Outcomes of Density-Targeted Supplemental Breast Magnetic Resonance Imaging Screening by Breast Cancer Risk: Long-Term Health and Economic Considerations.

INTRODUCTION
National breast density notification regulations went into effect in the United States (U.S.) in September 2024 (1). All women undergoing mammography screening are now notified of whether they have mammographically “dense” breasts or not. Those with “dense breasts,” an estimated 43% of the mammography screening population (2), are directed to discuss screening options with their healthcare providers.
Considerable uncertainty remains about the best screening approaches for individuals with dense breasts. This was underscored by the 2024 U.S. Preventive Services Task Force (USPSTF) guidelines, which found insufficient evidence to recommend supplemental screening like breast magnetic resonance imaging (MRI) to otherwise average-risk women based on breast density alone (3). However, it is possible that MRI targeted to subgroups of women based on combinations of both high breast density and breast cancer risk would be effective (4), and feasible for facilities with limited MRI resources (5).
Recent studies of screening breast MRI alone or as a supplement to mammography report both short-term benefits (increased cancer detection and sensitivity) and harms (increased recall and lower specificity) in women with dense breasts (6–11). However, most studies focused on women at very high risk of breast cancer or with a personal history of breast cancer—populations in which MRI as a supplement to mammography has been shown to be cost-effective, but did not consider density or long-term outcomes (12, 13).
The European DENSE trial found that supplemental breast MRI following negative biennial screening mammography among women with extremely dense breasts resulted in a significantly lower interval cancer rate than mammography alone after the first screening round (14). Based on data from that trial, adding MRI for 50–75 year old women with extremely dense breasts was found to be cost-effective in European settings at a 3- to 4-year screening interval (15). In the U.S., women ages 40–74 years with extremely dense breasts comprise 7.4% of the population (2).
The value of adding supplemental MRI within selectively targeted breast density subgroups across a range of breast cancer risks in U.S. settings remains uncertain (11, 16). To address this evidence gap, we used Breast Cancer Surveillance Consortium (BCSC) data with three Cancer Intervention and Surveillance Modeling Network (CISNET) breast cancer simulation models to extend our prior research (17, 18). We evaluated outcomes for digital breast tomosynthesis (DBT) screening strategies that target supplemental MRI use by breast density across breast cancer risk levels.

METHODS
We used computer simulations of a 1980 U.S. female birth cohort to project cancer screening outcomes associated with DBT alone or with supplemental MRI added by breast density subgroup compared to no screening across four exemplar breast cancer risk levels. Here, we report on average risk (relative risk [RR]=1) and the highest risk level considered (RR=4, meaning four times higher-than-average risk, e.g., a family history of breast cancer in two or more first-degree relatives). For risk levels 30% higher-than-average (RR=1.3, e.g., first birth at ages 25–29), and risks two-times higher-than-average (RR=2, e.g., a family history of breast cancer in one first-degree relatives) see eSupplement and eTable 1 for further information.

Strategies Evaluated
We evaluated screening strategies involving DBT that targeted MRI use by breast density and varied starting age (40, 45, 50 years) and screening interval (annual, biennial) across four relative risk levels. Screening ended at age 74 as currently recommended by the USPSTF (3, 17). To estimate best-case screening outcomes, analyses assumed full screening and recommended treatment adherence.

Breast Cancer Simulation Modeling
Breast cancers arise and may progress to become detectable via screening or clinical means. Screening has the potential to reduce breast cancer mortality by detecting tumors at smaller sizes or earlier stages when they may have better treatment response and longer survival. Treatment is assigned at cancer diagnosis based on current age, stage and subtype-specific guidelines. Collaborative modeling of breast cancer was conducted by three modeling teams: Model E (Erasmus Medical Center)(19), GE (Georgetown University/Albert Einstein College of Medicine)(20) and W (University of Wisconsin-Madison)(21). Details of the independently developed models described previously are available online (22–24). Briefly, the models begin with breast cancer natural history and overlay screening use and treatment to project breast cancer incidence and survival (25). Because epidemiological methods cannot provide the duration of time between when a cancer may become screen detectable versus symptomatic, CISNET models employ a variety of approaches to modeling the natural history of breast cancer (24). Independent modeling methods were used by the three groups, and all reproduced SEER incidence and U.S. mortality (face validity), and match AGE trial results (predictive validity) (23, 26). This study was determined not to be human subjects research by the Dartmouth institutional review board.

Key Model Inputs
The models used a common set of input parameters (Table 1).

Breast density
The strategies applied the American College of Radiology’s Breast Imaging Reporting and Data System (BI-RADS) density categories: almost entirely fatty (“a”), scattered fibroglandular densities (“b”), heterogeneously dense (“c”), or extremely dense (“d”). The models assume that age affects the prevalence of breast densities and the breast density-specific relative risk of breast cancer incidence (eTable 2) (25, 27). Accordingly, each simulated woman was assigned an initial density category (BI-RADS a-d) at age 40 and could experience a decrease in density at age 50 and 65 based on observed BCSC breast density prevalence. Supplemental screening strategies assign modality by a woman’s current breast density and thus screening modality could change over time. (25, 27)

Screening performance
The BCSC collects data from seven breast imaging registries involving academic and community-based radiology facilities in California, Illinois, New Hampshire, North Carolina, Vermont, and Washington (28). Screening mammograms performed between 2010–2017 were used to estimate sensitivity and stage distributions for DBT (eTables 3–6). Sensitivity was stratified by age group, screening round (first, subsequent) and interval (annual, biennial), and breast density. BCSC data were also used to estimate false-positive recall rates and the probability of a recommendation for short-interval follow-up or for a biopsy following a false-positive recall (eTable 7).
BCSC data were used to estimate the combined sensitivity of mammography and MRI and modality-specific sensitivities (eTable 8) (18). MRI sensitivity was stratified by screening round, but not density. BCSC data were used to estimate false-positive recall, short-interval follow-up, and biopsy recommendations following MRI (eTables 9–10). False-positive recall was adjusted by density and screening round.

Quality-adjusted life years
To estimate quality-adjusted life years (QALYs) (29), we weighted simulated life years using age-specific population utilities for U.S. women, adjusted for breast cancer diagnosis and treatment (30, 31). As done previously, we included a small disutility for screening participation (1 hour deduction) and a positive screen (3.7 day deduction) (Table 1) (32). Screening disutilities were doubled when MRI was added to DBT.

Costs
Costs were reported in 2023 US dollars (Table 1). Medicare reimbursement rates were used for screening test costs with initial diagnostic workup costs drawn from prior analyses (33, 34). Costs associated with false-positive screening recalls depended on screening modalities and whether diagnostic imaging alone or biopsy was required. Breast cancer treatment costs were assumed to vary by stage and phase of care based on SEER–Medicare data (35, 36).

Analysis
For each screening strategy and relative risk level, outcomes included screening benefits (breast cancer deaths averted and LYs gained) and harms (false-positive recall and false-positive biopsy recommendation) relative to no screening reported across the lifetimes of simulated cohorts. We report incremental changes within models for MRI screening as its use is extended across density categories and report the means and ranges across models per 1,000 women. The collaborative modeling approach provided implicit cross-validation, with cross-model ranges for benefits and harms providing a measure of uncertainty. Harm-to-benefit ratios portray tradeoffs in false-positive biopsy recommendations per breast cancer death averted across risk levels and compared with ratios associated with biennial DBT screening.
Cost and QALY outcomes were discounted at 3% annually and the analysis used a federal payer perspective (37). By risk level, these outcomes are reported for each screening strategy relative to “no screening”. Incremental cost-effectiveness ratios (ICERs) were estimated for each model as the change in cost between two strategies divided by the change in health (QALYs) when strategies were ranked by increasing cost. Dominated strategies were more costly and less effective than competing strategies. Weakly dominated strategies had ICERs that were less efficient than a more costly strategy. While there is no universal agreement on what constitutes a “cost-effective” intervention, we report results with attention to a $100K/QALY benchmark acknowledging that many accepted medical practices in the U.S. exceed this value (38). A priori, we designated strategies falling within 10% of the QALY gains associated with biennial DBT screening started at age 50 as being near-efficient (39). By risk level, strategies were plotted on an efficiency frontier, with shading used to designate near-efficient strategies.
To assess the impact of key parameters on cost-effectiveness, we conducted sensitivity analyses on breast MRI performance (false-positive biopsy recommendation rates) and breast MRI cost.

Role of Funding Source
This research was funded by the National Cancer Institute. Manuscript contents are solely the responsibility of the authors and do not necessarily represent the funder’s official views.

RESULTS

Clinical Outcomes
Clinical benefits of deaths averted and life years gained per 1000 women relative to no screening increased with earlier screening start age, more frequent screening, and supplemental MRI (Table 2, Figure 1, eTable 11) with even greater benefits for women at increased risk. For example, across start ages and screening intervals, the 7.4 to 10.5 deaths averted with DBT screening alone across all strategies for RR=1 increased to 23.2 to 33.6 for RR=4. Incremental gains from adding MRI by density group were small in comparison to the benefits of DBT screening alone. For example, when supplemental MRI was directed to women with extremely dense breasts (+MRI d) among average risk women, gains in deaths averted ranged from +0.1 to +0.2 across all start ages and intervals. When further extended to women with heterogeneously dense breasts (+MRI c), an additional +0.4 to +0.7 deaths could be averted.
Clinical harms of false-positive recalls and false-positive biopsy recommendations increased with earlier screening start age, more frequent screening, and use of supplemental MRI (Table 2, Figure 1, eTable 12). False-positive recalls per 1000 women for DBT screening ranged from 884 to 2,139 for average-risk women and were similar for RR=4. For MRI d, false-positive recalls increased by +35 to +276 per 1000 average risk women screened and from +27 to +242 for higher risk (RR=4). For false-positive biopsy recommendations, similar patterns were observed.
Clinical harm-to-benefit ratios portray false-positive biopsy recommendations per death averted and decrease with increasing relative risk (Figure 2). For average risk women (RR=1), DBT+MRId increased harms in comparison to tradeoffs associated with DBT screening alone (eTable 13). Expanding MRI use beyond density d markedly increased harms for both average and modestly-elevated (RR=1.3) risk groups. At higher risk levels (RR=2), directing MRI to all women with dense breasts beginning biennially at age 50 (DBT+MRIcd) was comparable to harm-to-benefit tradeoffs for DBT (gray bar in Figure 2).

Cost-effectiveness Outcomes
ICERs ranged widely within each risk group and generally decreased with increasing risk (Table 3, eTable 14). For all RRs, the most cost-effective strategy was biennial screening starting at age 50 with DBT followed by biennial DBT starting at age 45. Several biennial screening strategies involving supplemental MRI, however, were nearly as efficient as DBT alone (eFigure 1). These included biennial DBT with MRI starting at age 50 for women with extremely dense breasts (DBT+MRId). When excluding DBT alone starting at age 45 and comparing with no screening, this DBT+MRId strategy cost less than $50K per QALY gained across all RRs. For women at higher-than-average breast cancer risk, DBT+MRId strategies initiated before age 50 were also near-efficient (Table 3). At RR=4, the ICER for DBT+MRId was $114K per QALY relative to annual DBT screening at age 40. For 1,000 women starting biennial DBT screening at age 40, the total added cost of extending supplemental MRI to all women with dense breasts (DBT+MRIcd) over and above limiting MRI to women with extremely dense breasts (DBT+MRId) ranged from $1.96M at RR=1 to $1.7M at RR=4 (eTable 15).
In sensitivity analyses, DBT+MRI d strategies remained near-efficient for women of modestly-elevated-to-moderate risk (RR=1.3, 2) when the MRI cost was reduced by half. For higher risk women (RR=4), annual DBT+MRId starting at age 40 would cost $72,400 per QALY gained. When MRI costs are reduced by half and MRI false-positive biopsy recommendations are reduced by one third, the cost-effectiveness of MRI strategies improves (eFigure 1, eTable 16). Under this scenario, some DBT+MRId strategies are efficient in women at moderate increased risk (RR=2). Further, for high-risk women (RR=4), annual DBT plus MRI directed to all women with dense breasts (DBT+MRIcd) also has an ICER <$100K.

DISCUSSION
This collaborative modeling study combined real-world evidence from the BCSC with CISNET simulation modeling to facilitate comparison of clinical and economic population-level outcomes of density-targeted supplemental breast MRI for average to 4-times higher breast cancer risk levels. Our analysis highlights the importance of considering both age-group specific breast cancer risk and density prevalence in identifying strategies where adding MRI to DBT has similar or better harm-to-benefit tradeoffs as undergoing DBT screening alone. Over a lifetime of screening, supplemental MRI strategies could avert more deaths but also increase false-positive biopsy recommendations. When biennial supplemental breast MRI for women with extremely dense breasts (approximately 10% of the screening population (2)) was directed to women aged 40 and older with 1.3-times higher-than-average risk (RR=1.3), harm-to-benefit tradeoffs (false-positive biopsy recommendations per death averted) were somewhat higher than those experienced by average-risk women undergoing biennial DBT. For women at higher breast cancer risk (RR=2 or 4), the harm-to-benefit ratios for DBT+MRId were comparable and more favorable than those experienced by average-risk women undergoing DBT screening. This information is important for clinicians counseling women about supplemental screening.
Examining harm-to-benefit tradeoffs for supplemental MRI strategies across risk groups highlights the importance of starting age and screening interval. As a comparison, we considered biennial DBT screening as having an acceptable harm-to-benefit ratio (i.e., gray bar in Figure 2, with strategies below the gray bar showing the added value of supplemental MRI). All DBT+MRId strategies at RR=1.3, except annual screening starting at 40, have comparable ratios to screening with DBT alone (see eTable 1). However, for women with moderately-elevated risk (RR=2), such as women with a family history of breast cancer in one first degree relative, all MRI strategies for women with extremely dense breasts (DBT+MRId), with the exception of annual screening beginning at age 40, have similar or better harm-to-benefit ratios than those undergoing DBT alone.
When cost-effectiveness outcomes were considered, biennial supplemental breast MRI for women with extremely dense breasts started at age 50 was near efficient for all risk groups. This highlights that, at a population level, it is challenging for strategies that direct high-cost modalities to a limited subgroup of that population to achieve the same efficiencies as those that direct more intensive screening to a larger number of women (e.g., screening all women at younger start ages or with increased intensity). DBT+MRId strategies were not more cost-effective than DBT strategies at earlier start ages unless higher-risk women (RR≥2) were considered and both MRI costs and associated false-positive biopsy recommendations were reduced by 50% and 33%, respectively.
While the population as a whole may benefit from more extensive DBT screening (earlier age or shorter interval), women with extremely dense breasts (a small, but identifiable population) could have suboptimal screening outcomes (40). This finding underscores recognized limitations of cost-effectiveness analysis, which prioritizes population outcomes as a whole without a focus on who benefits (41). In this context, the near-efficient (DBT+MRId) strategies for above-average risk women merit consideration as their value could be considered comparable or better than DBT alone.
Our sensitivity analyses addressing MRI cost and false-positive biopsies were motivated by both the emergence of abbreviated-protocol MRI and artificial intelligence (AI) algorithms for mammography interpretation, respectively (5). Such improvements in false-positive recall are within the realm of possibility with intensive ongoing research, including trials involving both abbreviated MRI and AI-driven DBT screening (15, 42). In addition, the emergence of contrast-enhanced mammography holds some promise for improved performance for women with dense breasts (43, 44).
Our estimates of the value of strategies involving supplemental breast MRI are not directly comparable to economic evaluations based on the DENSE trial for several reasons (15). First, the DENSE trial only enrolled women with extremely dense breasts rather than our analysis of populations of women with a range of densities. Second, mammography performance in Europe differs markedly from performance in the U.S., particularly regarding recall rates (45, 46). Lastly, the cost-effectiveness analyses of the DENSE trial found MRI screening every 4 years, a strategy we did not evaluate, to be less costly and more effective than biennial supplemental breast MRI screening (15).
Several limitations warrant comment. First, our assumptions about full screening uptake and treatment adherence provide best-case estimates of the value of density- and risk-targeted supplemental breast MRI. Second, our analysis is for a population of women at each risk level, rather than from the perspective of an individual woman and was not based on absolute risk thresholds nor linked with a specific breast cancer risk prediction tool. Third, our analysis did not consider population subgroups defined by factors other than risk and breast density and is subject to limitations inherent in simulation modeling studies.

Conclusion
In the context of the recently enacted national breast density reporting mandate and renewed interest in risk-based screening (47), our study provides needed modeling evidence to inform clinicians, women, and policy makers regarding the potential value of supplemental breast MRI based on density and risk. Our findings on harm-to-benefit trade-offs suggest that in women with extremely dense breasts at higher risk (RR≥2), supplemental MRI could be considered biennially beginning at age 40. Annual intervals should be reserved for those with higher risk (RR≥4). Further, supplemental breast MRI directed to all higher-risk women (RR≥2) with dense breasts could provide reasonable value (i.e., ICER<$100K) if both MRI costs and false-positive biopsy recommendation rates are reduced.

Supplementary Material
Supplemental Material