Longitudinal Change in Mammographic Density among ER-Positive Breast Cancer Patients Using Tamoxifen

Mammographic density declines in 30% to 60% of women treated with tamoxifen (1–6). Mean absolute reductions are 5% to 10% after a mean of 12 to 18 months (1–6). Such density reductions have been associated with lower risks of breast cancer recurrence (2, 3) and breast cancer–specific death (5, 7). It has been suggested that the density reduction may be a biosensor of tamoxifen effectiveness. Although there are several hypothesized mechanisms (8), including that the reduction is a marker of tamoxifen metabolism, many properties of tamoxifen-associated density declines are unknown. In particular, it is unclear whether additional information might be gained by characterizing the density decline at multiple time points.

Most studies describing density change associated with adjuvant tamoxifen examined one (4, 5, 9, 10) or two (6) posttamoxifen mammograms, despite the fact that tamoxifen was historically prescribed for 5 years of use. Two studies (11, 12) evaluated more than two posttamoxifen mammograms, but neither provided yearly change data that included periods both shortly after and up to 5 years after tamoxifen initiation. Furthermore, little is known about tamoxifen discontinuation and density decline. 40% to 50% of patients discontinue tamoxifen before completing 5 years of treatment (13–15), but it is unclear whether density returns to baseline levels after early discontinuation.

We conducted an exploratory pilot study of quantitative mammographic density using five successive mammograms following the initiation of tamoxifen treatment in 40 patients with estrogen receptor (ER)–positive breast cancer. Our goal was to clarify patterns of longitudinal mammographic density change among patients treated with tamoxifen, accounting for the potential effects of tamoxifen discontinuation.

We previously conducted a study of change in mammographic density after tamoxifen initiation and breast cancer–specific death (5). Briefly, that study included female Kaiser Permanente Northwest (KPNW) health plan members who were diagnosed with localized or regional stage ER-positive breast cancer between 1990 and 2008, ages ≥18 years at diagnosis, and treated with tamoxifen. Vital status was observed through December 31, 2010, and a sample of 97 women who died from breast cancer (cases) and 252 who did not die from breast cancer (controls) were included in analyses.

For this study, we identified women from the case–control population who were alive ≥5 years after diagnosis (N = 305), and selected 40 subjects who had ≥5 post-diagnostic mammograms within strata of tamoxifen use duration and age at diagnosis: <4 years tamoxifen and ≤50 years (N = 6) or >50 years (N = 10) old; ≥4 years tamoxifen and ≤50 years (N = 13) or >50 years (N = 11) old. We included uneven numbers in each group to maintain the overall sample size of 40. This study was approved by the National Cancer Institute's Special Studies Institutional Review Board (IRB) and the KPNW IRB.

Six film mammograms of the contralateral (unaffected) breast per patient were identified from medical records (Fig. 1). The baseline mammogram was a mammogram occurring ≤720 days before diagnosis and before the first tamoxifen prescription. Five follow-up mammograms (T1 to T5) obtained approximately yearly after tamoxifen initiation were also selected. The first follow-up mammogram (T1) must have been >90 days after the first tamoxifen prescription and women must have had tamoxifen prescription coverage within 90 days of the T1 mammogram. One post-tamoxifen mammogram was unavailable at the appropriate time interval for one patient resulting in five mammograms for that patient. Six mammograms each were evaluated for all other patients.

Craniocaudal views from the selected films were digitized, as described previously (5). Total breast area and dense area were measured by a single reader using computer-assisted thresholding (16). Percent density was calculated as the ratio of dense area to total breast area. All mammograms from the same patient were evaluated within the same session. A reliability study evaluating masked duplicate images conducted within the parent study resulted in intraclass coefficients >95% and coefficients of variation <10% for dense area, total breast area, and percent density (5).

Patient characteristics (Supplementary Table S1) were obtained from medical records (5). Body mass index was calculated as weight (kg) divided by squared height (m²). Individual prescription records for tamoxifen, antidepressants (selective serotonin reuptake inhibitors or serotonin-norepinephrine reuptake inhibitors), and estrogen or estrogen plus progestin hormone therapy dispensed between 1986 and 2010 were obtained from KPNW databases. Duration of tamoxifen use was calculated by subtracting the date of the first tamoxifen prescription from the last day of the last tamoxifen prescription before discontinuation (indicated by no prescriptions for >1 year).

Mean and standard deviation (SD) percent density were calculated at each time point. Change in percent density was calculated in two ways: As the difference from the previous year and as the difference from baseline. Linear regression models were used to examine longitudinal change in density. Generalized estimating equations accounted for correlations among measurements from the same woman (17). Associations between patient characteristics (Supplementary Table S1) and percent density were evaluated by modeling associations with percent density and density change from the previous year. Percent density was square-root transformed and yearly density change was scaled and raised to the 1.75th power to approximate normality. The baseline percent density (tertiles: ≤15.4%, 15.5%–32.1%, >32.1%) was associated with density change from the previous year at P < 0.05; thus, we estimated models adjusted for this factor. Other characteristics were not associated with percent density or density change at P < 0.05 and were not included in adjusted models. Trend tests were not conducted due to the non-linear pattern of density change over time.

We compared density change according to varying intervals between baseline and T1 mammograms (9–13, >13–18, >18–22, and >22 months) using the Kruskal–Wallis test and by testing for an interaction between the baseline to T1 interval and mammogram timing (e.g., T1, T2, etc.) with respect to density change in a linear regression model. We examined the effect of excluding two women who experienced a recurrence either before the T5 mammogram or at an unknown date, as an unmeasured factor could be associated with recurrence and density change.

Interactions with age at diagnosis (≤50 and >50 years) and tamoxifen duration (<4 and ≥4 years) were estimated by including cross-product terms with indicators for mammogram timing in regression models. We additionally tested for differences related to tamoxifen use at each mammogram by evaluating a model with cross-product terms between mammogram timing and an indicator for a tamoxifen prescription within 60 days of the mammogram of interest. We evaluated interaction between the baseline percent density (tertiles) and mammogram timing based on our previously reported association between baseline density and density change after 1 year of tamoxifen (5). Interaction results from unadjusted and adjusted models were similar; unadjusted P values are presented.

Analyses were conducted using SAS v9.2 (SAS) and R (18). Two-sided P values <0.05 were statistically significant.

Patients were 41 to 77 years old at baseline and used tamoxifen for a mean of 51 months (SD 18). Mean mammographic density at baseline was 26.5% (SD 18.0%; Table 1). Despite individual variation (Supplementary Fig. S1), percent density decreased over time with a cumulative reduction in mean density of 7.2% by T5. The largest reduction was observed at T1; by T3, the yearly mean change was approximately 1% or less. Baseline to T1 density decline was slightly greater among women with >22 months between baseline and T1, but overall, mean change did not differ significantly according to time between baseline and T1 (Kruskal–Wallis P = 0.55) or through T5 (P_interaction = 0.74; Supplementary Fig. S2). Early density reductions were attenuated after adjustment for baseline density (Table 1); results were similar after additional adjustment for time between baseline and T1 mammograms (data not shown). Results were also similar after excluding women with a recurrence (data not shown).

Patterns of density change differed by baseline percent density (P_interaction < 0.01; Fig. 2). Density declined only among women in the highest tertile of baseline density, where the greatest reduction was observed between baseline and T1 (mean 14.9%; Supplementary Fig. S3). The slight increase in adjusted density change at T5 observed in the overall population was not seen among women in the highest baseline density tertile (Fig. 2). There were no differences in density change by tamoxifen use, whether comparing <4 with ≥4 year users (P_interaction = 0.42; Supplementary Fig. S4) or users to discontinuers at each mammogram (all P_interaction > 0.05). Patterns of density change were not statistically different by age (P_interaction = 0.36); however, graphical analysis revealed a modest decline among women ≤50-years-old that was not evident in women >50 years (Supplementary Fig. S5).

In this pilot study, we examined mammographic density at five time points following tamoxifen initiation and observed that the majority of the density decline occurred a mean 12 months after initiating tamoxifen. There was little additional change in subsequent years, suggesting that a mammogram obtained 1 year following treatment initiation is sufficient to classify the tamoxifen-associated density response.

Previous studies of this topic are few and have provided conflicting evidence. In the IBIS-I Chemoprevention Trial, density declines among high-risk women treated with tamoxifen appeared to decline monotonically (1), but it is unclear how those results relate to adjuvant use, given that breast cancer patients have higher mammographic densities than cancer-free women (19). Meggiorini and colleagues (12) evaluated qualitative and semiquantitative density from six mammograms in breast cancer patients and reported that density declined after the first year and that “variations were stable” throughout follow-up; however, data related to later time points were not presented. In contrast, Konez and colleagues (11) did not observe a reduction in density when first assessed (two to three years after adjuvant tamoxifen initiation), but did note a reduction after 5 years. Our results are consistent with aspects reported by Meggiorini and colleagues and IBIS-I, supporting a model where density declines occur early and persist over time.

Our analysis extends two findings observed previously. First, the cumulative density reduction a mean of 5 years after tamoxifen initiation was greatest among women with the highest baseline density, consistent with the identification of baseline density as a significant predictor of density change in studies that measured change over 1 to 2 years (2, 3, 5). Second, the pattern of density change over time did not appear to be influenced by tamoxifen discontinuation, consistent with a previous report that density did not change following tamoxifen cessation after 5 years of use (11). These data suggest that among women with a reduction, density does not revert to baseline levels once tamoxifen administration ends and that it may not be necessary to have information on long-term tamoxifen use when evaluating the prognostic meaning of density change. It is unclear why density did not continue to decline among women taking tamoxifen for ≥4 years. Research investigating the biologic mechanisms involved are necessary to understand the relationship between the pattern of density decline and tamoxifen's effects.

Younger age and chemotherapy receipt showed weak, nonsignificant associations with larger reductions in density. These associations are consistent with what we observed in the parent case–control study examining change after approximately one year of tamoxfien (5) and with what others reported in similar analyses (1–4, 12, 20).

Strengths of this analysis include the use of quantitative, reproducible density measurements, and digitization of all films from a patient during the same session, both of which reduced random variability. We used prospectively collected medical record data to assess the effects of patient characteristics and treatments. Individual prescription records allowed us to accurately determine periods of tamoxifen use in relation to mammography timing. There were also limitations. Women were selected to have survived at least 5 years, and therefore may not be representative of all ER-positive patients treated with tamoxifen. We lacked information on menopausal status, which is associated with mammographic density (21). We did have information on surgical menopause after diagnosis and found that it was not a predictor of density change, though few women had the procedure during the study period. There was a wide range in time between baseline and T1, although this did not affect the main conclusions regarding patterns of density change. We examined a large number of covariates relative to the number of subjects and some associations we observed may be due to multiple comparisons. Finally, post hoc calculations show that we had 80% power to detect a minimum difference in density change of 11%, which is greater than the observed differences in density change among the age or tamoxifen duration groups; therefore, conclusions related to statistically significant differences according to these factors should be interpreted with caution.

In summary, we found that density change 1 year after tamoxifen initiation was representative of density change 5 years after tamoxifen initiation. Replication of these results in a larger dataset is needed to confirm that assessment of density change 1 year following tamoxifen initiation is a suitable measure of exposure in studies of density change and tamoxifen response.

No potential conflicts of interest were disclosed.

Conception and design: S.J. Nyante, M.E. Sherman, A. Berrington de Gonzalez, A. Glass, G.L. Gierach

Development of methodology: S.J. Nyante, A. Glass, G.L. Gierach

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.J. Nyante, A. Berrington de Gonzalez, E.J. Aiello Bowles, A. Glass

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.J. Nyante, M.E. Sherman, R.M. Pfeiffer, A. Berrington de Gonzalez, R.N. Hoover, G.L. Gierach

Writing, review, and/or revision of the manuscript: S.J. Nyante, M.E. Sherman, R.M. Pfeiffer, A. Berrington de Gonzalez, L.A. Brinton, E.J. Aiello Bowles, R.N. Hoover, A. Glass, G.L. Gierach

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.J. Nyante

Study supervision: S.J. Nyante, G.L. Gierach

The authors thank Brenda Rush and Kathy Pearson of KPNW (Portland, OR) for their invaluable assistance in conducting this study.

This research was funded by the Intramural Research Program of the National Cancer Institute, NIH (to S.J. Nyante, M.E. Sherman, R.M. Pfeiffer, A. Berrington de Gonzalez, L.A. Brinton, R.N. Hoover, and G.L. Gierach) and National Cancer Institute federal funds awarded under contract HHSN261201100441P (to E. Aiello Bowles) and HHSN261200900017C/N02CP-90017 subcontract no. 10-312-0212208 (to A. Glass).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.