This model-based analysis of screening showed that mammography threshold and doubling time varied with age. This analysis also showed that age-specific differences in the mammographic threshold contributed more than the age-specific differences in doubling time to poorer outcomes associated with screening women aged 40–49 years compared with women aged 50–69 years. These findings are consistent with the empiric findings of Buist et al. (14
) regarding the effect of mammography breast density and tumor growth on screening sensitivity. In addition, we showed that these relative contributions extended to other screening outcomes, including mean detectable tumor size, lifetime gained, and breast cancer mortality.
In a sensitivity analysis, we varied the mammography threshold and doubling time across a broad range of values and analyzed their effect on the joint relationship between sensitivity and breast cancer mortality. On doing so, when comparing annual with biennial screening, we found that the largest sensitivity gain was associated with the fastest growing tumors but these tumors were not associated with the largest mortality reduction. Because fast growing tumors are more likely to be screen detected close to the time that they would have been symptomatically detected in the absence of screening, early detection of such tumors produces a small effect on breast cancer mortality reduction. Therefore, when comparing annual with biennial screening, a high sensitivity gain should not be used to infer a high reduction in breast cancer mortality; similarly, a moderate or low sensitivity gain should not be used to infer a low reduction in breast cancer mortality. We found that the greatest reduction in breast cancer mortality corresponded to moderate sensitivity gains and was achieved at low mammography thresholds and moderately fast growing tumors, further highlighting the greater impact of mammography threshold compared with tumor growth rate on the reduction in breast cancer mortality attributed to screening.
Our results underscore the importance of continued efforts to improve technologies for early detection of breast cancer in younger women, particularly in women with dense breast tissue. Digital mammography and ultrasound are two technologies that recently demonstrated higher detection rates in dense breast tissue (19
). In the Digital Mammographic Imaging Screening Trial (DMIST), digital mammography performed better than film mammography for women less than 50 years old with dense breasts (19
). The sensitivity of annual screening with digital mammography in women less than 50 years old was reported to be 0.68 (95% confidence interval = 0.56 to 0.77) (20
); the point estimate of 0.68 is close to our estimated sensitivity of 0.7 in scenario 3. This indicates that digital mammography in younger women is performing as well as conventional mammography performs in older women for invasive disease; although this comparison is not completely accurate, because unlike our study, DMIST included ductal carcinoma in situ.
In the American College of Radiology Imaging Network National Breast Ultrasound Trial (ACRIN-6666), addition of a single screening ultrasound examination to a screening mammogram increased the detectability of breast cancer, compared with mammography alone, among women who were at increased risk of breast cancer and also had dense breast tissue (21
). However, although the supplemental ultrasound screening uncovers more breast cancer, it also substantially increases the risk of a false-positive finding and unnecessary biopsy. Hence, ACRIN-6666 and DMIST trials highlight a key aspect that was not considered in this analysis—an increase in sensitivity may be accompanied by a decrease in specificity and an increase in costs. When evaluating a new screening technology to an existing one, the benefits are often compared with the harms and cost. A recent cost-effectiveness analysis of DMIST found that digital mammographic screening, compared with film-based screening, results in sufficient health gains in younger women to warrant its increased cost (20
This study on the use of a computer simulation model to analyze screening mammography outcomes has a few limitations. The findings depend on estimated age-specific differences in mammography threshold and doubling time, but admittedly, these estimates are subject to biases. We assumed that missing BCSC data on screen-detected tumor sizes were randomly distributed. Also, when using the BCSC data, we defined a cancer as screen detected if the diagnosis was made within 4 months of the last screening mammography, but alternative definitions could vary our estimates. Despite these limitations, our age-specific estimates of doubling times are similar to estimates of doubling times in other studies (22
). A further limitation of our analysis is that we did not consider the possibility that low mammographic tumor detectability, which we used as a surrogate for high breast density, could be considered a breast cancer risk factor in itself (25
) and may even be associated with a specific subtype of breast cancer. To incorporate these aspects, more research is needed to not only establish a better relationship between mammographic breast density and breast cancer risk but also understand the differences in the tumor characteristics in dense vs nondense breast tissue.