Epidemiological data from Japanese atomic bomb survivors and from children exposed to radiation for medical reasons suggest that excess relative risks (ERRs) for radiation-induced cancers at a given attained age are substantially higher for individuals who are exposed during childhood than for those exposed at older ages (1–12). This finding is reflected in recent estimates of radiation-induced cancer risks as a function of age at exposure produced by the International Commission on Radiological Protection (13) and the National Academy of Sciences Committee on the Biological Effects of Ionizing Radiation [BEIR Committee; (14)]. In these evaluations (13,14), the epidemiological data were described by empiric relationships that predicted that the eventual risks of developing radiation-induced cancer would continue to decrease with increasing age at exposure, for any exposure age.

Whereas the cancer risks due to radiation exposures in childhood have been extensively documented in the literature (7–12), the relationship between radiation-induced cancer risk and age at exposure in adulthood is less clear. Relevant epidemiological data—typically from adults treated with radiotherapy (15–22)—generally do not have sufficient statistical power to assess the dependence of radiation-induced cancer risk on age at exposure in exposed adult populations. More recent analyses of Japanese atomic bomb survivors (1,5,6,14) have shed some light on this issue: As illustrated in Figure 1, these analyses (1,5,6) suggest that the radiation-related ERR for cancer induction decreases with increasing age at exposure only until exposure ages of 30–40 years; at older ages at exposure, the ERR does not decrease further and, for many individual cancer sites (as well as for all solid cancers combined), the ERR may actually increase.

Patterns of radiation risk that do not decrease monotonically as a function of increasing age at exposure represent a challenge to our conceptual understanding of the mechanisms of cancer induction. From a mechanistic perspective, the commonly used biologically based models of carcinogenesis such as the Armitage–Doll model (24), the standard two-stage clonal expansion model (4), and multistage extensions of these models (25–27), all predict that ERRs should decrease continuously with increasing age of radiation exposure. This is because they essentially model the effects of radiation-induced initiation—the induction of irreversibly altered premalignant cells—that are frequently equated with a mutational event in a critical gene. As schematized in Figure 2, A, such initiation processes result in a predicted monotonic decrease in risk with increasing age at radiation exposure, because premalignant cells formed at earlier ages have longer times available to exploit their growth advantage during tumorigenesis, and also potentially through the more rapid cellular proliferation rates in children compared with adults (28).

More recent biologically based models of radiation carcinogenesis (3,23,29,30) not only allow ionizing radiation to act as an initiator of premalignant clones but also consider ionizing radiation as a promoter of preexisting premalignant damage. Promotion is the process by which an initiated cell (radiation-induced initiation or otherwise) clonally expands, proportionately increasing the preexisting average number of premalignant stem cells per clone. As discussed above, initiation processes would be expected to result in decreasing excess lifetime cancer risks with increasing ages at exposure. By contrast, as illustrated in Figure 2, B, promotional processes can result in increasing excess lifetime cancer risks with increasing ages at exposure because the number of preexisting premalignant clones on which promotional processes can act increases with age (31). Overall, because initiation effects are expected to dominate radiation-induced premalignant clone production at younger ages at exposure, whereas promotional effects will dominate for older ages at exposure, a combination of these two effects would produce cancer risk patterns as a function of age at radiation exposure that are schematized in Figure 2, C.

In this study, we analyzed radiation-induced cancer risks as a function of age at radiation exposure of Japanese atomic bomb survivors (1,5,6). Because these data are prima facie inconsistent with a standard initiation-based model predicting monotonically decreasing radiation risks with increasing age at exposure (Figure 1), our primary goal was to assess whether these data are consistent with a model of radiation-induced cancer that includes both initiation and promotion (23,30) components (Figure 2). Then, we used such a combined model to estimate age-dependent lifetime radiation-induced cancer risks after adult radiation exposure and compared these risk estimates with those generated from standard empirical models in which radiation risks are constrained to decrease with increasing age at exposure.

We conclude that the observed patterns of radiation-induced cancer risks as a function of age of exposure are not consistent with standard models of radiation carcinogenesis in which radiation solely initiates premalignant cells but are consistent with models of radiation carcinogenesis that include both radiation-induced initiation and promotion. This conclusion is conceptually important because many commonly used biologically based models of radiation-induced carcinogenesis, such as various derivatives of the original Armitage–Doll model (24,27), describe only the initiating component of radiation carcinogenesis, and do not describe potential radiation-induced promotional effects. More recently, however, several investigators have also published mechanistically based models of radiation-induced carcinogenesis that consider, in effect, both radiation-induced initiation and promotion (3,29).

Our second goal was to use an initiation- and promotion-based model, with radiation parameters estimated from fitting the atomic bomb survivor ERR data, to generate absolute lifetime radiation-induced cancer risks per unit dose in the US population, as a function of age at exposure. The results, summarized in Figure 3, suggest that the radiation-related cancer risk for an exposure at, for example, age 50 years, could be twice as high as estimated using standard models in which radiation-induced cancer risks are constrained such that the ERRs cannot increase with increasing age at exposure.

Practically speaking, there could be considerable societal consequences if the excess lifetime cancer risks for radiation exposure in middle age are somewhat higher than previously estimated, for example, in the recent BEIR-VII (14) or International Commission on Radiological Protection (13) reports. The majority of the radiation exposures in the population, both medical and occupational radiation exposures, occur in individuals who are older than 30 years (41). The relevant regulatory occupational exposure limits for ionizing radiation are derived almost entirely from analyses of atomic bomb survivors who were exposed in adulthood (13,42–44). Thus, an increase in the best estimate of the excess lifetime cancer risk after radiation exposure in middle age might be reflected in a corresponding change in occupational radiation exposure limits. The practical implications of such a change would probably not be wide ranging because it is unusual for radiation workers to be exposed to doses close to the regulatory limits; however, there could be implications for those activities in which individuals are occasionally exposed to doses near these limits, such as for staff in interventional radiology facilities (45) or for some emergency responder scenarios (46).

A far more common source of radiation exposure in middle age is from diagnostic radiology (41). The medically related component of the US population exposure to ionizing radiation has increased sixfold in the past three decades (47), mostly because of the rapid increase in computerized tomography (CT) imaging (48). The most common ages at which individuals undergo CT examinations are approximately 35 to 50 years (49). When a CT scan is medically warranted, its benefits far outweigh any radiation risks, so that even increasing the estimated risks by a factor of 2 would not materially affect the risk–benefit balance (48). However, the risk–benefit balance is potentially relevant for CT-based screening of asymptomatic “healthy” adults. Specifically, routine CT screening of the colon (50,51), lung (52–55), and heart (56–58) are increasingly being advocated. Lung and cardiac CT screening are of particular relevance here because in both cases, the most important organ in terms of radiation risk is the lung, and we have shown here that there is good evidence (Figures 1 and ​and3)3) that the excess lifetime risks of lung cancer do not decrease in middle age and indeed may peak at around age 50 years, the most likely age for individuals to undergo CT screening.

This study has several limitations, the primary one being the statistical uncertainties associated with the underlying data (5,6) from the atomic bomb survivors (see error bars in Figure 1). As can be seen in Figure 1, because the data are stratified by age at exposure, the individual confidence intervals are quite wide, particularly when further stratification by cancer site is made. Little (6) has shown that the data for all solid cancer cancers combined, as shown in Figure 1, are inconsistent with an ERR that decreases monotonically with increasing age at exposure. Although data for cancer mortality are not analyzed here, similar empiric conclusions have been reached for the variation in radiation-induced cancer mortality with age at radiation exposure (6). The corresponding atomic bomb survivor data for individual cancer sites (Figure 1) did not reach statistical significance, which is not surprising given the decreased statistical power (6) but, as shown in Figure 1, the same trends, for adult exposure (radiation risks not monotonically decreasing with increasing age at adult exposure), were consistently seen for liver, colon, lung, stomach, and bladder, but not for breast cancer. A recently detailed analysis of combined radiation and smoking effects among atomic bomb survivors (59) strongly suggests that the observed increase in ERR with age at exposure for lung cancer is present irrespective of smoking status. This same pattern of increasing ERR with increasing age at adult exposure was seen in a large study of cancer risks in more than 400000 radiation workers in the nuclear industry (30), where a lower ERR for all solid cancers was observed in workers exposed at ages 20–35 years compared with workers exposed at older ages (P = .09). This pattern of increasing ERR with increasing age at adult exposure was not, however, seen in a smaller study of cancer mortality in 20000 radiation workers at the Russian Mayak nuclear complex (60), although a study of approximately 30000 individuals exposed to protracted radioactive contamination from the Mayak complex (61) showed increased (P = .08) cancer mortality per unit dose with increasing age at first exposure. Overall, the weight of the epidemiological evidence suggests that for adult exposures, radiation risks do not generally decrease with increasing age at exposure, and the mechanistic underpinning described here provides this conclusion with some biological plausibility.