Collaborative Modeling to Compare Different Breast Cancer Screening Strategies: A Decision Analysis

Abstract

Importance:

The effects of breast cancer incidence changes and advances in screening and treatment on outcomes of different screening strategies are not well known.

Objective:

Estimate outcomes of various mammography screening strategies.

Design:

Compare outcomes using six Cancer Intervention and Surveillance Modeling Network (CISNET) models and national data on breast cancer incidence, mammography performance, treatment effects, and other-cause mortality.

Setting:

United States

Population:

Women without previous breast cancer diagnoses.

Exposures:

Thirty-six screening strategies with varying start (40, 45, 50) and stop (74, 79) ages with digital mammography (DM) or digital breast tomosynthesis (DBT) annually, biennially, or a combination of intervals. Strategies were evaluated for all women and for Black women, assuming 100% screening adherence and “real world” treatment.

Outcomes and Measures:

Estimated lifetime benefits (breast cancer deaths averted, percent reduction in breast cancer mortality, life-years gained [LYG]), harms (false-positive recalls, benign biopsies, over-diagnosis), and number of mammograms per 1000 women.

Results:

Biennial screening with DBT starting at age 40, 45, or 50 until age 74 averted a median of 8.2, 7.5, or 6.7 breast cancer deaths per 1,000 women screened, respectively, versus no screening. Biennial DBT screening 40–74 (versus no screening) was associated with a 30.0% breast cancer mortality reduction, 1,376 false-positive recalls, and 14 over-diagnosed cases per 1,000 women screened. DM screening benefits were like DBT but had more false-positive recalls. Annual screening increased benefits but resulted in more false-positive recalls and over-diagnosed cases. Benefit-to-harm ratios of continuing screening until age 79 were similar or superior to stopping at age 74. In all strategies, women with higher-than-average breast cancer risk, higher breast density, and lower comorbidity level experienced greater screening benefits than other groups. Annual screening of Black women from age 40 to 49 with biennial screening thereafter reduced breast cancer mortality disparities while maintaining similar benefit-to-harm tradeoffs as for all women.

Conclusions:

This modeling analysis suggests that biennial mammography screening starting at age 40 reduces breast cancer mortality and increases life-years gained per mammogram. More intensive screening for women with greater risk of diagnosis or death can maintain similar benefit-to-harm tradeoffs and reduce mortality disparities.

Introduction

Since 2009, the US Preventive Services Task Force (USPSTF) has recommended biennial mammography screening at ages 50 to 74, with clinical recommendations for discussion between patients and their providers about individual risks and preferences for starting screening before age 50.^1,2 The USPSTF concluded in 2016 that the evidence was insufficient to assess the benefits and harms of digital breast tomosynthesis (DBT) as a primary screening method. In contrast to digital mammography (DM) which uses a single x-ray projection per view, DBT involves multiple projections which are used to construct image slices, reducing tissue overlap. Screening facilities have been transitioning from DM to DBT because of lower false-positive recall rates and higher cancer detection rates for DBT compared with DM,^3,4 even though data do not show a reduction in rates of advanced cancer diagnosis.^5,6 Other changes since the 2016 recommendation include increasing breast cancer incidence among younger women and advances in treatment.⁷ Importantly, Black and African American women (hereafter referred to as Black women) continue to experience higher breast cancer mortality than White women despite similar rates of mammography screening and lower (but steadily increasing) rates of breast cancer incidence.⁸ The impact of these new data on the net benefit of screening mammography is unknown.

Population simulation models are a valuable tool for synthesizing evidence from observational and trial data to estimate the impact of different screening strategies. We used well-established Cancer Intervention and Surveillance Modeling Network (CISNET) models to estimate the benefits and harms of breast cancer screening strategies that varied by the ages to start and stop screening, modality, and interval for women overall and for Black women, including the impact of screening strategies on breast cancer mortality disparities for Black women. The results are provided to inform discussions about U.S. breast cancer screening strategies by the USPSTF and other groups.

Methods

Model Overview

Six CISNET breast cancer models were used to estimate benefits and harms of mammography screening: Dana-Farber Cancer Institute (Model D), Erasmus University Medical Center (Model E), Georgetown Lombardi Comprehensive Cancer Center-Albert Einstein College of Medicine (Model GE), University of Texas MD Anderson Cancer Center (Model M), Stanford University (Model S), and University of Wisconsin-Madison-Harvard Medical School (Model W). These models were included in the two previous decision analyses conducted for the USPSTF.^9,10 Since the 2016 analysis, the models have incorporated several updates to inputs including screening performance characteristics for DM and DBT, current breast cancer incidence trends, updated breast cancer stage and hormone receptor distributions, “real world” treatment assignment and effects for women overall and for Black women. Detailed descriptions of each model are available elsewhere^11–17 and in an online Technical Report.¹⁸ The University of Wisconsin Health Sciences Institutional Review Board determined that this study was not human subjects research.

Population for Analysis

These analyses modeled a single cohort of U.S. women with no personal history of breast cancer born in 1980 (i.e., age 40 in 2020) excluding women at the highest risk (i.e., genetic susceptibility mutations or chest radiation at a young age). The models began with women at birth or age 20 or 25 (since breast cancer is rare before this age; the initiation age varied by model) and accumulated all outcomes until death. The models evaluated women overall and Black women, and strata according to breast density, elevated risk, or comorbidity level. The term “women” was used while recognizing that not all individuals eligible for mammography screening self-identify as women.¹⁹ Since model results are based on data for sex (i.e., female) rather than gender identity, models apply to cisgender women and may not accurately reflect breast cancer risk for transgender men who have not undergone bilateral mastectomy and nonbinary persons. This modeling analysis treated race as a social construct and aimed to provide evidence regarding the tradeoffs of mammography screening strategies for self-identified Black women as an approach to reduce the observed disparities in breast cancer mortality.²⁰

Model Input Parameters

All six models used a common set of data inputs for women overall and four models included race-specific variables for Black women for breast cancer incidence, breast density, DM and DBT performance, treatment assignment and efficacy, and non-breast cancer causes of death (Table 1).¹⁸ In addition, model-specific parameters were employed to represent preclinical detectable times, lead-time, and age- and estrogen receptor (ER)/human epidermal growth factor receptor 2 (HER2)-specific stage distribution in screen- vs. non-screen-detected cases on the basis of each model’s structure.

Table 1.

Common CISNET Breast Cancer Model Input Parameters

Input	Description	Updated since 2016	Race-specific	Sourcesa
Breast cancer incidence without screening	Age-period-cohort model using SEER breast cancer incidence with a period effect for mammography removed	Yes. Recent years added, 1980 instead of 1970 birth cohort.	Yes; incidence varied by race. Same data source.	Gangnon,21 Holford22
Breast density	Prevalence of breast density (BI-RADS a, b, c, d) by age group (40–44, 45–49, 50–64, 65–74, 75–89)	Yes	Yes; density varied by race. Same data source.	BCSC
Mammography performance b	Sensitivity and false-positive recall of initial and subsequent mammography by age (40–44, 45–49, 50–64, ≥65) and screening interval (annual, biennial) and density (a,b,c,d) for DM and DBT	Yes	Screening sensitivity did not vary by race. False-positive recalls did vary by race. Same data source.	BCSC6
Breast cancer stage distribution (AJCC or SEER Summary Stage)	Stage distributions by mode of detection, age group (40–44, 45–49, 50–64, 65–74, 75–89), screening round/interval (first, annual, biennial) for screen-detected cancers, and density (a, b, c, d)	Yes	Yes; stage distributions varied by race. Same data source.	BCSC
ER/HER2 joint distribution	The distribution of ER/HER2 subtypes by age (40–49, 50–74, 75–89) and stage at diagnosis	Yes	Yes; subtype distributions varied by race. Same data source.	BCSC
Survival in the absence of screening and treatment	25-y breast cancer survival before systemic treatment by joint ER/HER2 status, age group, AJCC/SEER stage or tumor size	No	No; base survival did not vary by race.	Munoz,65 Plevritis66
Treatment dissemination	Treatments and rates of use by time period, ER/HER2, stage and age for initial breast cancer diagnosis	Yes	No; treatment assignment did not vary by race.	Caswell-Jin,25 Mandelblatt,26 Plevritis66
Treatment effects	Meta-analyses of clinical trial results by ER/HER2 for initial local therapy. Clinical trial reports for efficacy of systemic primary and metastatic therapy, and of newer targeted therapies.	Yes	Yes; treatment effectiveness reduced for Black patients based on published NCCN data.34	Caswell-Jin,25 Early Breast Cancer Trialists’ Collaborative,27,28,67–70 Plevritis,66 Warner34
Other-cause mortality	Age- and cohort-specific mortality rates from non-breast cancer causes by year and level of comorbidity	Yes	Yes; other-cause mortality rates varied by race. Same data source.	Cho,71 Gangnon,35 Lansdorp-Vogelaar36
Quality of life	Utility weights for general health and decrements for screening, diagnostic evaluation, and stage-specific treatment	No	No; utility weights did not vary by race.	de Haes,45 Hamner,43,44 Stout46

Abbreviations: AJCC, American Joint Committee on Cancer; BCSC, Breast Cancer Surveillance Consortium; BI-RADS, Breast Imaging Reporting and Data Systems; CISNET, Cancer Intervention and Surveillance Modeling Network; DM, digital mammography; DBT, digital breast tomosynthesis; ER, estrogen receptor; HER2, Human epidermal growth factor receptor 2; NCCN, National Comprehensive Cancer Network; SEER, Surveillance, Epidemiology, and End Results.

^aAdditional information regarding model inputs including BCSC data is available in the online Technical Report.¹⁸

^bWith treatment, screen detection of breast cancer at an earlier stage could lead to improve survival, reduced risk of death, and/or greater chance of cure with a small tumor size, depending on model.

Five of the six models adapted an age-period-cohort modeling approach to estimate breast cancer incidence in the absence of screening among the overall and Black female population;^21,22 Model M used SEER rates with a linear model based on rates in 1975 and calibrated over time.¹² Incidence was increased for subgroups with elevated risk or with greater breast density. Density was modeled by Breast Imaging Reporting and Data Systems (BI-RADS) categories: almost entirely fatty (“a”), scattered fibroglandular densities (“b”), heterogeneously dense (“c”), and extremely dense (“d”).²³ Density category was assigned at age 40 and remained the same or decreased by one level at age 50 and again at age 65, based on observed age-specific prevalence rates in the Breast Cancer Surveillance Consortium (BCSC).¹⁸ Density was related to breast cancer risk and screening performance but was assumed to not affect molecular subtype or disease natural history (e.g., tumor growth rates). Models incorporated screening sensitivity applied to each mammogram a woman received. Age-specific sensitivity values for DM and DBT (hereafter referred to collectively as mammography) overall and by density category were also based on data from the BCSC.¹⁸ Data for the BCSC reflects breast imaging in community practice across the U.S.²⁴

With treatment, screen detection at an earlier stage could lead to improve survival, reduced risk of death, and/or greater chance of cure with a smaller tumor size, depending on model. Treatment was assigned based on age, stage, and molecular subtype. To reflect “real world” patterns of breast cancer care, the probability of receiving specific types of systemic treatment was based on data from the National Comprehensive Cancer Network as previously reported and, for newer therapies, expert opinion.^25,26 Efficacy of systemic therapy was based on the most recent published meta-analysis of clinical trials and, for newer therapies, clinical trial reports;^27,28 treatment efficacy (in the setting of optimal stage- and tumor subtype-based treatment) was assumed to be equal by race.²⁹ In contrast to efficacy, treatment effectiveness was modeled as lower for Black women due to multiple factors which may arise from systemic racism and lead to worse treatment quality (e.g., delayed initiation, suboptimal regimens, dose reductions, and incomplete cycles).^30–33 Based on published data, treatment benefit was therefore reduced by 28% for ER-negative tumors and 56% for ER-positive tumors in models restricted to Black women.³⁴

Probability of death from non-breast cancer causes was derived from CDC Wide-ranging ONline Data for Epidemiologic Research (WONDER) and the Human Mortality Database; these values were replaced by comorbidity-specific values in subgroup analyses.^35,36

Screening Strategies

We compared model results for 36 mammography screening scenarios that varied by modality (DM or DBT performed with concurrent or synthetic DM),^37–42 starting (40, 45, or 50 years) and stopping age (74 or 79), and interval (annual, biennial, or hybrid intervals). The three hybrid screening scenarios were: 1) annual from 40–49 then biennial at 50; 2) annual from 45–54 then biennial at 55; and 3) annual from 45–49 then biennial at 50. The models assumed 100% adherence to screening.

Outcomes

Benefits included percent reduction in breast cancer mortality, breast cancer deaths averted, and life-years gained (LYG) over the lifetimes of 1,000 women screened compared with no screening. We also examined quality-adjusted life-years (QALYs) gained, which were calculated using age-specific utilities for women in the general population,^43,44 with disutilities applied for undergoing screening, diagnostic evaluation, and breast cancer treatment based on the stage at diagnosis (eTable 1 in the Supplement^45,46).

Harms accumulated over the lifetime included recalls for additional imaging in women without cancer (hereafter referred to as false-positive recalls), negative biopsies performed for findings on screening mammography (hereafter referred to as benign biopsies), and over-diagnosed cases of ductal carcinoma in situ (DCIS) and invasive breast cancer. Over-diagnosis was defined as the excess breast cancer cases diagnosed in the presence of screening that were not diagnosed in the absence of screening over the lifetime. The harm of over-treatment after over-diagnosis was captured by the treatment-related decrement in utility without a change in life expectancy.

Analysis

Outcomes were tallied from age 40 (the youngest age to start screening across strategies) to death and expressed per 1,000 women. Results were summarized by the median and range across models for each outcome. We also generated efficiency frontiers by plotting the sequence of strategies that represented the largest incremental percent breast cancer mortality reduction (or LYG) per mammogram performed. Screening strategies on this frontier were considered the most efficient (i.e., no alternative existed that provided equal or greater benefit with fewer screens or harms). Because a strategy providing outcomes that was very similar to an efficient strategy may be still be considered by decision-makers for other reasons (e.g., consistency of starting and stopping ages across screening modalities),⁴⁷ we also identified “near-efficient” strategies⁴⁸ defined as a strategy within 5% of the value for screening biennially from 50–74 with DBT. Strategies that had more harms and/or fewer benefits were referred to as “inferior” to (inefficient or dominated by) other strategies.

Analyses were repeated for Black women and for strata according to density category, elevated relative risk of breast cancer, or comorbidity level.

In sensitivity analyses, for comparison with previous modeling in 2009 and 2016, we repeated analysis assuming all women with cancer received the most effective therapy (vs. the “real-world” patterns used in the primary analyses).

RESULTS

Screening Strategies for the Overall Population

The six models produced consistent results for the screening strategies (eTable 2, 3). For instance, biennial screening with DBT from ages 40 to 74 yielded a median 30.0% (range: 24.0–33.7%) reduction in breast cancer mortality vs. no screening with 1,376 (range: 1,354–1,384) false-positive recalls per 1,000 women screened (Table 2). Compared to biennial screening with DBT from ages 50 to 74, starting at age 40 averted 1.3 (range: 0.9–3.2) additional breast cancer deaths with 503 (range: 493–506) additional false-positive recalls, 65 (range 62–66) additional benign biopsies, and 2 (range 0–4) more over-diagnosed cases per 1,000 women screened (Table 3).

Table 2.

Median Lifetime Benefits and Harms (and Range across Models) of Mammography Screening Strategies per 1000 Women Screened Compared with No Screening According to Screening Modality, Interval, Starting Age, and Stopping Age

Strategy	Mammograms	Median Lifetime Benefits			Median Lifetime Harms
		Breast Cancer Mortality Reduction, %	Breast Cancer Deaths Averted	Life-Years Gained	False-positive Recalls	Benign Biopsies	Over-diagnosed Cases b
Digital Mammography until Age 74 a
Biennial
50–74	11,192 (10,999–11,278)	24.3 (18.3–27.5)	6.9 (4.8–8.6)	114.6 (109.8–165.0)	1,021 (1,003–1,027)	148 (146–149)	10 (4–29)
45–74	13,283 (13,078–13,380)	26.4 (20.4–29.3)	7.8 (5.1–9.2)	140.0 (125.0–187.7)	1,230 (1,212–1,238)	173 (170–174)	11 (4–30)
40–74	16,092 (15,863–16,215)	28.4 (22.3–31.7)	8.4 (5.6–10.1)	170.1 (141.2–214.1)	1,540 (1,520–1,551)	210 (207–212)	12 (4–33)
Hybrid
A45–49, B50–74	15,992 (15,807–16,164)	29.3 (22.4–30.5)	8.6 (5.7–9.6)	151.3 (140.8–194.5)	1,416 (1,400–1,430)	189 (187–191)	19 (4–33)
A45–54, B55–74	18,006 (17,804–18,197)	29.3 (23.0–30.2)	8.8 (5.8–9.4)	159.3 (148.6–195.5)	1,514 (1,497–1,530)	195 (193–197)	19 (4–33)
A40–49, B50–74	20,898 (20,705–21,133)	31.7 (24.4–33.1)	9.3 (6.2–10.7)	178.9 (161.9–234.6)	1,896 (1,879–1,916)	236 (234–239)	21 (4–35)
Annual
50–74	21,439 (21,010–21,650)	29.4 (24.7–31.7)	9.2 (6.8–9.5)	153.2 (134.0–181.4)	1,543 (1,513–1,557)	192 (188–194)	16 (5–39)
45–74	26,272 (25,776–26,526)	33.4 (29.8–35.4)	10.4 (7.5–11.8)	187.3 (163.6–230.1)	1,943 (1,907–1,960)	233 (229–235)	18 (5–43)
40–74	31,178 (30,649–31,493)	35.2 (31.8–37.6)	11.0 (8.0–13.1)	208.7 (200.7–275.5)	2,423 (2,385–2,446)	281 (276–283)	19 (5–45)
Digital Mammography until Age 79
Biennial
50–79	12,456 (12,223–12,560)	26.9 (22.2–30.2)	7.9 (5.6–9.4)	122.7 (118.5–172.8)	1,105 (1,084–1,113)	160 (157–161)	12 (6–34)
45–79	15,176 (14,907–15,297)	31.7 (24.8–33.3)	8.9 (6.3–11.9)	145.6 (137.8–202.5)	1,356 (1,333–1,366)	191 (187–192)	14 (6–37)
40–79	17,354 (17,081–17,494)	32.9 (25.3–34.9)	9.1 (6.4–12.3)	176.8 (149.8–233.9)	1,624 (1,601–1,636)	222 (219–223)	14 (6–37)
Hybrid
A45–49, B50–79	17,242 (17,026–17,443)	31.8 (25.4–33.1)	9.4 (6.4–11.7)	156.7 (149.5–209.4)	1,499 (1,481–1,516)	200 (198–203)	22 (6–37)
A45–54, B55–79	19,876 (19,627–20,112)	33.9 (27.5–34.2)	10.0 (6.9–12.4)	168.8 (158.7–217.2)	1,639 (1,618–1,658)	213 (210–215)	24 (6–40)
A40–49, B50–79	22,150 (21,921–22,412)	34.9 (27.4–36.2)	10.1 (6.9–13.1)	187.9 (170.5–257.0)	1,979 (1,960–2,002)	248 (245–251)	24 (6–40)
Annual
50–79	24,563 (24,014–24,831)	33.7 (32.1–35.8)	10.5 (7.9–12.2)	172.7 (145.8–192.7)	1,716 (1,678–1,733)	212 (208–214)	19 (7–46)
45–79	29,389 (28,767–29,702)	38.1 (35.1–39.5)	11.6 (8.9–14.8)	202.9 (172.0–256.1)	2,115 (2,072–2,136)	253 (248–256)	21 (7–50)
40–79	34,289 (33,633–34,667)	41.7 (37.2–42.9)	12.2 (9.4–16.1)	224.3 (211.4–300.6)	2,595 (2,550–2,621)	301 (295–304)	23 (7–52)
Digital Breast Tomosynthesis until Age 74 c
Biennial
50–74	11,208 (10,976–11,278)	25.4 (18.8–29.4)	6.7 (5.1–9.2)	120.8 (115.1–175.8)	873 (855–878)	136 (133–137)	12 (4–33)
45–74	13,299 (13,051–13,380)	27.5 (21.7–31.2)	7.5 (5.5–9.8)	141.3 (133.9–200.1)	1,080 (1,061–1,086)	164 (161–165)	13 (4–34)
40–74	16,116 (15,826–16,214)	30.0 (24.0–33.7)	8.2 (6.1–10.6)	165.2 (152.4–221.9)	1,376 (1,354–1,384)	201 (198–203)	14 (4–37)
Hybrid
A45–49, B50–74	16,053 (15,775–16,164)	29.5 (23.9–32.5)	8.0 (6.0–10.2)	153.5 (146.3–207.2)	1,242 (1,221–1,250)	184 (180–185)	19 (4–37)
A45–54, B55–74	18,072 (17,772–18,197)	29.9 (24.4–32.1)	8.2 (6.2–10.0)	161.1 (148.2–207.9)	1,317 (1,296–1,326)	193 (189–194)	20 (4–37)
A40–49, B50–74	20,979 (20,662–21,133)	32.2 (26.1–34.4)	8.8 (6.6–11.0)	181.2 (163.9–240.1)	1,691 (1,667–1,703)	238 (233–240)	21 (4–39)
Annual
50–74	21,500 (20,963–21,650)	30.6 (24.7–32.8)	8.6 (7.0–10.1)	155.6 (137.1–191.7)	1,277 (1,246–1,285)	186 (182–187)	18 (5–42)
45–74	26,349 (25,716–26,526)	34.1 (31.4–36.5)	9.7 (7.9–11.8)	193.3 (165.7–230.1)	1,647 (1,610–1,657)	234 (229–235)	20 (5–46)
40–74	31,273 (30,572–31,492)	37.0 (33.6–38.9)	10.3 (8.5–13.1)	216.6 (190.1–274.9)	2,096 (2,055–2,110)	288 (283–290)	21 (5–48)
Digital Breast Tomosynthesis until Age 79
Biennial
50–79	12,488 (12,193–12,560)	28.0 (23.6–32.2)	7.6 (6.0–10.1)	129.3 (119.6–184.1)	937 (916–943)	144 (141–145)	14 (6–38)
45–79	15,218 (14,871–15,297)	32.1 (26.5–35.5)	8.6 (6.7–12.1)	153.4 (147.7–213.1)	1,176 (1,153–1,183)	176 (173–177)	16 (6–41)
40–79	17,397 (17,037–17,494)	33.3 (27.2–36.5)	8.9 (6.9–12.5)	173.9 (161.7–237.8)	1,440 (1,415–1,449)	210 (206–211)	17 (6–42)
Hybrid
A45–49, B50–79	17,325 (16,987–17,443)	32.5 (27.2–35.3)	8.9 (6.9–11.9)	160.5 (152.8–215.4)	1,306 (1,282–1,315)	192 (188–193)	22 (6–42)
A45–54, B55–79	19,980 (19,585–20,112)	34.1 (29.2–36.4)	9.2 (7.4–12.6)	172.7 (161.0–220.8)	1,413 (1,387–1,423)	205 (202–207)	24 (6–44)
A40–49, B50–79	22,255 (21,870–22,412)	35.3 (29.4–37.2)	9.5 (7.4–13.3)	188.7 (173.4–260.1)	1,755 (1,728–1,768)	247 (242–248)	24 (6–44)
Annual
50–79	24,687 (23,953–24,831)	34.5 (32.6–36.9)	9.8 (8.0–12.2)	173.2 (148.2–203.6)	1,405 (1,367–1,417)	202 (197–204)	22 (7–50)
45–79	29,517 (28,692–29,701)	39.1 (37.1–40.8)	10.9 (9.0–14.8)	207.1 (176.1–255.8)	1,774 (1,730–1,789)	250 (244–252)	24 (7–54)
40–79	34,441 (33,538–34,666)	41.7 (39.2–43.0)	11.5 (9.9–16.1)	229.7 (200.4–300.7)	2,224 (2,175–2,240)	304 (298–307)	25 (7–56)

Abbreviations: A, annual; B, biennial.

^aDigital mammography strategies show results for Models D, E, GE, M, and W.

^bOver-diagnosed cases are in situ and invasive breast cancer cases that would not have been clinically detected in the absence of screening. Overdiagnosis is calculated by subtracting the number of cases detected in the screening scenario from the number of cases detected in the no-screening scenario. Model S (Stanford University) is excluded because it does not include DCIS.

^cDigital breast tomosynthesis strategies show results for Models D, E, GE, M, S, and W.

Table 3.

Lifetime additional benefits and harms of screening mammography starting at age 40 instead of 50 until age 74 per 1,000 women from six models

Initial Strategy	Comparison Strategy	Model
		D	E	GE	M	S	W	Median	Range
Mammograms
DM B50–74	DM B40–74	4,936	4,900	4,924	4,869	a	4,864	4,900	4,864 – 4,936
DBT B50–74	DBT B40–74	4,936	4,895	4,924	4,870	4,920	4,850	4,907	4,850 – 4,936
Percent breast cancer mortality reduction
DM B50–74	DM B40–74	4.1%	4.1%	8.6%	6.5%	a	3.1%	4.1%	3.1 – 8.6%
DBT B50–74	DBT B40–74	4.4%	4.3%	8.6%	6.4%	4.9%	3.6%	4.6%	3.6 – 8.6%
Breast cancer deaths averted
DM B50–74	DM B40–74	1.3	1.2	3.2	1.5	a	0.8	1.3	0.8 – 3.2
DBT B50–74	DBT B40–74	1.4	1.3	3.2	1.5	1.2	0.9	1.3	0.9 – 3.2
Life-years gained
DM B50–74	DM B40–74	43.1	35.3	102.7	53.3	a	31.4	43	31.4 – 102.7
DBT B50–74	DBT B40–74	46.1	36.5	102.9	52.3	27.0	36.3	41	27.0 – 102.9
False-positive recalls
DM B50–74	DM B40–74	523	520	521	514	a	517	520	514 – 523
DBT B50–74	DBT B40–74	506	502	504	493	505	499	503	493 – 506
Benign biopsies
DM B50–74	DM B40–74	63	62	62	60	a	62	62	60 – 63
DBT B50–74	DBT B40–74	66	65	66	62	66	65	65	62 – 66
Overdiagnosed cases (DCIS and invasive)
DM B50–74	DM B40–74	1	2	0	3	a	4	2	0 – 4
DBT B50–74	DBT B40–74	1	2	0	3	a	4	2	0 – 4

Abbreviations: A, annual; B, biennial; D, Dana-Farber Cancer Institute; DBT, digital breast tomosynthesis; DM, digital mammography; E, Erasmus Medical Center; GE, Georgetown Lombardi Comprehensive Cancer Center-Albert Einstein College of Medicine; M, University of Texas MD Anderson Cancer Center; S, Stanford University; W, University of Wisconsin-Madison and Harvard Pilgrim Health Care Institute.

^aNot available

Annual screening led to greater reductions in mortality than biennial strategies, with a 37.0% median (range: 33.6–38.9%, Table 2) reduction with screening annually from ages 40–74 with DBT, but resulted in more false-positive recalls, benign biopsies, and over-diagnosed cases.

With biennial screening from ages 40 to 74, DM resulted in 1,540 false-positive recalls and 210 benign biopsies per 1000 women screened versus 1,276 and 201, respectively, with DBT (Table 2). Use of DBT instead of DM further decreased breast cancer mortality by approximately 1 percentage point and averted less than 1 additional breast cancer death per 1,000 women and reduced false-positive recalls by approximately 150–300 per 1000 women over their lifetimes among nine screening strategies stopping at age 74 (eTable 4).

Stopping screening at age 79 versus 74 generally resulted in an additional 3–5 percentage point mortality reduction, one additional breast cancer death averted, 64–172 more false-positive recalls per 1000 women and 2–4 added over-diagnosed cases, depending on strategy (eTable 5).

Among all possible strategies, five DBT screening strategies were identified as efficient or near-efficient for both percent mortality reduction and LYG in at least five of six models, including one with stopping age 74 (biennial starting at age 50) and four with stopping age 79 (biennial starting at age 40; biennial starting at age 45; annual from ages 40 to 49 with biennial thereafter; and annual starting at age 40) (Figure 1, eFigure 1–2, eTable 6). Efficient strategies ranged from 1.7–4.3 more breast cancer deaths averted and 41–168 more false-positive recalls than screening biennially from ages 50 to 74 per 1000 women (Figure 2). Five similar strategies were identified as efficient when limited to the 18 options with stopping age 74 (biennial starting at age 40, biennial starting at age 45, biennial starting at age 50, annual at ages 40 to 49 with biennial at ages 50 to 74, and annual at ages 40 to 74; eFigure 3).

Figure 1.

Lifetime Number of Screening Mammograms, Life-Years Gained, and Breast Cancer Mortality Reduction (%) per 1000 Women Screened from Model D (Dana-Farber Cancer Institute) According to Screening Strategy for Women Overall and for Black Women

Abbreviations: A, annual; B, biennial; DBT, digital breast tomosynthesis; DM, digital mammography.

Each point represents a different screening strategy, and the line represents the estimated efficiency frontier for Model D. Efficiency frontier graphs for all models are shown in the Supplement. Labels for each point are provided for efficient and near-efficient strategies. Grey shading, in which near-efficient strategies are located, shows area within 5% of the value for screening biennially during ages 50–74 with DBT. For plots of life-years gained, near-efficient strategies included those within 2.20 days of life gained per woman of the efficient frontier for all women and 3.22 days of life per Black woman. For plots of the percent reduction in breast cancer mortality, near-efficient strategies included those within 5% of the efficiency frontier on a relative scale, which is equivalent to 1.27 percentage points on an absolute scale for women overall and 1.21 percentage points on an absolute scale for Black women. Strategies vary by age at starting and stopping screening, interval between mammograms, and screening modality.

*Near-efficient.

Figure 2.

Breast Cancer Deaths Averted and Benign Biopsies per 1,000 Women Screened with Various Digital Breast Tomosynthesis Mammography Screening Strategies

Abbreviations: A, annual; B, biennial.

All strategies use digital breast tomosynthesis. Blue bars represent the initial strategy (B50-74, biennial screening at ages 50–74), with the orange bars showing the increase in breast cancer deaths averted and benign biopsies by screening more frequently, starting screening earlier, and/or stopping screening later. Results shown as medians across six models of women overall (D, E, GE, M, S, and W) and across four models of Black women (D, GE, M, and W).

Note that the differences in medians calculated by subtracting values in Table 2, Table 3, eTable 8, and eTable 9 may not be equivalent to the median of the differences across models, as shown in this figure.

Screening Strategies for Black Women

Seven screening strategies were efficient or near-efficient for LYG or breast cancer mortality reduction among Black women (Figure 1, eFigure 4–5, eTable 7). Three strategies were efficient or near-efficient for both metrics among most models including biennial from ages 40 to 79, biennial from ages 45 to 79, and annual from ages 40 to 79. Expanding biennial screening with DBT from ages 50–74 to ages 40–74 or 79 averted a median of 1.8 and 2.8 additional breast cancer deaths across models, respectively (Figure 2).

Tradeoffs between benefits and harms of different screening strategies for Black women followed similar patterns as for all women combined (eTables 8–10). All strategies resulted in more breast cancer deaths averted and LYG for Black women compared with the same strategies for women overall. However, this gain in averted breast cancer deaths was insufficient to reduce breast cancer mortality disparities for Black women as compared with women overall. Specifically, if Black women were screened with the same strategy as for women overall, breast cancer mortality for Black women would remain over 40% greater than for women overall (Table 4). Alternatively, if Black women were screened annually from ages 40 to 49 with biennial screening from ages 50 to 79 and the overall population was screened biennially from ages 40 to 74, the ratio of breast cancer mortality rate for Black women versus women overall would be reduced from 1.44 (28.8/20.0) to 1.34 (26.8/20.0; a disparity reduction of 23%). Notably, Black women screened annually at ages 40 to 49 and biennially at ages 50 to 79 would experience fewer false-positives and mammograms per breast cancer death averted with greater life-years gained than women overall screened biennially at ages 40 to 74 (eTable 10).

Table 4.

Ratios of breast cancer deaths and life-years for 1,000 Black women versus 1,000 women overall by screening strategy

Strategies	All Womena	Screening Strategies for Black Women
		No screening	B50–74 c	B40–74 c	A40–49, B50–74 c	B45–79	B40–79	A40–49, B50–79 c	A40–74 c	A40–79
Breast Cancer Deaths		39.3	30.0	28.8	28.3	27.5	27.3	26.8	26.0	23.7
No screening	28.3	1.39	1.06	1.02	1.00	0.97	0.97	0.95	0.92	0.84
B50–74	21.1	1.86	1.42	1.36	1.34	1.30	1.29	1.27	1.23	1.12
B40–74 b	20.0	1.97	1.50	1.44	1.42	1.38	1.37	1.34	1.30	1.19
A40–49, B50–74 b	19.6	2.01	1.53	1.47	1.44	1.41	1.39	1.37	1.33	1.21
B45–79	19.4	2.03	1.55	1.48	1.46	1.42	1.41	1.39	1.34	1.23
B40–79	19.1	2.05	1.57	1.50	1.48	1.44	1.43	1.40	1.36	1.24
A40–49, B50–79	18.7	2.10	1.60	1.53	1.51	1.47	1.46	1.43	1.39	1.27
A40–74 b	18.2	2.16	1.65	1.58	1.55	1.51	1.50	1.47	1.43	1.30
A40–79	16.9	2.33	1.78	1.71	1.68	1.63	1.62	1.59	1.54	1.41
Life-Years		No screening	B50–74 c	B45–79	B40–74 c	B40–79	A40–49, B50–74 c	A40–49, B50–79 c	A40–74 c	A40–79
		41.783	41.994	42.058	42.063	42.080	42.080	42.097	42.116	42.139
No screening	43.670	0.957	0.962	0.963	0.963	0.964	0.964	0.964	0.964	0.965
B50–74	43.789	0.954	0.959	0.960	0.961	0.961	0.961	0.961	0.962	0.962
B45–79	43.850	0.953	0.958	0.959	0.959	0.960	0.960	0.960	0.960	0.961
B40–74 b	43.866	0.953	0.957	0.959	0.959	0.959	0.959	0.960	0.960	0.961
B40–79	43.879	0.952	0.957	0.959	0.959	0.959	0.959	0.959	0.960	0.960
A40–49, B50–74 b	43.882	0.952	0.957	0.958	0.959	0.959	0.959	0.959	0.960	0.960
A40–49, B50–79	43.897	0.952	0.957	0.958	0.958	0.959	0.959	0.959	0.959	0.960
A40–74 b	43.907	0.952	0.956	0.958	0.958	0.958	0.958	0.959	0.959	0.960
A40–79	43.927	0.951	0.956	0.957	0.958	0.958	0.958	0.958	0.959	0.959

^aCalculations use the median values for breast cancer deaths from four models (D, GE, M, and W). Strategies limited to efficient and near-efficient strategies for both percent breast cancer mortality reduction and life-years gained versus no screening in most models for all women, listed in eTable 6, along with selected other strategies.

^bStrategy not efficient nor near-efficient for at least 5 of 6 models for both percent breast cancer mortality reduction and life-years gained versus no screening for women overall as shown in eTable 6.

^cStrategy not efficient nor near-efficient for at least 3 of 4 models for both percent breast cancer mortality reduction and life-years gained versus no screening for Black women as shown in eTable 7.

Density, Elevated Risk, and Comorbidity Sub-groups

Only three strategies were efficient in most models for women with dense breasts (BI-RADS category c and d) including biennial screening from ages 50 to 74, biennial screening from ages 40 to 79, and annual screening at ages 40 to 79 (eTable 11). Across all strategies efficient in at least one density category, breast cancer deaths averted using DBT for women with almost entirely fatty breasts ranged from 4.9 for biennial screening at ages 50 to 74 to 7.6 with annual screening at ages 40 to 79, and increased among women with extremely dense breasts from 8.3 to 14.6 (eTable 12).

Models showed greater benefits and fewer harms as breast cancer risk increased to 150% and 200% of average-risk, with the same three screening strategies efficient for both elevated risk levels as for dense breasts (eTable 13). Incremental benefits of screening after age 74 were reduced in the presence of severe comorbidities (eTable 14).

Sensitivity Analysis

When all breast cancer cases received the most effective treatment for their cancer subtype and screening stopped at age 74, the percent reduction in breast cancer mortality increased as compared with the primary analysis where cases received treatment based on “real world” treatment patterns (eTable 15).

Discussion

This study used six well-established models to estimate the potential benefits and harms of different breast cancer screening strategies in the U.S. The models demonstrated that screening initiation at age 40 had superior benefit-to-harm tradeoffs compared to no screening and other screening strategies. Benefits of DBT were comparable with DM but resulted in fewer false-positive recalls and similar benign biopsies. Annual screening would lead to greater reductions in breast cancer mortality than biennial strategies but correspondingly more false-positive recalls and over-diagnosed cases. Since breast cancer death rates are higher for Black women, all screening strategies generated greater survival and mortality benefits for Black women than for women overall. However, to reduce racial disparities in breast cancer mortality in the absence of improved equity in the treatment setting, an increase in screening intensity such as annual screening of Black women from ages 40 to 49 would also be needed. Benefits for women with elevated risk or higher breast density were higher than for women overall, but the rankings of strategies were similar to those for women overall. In addition, several strategies with a stopping age of 79 were efficient. For women aged 75 to 79, comorbidities may be an important factor in decisions about when to cease breast cancer screening.

Compared to our 2016 analysis,¹⁰ the predicted benefit-to-harm ratios with biennial strategies starting at age 40 or 45 have modestly improved. Due to recent increases in breast cancer incidence among women aged 40–49 (154.1 to 160.5 per 100,000 from 1999 to 2018), life-years gained were notably higher for screening strategies that started at age 40 or 45.^7,49 Past analyses assumed optimal treatment selection; starting screening earlier partially compensated for less-than-optimal “real world” treatment uptake in the current analysis. Also, with the growing evidence for lower false-positive recall rates with DBT than DM,^3,4 fewer harms were associated with earlier ages of screening initiation than occurred in prior analyses.

Prospective studies that include multiple rounds of breast cancer screening are needed to determine whether, compared with DM, DBT results in a shift toward detecting breast cancer at earlier stages with a concomitant decrease in advanced stage. Initial studies suggest that DBT leads to an increase in stage I invasive breast cancer as compared with DM, although a reduction in advanced stage has not yet been demonstrated.^6,50–52 Screening benefit related to reductions in breast cancer deaths depends on the advantage of beginning treatment in earlier versus more advanced stages.

This analysis extended findings published in 2021 for one model (GE) that evaluated strategies for reducing breast cancer mortality disparities and improving health equity between Black and White women.⁵³ Our models are intended to generate findings for individuals that self-identify as Black, defining race as a social construct where the socio-political environment influences biological processes over the lifecourse.^54–56 The current study showed that Black women gained more life-years per mammogram than woman overall for each screening strategy. This was due in part to Black women having higher breast cancer mortality, especially among younger women, and gaining less benefit from intended therapy due to worse quality of care. If Black women obtained annual mammography from age 40 to 49 with biennial screening afterwards, mortality disparities were projected to decline while also achieving similar benefit-to-harm tradeoffs as biennial screening starting at age 40 for women overall. These results are similar to those recently published by others using U.S. mortality data that more intensive screening could potentially reduce the Black/White disparity in breast cancer mortality.⁵⁷ If healthcare systems, policy makers, clinicians, and scientists work to fully eliminate disparities experienced by Black women, the balance of benefits and harms for screening could eventually change to the extent that more intensive screening strategies for Black women are no longer needed to increase equity. However, as described by Chapman et al,⁵³ until treatment disparities are substantially decreased or eliminated, screening Black women more intensively represents an immediate possible solution for improving equity. Optimal implementation of any strategy will also require improved equity in DBT access and timely diagnostic workup.⁵⁸

Our analysis considered breast cancer screening strategies using mammography, which has poorer performance in women with dense breasts compared with non-dense breasts. Our models estimated that for any given mammography screening strategy, women with dense breasts had more deaths averted and greater life years gained per mammogram than those with non-dense breasts, but false-positive recall rates were higher. Evidence on the benefits of supplemental screening with breast MRI or ultrasound for women with dense breasts is limited.^59,60 With federal legislation expanding breast density notification in September 2024 and the absence of clear clinical guidelines for supplemental screening,⁶¹ this is a critical area for future research and policymaking.

After accounting for recent trends in life expectancy (prior to the coronavirus-2019 pandemic) and improvements in breast cancer therapies, strategies with screening until age 79 were identified as efficient. This is consistent with a recent simulation study but contrasts with an emulated trial based on Medicare data showing that breast cancer mortality was not significantly reduced among women screened through age 79.^62,63 Unfortunately, current breast cancer screening trials in progress, including TMIST and WISDOM, are not recruiting women older than 74, and trials testing screening in older women are unlikely to be conducted. Evidence from other types of studies is needed to better understand outcomes of screening for older women.

Relative rankings of strategies were similar across the models. However, the models differ in meaningful ways in structure and assumptions. For example, some models incorporated a benefit to screening due to within-stage shift in detection and subsequent treatment (Model E, S, and W) while others required a stage shift (Model D and GE) or assigned greater benefit for screen-detected than clinically-detected cases within each stage at detection (Model M). Among the five models that included DCIS as well as invasive breast cancer, three models found that the overall number of overdiagnosed cases exceeded the number of breast cancer deaths averted for all screening strategies considered. Underlying incidence in the absence of screening and the proportion of tumors that were nonprogressive are unknown and unobservable; therefore, the different results across models with their respective assumptions about breast cancer natural history provide a range of possible estimates.

Limitations

This research has many important strengths, including the collaboration of six independent modeling teams with consistent results and use of the most current data on incidence, screening performance, and modern, real-world therapy. Several caveats should also be considered in interpreting our results. First, the models portray the entire lifetime of women in the 1980 birth cohort and assumes that future trends continued along the same trajectories as observed now. Second, we compared results for Black women to the overall female population, which leads to an underestimate of the impact of racism. This was a necessary simplification because these models did not produce estimates for other minoritized groups, White women, or non-Black women. In future research, models will be developed to examine results by racial and ethnic groups as well as interventions to improve health equity. Finally, some analyses were based on findings from fewer than six models for pragmatic reasons. In particular, some models were well-poised to examine analyses of racial disparities,⁵³ breast density,⁶⁴ or comorbidities³⁶ due to programming completed in previous projects.

Conclusions

Overall, this analysis suggests that biennial screening starting at ages 40 or 45 with DM or DBT and continuing through age 74 or 79 provides gains in life years and breast cancer mortality reduction per mammogram—and averts more deaths from breast cancer among Black women—than waiting to start screening at age 50. More intensive screening for populations of women with greater risk of diagnosis or death can maintain similar benefit-to-harm tradeoffs and reduce breast cancer mortality disparities. In the presence of recent changes in breast cancer incidence and improvements in screening technology and breast cancer therapy, mammography screening remains an important strategy to reduce the breast cancer burden.

Key Points.

Question:

What are the benefits and harms of different screening mammography strategies?

Findings:

Six validated CISNET models found that, compared to no screening, biennial mammography screening with digital breast tomosynthesis from age 40 to 74 yielded a median of 8.2 breast cancer deaths averted per 1,000 women screened, equal to a 30% reduction in breast cancer mortality, and 165 life-years gained, 1,376 false-positive recalls, 201 benign biopsies, and 14 over-diagnosed cases per 1,000 women screened. For each strategy, benefits were larger for Black women than for all women.

Meaning:

Biennial mammography from ages 40 to 74 has favorable benefit-to-harm tradeoffs.

From the Breast Cancer Working Group of the Cancer Intervention and Surveillance

Modeling Network and the Breast Cancer Surveillance Consortium. The small writing committee included Drs. Amy Trentham-Dietz, Jeanne Mandelblatt, Christina Chapman, Jinani Jayasekera, Kathryn Lowry, and Diana Miglioretti.

Data Sharing Statement:

Data used in this analysis come from published manuscripts and publicly available reports.