Benefits

Randomized controlled trials (RCTs) initiated 50 years ago provide evidence that screening mammography reduces breast cancer–specific mortality for women aged 60 to 69 years (solid evidence) and women aged 50 to 59 years (fair evidence). Population-based studies done more recently raise questions about the benefits for populations who participate in screening for longer time periods.

Magnitude of Effect: Based on a meta-analysis of RCTs, the number of women needed to invite for screening to prevent one breast cancer death depends on the woman’s age: for women aged 39 to 49 years, 1,904 women needed (95% confidence interval [CI], 929–6,378); for women aged 50 to 59 years, 1,339 women needed (95% CI, 322–7,455); and for women aged 60 to 69 years, 377 women needed (95% CI, 230–1,050).[9]

The validity of meta-analyses of RCT demonstrating a mortality benefit is limited by improvements in medical imaging and treatment in the decades since their completion. The 25-year follow-up from the Canadian National Breast Screening Study (CNBSS),[10] completed in 2014, showed no mortality benefit associated with screening mammograms.

Harms

Based on solid evidence, screening mammography may lead to the following harms:

For all of these conclusions regarding potential harms from screening mammography, internal validity, consistency, and external validity are good.

Benefits

The CNBSS trial did not study the efficacy of CBE versus no screening. Ongoing randomized trials, two in India and one in Egypt, are designed to assess the efficacy of screening CBE but have not reported mortality data.[3-8] Thus, the efficacy of screening CBE cannot be assessed yet.

Harms

Screening by CBE may lead to the following harms:

Benefits

BSE has been compared with no screening and has been shown to have no benefit in reducing breast cancer mortality.

Harms

There is solid evidence that formal instruction and encouragement to perform BSE leads to more breast biopsies and more diagnoses of benign breast lesions.

Invasive breast cancer

Breast cancer can be diagnosed when breast tissue cells removed during a biopsy are studied microscopically. The breast tissue to be sampled can be identified by an abnormality on an imaging study or because it is palpable. Breast biopsies can be performed with a thin needle attached to a syringe (fine-needle aspirate), a larger needle (core biopsy), or by excision (excisional biopsy). Image guidance can improve accuracy. Needle biopsies sample an abnormal area large enough to make a diagnosis. Excisional biopsies aim to remove the entire region of abnormality.

Ductal carcinoma in situ (DCIS)

DCIS is a noninvasive condition that can be associated with, or evolve into, invasive cancer, with variable frequency and time course.[9] Some authors include DCIS with invasive breast cancer statistics, but others argue that it would be better if the term were replaced with ductal intraepithelial neoplasia, similar to the terminology used for cervical and prostate precursor lesions, and that excluding DCIS from breast cancer statistics should be considered.

DCIS is most often diagnosed by mammography. In the United States, only 4,900 women were diagnosed with DCIS in 1983 before the adoption of mammography screening, compared with approximately 56,500 women who are expected to be diagnosed in 2024.[1,9,10] The Canadian National Breast Screening Study-2, which evaluated women aged 50 to 59 years, found a fourfold increase in DCIS cases in women screened by clinical breast examination (CBE) plus mammography, compared with those screened by CBE alone, with no difference in breast cancer mortality.[11] For more information, see Breast Cancer Treatment.

The natural history of DCIS is poorly understood because nearly all DCIS cases are detected by screening and nearly all are treated. Development of breast cancer after treatment of DCIS depends on the pathological characteristics of the lesion and on the treatment. In a randomized trial, 13.4% of women whose DCIS was excised by lumpectomy developed ipsilateral invasive breast cancer within 90 months, compared with 3.9% of those treated by both lumpectomy and radiation.[12] Among women diagnosed and treated for DCIS, the percentage of women who died of breast cancer is lower than that for the age-matched population at large.[13,14] This favorable outcome may reflect the benign nature of the condition, the benefits of treatment, or the volunteer effect (i.e., women who undergo breast cancer screening are generally healthier than those who do not do so).

Atypia

Atypia, which is a risk factor for breast cancer, is found in 4% to 10% of breast biopsies.[15,16] Atypia is a diagnostic classification with considerable variation among practicing pathologists.[17]

Variability of pathologists’ diagnoses on the interpretation of breast biopsy specimens

The range of pathologists' diagnoses of breast tissue includes benign without atypia, atypia, DCIS, and invasive breast cancer. The incidence of atypia and DCIS breast lesions has increased over the past three decades as a result of widespread mammography screening, although atypia is generally mammographically occult.[18,19] Misclassification of breast lesions may contribute to either overtreatment or undertreatment of lesions—with variability especially in the diagnoses of atypia and DCIS.[17,20-24]

The largest study on this topic, the B-Path study, involved 115 practicing U.S. pathologists who interpreted a single-breast biopsy slide per case, and it compared their interpretations with an expert consensus-derived reference diagnosis.[17] While the overall agreement between the individual pathologists’ interpretations and the expert reference diagnoses was highest for invasive carcinoma, there were markedly lower levels of agreement for DCIS and atypia.[17] As the B-Path study included higher proportions of cases of atypia and DCIS than typically seen in clinical practice, the authors expanded their work by applying Bayes’ theorem to estimate how diagnostic variability affects accuracy from the perspective of a U.S. woman aged 50 to 59 years having a breast biopsy.[20] At the U.S. population level, it is estimated that 92.3% (95% confidence interval [CI], 91.4%–93.1%) of breast biopsy diagnoses would be verified by an expert reference consensus diagnosis, with 4.6% (95% CI, 3.9%–5.3%) of initial breast biopsies estimated to be overinterpreted and 3.2% (95% CI, 2.7%–3.6%) under interpreted. Figure 1 shows the predicted outcomes per 100 breast biopsies, overall and by diagnostic category.

Enlarge

To address the high rates of discordance in breast tissue diagnosis, laboratory policies that require second opinions are becoming more common. A national survey of 252 breast pathologists participating in the B-Path study found that 65% of respondents reported having a laboratory policy that requires second opinions for all cases initially diagnosed as invasive disease. Additionally, 56% of respondents reported policies that require second opinions for initial diagnoses of DCIS, while 36% of respondents reported mandatory second opinion policies for cases initially diagnosed as atypical ductal hyperplasia.[25] In this same survey, pathologists overwhelmingly agreed that second opinions improved diagnostic accuracy (96%).

A simulation study that used B-Path study data evaluated 12 strategies for obtaining second opinions to improve interpretation of breast histopathology.[26] Accuracy improved significantly with all second-opinion strategies, except for the strategy limiting second opinions only to cases of invasive cancer. Accuracy improved regardless of the pathologists’ confidence in their diagnosis or their level of experience. While the second opinions improved accuracy, they did not completely eliminate diagnostic variability, especially in the challenging case of breast atypia.

Women at increased risk who may benefit more from screening

Women with Hodgkin and non-Hodgkin lymphoma who were treated with mantle irradiation have an increased risk of breast cancer, starting 10 years after completing therapy and continuing life-long. Therefore, screening mammography has been advocated, even though it may begin at a relatively young age.[27,28]

Women who self-identify as Black in the United States have a lower overall lifetime risk of developing breast cancer than White women, although they have a slightly higher breast cancer incidence in their 30s and 40s. However, Black women have a 40% higher breast cancer mortality than White women, a finding that is attributed to multiple factors, such as delayed follow-up of abnormal mammograms, later stage at diagnosis, inferior breast cancer treatment, and more aggressive tumor types.

To inform the U.S. Preventive Services Task Force (USPSTF) 2024 breast cancer screening recommendations, a modeling study was commissioned. This study used the six Cancer Intervention and Surveillance Modeling Network (CISNET) models to assess the benefits and harms of mammography screening at different starting ages and frequencies in the average-risk population of U.S. women, overall, and for Black women, specifically. The models incorporated race-specific data on breast cancer incidence, tumor subtypes, stage distribution, treatment quality/effectiveness, and mortality. Because of Black women’s inferior breast cancer outcomes, the models found that Black women experienced a slightly greater absolute benefit (i.e., more breast cancer deaths prevented) from mammography screening compared with the general population.[29] For example, the models estimated that screening 1,000 women in the general population every 2 years between the ages of 50 years and 74 years (as recommended by previous USPSTF guidelines) would avert an estimated 6.7 breast cancer deaths, while biennial screening starting at age 40 years would avert an additional 1.3 breast cancer deaths. Among Black women, screening every 2 years between the ages of 50 years and 74 years would avert an estimated 9.2 breast cancer deaths per 1,000 women screened, while biennial screening starting at age 40 years would avert an additional 1.8 breast cancer deaths. (See Table 1.)

In response to these findings and to address inequities in breast cancer outcomes, the USPSTF recommended that all average-risk women initiate screening at age 40 years (instead of at age 50 years, as previously recommended) and be screened every 2 years until age 74 years. Although this approach may result in additional lives saved, the models demonstrate that earlier screening also increases the likelihood of harm from mammography screening. In the general population of women, biennial screening from age 40 to 74 years, rather than age 50 to 74 years, would result in 503 additional false-positive results, 65 additional biopsies, and 2 additional overdiagnosed breast cancers per 1,000 women screened. Among Black women specifically, biennial screening starting at age 40 years, rather than at age 50 years, would result in 439 additional false-positive results, 75 additional biopsies, and 2 additional overdiagnosed breast cancers per 1,000 women screened. (See Table 1.)

Although mathematical modeling is increasingly used to estimate mammography’s benefits and harms, it has a number of limitations, as described later in this summary. Limitations include models’ reliance on multiple assumptions and their inability to predict and incorporate factors that are as highly dynamic as breast cancer diagnosis and treatment. The assumptions and methods used by mathematical models are difficult for nonmodelers to understand. Therefore, it can be risky to base policy decisions on the findings of mathematical models. Further, as the USPSTF has noted, to address higher breast cancer mortality in Black women, systematic approaches are needed to address existing inequities in screening quality, diagnostic processes, and treatment quality. It is not clear whether earlier screening initiation in the general population will improve outcomes among Black women without dedicated efforts to address such documented inequities.

Individuals who benefit less from screening

Screening biases–concepts

Numerous uncontrolled trials and retrospective series have documented the ability of mammography to diagnose small, early-stage breast cancers, which have a favorable clinical course.[35] Individuals whose cancer is detected by screening show a higher survival rate than those whose cancers are not detected by screening even when screening has not prolonged any lives. This concept is explained by the following four types of statistical bias:

The impact of these biases is not known. A new randomized controlled trial (RCT) with cause-specific mortality as the end point is needed to determine both survival benefit and impact of overdiagnosis, lead time, length time, and healthy volunteer biases. This is not achievable; randomly assigning patients to screen and nonscreen groups would be unethical, and at least three decades of follow-up would be needed, during which time changes in treatment and imaging technology would invalidate the results. Decisions must therefore be based on available RCTs, despite their limitations, and on ecological or cohort studies with adequate control groups and adjustment for confounding. For more information, see Cancer Screening Overview.

Assessment of performance and accuracy

Performance benchmarks for screening mammography in the United States are described on the Breast Cancer Surveillance Consortium (BCSC) website. For more information, see Cancer Screening Overview.

Sensitivity

The sensitivity of mammography is the percentage of women with breast cancers detected by mammographic screening. Sensitivity depends on tumor size, conspicuity, hormone sensitivity, breast tissue density, patient age, timing within the menstrual cycle, overall image quality, and interpretive skill of the radiologist. Overall sensitivity is approximately 79% but is lower in younger women and in those with dense breast tissue (see the BCSC website).[37-39] Sensitivity is not the same as benefit because some woman with possible breast cancer are harmed by overdiagnosis. According to the Physician's Insurance Association of America (PIAA), delay in diagnosis of breast cancer and errors in diagnosis are common causes of medical malpractice litigation. PIAA data from 2002 through 2011 note that the largest total indemnity payments for breast cancer claims are for errors in diagnosis.[40]

Specificity and false-positive rate

The specificity of mammography is the percentage of all women without breast cancer whose mammograms are negative. The false-positive rate is the likelihood of a positive test in women without breast cancer. Low specificity and high rate of false-positives result in unnecessary follow-up examinations and procedures. Because specificity includes all women without cancer in the denominator, even a small percentage of false-positives turns out to be a large number in absolute terms. Thus—in screening—a good specificity must be very high. Even 95% specificity is quite low for a screening test.

Interval cancers

Interval cancers are cancers that are diagnosed in the interval between a normal screening examination and the anticipated date of the next screening mammogram. One study found interval cancers occurred more often in women younger than 50 years, and had mucinous or lobular histology, high histological grade, high proliferative activity with relatively benign mammographic features, and no calcifications. Conversely, screen-detected cancers often had tubular histology, small size, low stage, hormone sensitivity, and a major component of DCIS.[41] Overall, interval cancers have characteristics of rapid growth,[41,42] are diagnosed at an advanced stage, and carry a poor prognosis.[43]

Analysis of mammography screening length bias preferentially detects indolent cancers that grow more slowly (e.g., exist for a longer length of time in the preclinical phase). In contrast, the more aggressive cancers grow faster (e.g., spend a shorter length of time in the preclinical phase) and are often detected clinically in the intervals between screening sessions. For a more detailed explanation of length and lead-time bias in cancer screening, see Cancer Screening Overview.

In recent years, novel breast cancer screening technologies have been assessed in clinical trials with the interval cancer detection rate as the primary outcome of interest, and newer screening methods recommended on the basis of reductions in interval cancer detection rates. However, the interval cancer detection rate has not been validated as a proper surrogate for breast cancer mortality, and its use as a surrogate outcome measure in breast cancer screening trials remains controversial.

In breast cancer screening programs, screen-detected breast cancers tend to have a better prognosis than cancers detected during the intervals between screening sessions (interval breast cancers). This was confirmed in a registry-based cohort study from Manitoba in which interval breast cancers were more likely than were screen-detected breast cancers to be high-grade and estrogen receptor–negative, and associated with greater than a threefold increased risk of breast cancer death.[44]

The Nova Scotia Breast Screening Program defined missed cancers as those that were false-negatives on the previous screening exam, occurring less often than 1 per 1,000 women. It concluded that interval cancers occurred in approximately 1 per 1,000 women aged 40 to 49 years, and 3 per 1,000 women aged 50 to 59 years.[45]

Conversely, a larger trial found that interval cancers were more prevalent in women aged 40 to 49 years. Those appearing within 12 months of a negative screening mammogram were usually attributable to greater breast density. Those appearing within a 24-month interval were related to decreased mammographic sensitivity caused by greater breast density or to rapid tumor growth.[46]

Patient characteristics

The accuracy of mammography has been noted to vary with patient characteristics, such as a woman's age, breast density, whether it is her first or subsequent exam, and the time since her last mammogram. Younger women have lower sensitivity and higher false-positive rates than do older women.

The Million Women Study in the United Kingdom found decreased sensitivity and specificity in women aged 50 to 64 years if they used postmenopausal hormone therapy, had prior breast surgery, or had a body mass index below 25.[47] Increased time since the last mammogram increases sensitivity, recall rate, and cancer detection rate and decreases specificity.[48]

The United Kingdom Age Trial assessed the efficacy of mammography screening for women younger than 50 years. After a median follow-up of 22.8 years, there was no difference in breast cancer mortality between women randomly assigned to initiate screening at age 39 to 41 years until entry into the National Health Service (NHS) breast screening program at age 50 to 52 years, versus the group that did not initiate mammography screening until entry into the NHS breast screening program (RR, 0.98; 95% CI, 0.79–1.22; P = .86).[49]

Sensitivity may be improved by scheduling the exam after the initiation of menses or during an interruption from hormone therapy.[50] Obese women have more than a 20% increased risk of having false-positive mammography, although sensitivity is unchanged.[51]

Breast density

Dense breasts may obscure the detection of small masses on mammography, thereby reducing the sensitivity of mammography.[13] For women of all ages, high breast density is associated with 10% to 29% lower sensitivity.[38] High breast density is also associated with a modestly increased risk of developing breast cancer.[52] High breast density does not confer a higher risk of breast cancer death.

High breast density is an inherent trait, which can be inherited [53,54] or affected by age; endogenous [55] and exogenous [56,57] hormones;[58] selective estrogen receptor modulators, such as tamoxifen;[59] and diet.[60] Hormone therapy is associated with increased breast density, lower mammographic sensitivity, and an increased rate of interval cancers.[61]

Dense breast tissue is not abnormal. Breast density describes the proportion of dense versus fatty tissue in a mammographic image.[62] The American College of Radiology’s BI-RADS classifies breast density as follows:

The latter two categories are considered dense breast tissue, a description affecting 43% of women aged 40 to 74 years.[63] A radiologist's assignment of breast density is subjective and may vary over time in any woman.[63,64]

There is limited high-quality evidence to guide optimal breast cancer screening in individuals with dense breasts. For dense breasts, digital breast tomosynthesis has improved sensitivity and modestly lowers false-positive rates compared with conventional digital mammography.[65]

Supplemental imaging with ultrasonography or breast magnetic resonance imaging (MRI) has been suggested by some groups for screening women with dense breasts, but there are no data showing that this strategy results in lower breast cancer mortality. The potential harm of adding these supplemental screening tests is the likelihood of producing more false-positives, leading to additional imaging and breast biopsies, with resultant anxiety and cost.[66] Supplemental screening may also increase overdiagnosis of breast cancer with resultant overtreatment.

A study examining cancer detection end points in women with dense breasts undergoing supplemental screening (e.g., ultrasound, MRI, digital resources) showed higher breast cancer detection, but it is not known if that translates into cancer protection.[67] An RCT of supplemental MRI versus mammography only in 40,373 individuals aged 50 to 75 years with extremely dense breasts in the Netherlands was performed.[68] The study showed lower incidence of interval cancers at 2 years of follow-up in the MRI group (2.5 per 1,000 screenings in the group invited to receive MRI, 0.8 per 1,000 in the group that actually received MRI, and 5.0 per 1,000 in the group that received mammography only). This finding suggests that at least some of the excess cancers detected by MRI in the MRI group were earlier diagnoses of cancers that would have become clinically apparent. However, whether earlier diagnoses facilitated by MRI resulted in improved clinical outcomes has not been shown. As would be expected, cancers detected by MRI were more likely to have favorable tumor characteristics than interval cancers. MRI screening was associated with 79.8 false-positive results per 1,000 screenings.[68]

A prospective multicenter study, known as the Dense Breast Tomosynthesis Ultrasound Screening Trial (DBTUST), investigated whether ultrasound improved cancer detection after DBT in women with dense breasts.[69] Between December 2015 and June 2021, 6,179 women at three Pennsylvania locations underwent three rounds of annual screening with DBT and technologist-performed handheld ultrasounds. The images were interpreted by two radiologists at baseline, 12 months, and 24 months. The study concluded that technologist-performed ultrasound screening modestly improved detection of cancer in women with dense breasts by 1.3 cases per 1,000 in year 1 and by 1 case per 1,000 in years 2 to 3. This screening also increased the false-positive recall rate. In 3 years, 1,007 (16.3%) women had a false-positive recall based on DBT, and an additional 761 (12.3%) women had a false-positive recall based on ultrasound.

The FDA mandates that mammography facilities report breast density to patients and suggest that patients speak with their primary care clinician about supplemental screening.[70] However, limited evidence, inconsistent guidelines, and wording of breast density reports have generated confusion and anxiety among patients and health care providers.[71]

Tumor characteristics

Mucinous and lobular cancers are more easily detected by mammography. Rapidly growing cancers can sometimes be mistaken for normal breast tissue (e.g., medullary carcinomas, an uncommon type of invasive ductal breast cancer that is often associated with the BRCA1 mutation and aggressive characteristics, but that may demonstrate comparatively favorable responses to treatment).[41,72] Some other cancers associated with BRCA1/2 mutations, which may appear indolent, can also be missed.[73,74]

Physician characteristics

Radiologists’ performance is variable, affected by levels of experience and the volume of mammograms they interpret.[75] Biopsy recommendations of radiologists in academic settings have a higher positive PPV than do community radiologists.[76] Fellowship training in breast imaging may improve detection.[11]

Performance also varies by facility. Mammographic screening accuracy was higher at facilities offering only screening examinations than at those also performing diagnostic tests. Accuracy was also better at facilities with a breast imaging specialist on staff, performing single rather than double readings, and reviewing performance audits two or more times each year.[77]

False-positive rates are higher at facilities where concern about malpractice is high and at facilities serving vulnerable women (racial or ethnic minority women and women with less education, limited household income, or rural residence).[78] These populations may have a higher cancer prevalence and a lack of follow-up.[79]

Artificial intelligence algorithms

Artificial intelligence (AI) algorithms are being developed to interpret screening mammograms and breast biopsy specimens.[80-82] While such tools may improve interpretive speed and reproducibility in the future, it is unknown if they will exacerbate overdiagnosis [83] and how they might influence physicians’ final assessments.

International comparisons

International comparisons of screening mammography have found higher specificity in countries with more highly centralized screening systems and national quality assurance programs.[84,85]

The recall rate in the United States is twice that of the United Kingdom, with no difference in the rate of cancer detection.[84]

Prevalent versus subsequent examination and the interval between exams

The likelihood of diagnosing cancer is highest with the prevalent (first) screening examination, ranging from 9 to 26 cancers per 1,000 screens, depending on the woman’s age. The likelihood decreases for follow-up examinations, ranging from 1 to 3 cancers per 1,000 screens.[86]

The optimal interval between screening mammograms is unknown; there is little variability across the trials despite differences in protocols and screening intervals. A prospective U.K. trial randomly assigned women aged 50 to 62 years to receive mammograms annually or triennially. Although tumor grade and nodal status were similar in the two groups, more cancers of slightly smaller size were detected in the annual screening group than in the triennial screening group.[87]

A large observational study found a slightly increased risk of late-stage disease at diagnosis for women in their 40s who were adhering to a 2-year versus a 1-year schedule (28% vs. 21%; OR, 1.35; 95% CI, 1.01–1.81), but no difference was seen for women in their 50s or 60s based on schedule difference.[88,89]

A Finnish study of 14,765 women aged 40 to 49 years randomly assigned women to receive either annual screens or triennial screens. There were 18 deaths from breast cancer in 100,738 life-years in the triennial screening group and 18 deaths from breast cancer in 88,780 life-years in the annual screening group (HR, 0.88; 95% CI, 0.59–1.27).[90]

Randomized controlled trials (RCTs)

RCTs that studied the effect of screening mammography on breast cancer mortality were performed between 1963 and 2015, with participation by over half-a-million women in four countries. One trial, the Canadian NBSS-2, compared mammography plus clinical breast examination (CBE) to CBE alone; the other trials compared screening mammography with or without CBE to usual care. For a detailed description of the trials, see the Appendix of Randomized Controlled Trials section.

The trials differed in design, recruitment of participants, interventions (both screening and treatment), management of the control group, compliance with assignment to screening and control groups, and analysis of outcomes. Some trials used individual randomization, while others used cluster randomization in which cohorts were identified and then offered screening; one trial used nonrandomized allocation by day of birth in any given month. Cluster randomization sometimes led to imbalances between the intervention and control groups. Age differences have been identified in several trials, although the differences had no major effect on the trial outcome.[91] In the Edinburgh Trial, socioeconomic status, which correlates with the risk of breast cancer mortality, differed markedly between the intervention and control groups, rendering the results uninterpretable.

Breast cancer mortality was the major outcome parameter for each of these trials, so the attribution of cause of death required scrupulous attention. The use of a blinded monitoring committee (New York) and a linkage to independent data sources, such as national mortality registries (Swedish trials), were incorporated but could not ensure impartial attributions of cancer death for women in the screening or control arms. Possible misclassification of breast cancer deaths in the Two-County Trial biasing the results in favor of screening has been suggested.[92]

There were also differences in the methodology used to analyze the results of these trials. Four of the five Swedish trials were designed to include a single screening mammogram in the control group and were timed to correspond with the end of the series of screening mammograms in the study group. The initial analysis of these trials used an evaluation analysis, tallying only the breast cancer deaths that occurred in women whose cancer was discovered at or before the last study mammogram. In some of the trials, a delay occurred in the performance of the end-of-study mammogram, resulting in more time for members of the control group to develop or be diagnosed with breast cancer. Other trials used a follow-up analysis, which counts all deaths attributed to breast cancer, regardless of the time of diagnosis. This type of analysis was used in a meta-analysis of four of the five Swedish trials as a response to concerns about the evaluation analyses.[92]

The accessibility of the data for international audits and verification also varied, with a formal audit having been undertaken only in the Canadian trials. Other trials have been audited to varying degrees, but with less rigor.[93]

All of these studies were designed to study breast cancer mortality rather than all-cause mortality because breast cancer deaths contribute only a small proportion of total mortality in any given population. When all-cause mortality in these trials was examined retrospectively, only the Edinburgh Trial showed a difference attributable to the previously noted socioeconomic differences in the study groups. The meta-analysis (follow-up methods) of the four Swedish trials also showed a small improvement in all-cause mortality.

The relative improvement in breast cancer mortality attributable to screening is approximately 15% to 20%, and the absolute improvement at the individual level is much less. The potential benefit of breast cancer screening can be expressed as the number of lives extended because of early breast cancer detection.[94,95]

The RCT results represent experiences in a defined period of regular examinations, but in practice, women undergo 20 to 30 years of screening throughout their lifetimes.[89,96]

There are several problems with using these RCTs that were performed up to 50 years ago to estimate the current benefits of screening on breast cancer mortality. These problems include the following:

For these reasons, estimates of the breast cancer mortality reduction resulting from current screening are based on well-conducted cohort and ecological studies in addition to the RCTs.

Effectiveness of population-based screening programs

An estimate of screening effectiveness can be obtained from nonrandomized controlled studies of screened versus nonscreened populations, case-control studies of screening in real communities, and modeling studies that examine the impact of screening on large populations. These studies must be designed to minimize or exclude the effects of unrelated trends influencing breast cancer mortality such as improved treatment and heightened awareness of breast cancer in the community.

Three population-based, observational studies from Sweden compared breast cancer mortality in the presence and absence of screening mammography programs. One study compared two adjacent time periods in 7 of the 25 counties in Sweden and found a statistically significant breast cancer mortality reduction of 18% to 32% attributable to screening.[97] The most important bias in this study is that the advent of screening in these counties occurred over a period during which dramatic improvements in the effectiveness of adjuvant breast cancer therapy were being made, changes that were not addressed by the study authors. The second study considered an 11-year period comparing seven counties with screening programs with five counties without them.[98] There was a trend in favor of screening, but again, the authors did not consider the effect of adjuvant therapy or differences in geography (urban vs. rural) that might affect treatment practices.

The third study attempted to account for the effects of treatment by using a detailed analysis by county. It found screening had little impact, a conclusion weakened by several flaws in design and analysis.[99]

In Nijmegen, the Netherlands, where a population-based screening program was undertaken in 1975, a case-cohort study found that screened women had decreased mortality compared with unscreened women (OR, 0.48).[100] However, a subsequent study comparing Nijmegen breast cancer mortality rates with neighboring Arnhem in the Netherlands, which had no screening program, showed no difference in breast cancer mortality.[101]

A community-based case-control study of screening in high-quality U.S. health care systems between 1983 and 1998 found no association between previous screening and reduced breast cancer mortality, but the mammography screening rates were generally low.[102]

A well-conducted ecological study compared three pairs of neighboring European countries that were matched on similarity in health care systems and population structure, one of which had started a national screening program some years earlier than the others. The investigators found that each country had experienced a reduction in breast cancer mortality, with no difference between matched pairs that could be attributed to screening. The authors suggested that improvements in breast cancer treatment and/or health care organizations were more likely responsible for the reduction in mortality than was screening.[103]

A systematic review of ecological and large cohort studies published through March 2011 compared breast cancer mortality in large populations of women, aged 50 to 69 years, who started breast cancer screening at different times. Seventeen studies met inclusion criteria, but all studies had methodological problems, including control group dissimilarities, insufficient adjustment for differences between areas in breast cancer risk and breast cancer treatment, and problems with similarity of measurement of breast cancer mortality between compared areas. There was great variation in results among the studies, with four studies finding a relative reduction in breast cancer mortality of 33% or more (with wide CIs) and five studies finding no reduction in breast cancer mortality. Because only a part of the overall reduction in breast cancer mortality could possibly be attributed to screening, the review concluded that any relative reduction in breast cancer mortality resulting from screening would likely be no more than 10%.[104]

A U.S. ecological analysis conducted between 1976 and 2008 examined the incidence of early-stage versus late-stage breast cancer for women aged 40 years and older. To assess a screening effect, the authors compared the magnitude of increase in early-stage cancer with the magnitude of an expected decrease in late-stage cancer. Over the study, the absolute increase in the incidence of early-stage cancer was 122 cancers per 100,000 women, while the absolute decrease in late-stage cancers was 8 cases per 100,000 women. After adjusting for changes in incidence resulting from hormone therapy and other undefined causes, the authors concluded (1) the benefit of screening on breast cancer mortality was small, (2) between 22% and 31% of diagnosed breast cancers represented overdiagnosis, and (3) the observed improvement in breast cancer mortality was probably attributable to improved treatment rather than screening.[105]

An analytic approach was used to approximate the contributions of screening versus treatment to breast cancer mortality reduction and the magnitude of overdiagnosis.[106] The shift in the size distribution of breast cancers in the United States (before the introduction of mammography) to 2012 (after its widespread dissemination), was investigated using SEER data in women aged 40 years and older. The rate of clinically meaningful breast cancer was assumed to be stable during this time. The authors documented a lower incidence of larger (≥2 cm) tumors as well as a reduction in breast cancer case fatality. The lower mortality for women with larger tumors was attributed to improvements in therapy. Two-thirds of the decline in size-specific case fatality was ascribed to improved treatment.

Enlarge

A prospective cohort study of community-based screening programs in the United States found that annual compared with biennial screening mammography did not reduce the proportion of unfavorable breast cancers detected in women aged 50 to 74 years or in women aged 40 to 49 years without extremely dense breasts. Women aged 40 to 49 years with extremely dense breasts did have a reduction in cancers larger than 2.0 cm with annual screening (OR, 2.39; 95% CI, 1.37–4.18).[107]

An observational study of women aged 40 to 74 years conducted in 7 of 12 Canadian screening programs compared breast cancer mortality in those participants screened at least once between 1990 and 2009 (85% of the population) with those not screened (15% of the population). The abstract reported a 40% average breast cancer mortality among participants; however, it was likely intended to report a 40% reduction in breast cancer mortality on the basis of language used in the Discussion section.[108]

Limitations of this study included the lack of all-cause mortality data, the extent of screening, screening outside of the study, screening prior to the study, the method used for calculating expected mortality and the referent rates of nonparticipants, nonparticipant survival, province-specific population differences, the extent to which limitations of the database prevented correcting for age and other differences between participants, the generalizability of the substudy data of a single province (British Columbia), and the potentially large impact of selection bias. Overall, the study lacked important data and had limitations in methodology and data analysis.

Statistical modeling of breast cancer incidence and mortality in the United States

The optimal screening interval has been addressed by modelers. Modeling makes assumptions that may not be correct; however, the credibility of modeling is greater when the model produces overall results that are consistent with randomized trials and when the model is used to interpolate or extrapolate. For example, if a model’s output agrees with RCT outcomes for annual screening, it has greater credibility to compare the relative effectiveness of biennial versus annual screening.

In 2000, the National Cancer Institute formed a consortium of modeling groups (Cancer Intervention and Surveillance Modeling Network [CISNET]) to address the relative contribution of screening and adjuvant therapy to the observed decline in breast cancer mortality in the United States.[109] These models predicted reductions in breast cancer mortality similar to those expected in the circumstances of the RCTs but updated to the use of modern adjuvant therapy. In 2009, CISNET modelers addressed several questions related to the harms and benefits of mammography, including comparing annual versus biennial screening.[89] Women aged 50 to 74 years received most of the mortality benefit of annual screening by having a mammogram every 2 years. The reduction in breast cancer deaths that was maintained because of the move from annual to biennial screening ranged across the six models from 72% to 95%, with a median of 80%.

Data are limited as to how much of the reduction in mortality, seen over time from 1990 onward, is attributable to advances in imaging techniques for screening and as to how much is the result of the improved effectiveness of therapy. In one CISNET study of six simulation models, about one-third of the decrease in breast cancer mortality in 2012 was attributable to screening, with the balance attributed to treatment.[110] In this CISNET study, the mean estimated reduction in overall breast cancer mortality rate was 49% (model range, 39%–58%), relative to the estimated baseline rate in 2012 if there was no screening or treatment; 37% (model range, 26%–51%) of this reduction was associated with screening, and 63% (model range, 49%–74%) of this reduction was associated with treatment.

Overdiagnosis

Overdiagnosis occurs when screening procedures detect cancers that would never become clinically apparent in the absence of screening. It is a special concern because identification of the cancer does not benefit the individual, while the side effects of diagnostic procedures and cancer treatment may cause significant harm. The magnitude of overdiagnosis is debated, particularly regarding DCIS, a cancer precursor whose natural history is unknown. By reason of this inability to predict confidently the tumor behavior at time of diagnosis, standard treatment for invasive cancers and DCIS can cause overtreatment. The related harms include treatment-related side effects and the number of harms associated with a cancer diagnosis, which are immediate. Conversely, a mortality benefit would occur at an uncertain point in the future.

One approach to understanding overdiagnosis is to examine the prevalence of occult cancer in women who died of noncancer causes. In an overview of seven autopsy studies, the median prevalence of occult invasive breast cancer was 1.3% (range, 0%–1.8%) and of DCIS was 8.9% (range, 0%–14.7%).[119,120]

Overdiagnosis can be indirectly measured by comparing breast cancer incidence in screened versus unscreened populations. These comparisons can be confounded by differences in the populations, such as time, geography, health behaviors, and hormone usage. The calculations of overdiagnosis can vary in their adjustment for lead-time bias.[121,122] An overview of 29 studies found calculated rates of overdiagnosis to be 0%–54%, with rates from randomized studies between 11% and 22%.[123] In Denmark, where screened and unscreened populations existed concurrently, the rate of overdiagnosis of invasive cancer was calculated to be 14% and 39%, using two different methodologies. If DCIS cases were included, the overdiagnosis rates were 24% and 48%. The second methodology accounts for regional differences in women younger than the screening age and is likely more accurate.[124]

Theoretically, in a given population, the detection of more breast cancers at an early stage would result in a subsequent reduction in the incidence of advanced-stage cancers. This has not occurred in any of the populations studied to date. Thus, the detection of more early-stage cancers likely represents overdiagnosis. A population-based study in the Netherlands showed that about one-half of all screen-detected breast cancers, including DCIS, would represent overdiagnosis and is consistent with other studies, which showed substantial rates of overdiagnosis associated with screening.[125]

A cohort study in Norway compared the increase in cancer incidence in women who were eligible for screening with the cancer incidence in younger women who were not eligible for screening, eligibility was based on age and residence. Eligible women experienced a 60% increase in incidence of localized cancers (RR, 1.60; 95% CI, 1.42–1.79), while the incidence of advanced cancers remained similar in the two groups (RR, 1.08; 95% CI, 0.86–1.35).[126]

A population study that compared different counties in the United States showed that higher rates of screening mammography use were associated with higher rates of breast cancer diagnoses, yet there was no corresponding decrease in 10-year breast cancer mortality.[127] The strengths of this study include its very large size (16 million women) and the strength and consistency of correlation observed across counties. The limitations of this study include the self-reporting of mammograms, the use of a 2-year window to estimate screening prevalence, and the period of analysis (when menopausal hormone use was present).[127]

The extent of overdiagnosis has been estimated in the Canadian NBSS, a randomized clinical trial. At the end of the five screening rounds, 142 more invasive breast cancer cases were diagnosed in the mammography arm, compared with the control arm.[128] At 15 years, the excess number of cancer cases in the mammography arm versus the control arm was 106, representing an overdiagnosis rate of 22% for the 484 screen-detected invasive cancers.[128]

As a consequence of screening mammography, greater numbers of breast cancers with indolent behavior are now identified, resulting in potential overtreatment. In a secondary analysis of a randomized trial of tamoxifen versus no systemic therapy in patients with early breast cancer, the authors utilized the 70-gene MammaPrint assay and identified 15% of patients at ultra-low risk, with 20-year disease-specific survival rates of 97% in the tamoxifen group and 94% in the control group. Thus, these patients would likely have extremely good outcomes with surgery alone. The frequency of such ultra-low risk cancers in the screened population is likely around 25%. Tools such as the 70-gene MammaPrint assay might be utilized in the future to identify these cancers, and thereby, reduce the risk of overtreatment. However, additional studies are needed to confirm these findings.[129]

In 2016, the Canadian NBSS, a randomized screening trial with 25-year follow-up, re-estimated overdiagnosis of breast cancer from mammography screening by age group and concluded that approximately 30% of invasive screen-detected cancers in women aged 40 to 49 years and up to 20% of those detected in women aged 50 to 59 years were overdiagnosed. When in situ cancers are included, the estimated risks of overdiagnosis are 40% aged 40 to 49 years and 30% in women aged 50 to 59 years. Overdiagnosis was calculated as the persistent excess incidence in the screened arm versus the control arm divided by the number of screen-detected cases (excess incidence method). Requirements for adequate estimation of overdiagnosis utilizing this method included the following:

These conditions were largely met in the CNBSS because population-based screening did not become available throughout Canada until a minimum of 2 years later and in most instances 5 to 10 years later (thereby, allowing for cessation of screening after the trial screening period and follow-up longer than most estimates of lead time), because contamination is documented to have been minimal, and because individual randomization resulted in 44 almost identically distributed demographic factors and risk factors between the two trial arms.

Since the conclusion of the trial screening period in 1988, differences in screening quality, intensity, invited age range, and biopsy thresholds decrease the generalizability of these results. These factors and improved imaging technique/quality and low threshold for biopsy, likely contribute to lower estimates of overdiagnosis of in situ cancer than that of invasive cancer.[130]

Table 2 shows results from a 10-year period of screening 10,000 women, estimating the number of women with breast cancer or DCIS that would never become clinically important (overdiagnosis). There was likely no overdiagnosis in the Health Insurance Plan study, which used old-technology mammography and CBE. Overdiagnosis has become more prominent in the era of improved-technology mammography. The improved technology has not, however, been shown to make further reductions in mortality than the original technology. In summary, breast cancer overdiagnosis is a complex topic. Studies that used many different methods reported a wide range of estimates, and there is currently no way to assess whether new cancer cases are overdiagnosed or are of real harm to patients.[111]

False-positives leading to additional interventions

Because fewer than 5 per 1,000 women screened have breast cancer, most abnormal mammograms are false-positives, even given the 90% specificity of mammography (i.e., 90% of all women without breast cancer will have a negative mammogram).[86]

This high false-positive rate of mammography is underestimated and can seem counterintuitive because of a statistically based cognitive bias known as the base rate fallacy. Because the base rate of breast cancer is low, (5/1000), the false-positive rate vastly exceeds the true-positive rate, even when using a very accurate test.

Mammography’s true-positive rate of approximately 90% means that, of women with breast cancer, approximately 90% will test positive. The true-negative rate of 90% means that, of women without breast cancer, 90% will test negative. A 10% false-positive rate over 1,000 people means that there will be 100 false-positives in 1,000 people. If 5 in 1,000 women have breast cancer, then 4.5 women with breast cancer will have a positive test. In other words, there will approximately 100 false-positives for every 4.5 true positives.

Further, abnormal results from screening mammograms prompt additional tests and procedures, such as mammographic views of the region of concern, ultrasound, MRI, and tissue sampling (by fine-needle aspiration, core biopsy, or excisional biopsy). Overall, the harm from unnecessary tests and treatments must be weighed against the benefit of early detection.

A study of breast cancer screening in 2,400 women enrolled in a health maintenance organization found that over a decade, 88 cancers were diagnosed, 58 of which were identified by mammography. One-third of the women had an abnormal mammogram result that required additional testing: 539 additional mammograms, 186 ultrasound examinations, and 188 biopsies. The cumulative biopsy rate (the rate of true positives) resulting from mammographic findings was approximately 1 in 4 (23.6%). The PPV of an abnormal screening mammogram in this population was 6.3% for women aged 40 to 49 years, 6.6% for women aged 50 to 59 years, and 7.8% for women aged 60 to 69 years.[131] A subsequent analysis and modeling of data from the same cohort of women, estimated that the risk of having at least one false-positive mammogram was 7.4% (95% CI, 6.4%–8.5%) at the first mammogram, 26.0% (95% CI, 24.0%–28.2%) by the fifth mammogram, and 43.1% (95% CI, 36.6%–53.6%) by the ninth mammogram.[132] Cumulative risk of at least one false-positive result depended on four patient variables (younger age, higher number of previous breast biopsies, family history of breast cancer, and current estrogen use) and three radiologic variables (longer time between screenings, failure to compare the current and previous mammograms, and the individual radiologist’s tendency to interpret mammograms as abnormal). Overall, the factor most responsible for a false-positive mammogram was the individual radiologist’s tendency to read mammograms as abnormal.

A prospective cohort study of community-based screening found that a greater proportion of women undergoing annual screening had at least one false-positive screen after 10 years than did women undergoing biennial screening, regardless of breast density. For women with scattered fibroglandular densities, the difference was 68.9% (annual) versus 46.3% (biennial) for women in their 40s. For women aged 50 to 74 years, the difference for this density group was 49.8% (annual) versus 30.7% (biennial).[107]

As shown in Table 2, the estimated number of women out of 10,000 who underwent annual screening mammography during a 10-year period with at least one false-positive test result is 6,130 for women aged 40 to 50 years and 4,970 for women aged 60 years. The number of women with a false-positive test that results in a biopsy is estimated to range from 700 to 980, depending on age.[111]

Relationship between prior screening results and subsequent breast cancer diagnosis

A longitudinal Norwegian study correlated benign abnormal screening results with long-term breast cancer outcomes. Women with any abnormal screening examination had an increased risk of subsequent breast cancer, despite a negative evaluation (see Table 3). The features of the subsequent breast cancer were more favorable for the women who had prior screening abnormalities, possibly because the preexisting breast abnormality was a marker for slow-growing premalignant disease.[133]

False-negatives leading to a false sense of security

The sensitivity of mammography ranges from 70% to 90%, depending on characteristics of the interpreting radiologist (level of experience) and characteristics of the woman (age, breast density, hormone status, and diet). Assuming an average sensitivity of 80%, mammograms will miss approximately 20% of the breast cancers that are present at the time of screening (false-negatives). Many of these missed cancers are high risk, with adverse biological characteristics. If a normal mammogram dissuades or postpones a woman or her doctor from evaluating breast symptoms, she may suffer adverse consequences. Thus, a negative mammogram should never dissuade a woman or her physician from additional evaluation of breast symptoms.

Discomfort

Positioning of the woman and breast compression reduce motion artifact and improve mammogram image quality. Pain and/or discomfort was reported by 90% of women undergoing mammography, with 12% of women rating the sensation as intense or intolerable.[134] A systematic review of 22 studies investigating mammography-associated pain and discomfort found wide variations, some of which were associated with menstrual cycle stage, anxiety, and premammography anticipation of pain.[135]

Radiation exposure

The major risk factors for radiation-associated breast cancer are young age at exposure and dose; however, rarely there are women with an inherited susceptibility to radiation-induced damage who must avoid radiation exposure at any age.[136,137] For many women older than 40 years, the likely benefits of screening mammography outweigh the risks.[138,136,139] Standard two-view screening mammography exposes the breasts to a mean dose of 4 mSv, and the whole body to 0.29 mSv.[137,140] Thus, up to one breast cancer may be induced per 1,000 women undergoing annual mammograms from ages 40 to 80 years. Such risk is doubled in women with large breasts who require increased radiation doses and in women with breast augmentation who require additional views. Radiation-induced breast cancers may be reduced fivefold for women who begin biennial screening at age 50 years rather than annually at age 40 years.[141]

Psychological harms of false-positives

A telephone survey of 308 women performed 3 months after screening mammography revealed that about one-fourth of the 68 women recalled for additional testing were still experiencing worry that affected their mood or functioning, even though that testing had ruled out cancer.[142] Research into whether the psychological impact of a false-positive test is long-standing yields mixed results. A cohort study in Spain in 2002 found immediate psychological impact to a woman after receiving a false-positive mammogram, but these results dissipated within a few months.[143] A cohort study in Denmark in 2013 that measured the psychological effects of a false-positive test result several years after the event found long-term negative psychological consequences.[144] Several studies have shown that the anxiety after evaluation of a false-positive test leads to increased participation in future screening examinations.[145-148]

Financial strain and opportunity costs

These potential harms of screening have not been well researched, but it is clear that they exist.