, 95% CI: 0.67–7.27) but not statistically significant. When the 47 preconception-exposed fathers were asked for the cumulative preconception dose, 25 could not give any details, 10 said that they had been exposed below the threshold of the dosimeter, and 12 reported doses above 0 mSv, the maximum dose being 29.9 mSv. It is also interesting to note that of the 47 fathers who had ever been monitored in the preconception period, only 18 reported that they had been exposed to ionizing radiation in the year preceding conception.

Unfortunately, X-ray examinations of the fathers in the 2 years preceding the birth of the child could not be partitioned into examinations prior to pregnancy and during pregnancy. A statistically significant association between prenatal paternal X-ray examinations was found with leukemias (OR = 1.33, 95% CI: 1.10–1.61; Table 4). A similar but statistically nonsignificant OR of 1.39 was observed for NHLs, whereas in the group of solid tumors, an OR of 1.15 was found. When the analyses were restricted to X-ray examinations of the abdomen and the intestinal tract, the OR for leukemias and NHLs were 1.76 and 2.06, respectively, neither being statistically significant (Table 4). Associations for the group of leukemia cases diagnosed in their first 18 months of life were not more pronounced than for the total group of leukemia cases (Table 5). A closer examination of the questionnaire data revealed that the proportion of parents who answered the question about paternal X-ray examinations with “unknown” was somewhat higher among cases of leukemias (14.8%) and NHLs (16.1%) than among solid tumors (12.1%) and controls (12.2%). When we recoded the answer “unknown” as “unexposed,” ORs for leukemias and NHLs decreased to values of 1.24 (95% CI: 1.04–1.52) and 1.28 (95% CI: 0.87–1.89), respectively. No association between diagnostic X-ray examinations of the mother prior to or during pregnancy and any of the diagnostic groups was observed (Table 4). X-ray examinations of the abdomen, the pelvis, or the intestinal tract during pregnancy were received by only 14 mothers (3 leukemias, 2 NHLs, 2 solid tumors, 7 controls).

In a previous study by the GCCR, a statistically significantly elevated OR of 6.96 (90% CI: 1.18–41.1) for cases of leukemia who had received more than four X-ray examinations was found in an exploratory analysis (11). To test that finding with independent data, the same classification of number of X-ray examinations was used in the present study. X-ray examinations in the year before the date of diagnosis (0.5 year for children diagnosed in their 1^st year of life) were not taken into account. Table 6 shows that more than four X-ray examinations did not seem to pose a risk. The unexpected finding that one-to-four X-ray examinations are negatively associated with risk caused us to perform some additional sensitivity analyses. To investigate whether residual confounding by factors related to SES could have caused these results, analyses were repeated with adjustment for parental education and family income instead of SES. This adjustment, however, changed the results only marginally (data not shown). We, furthermore, speculated whether differential recall among parents of cases versus controls could have influenced our results, inasmuch as the information given by parents of controls could have been more imprecise than the information given by parents of cases, especially when the X-ray examination had been performed several years ago. Therefore, the sample was split into subjects born between 1975 and 1987 (n = 2556) and subjects born between 1988 and 1994 (n = 2390), and stratified analyses were performed. These analyses show that the inverse association between risk and X-ray examinations is particularly pronounced in children born between 1975 and 1987 but is not observed in children born between 1988 and 1994 (Table 5). For the latter group, ORs for one-to-four X-ray examinations are relatively close to unity, whereas ORs for more than four X-ray examinations are greater than 2.0 for any diagnostic group; however, these risk elevations are statistically nonsignificant. A possible explanation for this phenomenon could be that parents of cases could remember clearly whether their children’s X-ray examinations were performed before or after the date of diagnosis, whereas parents of controls could have dated examinations after the corresponding reference date into the time period between birth and reference date.

As a surrogate for X-ray examinations of the child, a treatment of the newborn child in an intensive care unit was also analyzed; however, this factor was not related to childhood malignancies in our study.

The strength of our study is its population basis and the large number of cases who were identified by a cancer registry with almost complete case ascertainment. Controls were drawn at random from the files of population-based registries. Response rates were more than 80% for cases and about 70% for controls; however, among controls, the readiness to participate obviously depended on socioeconomic factors, resulting in a higher income and a better parental education among controls (Table 2). Therefore, all of the analyses were adjusted for SES, and additional analyses with adjustment for parental education and family income were carried out routinely. The analyses based on adjustment for parental education and family income were not presented as a standard because this information was not provided by all of the subjects.

The two parts of our study differ regarding the periods of diagnosis and the one part’s being restricted to specific geographical areas. Otherwise, the designs were identical, we used the same questionnaires, and the interviews were done simultaneously by the same regularly trained interviewers. Thus, technically the data of the two parts of our study could be pooled without problems. However, it has to be kept in mind that—for the NW-part—the interviews were performed a short time after the date of diagnosis, whereas—for the NI-part—the time interval between the date of diagnosis and the interview date may have been as much as 15 years. Hence, recall bias could have been a problem for exposures that occurred some years ago and were not easy to remember. The median (minimum–maximum) number of months between diagnosis (reference date) and arrival of the questionnaire at the GCCR were 7 (0–44) for the nationwide cases, 14 (2–46) for the nationwide controls, 27 (2–186) for the nuclear-installation cases and 51 (2–197) for the nuclear-installation controls. However, analyses carried out separately for both parts of the study did not reveal noticeable inconsistencies (data not shown).

Most comparisons between cases and controls were carried out using a post hoc frequency-matched approach, which has the advantage of including a larger number of cases and controls, resulting in more stable estimates and smaller CIs. This is especially useful for the relatively small NHL group. Furthermore, risk estimates for different diagnoses are more comparable using the m:n matched analyses because the three diagnostic groups are compared with the same controls, namely the whole group of 2588 children. However, if the prevalence of a risk factor like, for instance, occupation in a nuclear-installation area, depends not only on degree of urbanization but on exact geographic location, then the m:n matched approach may underestimate the OR because it does not fully take into account the matching for district. Therefore, the results presented for paternal occupation in a nuclear installation are based on 1:1 matched analyses. Individually matched (but less detailed) analyses carried out separately for both parts of the study are published in the official report of our study.³ In general, those results differ only marginally from the results published in this paper.

The so called “Gardner hypothesis,” which suggests that preconception paternal exposure to ionizing radiation, particularly among men occupied in the nuclear industry, is a risk factor for childhood leukemia and NHL has been repeatedly investigated (e.g., Refs. 12, 13, 14). In their reviews, Doll et al. (5) as well as Wakeford (6) conclude that current evidence suggests that the Gardner hypothesis is not valid. A recent study conducted by Draper et al. (15) linked data of almost 36,000 cases of childhood cancer and matched controls to the British National Registry for Radiation Workers, which keeps exposure data of individuals who have worked under dosimetric surveillance. Draper et al. found a relative risk of 1.77 (95% CI: 1.05–3.03) for children diagnosed with leukemia or NHL whose fathers were radiation workers, irrespective of the dose the fathers had received. It is interesting to note that, in our study, the finding for preconception paternal occupation including dosimetric monitoring is very similar in the group of leukemia cases (OR = 1.80, 95% CI: 0.71–4.58). The information on preconception doses of the fathers in our study was too sparse to perform any meaningful analyses; however, no father reported a dose of more than 30 mSv. When Draper et al. took into account the preconception dose of the fathers, they found the highest risk was for those exposed below the level of detection (15). Earlier British studies also found that fathers of cases had been monitored more frequently than fathers of controls (13, 14), whereas, in a study performed in Canada, no differences were found between cases and controls in relation to occupations of their fathers in the nuclear industry (16).

Apart from occupations in nuclear installations involving dosimetric monitoring, we asked both parents about occupational radiation exposures. Such exposures were reported relatively frequently. For example, 194 fathers reported an occupational exposure in the year preceding conception, although there were only 47 nuclear-installation workers who had ever been monitored in the preconception period and only 18 in the year preceding conception. The vast majority of mothers who reported occupational radiation exposures had worked in medical professions, which made an exposure plausible. However, in some cases the reported job titles suggested that radiation exposure was not very likely. Hence, a certain degree of overreporting is possible. Occupationally exposed fathers worked in a great variety of different professions (mainly blue collar jobs), which implied exposures to other possible risk factors also. Therefore, our finding that prenatal occupational exposures of both parents are more common among cases of childhood leukemia who were diagnosed in the first 18 months of life must be interpreted with caution. On the other hand, there were no such associations in the group of solid tumors who were diagnosed within the first 18 months of life (data not shown), which suggests that the finding for the leukemia cases was not caused by recall bias. Shu et al. (17) conducted a case-control interview study on 302 cases of leukemia diagnosed in the first 18 months of life. Similar to our findings, they observed that more fathers of cases than fathers of controls were occupationally exposed to radioactive materials (OR = 1.7, 95% CI: 1.07–2.71) but did not report similar results for occupational exposures of the mother. Furthermore, paternal preconception diagnostic X-rays of the lower gastrointestinal tract and the lower abdomen were a statistically significant risk factor (OR = 3.78; 95% CI: 1.49–9.64) in the study by Shu et al. (17). Such strong findings could not be observed in our study although one could argue that the unfortunate mixing of preconception paternal X-ray examinations with irrelevant examinations during pregnancy may have biased OR estimates toward unity. Nevertheless, a significant association was observed with prenatal X-ray examinations of the father (OR = 1.33; 95% CI: 1.10–1.31; P = 0.0077), and again, although this finding is based on small numbers, the null result for the group of solid tumors suggests that the finding for the leukemia group was not caused by recall ias. If we restrict this analysis to the more concurrent NW-part of the study and use the 1:1 matched analysis, the OR is 1.66 (95% CI: 1.25–2.20; P = 0.0004).

A last argument that weakens our findings with respect to the effects of prenatal exposures to ionizing radiation on childhood leukemia is their internal inconsistency. It does not seem plausible that, for leukemia cases diagnosed in the first 18 months of life, occupational exposures of both parents pose a risk factor and parental X-ray examinations do not; whereas, in the whole group of leukemia cases, X-ray examinations of the father but neither X-ray examinations of the mother nor occupational parental radiation exposures are related to the child’s disease.

Diagnostic X-ray examinations of the child were not positively associated with any of the diagnostic groups in our study. On the contrary, one to four X-ray examinations were apparently inversely associated with risk. These unexpected results may be explained by residual confounding, inasmuch as X-ray examinations were more frequent among families having a higher income and a better education, and the readiness of control families to participate in the study depended on these factors. We believe that a more refined recoding of socioeconomic factors and a corresponding adjustment for such factors would possibly lead to a disappearance of the observed inverse associations in our study, at least partially. Another possible explanation is that parents of controls erroneously dated examinations that happened after the corresponding reference date into the time period between birth and reference date. This explanation is supported by the observation that the suspected bias is particularly pronounced in the group of children with a date of birth between 1975 and 1987, i.e., several years before the study was conducted, but not in the group of children born 1988 or later (Table 6). To our knowledge, there is no recent study that could convincingly demonstrate an association between postnatal X-ray examinations and childhood cancer. In 1966, Graham et al. (3) compared 319 leukemia cases with 884 controls and found a significant relative risk of 2.3 for children who had received X-ray examinations at more than one site. A study of 309 childhood leukemia cases and 618 controls that was conducted in Shanghai by Shu et al. (18) found a statistically nonsignificantly elevated leukemia risk of 2.4 for children who had received more than five X-ray examinations, whereas the OR for one-to-five examinations was 0.8. Those results are similar to our data when only children born after 1987 are considered (Table 5). In a subsequent study that included 166 childhood leukemia cases and was also carried out in Shanghai, Shu et al. used different cutoff points and observed a statistically nonsignificantly elevated OR of 2.0 for children who had received more than two X-ray examinations. Taken together, the current evidence suggests that, at present, postnatal X-ray examinations do not play a noticeable role in the development of childhood cancers.

To summarize, our results do not provide convincing evidence that prenatal or postnatal X-ray examinations pose a relevant public health hazard for the childhood malignancies studied here. An OR of 2.0 for a hypothetical exposure with a prevalence of 2% would correspond to an etiological fraction of less than 2%. In Germany, this number would correspond to 12 of the 600 children per year (incidence: 4/100,000) who are diagnosed with leukemia. This finding may be explained by (a) technical improvements in X-ray units resulting in lower radiation doses than in earlier years; (b) a more prudent attitude toward X-ray examinations during pregnancy; and (c) the extended use of other diagnostic techniques (e.g., ultrasound). Prenatal occupational exposures to ionizing radiation were more frequent among both parents of leukemia cases who were diagnosed in the first 18 months of life; however, this finding, although biologically plausible, has to be interpreted with care because of the possible overreporting of exposure among parents of cases. With respect to the Gardner hypothesis, similar to other studies, our data provide only weak and inconsistent findings.