Effectiveness of Breast Cancer Screening: Systematic Review and Meta-analysis to Update the 2009 U.S. Preventive Services Task Force Recommendation

In 2009, the U.S. Preventive Services Task Force (USPSTF) recommended biennial mammography screening for women aged 50 to 74 years (1) on the basis of evidence of benefits and harms (2). The USPSTF concluded that screening decisions for women aged 40 to 49 years should be based on individual considerations, and that evidence was insufficient to assess benefits and harms for women aged 75 years or older (1). Mammography screening in the United States is generally opportunistic, unlike many screening programs organized as public health services in other countries. Despite changes in practice guidelines and variation in clinical practices (3), overall screening rates in the United States have remained relatively stable for the past decade (4, 5). Data from the Healthcare Effectiveness Data and Information Set indicate that mammography screening in 2014 in HMOs was performed for 74% of eligible women covered by commercial plans, 72% by Medicare, and 59% by Medicaid (6). This systematic review updates evidence for the USPSTF on the effectiveness of mammography screening in reducing breast cancer mortality, all-cause mortality, and advanced breast cancer for women at average risk; and how effectiveness varies by age, risk factors, screening intervals, and imaging modalities. Systematic reviews of harms of screening (7), performance characteristics of screening methods (8), and accuracy of breast density determination and use of supplemental screening technologies (9) are provided in separate reports. Methods Scope, Key Questions, and Analytic Framework The USPSTF determined the scope and key questions for this review by using established methods (10, 11). A standard protocol was developed and publicly posted on the USPSTF Web site. A technical report further describes the methods and includes search strategies and additional information (7). Investigators created an analytic framework outlining the key questions, patient populations, interventions, and outcomes reviewed (Appendix Figure 1). Key questions include the effectiveness of screening in reducing breast cancer mortality, all-cause mortality, and advanced breast cancer, and how effectiveness differs by age, risk factors, screening intervals, and modalities (mammography [film, digital, tomosynthesis], magnetic resonance imaging [MRI], and ultrasonography). Appendix Figure 1. Analytic framework and key questions. KQ = key question. * Excludes women with preexisting breast cancer; clinically significant BRCA1 or BRCA2 mutations, LiFraumeni syndrome, Cowden syndrome, hereditary diffuse gastric cancer, or other familial breast cancer syndromes; high-risk lesions (ductal carcinoma in situ, lobular carcinoma in situ, atypical ductal hyperplasia, atypical lobular hyperplasia); or previous large doses of chest radiation (20 Gy) before age 30 y. Risk factors include family history; breast density; race/ethnicity; menopausal status; current use of menopausal hormone therapy or oral contraceptives; prior benign breast biopsy; and, for women aged >50 y, body mass index. Morbidity includes physical adverse effects of treatment, quality-of-life measures, and other measures of impairment. Screening modalities include mammography (film, digital, tomosynthesis), magnetic resonance imaging, ultrasonography, and clinical breast examination (alone or in combination). The target population for the USPSTF recommendation includes women aged 40 years or older, and excludes women with known physical signs or symptoms of breast abnormalities and those at high-risk for breast cancer whose surveillance and management are beyond the scope of the USPSTF's recommendations for prevention services (i.e., preexisting breast cancer or high-risk breast lesions, hereditary genetic syndromes associated with breast cancer, or previous large doses of chest radiation before age 30 years). Risk factors considered in this review are common among women who are not at high risk for breast cancer (12) (Appendix Figure 1). Data Sources and Searches A research librarian conducted electronic database searches of the Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, and Ovid MEDLINE to 4 June 2015. Searches were supplemented by references identified from additional sources, including reference lists and experts. Additional unpublished data were provided by the investigators of the Canadian National Breast Screening Study (CNBSS) and Swedish Two-County Trial. Study Selection Two investigators independently evaluated each study to determine inclusion eligibility on the basis of prespecified criteria. Discrepancies were resolved through consensus. We included randomized, controlled trials (RCTs); observational studies of screening cohorts; and systematic reviews that compared outcomes of women exposed to screening versus not screening. For advanced cancer outcomes, studies that reported the incidence of late-stage disease among screened and unscreened populations were included, whereas those reporting comparisons of detection methods that did not capture a woman's longitudinal screening experience were not included (e.g., rates of screen-detected vs. nonscreen-detected cancer). Studies providing outcomes specific to age, risk factors, screening intervals, and modalities were preferred over studies providing general outcomes, when available. Studies most clinically relevant to practice in the United States were selected over studies that were less relevant. Relevance was determined by practice setting, population, date of publication, and use of technologies and therapies in current practice. Studies meeting criteria for high quality and those with designs ranked higher in the study designbased hierarchy of evidence were emphasized because they are less susceptible to bias (e.g., RCTs over observational studies). Data Extraction and Quality Assessment Details of the study design, patient population, setting, screening method, interventions, analysis, follow-up, and results were abstracted by one investigator and confirmed by a second. Two investigators independently applied criteria developed by the USPSTF (10, 11) to rate the quality of each study as good, fair, or poor for studies designed as RCTs, cohort studies, casecontrol studies, and systematic reviews; criteria to rate other study designs included in this review are not available. Discrepancies were resolved through consensus. Data Synthesis We conducted several meta-analyses to determine more precise summary estimates when adequate data were reported by trials rated as fair- or good-quality. In each meta-analysis, the number of included trials was counted as the number of discrete data sources contributing to the summary estimate using their most recent results. To determine the appropriateness of meta-analysis, we considered clinical and methodological diversity and assessed statistical heterogeneity. All outcomes were binary (breast cancer mortality, all-cause mortality, and advanced cancer incidence defined by stage and tumor size). We used a random-effects model to combine relative risks (RRs) as the effect measure of the meta-analyses, while incorporating variation among studies. A profile-likelihood model was used to combine studies in the primary analyses (13). We assessed the presence of statistical heterogeneity among the studies by using the standard Cochran chi-square test, and the magnitude of heterogeneity by using the I 2 statistic (14). To account for clinical heterogeneity and obtain clinically meaningful estimates, we stratified the analyses by age group whenever possible (39 to 49 years, 50 to 59 years, 60 to 69 years, 70 to 74 years, and 50 years). We obtained additional age-stratified data for the meta-analysis from the investigators of 3 trials (15, 16) (Tabr L. Personal communication). For breast cancer mortality, we used 2 methods of including cases to help clarify discrepancies between estimates. The long case accrual method counts all breast cancer cases contributing to breast cancer deaths. In this method, the case accrual time is equivalent to or close to the follow-up time. The short case accrual method includes only deaths that occur among cases of breast cancer diagnosed during the screening intervention period, and in some trials, within an additional defined case accrual period. The longest follow-up times available for each trial were selected for inclusion in the initial meta-analyses, and sensitivity analyses were conducted by using results of short case accrual methods. We calculated the absolute rate reduction for 100000 woman-years of follow-up (i.e., 10000 women followed for 10 years) for each age group on the basis of the combined RR and the combined cancer rate of the control group. We estimated combined cancer rates for each age group for controls with a random effects Poisson model using data from the trials. All analyses were performed by using Stata/IC, version 13.1 (StataCorp). We assessed the aggregate internal validity (quality) of the body of evidence for each key question as good, fair, or poor by using methods developed by the USPSTF that are based on the number, quality, and size of studies; consistency of results between studies; and directness of evidence (10, 11). Role of the Funding Source This research was funded by the Agency for Healthcare Research and Quality (AHRQ) under a contract to support the work of the USPSTF. The investigators worked with USPSTF members and AHRQ staff to develop and refine the scope, analytic framework, and key questions; resolve issues during the project; and finalize the report. The AHRQ had no role in study selection, quality assessment, synthesis, or development of conclusions. The AHRQ provided project oversight; reviewed the draft report; and distributed the draft for peer review, including to representatives of professional societies and federal agencies. The AHRQ performed a final review of the manuscript to ensure that the