Limits...
Low dose radiation risks for women surviving the a-bombs in Japan: generalized additive model

View Article: PubMed Central - PubMed

ABSTRACT

Background: Analyses of cancer mortality and incidence in Japanese A-bomb survivors have been used to estimate radiation risks, which are generally higher for women. Relative Risk (RR) is usually modelled as a linear function of dose. Extrapolation from data including high doses predicts small risks at low doses. Generalized Additive Models (GAMs) are flexible methods for modelling non-linear behaviour.

Methods: GAMs are applied to cancer incidence in female low dose subcohorts, using anonymous public data for the 1958 – 1998 Life Span Study, to test for linearity, explore interactions, adjust for the skewed dose distribution, examine significance below 100 mGy, and estimate risks at 10 mGy.

Results: For all solid cancer incidence, RR estimated from 0 – 100 mGy and 0 – 20 mGy subcohorts is significantly raised. The response tapers above 150 mGy. At low doses, RR increases with age-at-exposure and decreases with time-since-exposure, the preferred covariate. Using the empirical cumulative distribution of dose improves model fit, and capacity to detect non-linear responses. RR is elevated over wide ranges of covariate values. Results are stable under simulation, or when removing exceptional data cells, or adjusting neutron RBE. Estimates of Excess RR at 10 mGy using the cumulative dose distribution are 10 – 45 times higher than extrapolations from a linear model fitted to the full cohort. Below 100 mGy, quasipoisson models find significant effects for all solid, squamous, uterus, corpus, and thyroid cancers, and for respiratory cancers when age-at-exposure > 35 yrs. Results for the thyroid are compatible with studies of children treated for tinea capitis, and Chernobyl survivors. Results for the uterus are compatible with studies of UK nuclear workers and the Techa River cohort.

Conclusion: Non-linear models find large, significant cancer risks for Japanese women exposed to low dose radiation from the atomic bombings. The risks should be reflected in protection standards.

Electronic supplementary material: The online version of this article (doi:10.1186/s12940-016-0191-3) contains supplementary material, which is available to authorized users.

No MeSH data available.


Related in: MedlinePlus

Optimisation methods and Bayesian posterior confidence intervals. For all solid cancers in A- (0 - 20 mGy excluding NIC), the model P4se (see Table 2) is fitted with smoothing parameter optimisation by GCV.Cp, ML, and REML. At 35 years time-since-exposure, Relative Risk is shown by grey curves, and 95% Bayesian posterior confidence intervals (Appendix A) are shown by blue and red curves. Simulation gives estimates of coverage, Bootstrap-t CIs (triangles), and an alternative CI (large circles) stretched to achieve 95% coverage (Appendix B), with stretch factors as shown in the plot key
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC5121957&req=5

Fig2: Optimisation methods and Bayesian posterior confidence intervals. For all solid cancers in A- (0 - 20 mGy excluding NIC), the model P4se (see Table 2) is fitted with smoothing parameter optimisation by GCV.Cp, ML, and REML. At 35 years time-since-exposure, Relative Risk is shown by grey curves, and 95% Bayesian posterior confidence intervals (Appendix A) are shown by blue and red curves. Simulation gives estimates of coverage, Bootstrap-t CIs (triangles), and an alternative CI (large circles) stretched to achieve 95% coverage (Appendix B), with stretch factors as shown in the plot key

Mentions: For all solid cancers, no minimum smoothing parameters were imposed. Basis dimension k=10 was sufficient when tested by gam.check or modelling residuals against covariates. The ML method of smoothing parameter optimisation was preferred to alternatives, giving 95% CIs which performed well under simulation tests described in Appendix B and shown in Fig. 2. ML is discussed in [24].Fig. 2


Low dose radiation risks for women surviving the a-bombs in Japan: generalized additive model
Optimisation methods and Bayesian posterior confidence intervals. For all solid cancers in A- (0 - 20 mGy excluding NIC), the model P4se (see Table 2) is fitted with smoothing parameter optimisation by GCV.Cp, ML, and REML. At 35 years time-since-exposure, Relative Risk is shown by grey curves, and 95% Bayesian posterior confidence intervals (Appendix A) are shown by blue and red curves. Simulation gives estimates of coverage, Bootstrap-t CIs (triangles), and an alternative CI (large circles) stretched to achieve 95% coverage (Appendix B), with stretch factors as shown in the plot key
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC5121957&req=5

Fig2: Optimisation methods and Bayesian posterior confidence intervals. For all solid cancers in A- (0 - 20 mGy excluding NIC), the model P4se (see Table 2) is fitted with smoothing parameter optimisation by GCV.Cp, ML, and REML. At 35 years time-since-exposure, Relative Risk is shown by grey curves, and 95% Bayesian posterior confidence intervals (Appendix A) are shown by blue and red curves. Simulation gives estimates of coverage, Bootstrap-t CIs (triangles), and an alternative CI (large circles) stretched to achieve 95% coverage (Appendix B), with stretch factors as shown in the plot key
Mentions: For all solid cancers, no minimum smoothing parameters were imposed. Basis dimension k=10 was sufficient when tested by gam.check or modelling residuals against covariates. The ML method of smoothing parameter optimisation was preferred to alternatives, giving 95% CIs which performed well under simulation tests described in Appendix B and shown in Fig. 2. ML is discussed in [24].Fig. 2

View Article: PubMed Central - PubMed

ABSTRACT

Background: Analyses of cancer mortality and incidence in Japanese A-bomb survivors have been used to estimate radiation risks, which are generally higher for women. Relative Risk (RR) is usually modelled as a linear function of dose. Extrapolation from data including high doses predicts small risks at low doses. Generalized Additive Models (GAMs) are flexible methods for modelling non-linear behaviour.

Methods: GAMs are applied to cancer incidence in female low dose subcohorts, using anonymous public data for the 1958 – 1998 Life Span Study, to test for linearity, explore interactions, adjust for the skewed dose distribution, examine significance below 100 mGy, and estimate risks at 10 mGy.

Results: For all solid cancer incidence, RR estimated from 0 – 100 mGy and 0 – 20 mGy subcohorts is significantly raised. The response tapers above 150 mGy. At low doses, RR increases with age-at-exposure and decreases with time-since-exposure, the preferred covariate. Using the empirical cumulative distribution of dose improves model fit, and capacity to detect non-linear responses. RR is elevated over wide ranges of covariate values. Results are stable under simulation, or when removing exceptional data cells, or adjusting neutron RBE. Estimates of Excess RR at 10 mGy using the cumulative dose distribution are 10 – 45 times higher than extrapolations from a linear model fitted to the full cohort. Below 100 mGy, quasipoisson models find significant effects for all solid, squamous, uterus, corpus, and thyroid cancers, and for respiratory cancers when age-at-exposure > 35 yrs. Results for the thyroid are compatible with studies of children treated for tinea capitis, and Chernobyl survivors. Results for the uterus are compatible with studies of UK nuclear workers and the Techa River cohort.

Conclusion: Non-linear models find large, significant cancer risks for Japanese women exposed to low dose radiation from the atomic bombings. The risks should be reflected in protection standards.

Electronic supplementary material: The online version of this article (doi:10.1186/s12940-016-0191-3) contains supplementary material, which is available to authorized users.

No MeSH data available.


Related in: MedlinePlus