In the largest study to date of radiation exposure for the treatment of childhood cancers and subsequent thyroid cancer risk, we confirmed the downturn in risk of developing thyroid cancer after radiation doses exceeding approximately 20–25 Gy that was demonstrated in the previous case-control study conducted in the CCSS cohort (1
) and is thought to be attributable to cell killing (15
). We also found that age at radiation exposure significantly modified the ascending linear portion of the ERR dose response, for which the previous case-control study found only suggestive evidence. Unlike the previous study, we were able to evaluate the EAR, which demonstrated a similar dose–response pattern to the ERR; however, significant modifiers of the EAR linear term were sex and time since radiation exposure. Results were similar when restricting analyses to the 111 second primary thyroid cancers (results not shown) and when evaluating different lag periods (0, 10 and 20 years) for the radiation doses to the thyroid gland (results not shown). It is also important to note that the downturn in risk remained significant in the ERR and EAR models when accounting for effect modification of the linear portion of the dose response by age at exposure and time since radiation exposure, respectively (results not shown).
As with the previous case-control study, we found the linear exponential quadratic model (model 5) to best describe the ERR for thyroid cancer in relation to radiation dose, and the values of the parameters in our model (β1
= 1.4, β2
= −0.002) were nearly identical to the parameters estimated in the case-control study (β1
= 1.3, β2
= −0.002). While this model is the most consistent with radiobiological theory (14
), the quadratic exponential quadratic model (model 10) was also consistent with our data, particularly when examining the EAR. Both models 5 and 10, however, demonstrated similar patterns, and sex and time since radiation exposure were identified as modifiers of the EAR dose–response relationships in each of the models (results not shown). It is important to note that at low doses (<1 Gy) model 5 predicts an excess number of thyroid cancers that is nearly 12 times larger than those predicted by model 10 (). At larger doses, however, there were no substantial differences in the predicted number of excess cases between the models.
Also similar to the previous CCSS case-control study, we found neuroblastoma to be a risk factor for thyroid cancer, independent of radiation exposure, which has been observed in another cohort of childhood cancer survivors (3
). However, our finding was based on only nine cases, four of which were diagnosed with neuroblastoma before 1 year of age; only two patients were diagnosed with cancer before age 1 among the other 110 thyroid cancer cases. Given the strong interrelationship between type of first cancer, age at exposure and radiation dose (), our observation for neuroblastoma may be attributable to an age-at-exposure effect for which statistical adjustment was not completely possible. It is also worth mentioning that when patients diagnosed with Hodgkin lymphoma were excluded from the analysis, the thyroid cancer dose–response relationship remained virtually unchanged (results not shown), meaning that our findings are not entirely attributable to Hodgkin lymphoma patients.
We also observed suggestive evidence of an increased risk of thyroid cancer in association with chemotherapy that was independent of radiation exposure, which was not observed in the previous case-control study (1
). However, after adjustment for radiation treatment the association with chemotherapy was relatively weak, of the order of 1.6-fold increased risk, demonstrating that the risk of second primary thyroid cancer is typically dominated by the radiation effect, particularly at the highest radiation doses where cell killing would presumably remove cells from chemotherapy-related cancer risk. Chemotherapy was not found to be a confounder of the association between radiation and thyroid cancer risk. The similarity of the EAR/Gy estimates by chemotherapy status but suggestive heterogeneity of the ERR/Gy estimates by chemotherapy status points to a potentially additive rather than multiplicative interaction between chemotherapy and radiation for the risk of developing a second primary thyroid cancer. Because of the variety of drug regimens and combinations, doses and schedules of treatments, however, it was beyond the scope of the present analysis to assess in detail the effects of chemotherapy as a modifier of radiation-associated risk.
Small numbers of thyroid cancer cases have precluded previous studies of childhood cancer survivors from conclusively demonstrating a downturn in risk at high doses and from examining effect modifiers of the dose response (3
). There are larger studies in other settings with childhood radiation exposure, but doses to the thyroid gland were considerably lower (16
). These lower-dose studies have been restricted to the linear portion of the dose–response relationship, though results from a few studies have suggested a leveling or downward curvature of thyroid cancer risk at the upper dose ranges (4 to 10 Gy) (16
Results for effect modification by sex have varied among previous studies, including no difference in ERR/Gy by sex (16
), a higher ERR/Gy in men compared to women (19
) and a higher ERR/Gy in female subjects compared to male subjects (21
). The EAR/Gy, however, has consistently been shown to be elevated in women compared to men (two- to fourfold) (20
), as in our study, reflecting the higher background rate of thyroid cancer among women.
The radiation-related relative risk of thyroid cancer has also consistently been shown to decrease with increasing age at exposure (19
). When considering time since exposure, a pooled analysis of seven studies of thyroid cancer, primarily among those exposed to radiation as children, found significant heterogeneity in the ERR/Gy, with an increase up to about 30 years, after which the risk began to decline (21
). This pattern was also apparent in an updated analysis of a study of thyroid cancer risk after childhood treatment with radiation for tinea capitis that originally contributed to the pooled analysis (22
). However, a recent case-control study of children in Belarus exposed to radiation from the Chernobyl accident found no significant heterogeneity in radiation-related risk by time since the accident (16
A downturn in risk at high doses has generally not been observed for solid tumors other than thyroid cancer. Risks of breast cancer, central nervous system tumors, osteosarcoma and lung cancer have demonstrated no evidence for departure from linearity for organ doses in excess of 30 to 60 Gy (24
) among childhood cancer survivors and adults treated for Hodgkin lymphoma. Small study size, limited case numbers and narrow ranges of radiation exposure could explain why a downturn in risk was not observed in these studies. To our knowledge, the only other cancer type to demonstrate a downturn in risk at high doses is leukemia. Among patients treated for cervical cancer, an increased risk of leukemia was observed up to bone marrow doses of 4 Gy, after which the risk began to decrease (29
Our study had several strengths and limitations. Strengths included its large size, detailed treatment data and individual level dose estimates to the thyroid gland allowing for complex dose–response modeling. The dosimetry was not as detailed as for the previous case-control study because of the impracticality of evaluating each instance when the thyroid gland was under blocking, given the large number of patients who had radiotherapy. However, our findings are nearly identical to those of the case-control study, suggesting that the less detailed dosimetry was not a major limitation. We relied on self-reported thyroid cancers that were subsequently confirmed by study pathologists, and it may be possible that some cancer survivors neglected to report their second cancers or may not have completed the follow-up questionnaire. Unless such omissions are related to dose, it is unlikely to have introduced bias but would have added some imprecision to the risk estimates. Contrary to findings from the case-control study (2
), we did not observe a significant difference in mean tumor size by type of first cancer, such as might be expected if surveillance was greater after some types of cancer (e.g., Hodgkin lymphoma) than for others. While there appeared to be an independent effect of neuroblastoma, we were unable to discount confounding by age at exposure, given that neuroblastoma patients tended to be younger at time of diagnosis.
In the largest study to date of second primary thyroid cancers among childhood cancer survivors, we have confirmed and strengthened the results of previous studies of populations exposed to radiation during childhood. Sex, age at exposure and time since exposure were found to be significant modifiers of the radiation-related risk of thyroid cancer and as such are important factors to take into account for clinical follow-up and thyroid cancer risk estimation among childhood cancer survivors.