Our estimates suggest that for all women a decade of annual mammographic screening starting at age 20 years would cause more radiation-induced breast cancer deaths than it prevents, and starting at age 30 years it is unlikely to result in a net reduction in breast cancer mortality. However, a decade of annual screening starting at age 40 years could result in a material net decrease in breast cancer mortality if, among women screened, breast cancer mortality is reduced by about 20% or more. If the mortality reduction is 10% in women screened, then a decade of annual screening starting at age 40 years may have little or no net benefit. Results for women with first-degree relatives with breast cancer were generally in the same direction but, because their background incidence rates are higher, the net increases or decreases were greater.
These calculations were necessarily based upon a number of assumptions and parameter estimates; however, the sensitivity analysis suggested that, for the most part, the conclusions were unlikely to be significantly altered by varying the parameters within a feasible range. One exception was for screening starting at age 40 years for all women, where the direction of the net effect was altered by varying the assumptions, however the magnitude of the net effects were modest eitherway.
In this paper, we have focused on the comparison of the radiation-induced breast cancer deaths with the number of breast cancer deaths prevented by screening. However, the radiation-induced cancers are likely to occur later in life, on average, than the deaths that are prevented by screening. Therefore, we also compared the years of life lost and gained by each decade of mammographic screening. The conclusions from these analyses were generally similar to those from the analysis of numbers of deaths, with the exception of estimated small net gains in years of life for a decade of annual screening starting at age 30 years for all women if a 20% mortality reduction from screening was assumed, and also for all women for a decade of screening starting at age 40 years if a 10% mortality reduction was assumed (net gain=2 and 7 years of life per 1000 women screened, respectively).
In our calculations we did not assume that women attend for regular screening after the specific decade of interest, because the question under investigation was the net effect of each decade of screening. The reason for this is that we do not think that screening could be recommended to a certain age group on the basis of guaranteed future screening attendance. Future screening attendance could reduce the magnitude of the risk of radiation-induced breast cancer mortality, if some of these cancers were detected by screening. Therefore, we investigated the effect on the estimated risk of radiation-induced breast cancer mortality of assuming 100% future screening attendance by increasing the breast cancer survival probabilities by 10 or 20% for screening before age 50 years and by 35% after age 50 years. For a 20% mortality reduction due to a decade of annual screening, assuming 100% future screening attendance reduced the net increase in breast cancer deaths from 0.86 to 0.62 per 1000 women screened starting at age 20 years, from 0.37 to 0.16 starting at age 30 years, and for starting at age 40 years this changed the net decrease from 0.46 to 0.64 per 1000 women screened. Therefore, even if we assumed 100% future screening attendance, this would be likely to alter the magnitude of the net change, but not the direction of the result.
In our calculations we assumed women would be screened annually with a two-view mammogram. In the UK Age trial, two-view mammography was used for the first screen only, whereas subsequent rounds used single-view mammography (Moss et al, 2005
). The radiation dose from a single-view mammogram is 2.5
mGy (Young et al, 2005
) and so under this screening pattern, the estimated risk of radiation-induced breast cancer mortality for annual mammographic screening from age 40 to 47/48 years would approximately be halved (0.22 breast cancer deaths per 1000 women screened). Although reducing the number of views per screen, or the frequency of screening, will reduce the radiation risk it may also decrease the reduction in breast cancer mortality due to screening.
The estimates of the radiation-induced breast cancer risk were based on the linear no-threshold assumption,that is assuming the radiation risks are linear in dose down to very low doses and that there is no threshold dose below which there is no risk of cancer. A recent review of the available biological and epidemiological evidence concluded that there is direct epidemiological evidence of an excess cancer risk from fractionated radiation doses as low as 50
mGy (Brenner et al, 2003
), which is approximately the dose received from a decade of annual two-view screening mammograms (45
mGy). Therefore, the assumptions made in this article about the existence of cancer risks at these low dose levels are supported by epidemiological evidence, but the extrapolation is necessary because it is not feasible to quantify the risks using observational studies directly. Brenner et al (2003)
also reviewed the evidence regarding the most appropriate form of the extrapolation and concluded that the assumption of linearity was most consistent with the experimental evidence and, furthermore, that alternative forms could result in larger as well as smaller risk estimates. As Preston et al (2002)
found no evidence that fractionated exposure resulted in a lower risk of radiation-induced breast cancer than acute exposure, a dose rate reduction effectiveness factor was not included in these calculations.
The estimates for women with a family history of breast cancer were based upon the assumption that the excess relative risk of radiation-induced breast cancer per unit dose for these women is the same as for all women, that is, that there is no supra or submultiplicative interaction between radiation exposure and a family history of breast cancer. However, BRCA-1 and BRCA-2 mutations appear to be associated with a reduction in efficiency of DNA repair, which suggests that there may be an interaction between these two risk factors, at least for this subgroup of women with a family history of breast cancer (IARC, 2000
). To date, only one study has investigated the risk of radiation-induced breast cancer in women with a family history of breast cancer directly, and reported that women with a family history of breast cancer might have a greater relative risk of radiation-induced breast cancer (Ronckers, 2003
). Further research into this question is needed, because if this were true then the radiation risks reported in this paper for women with a family history of breast cancer could be underestimates.
The risk model that was used for these calculations was an excess relative risk model based on a pooled analysis of three cohorts, including two cohorts of women who were exposed to multiple fluoroscopy examinations (Preston et al, 2002
). In their pooled analysis, Preston et al
found no single excess relative or excess absolute risk model that adequately described the risk of radiation-induced breast cancer across all of the eight cohorts considered. For breast cancer risk estimation in general populations, they suggested the use of their pooled excess absolute risk model, which included four of the possible eight cohort studies. Formal statistical comparison of the fits of the excess relative risk and excess absolute risk models is not possible, but an informal comparison based on deviance values suggested that the excess relative risk model fitted the data marginally better. Furthermore, the assumption underlying the excess absolute risk model is that the risk of radiation-induced breast cancer is not related to the underlying breast cancer incidence rate in the population. This is equivalent to assuming that the relative increase in the risk of breast cancer associated with radiation is actually lower for women with a greater than average baseline risk of breast cancer, such as women with a family history, than it is for other women. As explained above, to date there is little reliable information on the risk of radiation-induced breast cancer specifically in women with a family history of the disease, but we do not think currently that such an assumption is justifiable. However, in the sensitivity analysis, we investigated the effect of using Preston et al
's pooled excess absolute risk model for all women, and although the radiation risk estimates were somewhat lower than those estimated using the excess relative risk model, the conclusions were not materially altered ().
Several previous studies have also estimated the radiation risks associated with mammographic screening of younger women. Feig and Hendrick (1997)
estimated the risks from screening women aged 40–49 years and, assuming mortality reductions of between 24 and 36% for screened women, concluded that the radiation risks would be small compared to the mortality benefits. Beemsterboer et al (1998)
and Mattsson et al (2000)
focused on the question of whether to start screening all women at age 40 years rather than at age 50 years. Both conclude that this strategy would reduce the net reduction in breast cancer mortality. Mattsson et al
also concluded that at least a 20% annual reduction in breast cancer mortality was necessary for the reduction in breast cancer mortality to outweigh the radiation risks if screening starts at age 40 years. Finally, Law and Faulkner (2001)
considered the question of screening all women younger than age 50 years and those with a family history of breast cancer by estimating the ratio of the number of cancers that might be detected by screening compared to the number of cases induced by radiation from a single mammogram. The interpretation of this ratio is much less straightforward than the comparison of deaths induced to deaths prevented, but the authors suggest that a ratio of 10
1 may be necessary to recommend screening and concluded therefore that mammographic screening should certainly not start before age 35 years.
The estimates for women with a family history of breast cancer were for women with one or two affected first-degree relatives, but these estimates could also be applied to other groups of women that have a similarly increased underlying risk of breast cancer. For example, for a woman with one first-degree relative who was diagnosed with breast cancer before age 40 years, the conclusions would be similar to those we have presented for women with two affected first-degree relatives. Furthermore, although our estimates are based on breast cancer incidence and mortality data from the UK, the other parameters that were used in the calculations including the radiation risk models and relative risks for a family history of breast cancer were all based upon data from international studies. Therefore, it is likely that our findings would be broadly applicable to other Western populations with broadly similar breast cancer incidence and mortality rates.
In conclusion, our estimates suggest that a decade of annual mammographic screening before age 40 years would result in a net increase in breast cancer deaths, and that starting at age 40 years could result in a material net decrease in breast cancer deaths if breast cancer mortality is reduced by about 20% or more in women screened. Although these calculations were based on a number of uncertain parameters, in general, the conclusions were not altered when the parameters were varied within a feasible range.