Second solid cancers after radiation therapy: a systematic review of the epidemiologic studies of the radiation dose-response relationship

Abstract

Rapid innovations in radiation therapy techniques have resulted in an urgent need for risk projection models for second cancer risks from high-dose radiation exposure, because direct observation of the late effects of newer treatments will require patient follow-up for a decade or more. However, the patterns of cancer risk after fractionated high-dose radiation are much less well understood than those after lower-dose exposures (0.1-5 Gy). In particular, there is uncertainty about the shape of the dose-response curve at high doses and about the magnitude of the second cancer risk per unit dose. We reviewed the available evidence from epidemiologic studies of second solid cancers in organs that received high-dose exposure (>5 Gy) from radiation therapy where dose-response curves were estimated from individual organ-specific doses. We included 28 eligible studies with 3434 second cancer patients across 11 second solid cancers. Overall, there was little evidence that the dose-response curve was nonlinear in the direction of a downturn in risk, even at organ doses of ≥60 Gy. Thyroid cancer was the only exception, with evidence of a downturn after 20 Gy. Generally the excess relative risk per Gray, taking account of age and sex, was 5 to 10 times lower than the risk from acute exposures of <2 Gy among the Japanese atomic bomb survivors. However, the magnitude of the reduction in risk varied according to the second cancer. The results of our review provide insights into radiation carcinogenesis from fractionated high-dose exposures and are generally consistent with current theoretical models. The results can be used to refine the development of second solid cancer risk projection models for novel radiation therapy techniques.

Figure 1

Relative risk (& 95% CI) for subsequent breast cancer according to the estimated absorbed radiation dose (Gy) Dotted black line indicates fitted linear dose-response for that study. Dashed grey line indicates the RR for similar age at exposure and attained age based on the pooled ERR model from Preston et al (31). *Travis et al (2003) the reference dose category was 0–3Gy, and the fitted linear dose-response was for women who did not receive alkylating agents or ovarian radiation (5+Gy).

Figure 2

Relative risk (& 95% CI) for subsequent brain cancer according to according to the estimated absorbed radiation dose (Gy) Dotted black line indicates fitted linear dose-response for that study. Dashed grey line indicates the RR for similar age at exposure and attained age based on the BEIR VII risk models (4,9). *Reference categories for Neglia et al (2006) were <1Gy rather than 0Gy, which is used for the fitted dose-response, and this accounts for any apparent discrepancy between the categorical estimates and fitted line.

Figure 3

Relative risk (& 95% CI) for subsequent lung cancer according to the estimated absorbed radiation dose (Gy) Dotted black line indicates fitted linear dose-response for that study. Dashed grey line indicates the RR for similar age at exposure and attained age based on the BEIR VII risk models (4).

Figure 4

Relative risk (& 95% CI) for subsequent thyroid cancer according to the estimated absorbed radiation dose (Gy) Dotted black line indicates fitted linear dose-response for that study. Dashed grey line indicates the RR for similar age at exposure and attained age based on the BEIR VII risk models (4). *For Tucker et al (1991) the reference category was <2Gy rather than 0Gy, which is used for the fitted dose-response, and this accounts for any apparent discrepancy between the categorical estimates and fitted line. The fitted linear dose-response model for Bhatti et al (2010) is based on the linear term from the linear-quadratic model.

Figure 5

Relative risk (& 95% CI) for subsequent bone sarcoma according to the estimated absorbed radiation dose (Gy) Dotted black line indicates fitted linear dose-response for that study. BEIR VII risk models not available. *No confidence intervals were available for Tucker et al. **Reference category for Le Vu et al (1998) and Boice et al (1988) was <1Gy rather than 0Gy, which is used for the fitted dose-response, and this accounts for any apparent discrepancy between the categorical estimates and fitted line. RR for Le Vu et al 40Gy dose category=184.