Longitudinal history of mammographic breast density and breast cancer risk by familial risk, menopausal status, and initial mammographic density level in a high risk cohort: a nested case–control study

Abstract

Background

Elevated mammographic density is associated with increased breast cancer risk. However, the contribution of longitudinal changes in mammographic density to breast cancer risk beyond initial mammographic density levels, considering familial breast cancer risk and menopausal status, remains uncertain but holds important clinical implications.

Methods

In a nested case–control study within the Sister Study (323 cases, 899 controls; 12,095 mammograms), a cohort enriched for family history of breast cancer, we examined case–control status in relation to the largest annual change in percent density and dense area using mammograms available spanning 5.4 years, on average, using multivariable logistic regression and to the rate of mammographic density change using linear mixed-effects models. We considered effect modification by: mammographic density level of the earlier mammogram, the extent of family history, Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation (BOADICEA) risk strata, and menopausal status.

Results

Cases (diagnosed < 60 years) had greater initial percent density and dense area levels and a slower rate of decline in dense area than controls. Women with stable mammographic density (≤ 10% annual change) had an increased breast cancer risk as compared with women whose largest mammographic density change was > 10% annual decline (e.g., Odds Ratio (OR) 2.34, 95% Confidence Interval (CI) 1.63–3.37 for dense area). Increasing vs. decreasing dense area was also associated with elevated risk, especially in women with the highest dense area levels at the earlier mammogram (OR: 2.56, 95%CI 1.50–4.36). Although generally similar across menopausal and familial risk categories, the associations of MD change with risk appeared stronger in pre-menopausal and lower-risk women.

Conclusions

Women who maintain higher levels of mammographic density (i.e. do not decrease over time) or have increasing mammographic density over time have a higher risk of subsequent breast cancer than women with high mammographic density that decreases over time. These findings suggest potential for incorporating mammographic density trajectories in clinical risk assessment, and the importance of additional breast cancer monitoring in women not experiencing declines in mammographic density over time.

Background

Mammographic density (MD) refers to the breast tissue composition, distinguishing between non-dense (primarily adipose) and dense (primarily fibroglandular) components on mammograms [1, 2]. Consistent evidence across MD measurement methods and populations show a two-to-four fold increase in the relative risk of breast cancer when comparing very high MD to very low MD [3, 4]. MD is not static and has been linked to growth, reproductive, and hormonal factors that influence breast tissue composition [5,6,7,8]. Mammographic density reductions in response to endocrine therapies in interventional studies have also been associated with subsequent breast cancer risk reduction [9, 10]. Together, these observations support that changes in MD may provide insights into breast cancer risk; yet studies of longitudinal MD changes remain limited.

The association between MD change (MDC) and breast cancer risk may depend on the initial MD levels and the magnitude of change in MD. Large changes in MD across the four clinical categories of Breast Imaging Reporting and Data Systems (BI-RADS) breast composition (typically reflecting ≥ 25% change) have been shown to correspond with breast cancer risk while results are less clear for smaller MDCs using semi- or fully-automated continuous computational methods [11,12,13,14,15,16,17]. For example, a study within the KARMA Cohort, using continuous measures of MD and smaller MDC between two mammograms (10% change), reported that annual MDC did not predict breast cancer risk beyond the MD measured during the earlier mammogram [14]. Investigations with only two mammograms over a short follow-up period, however, may miss the critical periods of MDC and fail to capture the overall patterns of change that require a larger number of mammograms. Findings from a recent nested case–control study within a health system mammography practice, which included serial mammograms per women, showed a slower rate of MD decline for women with breast cancer in the breast that later developed cancer (ipsilateral) [18]. This finding replicated an earlier study reporting smaller MD declines over time in contralateral breasts in women with breast cancer compared to women without breast cancer despite similar initial MD levels [19].

Despite familial breast cancer susceptibility and MD being strong risk factors [20, 21], there is a paucity of data on whether and how the two risk factors independently or jointly affect risk. Understanding MDC in women at high risk for breast cancer is relevant for clinical risk stratification and informing screening and risk reduction strategies for high-risk women based on their MD trajectories. Available research has primarily focused on average-risk and postmenopausal women, overlooking the more dynamic MDCs in the pre-menopausal years. We test the hypothesis that the increasing MD or maintaining higher levels of MD over time contributes to breast cancer risk beyond a single earlier measure of MD. Drawing upon a study population enriched for women with family history of breast cancer (FHBC), we use multiple mammograms for each woman to assess the longitudinal history of MD and the rate of MDC in relation to risk, with further evaluation of differences in these associations by earlier MD, familial breast cancer risk, and menopausal status.

Methods

Study population

We used data from the Integrating Mammograms in Analyses of Genes and Environment in Sisters (IMAGES), a nested case–control study within the prospective Sister Study cohort (for details see [22]). The cases in the IMAGES study were prospectively diagnosed with invasive breast cancer or ductal carcinoma in situ (DCIS) at or before age 60. Given the retrospective collection of pre-diagnostic mammograms, this age restriction facilitated adequate coverage of pre- and peri-menopause periods when MD tend to be higher and more likely to change [23]. Using incidence density sampling, controls were selected from the cohort of women who were being followed, remained breast cancer free at the time of case selection, and were comparable to cases on age at and year of enrollment.

A random sample of 3,506 participants (1263 cases, 2243 controls) meeting the case and control selection criteria were invited to provide information about prior mammograms and medical authorization for release of mammogram copies. Of the 2293 participants (65% of selected; 858 cases, 1435 controls) who provided authorization and the necessary facility information, we retrieved mammograms for 1917 (84% of women with authorization; 693 cases, 1224 controls). The final sample included 1222 women with at least two available cranio-caudal view mammograms of the same side breast obtained on two different dates at least nine months apart at ages ≤ 60 years (64% of women with mammograms; 323 cases, 899 controls), and that could be read through Cumulus software. A total of 12,095 mammograms, obtained between 1991 and 2019 (60.0% taken after 2010) were available for this sample, with 863 in analog/film format and the remainder (92.9%) in digital format. Mammograms pre-diagnosis (cases) or before age 60 (controls) spanned approximately 5 years (median 3.1 years for cases, 5.9 years for controls). We found minimal differences between women who were included in the analyses relative to women who were initially selected (Supplementary Table 1).

All participants provided signed informed consent, and both studies received Institutional Review Board approval (National Institutes of Health for the Sister Study, and Columbia University Irving Medical Center for the IMAGES study).

MD assessment

A single trained reader, blinded to all study data, used Cumulus software to read breast images in batches of approximately 100 images, assessing dense area in cm² and percent dense area (dense area/total breast area*100). Within-batch reliability scores (based on five randomly duplicated mammograms within the same batch) were 96% and 95% and between-batch reliability (based on five mammograms repeated in each batch) were 91% and 93% for percent density and dense area respectively. We used images for left breasts when images for both breasts were available and met the criteria for MDC measures.

MDC measurement

We assessed MD change using mammograms from the same side breasts and included only pre-diagnostic mammograms for cases, including images from the contralateral side or images from ipsilateral breasts taken at least one year before diagnosis when contralateral images were unavailable (seven cases). To measure the predominant change in MD, we identified the largest longitudinal MD change for each woman by subtracting MD between any two selected mammograms that were of the same format and taken ≥ nine months apart, divided by the number of years to obtain annual change, separately for percent density and dense area. Our primary MD change measure is the relative annual change in MD ([((mammogram 2 − mammogram 1)/mammogram 1)/(date 2 − date 1)]*365.25), using categories that correspond to decrease (> 10%), stable (± ≤ 10% change) or increase (> 10%) for both percent density and dense area. We further assessed absolute annual change in MD ((mammogram 2 − mammogram 1)/(date 2 − date 1)*365.25). Alternative cut-points for defining MDC change (e.g., 5% and 15% relative change) did not alter the overall findings. For analysis of rate of MD change, we included all digital mammograms available for a woman following the same selection parameters described above (n = 1,169, 293 Cases, 876 controls).

Statistical analysis

We considered baseline data for breast cancer risk factors and sociodemographic variables. We used age at menopause, assessed through baseline and follow-up questionnaires, to approximate the menopausal status at the time of each mammogram. We used the menopausal status closest to the time of the largest change for analysis of the largest MDC and used baseline menopausal status for analysis of rate of MDC. On average, mammogram pairs with the largest change were about 1.3 years apart (median: 1.0 years for cases, 1.1 years for controls). We used three measures of FHBC: number of first-degree relatives with breast cancer (0, 1, ≥ 2; half-sisters were considered second-degree relatives), number of first- or second-degree relatives with breast cancer (1, 2, ≥ 3), and the youngest age at breast cancer diagnosis for any affected first-degree relative (< 40, 40–49, ≥ 50). We calculated absolute 5-year risk scores (high risk: ≥ 1.67%) using the Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation (BOADICEA), version 5, based on detailed family history information including each affected relative’s age at diagnosis and at death [24]. Given the positive correlation between BOADICEA risk score and FHBC, we examined these variables separately in multivariable analyses.

To evaluate the associations for the largest MD change, we started with a minimally adjusted unconditional logistic regression model, which in addition to MD change (continuous and in categories of decline, stable and increase), included MD and age both at the time of the earlier mammogram, and body mass index (BMI) based on examiner-measured height and weight at baseline. We also considered additional BMI data at different time points as well as adjusting for all BMI measures but this did not substantively change the results; we therefore limit our presentation to models adjusting for baseline BMI. We then expanded this age and BMI adjusted model to include additional covariates and examined multiplicative interaction between largest MD change category and (a) the earlier MD, (b) familial breast cancer risk (BOADICEA risk score, FHBC), and (c) menopausal status. To examine the rate of MD change, we used linear mixed-effects models to account for the correlated multiple MD data per woman and incorporated an interaction term between the follow-up time since each mammogram and the case–control status to evaluate whether MD exhibited different rates of change over time for cases and controls by familial risk and menopausal status. Tests of significance were two-sided with statistical significance denoted at α = 0.05, and SAS (version 9.4, Cary, NC) was used.

Results

Relative to controls, cases had higher initial MD, higher BOADICEA risk scores and a stronger FHBC and therefore these variables were retained in multivariable analyses in addition to age at mammogram and BMI as a priori covariates. Other factors considered were not associated with case–control status and/or MD (Table 1, Supplementary Fig. 1).

Table 1 Participant baseline characteristics by case and control status

We observed similar associations when considering relative annual change in both percent density and dense area (Table 2). Compared to women for whom the largest relative change in MD was a decrease > 10%, women with stable MD (i.e., did not show > 10% change in MD in either direction over the mammograms compiled) had an increased breast cancer risk adjusted for age at earlier mammogram, FHBC, and baseline BMI (Odds Ratio (OR) 1.94, 95% Confidence Interval (CI) 1.39–2.69 for percent density; OR 2.34, 95%CI 1.63–3.37 for dense area). Increasing vs. decreasing DA was also associated with risk (OR: 1.55, 95%CI 1.12–2.17), as were the percent density and dense area of the earlier mammogram (OR for each unit increase: 1.02, 95%CI 1.01–1.03 for percent density and 1.01, 95%CI 1.00–1.01 for dense area). We observed similar associations for the largest absolute change in MD (Supplementary Table 2).

Table 2 Multivariable-adjusted associations of the largest annual relative change in mammographic density from any two mammograms^a with breast cancer risk

Cases were more likely to exhibit a stable or increasing MD than controls when considering the largest relative change in MD (70–74% vs. 64–65%) and had a slower rate of change in dense area (Fig. 1A). Similarly, the tertile of earlier dense area showed statistical evidence of positive multiplicative interaction with relative annual dense area change for breast cancer risk (Table 3). Among women with the highest tertile of dense area at the time of the earlier mammogram, stable dense area (OR: 4.35, 95% CI 2.53–7.50) and increasing dense area (OR: 2.56, 95%CI 1.50–4.36) were both associated with increased breast cancer risk. Women whose earlier mammograms were at low or middle tertiles of dense area and had stable or increased dense area did not show increased breast cancer risk. For absolute change in MD, neither percent density nor dense area had significant multiplicative interaction between categories of MD change and percent density/dense area at earlier mammogram (Supplementary Table 3).

Fig. 1

Longitudinal change in mammographic density for cases and controls by risk and menopausal status. A Total sample. B by BOADICCEA 5-year cancer risk categories. C by menopausal status at baseline. Adjusted for age closest to baseline and BMI at baseline, follow-up time, case–control status and product term between follow-up time and case status

Table 3 Multivariable-adjusted associations between the largest annual relative change in mammographic density from any two mammograms and breast cancer according to mammographic density levels of earlier mammogram

Among both low and high BOADICEA risk strata, women with stable MD were at higher risk of breast cancer than women with decreasing MD, with the interaction reaching statistical significance for percent density, showing stronger associations among low-risk women (OR for stable: 2.97, OR for increase: 2.63 relative to decrease in low-risk group) than among high-risk women (OR for stable: 1.58, OR for increase: 0.74 relative to decrease in high-risk group) (Fig. 2). While FHBC and percent density change showed statistically significant interaction (p_int = 0.002), the overall results still showed similar patterns of association of risk with percent density change across each category of FHBC, with different patterns observed among women with two first- or second-degree relatives with breast cancer whose MD increased (Supplementary Fig. 2). Results were consistent for the rate of change analysis (Fig. 1B, Supplementary Fig. 3). We did not observe a statistically significant interaction between change patterns in dense area or percent density and age at breast cancer diagnosis of first-degree relatives (Supplementary Table 4).

Fig. 2

Associations between the Largest Relative Annual Change in Percent Density and Breast Cancer by risk. Associations presented by 5-year BOADICEA risk category. Models adjusted for age at earlier mammogram, BMI at baseline, percent density at earlier mammogram, 5-year BOADICEA breast cancer risk score and interaction with BOADICEA score (p_int = 0.005)

Only 24% of participants had mammograms from both pre- and post-menopausal periods with the two mammograms comprising the largest change in MD spanning the menopausal transition for only 1.4%; for the rest, the mammogram pairs were concordant on menopausal status. We did not observe statistically significant differences in risk by menopausal status at the time of the largest change in percent density or the rate of percent density change, but found a statistically significant multiplicative interaction for dense area (p_int = 0.049) (Fig. 3). Compared to women whose largest change in dense area was a pre-menopausal decrease, women with stable or increasing dense area in either pre- or post-menopausal periods, showed greater breast cancer risk, with a stronger association observed for women with the largest relative change as ‘pre-menopausal stable’ compared to ‘post-menopausal stable’ (premenopausal stable OR: 4.07 95%CI 2.29–7.28 and postmenopausal stable OR: 2.08 95%CI 1.16–3.73, both relative to pre-menopausal decrease). However, women whose largest relative dense area change was a > 10% decrease were at similar breast cancer risk regardless of whether this change occurred in pre- or post-menopausal periods (OR for post-menopausal decrease vs. pre: 1.27, 95%CI (0.73–2.22)). Similarly, the rate of dense area change in the pre-menopausal period was statistically significantly slower in cases than in control, with smaller and nonsignificant differences for post-menopausal period (Fig. 1C).

Fig. 3

Associations between the largest Relative Annual Change in Density and breast cancer by Menopausal Status. Menopausal status is based on the timing of the largest relative annual change in dense area. Models are presenting change in dense area, adjusted for age at earlier mammogram, BMI at baseline, dense area at earlier mammogram, first and second degree family history of breast cancer, menopause status of MD change and interaction with menopause status of MD change (p_int = 0.049). Excluded cases where age of menopause fell between mammograms for largest change in MD

Discussion

We employed a nested case–control design among women with a family history of breast cancer to examine the contribution of longitudinal changes in MD to breast cancer risk beyond initial MD. We used several measurement approaches (dense area as absolute and percent density as relative MD measures; absolute and relative change in MD scales; rate of change) to enhance the interpretation of results and comparability with prior research. We also explored whether the change in MD and breast cancer risk associations vary according to familial risk and menopausal timing of MD change. Our findings suggest that women with higher initial MD levels and those whose MD does not appreciably decline over time are at higher risk for breast cancer. Although generally similar across menopausal and familial risk categories, the associations appeared stronger in pre-menopausal and lower-risk women.

Our results are broadly consistent with a meta-analysis of nine studies that suggested positive and inverse associations respectively between increased and decreased MD over time, and breast cancer [25]. However, the comparison groups in this meta-analysis, such as non-dense stable or those with stable but differing initial MD levels, made it challenging to discern the extent to which change in MD or stability contributes to breast cancer risk beyond the initial MD level. Our results support a role for both initial MD and change in MD. While our findings differ from prior studies that found earlier MD level rather than change in MD predictive of risk [14,15,16], they align with other studies, including among high-risk women [18, 19]. Variation in populations and measurement and scale of MD and MD change may explain differences across existing literature but future studies are needed to clarify these findings.

We observed the strongest and most consistent evidence of an association between change in MD and breast cancer risk for women who maintained stable MD over time, particularly for women with higher initial MD levels, consistent with existing literature emphasizing the importance of MD stability [19, 26]. Increasing vs. decreasing dense area, but not percent density, was associated with risk. Comparison of rate of change between cases and controls also revealed larger differences for dense area with minimal to no differences for percent density. Women experiencing increasing MD were more likely to have had lower MD at earlier mammograms, pointing to the possibility that the length of time with higher MD, rather than specific change patterns, may be more pertinent to breast cancer risk. Differences in results for dense area vs. percent density may involve the influence of BMI and non-dense breast area on the measurement of MD and/or on risk. Specifically, an increase in percent density may involve decreasing non-dense adipose breast tissue alongside increasing dense area. Non-dense area is highly positively correlated with BMI, which in turn may have other effects on breast cancer risk. Despite adjustment for BMI at multiple time points, it remains challenging to disentangle the association of percent density with risk from that of BMI [4, 26].

This study has several key strengths. Notably, we collected a substantial number of mammograms for each participant, increasing the likelihood of capturing any substantial change for a woman and enabling us to characterize the most predominant MD trajectory for each woman and examine the rate of MD change over time. The high MD measurement reliability (> 90%) and the inclusion of mammograms produced from many different machines at facilities across the nation enhances the precision and robustness of these findings, compared with single institution studies. By focusing on younger age at cancer diagnosis, we were able to assess effect modification by menopausal status with results highlighting the potential clinical importance of monitoring MD trajectories for both pre- and post-menopausal women. We were also able to assess familial risk through several methods including a risk model and the FHBC with variability across the risk spectrum. Among women at high familial risk, there were relatively small differences in the initial MD levels and in MD changes between cases and controls while among women at low familial risk, those with stable or increasing MD also had increased breast cancer risk relative to those with decreasing MD, suggesting that patterns of change in MD may be more relevant to risk among average-risk women than women with high breast cancer susceptibility. Nonetheless, even among women with decreasing MD, the risk remains statistically significantly higher among those at higher familial risk, highlighting the continuing importance of breast cancer susceptibility beyond MD changes.

We were unable to collect mammograms or assess MD for nearly half of the randomly selected participants, primarily due to the retrospective retrieval of mammograms (e.g., past mammography facilities were not recalled or located, pre-diagnostic mammograms were unavailable) and technical challenges in processing and assessment of images (e.g., unreadable images). Despite this constraint, we found similar characteristics between those with and without mammographic measures. Our study included mammograms spanning a broad period, potentially reflecting advancements in mammography technology. We would expect that incorporating more sensitive MD measures would only strengthen our findings. Study participants were predominantly non-Hispanic white and highly educated women, reflecting the overall racial and ethnic composition but higher educational distribution than the overall Sister Study cohort. Inadequate social diversity in studies of MD and high-risk populations is a significant gap that merits focused research to improve the validity of these findings for other populations [27].

In summary, we found that women who do not experience MD declines over time are at elevated breast cancer risk, with some indication of stronger associations for average-risk and pre-menopausal women. As automated MD assessment and mammogram-based artificial intelligence methods advance, incorporating longitudinal mammographic measures may become more feasible and continue to improve risk stratification. It may be possible to foresee the use of women’s MD changes as health providers currently consider other clinical measures of variability such as blood pressure variability as additional information for improved prediction and risk stratification for breast cancer.