20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Received: 2 October 2020 | Revised: 12 January 2021 | Accepted: 8 February 2021 DOI: 10.1002/cam4.3803 REVIEW Digital breast tomosynthesis compared to diagnostic mammographic projections (including magnification) among women recalled at screening mammography: a systematic review for the European Commission Initiative on Breast Cancer (ECIBC) Carlos Canelo-­Aybar1,2 | Lourdes Carrera3 | Jessica Beltrán2 | Margarita Posso2,4 | David Rigau2 | Annette Lebeau5 | Axel Gräwingholt6 | Xavier Castells4 | Miranda Langendam7 | Elsa Pérez8 | Paolo Giorgi Rossi9 | Ruben Van Engen10 | Elena Parmelli11 | Zuleika Saz-­Parkinson11 | Pablo Alonso-­Coello1,2 1 CIBER de Epidemiología y Salud Pública (CIBERESP), Madrid, Spain 2 Department of Clinical Epidemiology and Public Health, Biomedical Research Institute Sant Pau (IIB Sant Pau), Barcelona, Spain 3 Universidad Nacional Mayor de San Marcos, Lima, Perú 4 Department of Epidemiology and Evaluation, Hospital del Mar Medical Research Institute (IMIM), Barcelona, Spain 5 Institute of Pathology, University Medical Center Hamburg-­Eppendorf, Hamburg, Germany 6 Radiologie am Theater, Paderborn, Germany 7 Department of Clinical Epidemiology, Biostatistics and Bioinformatics, Amsterdam UMC, University of Amsterdam, Amsterdam Public Health Institute, Amsterdam, The Netherlands 8 University Hospital Dr. Josep Trueta, Girona, Spain 9 Epidemiology Unit, Azienda USL –­IRCCS di Reggio Emilia, Reggio Emilia, Italy 10 LRCB, Dutch Expert Centre for Screening, Nijmegen, The Netherlands 11 European Commission, Joint Research Centre (JRC), Ispra, Italy Correspondence Carlos Canelo-­Aybar, Iberoamerican Abstract Cochrane Centre -­Department of Background: Diagnostic mammography projections (DxMM) have been tradition- Clinical Epidemiology and Public Health, ally used in the assessment of women recalled after a suspicious screening mam- Biomedical Research Institute Sant Pau (IIB Sant Pau), Sant Antonio María Claret mogram. Digital breast tomosynthesis (DBT) reduces the tissue overlap effect, thus 167, 08025 Barcelona, Spain. improving image assessment. Some studies have suggested DBT might replace Email: ccanelo@santpau.cat DxMM with at least equivalent performance. Zuleika Saz-­Parkinson, European Objective: To evaluate the replacement of DxMM with DBT in women recalled at Commission, Joint Research Centre (JRC), Via E. Fermi, 2749. TP127, screening. I-­21027 Ispra (VA), Italy. Methods: We searched PubMed, EMBASE, and the Cochrane Library databases to Email: Zuleika.SAZ-PARKINSON@ identify diagnostic paired cohort studies or RCTs comparing DBT vs DxMM, pub- ec.europa.eu lished in English that: reported accuracy outcomes, recruited women recalled for as- sessment at mammography screening, and included a reference standard. Subgroup This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2021 The Authors. Cancer Medicine published by John Wiley & Sons Ltd. Cancer Medicine. 2021;10:2191–2204.  wileyonlinelibrary.com/journal/cam4 | 2191 | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2192    CANELO-­AYBAR et al. Funding information The systematic review was carried out by analysis was performed over lesion characteristics. We provided pooled accuracy es- Iberoamerican Cochrane Collaboration timates and differences between tests using a quadrivariate model. We assessed the under the Framework contract 443094 certainty of the evidence using the GRADE approach. for procurement of services between the European Commission's Joint Research Results: We included ten studies that reported specificity and sensitivity. One study Centre and Asociación Colaboración included 7060 women while the remaining included between 52 and 738 women. Cochrane Iberoamericana. DBT compared with DxMM showed a pooled difference for the sensitivity of 2% (95% CI 1%–­3%) and a pooled difference for the specificity of 6% (95%CI 2%–­11%). Restricting the analysis to the six studies that included women with microcalcification lesions gave similar results. In the context of a prevalence of 21% of breast cancer (BC) in recalled women, DBT probably detects 4 (95% CI 2–­6) more BC cases and has 47 (95%CI 16–­87) fewer false-­positive results per 1000 assessments. The cer- tainty of the evidence was moderate due to risk of bias. Conclusion: The evidence in the assessment of screen-­recalled findings with DBT is sparse and of moderate certainty. DBT probably has higher sensitivity and speci- ficity than DxMM. Women, health care providers and policymakers might value as relevant the reduction of false-­positive results and related fewer invasive diagnostic procedures with DBT, without missing BC cases. KEYWORDS breast neoplasms, digital breast tomosyntheses, mass screening, practice guidelines, systematic review 1 | IN T RO D U C T IO N is usually returned to a new screening round. Recall for fur- ther invasive assessment (ie, biopsy) leads to additional costs Breast cancer (BC) is the second most prevalent cancer in the and anxiety, not only during the diagnostic work-­up but also world and the most frequent among women. In the European through subsequent screening mammogram rounds (despite Union, 404,920 women were diagnosed with breast can- a negative result).8 cer in 2018.1 BC mortality has decreased over the last de- Digital breast tomosynthesis (DBT) is a mammographic cades, partially due to the implementation of mammography technique that acquires low-­dose projection images of the screening programs,2,3 which are recognized as an effective breast at different angles utilizing a moving X-­ray source. A method to detect early-­stage breast cancers.4 Consequently, stack of thin slices is reconstructed, overcoming the influ- the European Breast Guidelines on Screening and Diagnosis ence of overlapping breast tissue.9 DBT improves the visu- recommends mammography screening for asymptomatic alization of BC and thus may enhance the interpretation of women aged 50–­69, with an average risk of BC, and sug- mammography. Several studies suggest that DBT may be a gests mammography screening for women aged 45–­49 and promising technique as a screening modality.10 In the “fur- 70–­74.5,6 ther assessment” setting, however, the evidence is scarce. Li When an abnormality is found at mammography screen- et al. reviewed the evidence of DBT compared to conven- ing, women are recalled for assessment, which might tional imaging in the assessment of screen-­recalled findings, mean they need to undergo additional imaging testing. reporting limited evidence for a higher specificity of DBT.11 Mammography for assessment of suspicious abnormalities In 2015, the European Commission Initiative on Breast also referred to as “diagnostic mammography projections,” Cancer (ECIBC) was launched to develop the European usually consists of additional mammographic views (ie, spot Guidelines on Breast Cancer Screening and Diagnosis.5 This compression, Cleopatra view, cleavage view, among others) systematic review informed the recommendations on the use or magnification if two-­view mammography from the pre- of digital breast tomosynthesis (DBT) compared to diagnos- vious screening examination is available. During this imag- tic mammography projections (DxMM), for the assessment ing assessment, the suspicious finding from mammography of suspicious abnormalities in women recalled for further as- screening can be either confirmed, in a minority of women sessment at mammography screening in average risk asymp- who are then referred to an invasive assessment to obtain a tomatic women. During the guideline´s development process,5 sample of tissue or cells,7 or not confirmed, and the woman the Guidelines Development Group (GDG) made detailed | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License CANELO-­AYBAR et al.    2193 considerations on the evidence to decide about the balance be- This diagnostic question was framed as a replacement tween desirable and undesirable effects of the interventions to scenario among women of average risk of BC who have any issue a recommendation. The GDG also considered other crite- abnormality at mammography screening that leads to a re- ria such as values and preferences, equity, acceptability, and fea- call for assessment. These women are usually examined with sibility while upholding independence of commercial, private, further diagnostic mammography views (including magnifi- and national interests. We encourage readers to refer to these cations), but these may be replaced with DBT images. Both detailed considerations in the published recommendations on index tests (DBT or DxMM) can be performed alongside with the ECIBC website (https://healt​hcare​-­quali​ty.jrc.ec.europa.eu/ other exams if required (ie, clinical examination, ultrasound). europ​ean-­breas​t-­cance​r-­guide​lines/​diagn​osis/DBT) 2.3 | Eligibility criteria 2 | M ET H O D S Studies were included if: (1) they compared the accuracy of 2.1 | Guidelines development group (GDG) DBT images (one or two views), including either synthesized mammography or the previous screening mammography that The European Commission selected, via an open call, a panel triggered the assessment, with diagnostic mammography with a broad representation of different expertise, knowl- projections or magnification; (2) participants were women of edge, and background, that includes patients, healthcare average risk for BC with abnormal findings at mammogra- professionals, epidemiologists, guideline methodologists, phy screening; (3) there was a minimum sample size of 30 and others (the complete list of experts is available from the participants; (4) the study provided enough data to construct ECIBC website) who voluntarily participate in the ECIBC. a two by two contingency table; (5) all participants either re- ceived both DBT and diagnostic mammography projections (or magnification) or were randomly allocated to the index 2.2 | Structured question and outcome comparison tests; (6) a histopathological examination and/or prioritization imaging follow-­up for at least one year was used as a refer- ence standard. The clinical question “Should digital breast tomosynthesis Included cohort studies could either prospectively recruit (DBT) vs. diagnostic mammography projections (including and assess women recalled at screening or perform a blind magnification) be used in the assessment of recalled women re-­interpretation of the two index tests from retrospective of average risk of breast cancer due to suspicious lesions at case series that had both tests recorded during the assess- mammography screening?” was prioritized by the GDG, and ment. Both index tests could have been interpreted with or here, we describe the test accuracy assessment. The ques- without the additional help of breast ultrasound or clinical tion was structured following a diagnostic framework format examination. The following exclusion criteria were applied: of Population, Index test, Comparator (standard index test), (1) studies reporting DBT performed as a screening test in Target condition, and Reference standard (Box 1). the general population; (2) diagnostic studies of case-­control BOX 1 Clinical question definition for assessment of accuracy between two tests* Population Index test Comparison test Reference standard Target condition Women of average risk DBT images (one Diagnostic Histopathological exam Breast cancer lesions for breast cancer with or two views), mammography and/or imaging (DCIS or invasive) abnormal findings including either projections or follow-­up for at least at mammography synthesized magnification one year screening who are mammography recalled for further or the previous imaging assessment screening mammography that triggered the assessment DBT, digital breast tomosynthesis; DCIS, ductal carcinoma in situ.* Both tests could have been interpreted with or without the additional help of breast ultrasound or clinical examination. | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2194    CANELO-­AYBAR et al. design or non-­paired cohorts studies; (3) abstracts or con- with one index test and the other was less than 1 month; 4) if ference communications not published as complete articles, there were no unbalanced number of additional examinations and; (4) studies published in a language other than English. performed together with the index tests; (5) the reference stan- All citations retrieved were imported into a bibliographic dard was the same for both index tests, even if studies might reference software (EndNote X5; Thomson Reuters) to dis- have used an incomplete verification (ie, biopsies for positive card duplicates, and record screening decisions. Initially, at results but clinical follow-­up for negative ones). the title and abstract level, two previously calibrated review- ers (CCA and LC) assessed eligibility. In a second step, two reviewers independently reviewed the full text of all selected 2.6 | Data analysis references. Discrepancies were solved either by discussion or with the help of a third reviewer (DR). We classified the index test results as positive or negative ac- cording to the reporting system used in each study. When we were able to reconstruct the 2 by 2 contingency table based on 2.4 | Data sources and searches disaggregated data, we considered undetermined results as pos- itive, as they imply recalling the women for further assessment We searched MEDLINE (via PubMed, May 2018), (ie, BI-­RADS 0). If results were provided for multiple readers EMBASE (via Ovid, May 2018), and CENTRAL (via The within a study, readers´ data were averaged before analysis. Cochrane Library, May 2018) databases using predefined al- To perform a meta-­analysis of diagnostic tests, the current gorithms for individual studies. We adapted the search terms recommended method is the bivariate random-­effects model. to each database and used validated filters to retrieve appro- This model takes into account the correlation between sensi- priate designs. Additionally, we updated our initial search tivity and specificity at the study level, as well as underlying in MEDLINE (via PubMed) and EMBASE (via Ovid) in variability across studies due to difference in the (implicit) February 2020 (Table S1: Search strategy). We also reviewed thresholds used to interpret index test results or to the stud- lists of references of the included studies, and members of the ies’ designs.13 In this review, we implemented a quadrivariate GDG were consulted about potential missing studies. generalized linear mixed random model (GLMM). This is an extension of the bivariate model described before, to jointly account for the sensitivities and specificities parameters of 2.5 | Data extraction and risk of two diagnostic tests (and the correlation between them), com- bias assessment pared to a common reference standard.14 We present the hierarchical summary receiver operating Two reviewers (LC, CCA) independently extracted data and characteristic curves (HSROC) using the model parameters assessed risk of bias from the included studies. We collected from the bivariate model for each index test with the “madas” the following information from each study: accuracy results package in RStudio 3.5.1. We used the graphical functions of (TP, TN, FP, and FN), the total number of participants, coun- the metafor package in RStudio to display the forest plots for try, year of publication, first screening modalities, type of sensitivity and specificity, as well as their differences. We did suspicious lesions that triggered the recall for assessment (ie, not statistically evaluate publication bias as there is currently masses, architectural distortions, asymmetry, calcifications), no accepted method for comparative test analysis. participants´ mean age, study design, characteristics of the We assessed heterogeneity by visual inspection of forest index tests at assessment (ie, the number of views, type of addi- plots, as statistical approaches (ie, I-­squared percentage) may tional views, use of additional tests), classification system, and overestimate the variability across studies and do not con- cutoff used to interpret the index test´s results (ie, BI-­RADS), sider the clinical relevance of the results. Predefined sub- and type of reference standard used. Discrepancies were solved group analysis included: risk of bias (low versus high), type either by discussion or with the help of a third reviewer (DR). of further mammographic assessment (magnification vs ad- We assessed the risk of bias of the included studies using ditional views), use of additional ultrasound in either index the Quality Assessment of Diagnostic Accuracy Studies test, the extension of microcalcification, and type of system (QUADAS-­2) tool12 which includes the following four do- used to classify the index test results (BI-­RADS vs others). mains: patient selection, index test, reference standard, flow, All statistical analyses were performed in SAS University and timing. To adapt the QUADAS-­2 tool to the comparison of Edition, using the PROC GLIMMIX with a penalized two index tests, we also assessed whether: (1) a random alloca- quasi-­ likelihood (PQL) method, a logit link, and the tion of participants to each index test or paired comparison was Newton-­Raphson Ridge Optimization technique. Due to the implemented; (2) radiologists reading any of the index tests non-­convergence of the quadrivariate model with sparse data, were blinded to the results of the alternative test or the refer- we only performed a sensitivity analysis without the studies ence standard; (3) the timing elapsed between the examination that explicitly excluded women with calcified lesions. | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License CANELO-­AYBAR et al.    2195 2.7 | Certainty of the evidence that either: included a different population (symptomatic women, screening population, or invasive BC under staging We rated the overall certainty of the evidence as high, mod- assessment), reported a different comparison to diagnostic erate, low or very low according to the GRADE Working mammography (eg, ultrasound), or used DBT in the screen- group´s guidance for the assessment of accuracy studies ing setting (see supplementary file Table S3). In total, we in- which includes the following domains: risk of bias, impreci- cluded 10 primary accuracy studies,9,16-­24 of which nine were sion, inconsistency, indirectness, and publication bias.15 We designed as prospective cohorts9,17-­24, and one was a blinded considered a comparative accuracy approach to rate the cer- re-­ assessment of retrospective case series imaging from tainty of evidence (ie, the heterogeneity was not assessed for several breast screening centers.16 Finally, after the update a single index test but relative to the comparator index test). search in February 2020, we did not identify any additional studies that fulfilled our eligibility criteria (see supplemen- tary file Table S3). 3 | R E S U LTS 3.1 | Search result 3.2 | Study characteristics A total of 5,978 unique citations were initially retrieved. We The included studies were conducted in the United States excluded 757 duplicate records together with 5,179 citations of America, Italy, Germany, Switzerland, and the United based on title or abstract review and selected 42 for a detailed Kingdom. Five studies did not describe the age of the in- appraisal of the full text (Figure 1). We excluded 32 studies cluded women,9,20,22-­24 The mean age of women was between Records identified through multiple databases searching [May 2018] (n = 5,978) Identification Duplicated records removed (n = 757) Records screened (n = 5,221) Records excluded Screening (n = 5,179) Full-text articles assessed for eligibility (n = 42) Full-text articles excluded (n = 32): Screening seng: 5 Different popula on: 18 Update search in Systema c reviews: 2 Medline/Embase from Eligibility May 2018 to February Different interven on / 2020 comparison: 6 (n = 928) Duplicate analysis: 1 Studies included up to April 2018 (n = 10) Studies included from references surveillance (n = 0) Included Total studies included in the review FIGURE 1 Flow chart for the evidence (n = 10) of effects. | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2196    CANELO-­AYBAR et al. 51 and 58 in three studies,16,19,21 and the age range was be- a much higher sample size (n = 7060),16 while the number of tween 50 and 69 in another one.18 One study stated that only women recalled for assessment in the remaining studies was women older than 40 were included.20 Women were recalled between 5219 and 7389 (Table 1). for assessment after an abnormal finding during screening All studies enrolled women recalled from screening mam- with digital mammography in five studies and film mam- mography with abnormalities such as masses, areas of distor- mography in one study, while in four studies the type of tion, asymmetries, or other non-­calcified lesions.9,16-­24 Five screening mammography was not described. One study had studies also included women with calcified lesions,9,16,20,22 TABLE 1 Characteristics of the included studies in the systematic review Number of Age Author (year), country patients Mean (range) Type of lesions Digital breast tomosynthesis 17 Brandt (2013), US 146 >40 -­Possible masses 20% Two-­view DBT -­Areas of distortion 18% Ultrasound in selected cases -­Asymmetries 63% -­Calcifications: excluded Cornford (2016),20 UK 324 ND -­Masses (50%) Two views DBT -­Distortions (13.5%) -­Asymmetric densities (36.5%) -­Microcalcification (3.8%) Gilbert (2015),16 UK 7,060 56 (29-­85) -­Mass (69%) Two views DBT (plus synthesized -­Microcalcifications (13%) mammography) -­Distortion or ASD (18%) Heywangkbrunner 284 patients with 50–­69 years Screen-­detected abnormalities Single view with a wide-­angle DBT (2017)a ,18 Germany 311 lesions that needed further view CC projection for lesions mammographic assessment. not visible in one view Heywangkbrunner (2018),24 241 ND Cases with microcalcifications Single view with a wide-­angle DBT Germany were excluded view CC projection for lesions not visible in one view Ultrasound in selected cases Michell (2012),9 UK 738 ND Mammographic abnormality: Two views circumscribed, spiculate, microcalcifications (21%), parenchymal distortion, asymmetry. Poplack (2007) a ,23 US 100 ND Abnormal digital screening Up to 3 projections resembling the mammogram interpreted mammography exam prospectively Tagliafico (2012),19 Italy 52 51 ± 12 Women with BI-­RADS 0 lesions Single view (calcifications were excluded) Waldherr (2013),22 66 ND Abnormalities at the screening Single view Switzerland included: masses, microcalcifications Whelehan (2017),21 238 57.6 (50-­69) Microcalcifications were Single view Germany excluded AV, additional views; ND, not described; DBT, digital breast tomosynthesis; BI-­RADS, breast imaging reporting and data system; UK-­RCS, UK royal College of Radiologist score. a Data were categorized as positive (BI-­RADS 0, 3-­5) and as negative BI-­RADS (1 and 2). | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License CANELO-­AYBAR et al.    2197 but two of them did not report the exact proportion of patients included additional or supplementary views such as digi- with calcifications,18,22 and another one had a low proportion tal spot compression views, rolled view, lateral views with (3.8%) because women with calcifications as the predomi- or without magnification. One study did not describe what nant lesion were not invited to participate in the study.20 type of supplementary views was used.22 Ultrasound was Five studies used a two-­ view DBT,9,16,17,20,23 while also used in five studies either to assess selected cases at the other five used a single view DBT during the assess- the clinician discretion or in the assessment of all cases ment.18,19,21,22,24 Diagnostic mammography projections included.17,18,20,23,24 Diagnostic mammography projections Index test threshold Number of radiologists and experience Reference test May include digital spot Positive: BI-­RADS 4, 5 DBT: three independent readers* with Positive lesions: biopsy-­proven compression (with or Negative: BI-­RADS 1 -­3 8 hours-­training cancers or high-­risk abnormality without magnification), 2DM: ND Negative: biopsy-­proven benign, or rolled views, lateral no BC at follow-­up views, exaggerated views Ultrasound in selected cases May include extended CC Positive: UK-­RCS M3-­M5 8 specialist Brest radiologists. (7-­24 Biopsy lesion histopathology. views, lateral projections Negative: UK-­RCS M1-­M2 years of experience) or magnification views DBT and 2DM: 1-­day training in interpretation of DBT images (blinded) May include magnification Positive: “reader decided to Average of 10 years of experience Biopsy lesion histopathology and recall” (range 3-­25) imaging re-­assessment before Negative: “reader decided discharge not to recall” May include 1 or 2 Positive: BI-­RADS 0, 3-­5 Consensus reading BIRAD 4 OR 5 histopathological mammographic additional Negative: BI-­RADS 1,2 Reader 1 (the principal investigator) proof views Reader 2 experience with DBT on > and 2-­year follow-­up Ultrasound in selected cases 200 proven cases. May include additional views Positive: BI-­RADS 0, 3-­5 Blinded reading Needle biopsy Histology and Ultrasound in selected cases Negative: BI-­RADS 1, 2 follow-­up > 2 years May include magnification, Positive: UK-­RCS M3-­M5 One specialist breast radiologist Positive: final surgical histology. black/with inversion (for Negative: UK-­RCS M1-­M2 Negative: benign findings in needle both exams) biopsy May include focal Results reported by Seven mammography technologists Core needle biopsy was performed compression (film BI-­RADS categories trained to perform tomosynthesis with stereotactic mammography) examinations before the study. guidance Ultrasound Spot compression: Positive: BI-­RADS 3-­5 Blinded Ultrasound + fine needle aspiration Negative: BI-­RADS 1, 2 two experienced breast radiologists and core biopsy + follow > 12 with 20 and 7 years of experience. months ND Positive: BI-­RADS 4-­5 Blinded reading Histopathology + follow -­up for 12–­ Negative: BI-­RADS 1-­3, 0 16 months Supplementary views Positive: biopsy requested 2DM: 10.9 mean years of experience. Histopathology + follow -­up for two Negative: biopsy not DBT: 1.5 mean years of experience years requested | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2198    CANELO-­AYBAR et al. F I G U R E 2 Forest plots for the difference in sensitivity comparing digital breast tomosynthesis vs. diagnostic mammography in recalled women. In six studies, the screening mammography images used for recall were available during the DBT assessment.9,17,18,20,21,24 T A B L E 2 Pooled estimates (quadrivariate model) for sensitivity, specificity, and the difference between the index tests. In three studies the readers either based their assessment on DBT alone or did not clearly describe if they had access to the Diagnostic previous mammography images.19,22,23 One study interpreted Digital breast mammography the DBT images together with synthesized mammography,16 tomosynthesis projections Population (N which was compared to a DxMM assessment that included of studies) 95% CI 95% CI magnification views if women had microcalcifications. 9,16-­24 All studies (n = 10) Regarding the classification system used to interpret the DBT and mammography results, six studies used the breast Sensitivity 0.94 0.91–­0.97 0.92 0.89–­0.95 imaging reporting and data system (BI-­RADS). Three of Difference: 0.02 (95% CI 0.01–­0.03) these defined positive result as BI-­RADS of 3 to 5 (two of Specificity 0.74 0.69–­0.78 0.67 0.62–­0.73 them including BI-­RADS 0).18,19,24 Two studies considered Difference: 0.06 (95% CI 0.02–­0.11) positive results as BI-­RADS 4 to 5,17,22 and one reported Studies including calcified lesionsa (n = 6)9,16,18,20,22,23 the results according to each BI-­RADS category.23 The UK Sensitivity 0.94 0.90–­0.97 0.89 0.84–­0.93 Royal College of Radiologist score (UK-­RCS) was used in Difference: 0.05 (95% CI 0.01–­0.09) two studies and categorized results as positive if classified Specificity 0.71 0.66–­0.75 0.60 0.54–­0.65 between UK-­RCS M3 to M5.9,20 In the remaining two stud- Difference: 0.12 (95% CI 0.06–­0.17) ies, a result was considered positive if recall or biopsy was Calcified lesions onlyb (n = 1)16 requested by the imaging assessor.16,21 The reference standard also differed across studies but not Sensitivity 0.85 0.81–­0.89 0.88 0.84–­0.92 between the index tests. Six studies included a clinical or im- Difference: ‒­0.03 (95% CI −0.09 to −0.03) aging follow-­up for negative results at assessment17-­19,21,22,24 Specificity 0.44 0.40–­0.48 0.31 0.28–­0.34 which lasted between 12 months19 to more than two years18. Difference: 0.13 (95% CI 0.08–­0.18) In the remaining four studies, results were confirmed based n = number of studies included in the analysis. only on histopathological results9,20,23 or together with re-­ a Studies that included women with or without calcified lesions. assessment of the breast images before discharge.16 b Subgroup analysis of women with pure calcified lesions. The estimate was calculated using a standard difference of proportions formula for a single comparison. 3.3 | Sensitivity difference was 2% (95% CI 1–­3%) in favor of DBT (p < 0.001) A total of 1,592 cases of BC lesions were included in the (Table 2). For a BC prevalence of 21% in women recalled for analysis. The sensitivity difference between DBT and DxMM assessment, DBT probably identifies 4 more (95% from 2 to ranged from −25%23 to 17% across studies, while six studies 6 more) true BC cases per 1000 assessments (Table 3). showed a probably higher sensitivity for DBT. The visual assessment of the forest plots for each test independently did not show a relevant heterogeneity for either DBT or DxMM 3.4 | Specificity (Figure S1). A relatively homogenous pattern was also ob- served when the sensitivity difference was plotted (Figure 2). A total of 6096 negative cases were included in the analysis. The pooled sensitivity was 94% (95% CI 91–­97%) for The specificity difference between DBT and DxMM ranged DBT, and 92% (95% CI 89–­95%) for DxMM. The pooled from −10% to 23% across studies, while seven studies showed | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License CANELO-­AYBAR et al.    2199 T A B L E 3 Summary of findings (GRADE) for digital breast tomosynthesis versus diagnostic mammography. *Pooled sensitivity digital breast tomosynthesis: 0.94 (95% CI: 0.91–­0.97) | Pooled specificity digital breast tomosynthesis: 0.74 (95% CI: 0.69–­0.78). Pooled sensitivity assessment mammography: 0.92 (95% CI: 0.89–­0.95) | Pooled specificity assessment mammography: 0.67 (95% CI: 0.62–­0.73). Pooled sensitivity difference: 0.02 (95% CI: 0.01–­0.03) | Pooled specificity difference: 0.06 (95% CI: 0.02–­0.11)# Number of results per 1000 patients tested (95% CI) Prevalence 21% Number of Typically seen in European screening programs participants Certainty of the Test result Digital breast tomosynthesis Diagnostic mammography (studies) Evidence (GRADE) True positives 197 (191–­204) 193 (187–­199) 1584 ⨁⨁⨁◯ (10)9,16-­24, a MODERATE b,c,d,e,f,h 4 more (from 2 more to 6 more) TP in digital breast tomosynthesis False negatives 13 (6–­19) 17 (11–­23) 4 fewer (from 2 fewer to 6 fewer) FN in digital breast tomosynthesis True negatives 585 (545–­616) 529 (490–­577) 6096 ⨁⨁⨁◯ (10) 9,16-­24, a MODERATE b,e,f,g,h 47 more (from 16 more to 87 more) TN in digital breast tomosynthesis# False positives 205 (174–­245) 261 (213–­300) 47 fewer (from 16 fewer to 87 fewer) FP in digital breast tomosynthesis# Abbreviations: CI, confidence interval; TP, true positives; FN, false negative; TN, true negative; FP, false positive. a The absolute differences are the additional cases identified or missed with digital breast tomosynthesis compared to diagnostic mammographic views, among those women, recalled at the screening mammography assessment. b In some of the included studies, there was a non-­blinded reading of the index tests. There was variability in how the evaluations were performed; in some cases, they included additional tests such as ultrasound. Those additional exams could be requested at clinical discretion and therefore could be a potential source of differential misclassification in the test accuracy estimates. c There was no important unexplained heterogeneity in the sensitivity difference between DBT and DxMM d One study (Poplack 2012) showed largely inconsistent results. However, a sensitivity analysis excluding this study did not show relevant differences in the pooled estimate. e One study (Gilbert 2015), performed a retrospective analysis with DBT plus synthesized two-­view mammography as the intervention index test. In the remaining studies, the intervention index test was DBT plus screening mammography. f Publication bias was not suspected by the expert panel. g There was no important unexplained heterogeneity in the specificity difference between DBT and DxMM. h Observed variability might be explained using different thresholds to define positivity, use of diverse additional imaging tests, different reference standards and the imprecision of the estimates. The panel decided not to downgrade the certainty of evidence due to this variability. *The SoF table have been adapted ad hoc for a two index test comparison. #Due to rounding the estimated specificity difference might differ from the manual subtraction between each test´s estimates. probably higher specificity for DBT. The visual assessment of 0.913 and 0.895 for DBT and DxMM, respectively of the forest plots for each test independently showed a rel- (Figure 4), with overlapping confidence regions, which evant heterogeneity for both DBT and DxMM (Figure S2). were wider for DxMM. However, the specificity difference between tests showed a more homogenous pattern (Figure 3). The pooled specificity was 74% (95% CI 69–­78%) for 3.6 | Subgroup and sensitivity analysis DBT and 67% (95% CI 62–­73%) for DxMM. The pooled difference was 6% (95% CI 2–­11%; p = 0.001) in favor of Due to non-­convergence of the quadrivariate model with DBT (Table 2). Considering the same 21% prevalence of BC sparse data we only performed a sensitivity analysis for lesions used for sensitivity, DBT probably reduces the false-­ microcalcification lesions. Excluding the studies that positive results by 47 (from 16 to 87 fewer) cases per 1000 did not include women with calcified lesions at screen- women assessed (Table 3). ing mammography, we found a pooled difference of 5% (95% CI 1–­9%) and 11% (95% CI 6–­17%) for sensitivity and specificity, respectively (Table 2). Additionally, we 3.5 | HSROC curve did a sensitivity analysis excluding the study by Poplack et al., which had inconsistent results compared to the The larger the area under the curve (AUC) for the HSROC, other included studies, but this did not change the original the better the diagnostic performance. We found an AUC estimates. | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2200    CANELO-­AYBAR et al. F I G U R E 3 Forest plots for the difference in specificity comparing digital breast tomosynthesis vs. diagnostic mammography in recalled women. F I G U R E 4 Hierarchical summary receiver operating characteristic (HSROC) curves for each test modality. Two studies reported accuracy estimates for patients with follow-­up for negative results, but these were similar in both only calcified lesions. Michel et al, reported no significant index tests (Figure 5). differences among tests, with an area under the ROC curve We considered that risk of bias due to these aspects was of 79% (95% CI 73–­86%) for DBT and of 78% (95% CI 72–­ serious and therefore judged the certainty of the evidence as 85%) for DxMM.9 Gilbert et al., in a sample of 1027 cases moderate. We observed some heterogeneity in the sensitiv- with calcified lesions showed a sensitivity difference of −3% ity and specificity differences but it was probably explained (95% CI −9% to 3%) and a specificity difference of 13% by the use of different thresholds to define positive results (95%CI 8–­18%).16 across studies and, to a lesser extent, due to the diversity of additional imaging tests or reference standards used (Table 3; Table S2). 3.7 | Risk of bias and certainty of the evidence 4 | DISCUSSION In all the included studies there was a high or unclear risk of bias regarding the index test comparisons due to: (1) not 4.1 | Main findings clearly implementing a protocol to blind the assessment, the index tests or the assessors to the results of the histological Our systematic review found that in women of average risk or clinical exams,9,18,23 and (2) an imbalance in the experi- of BC with suspicious lesions at mammography screening, ence of the reader of the index test results, as most of them DBT probably has a higher specificity difference (6%) than had little training in the assessment of DBT images (ie, less diagnostic views of 2D mammography in women recalled for than 1-­year experience or attending 1-­day training / 80 cases assessment (moderate certainty evidence). The sensitivity is before the study)16-­18,20,21,24 or did not report the experience also probably higher with DBT though in this case the dif- of DBT readers.9,19,22,23 ference was not so large (2%). The clinical impact of these Four studies inappropriately excluded participants with results in a screening program is a probable reduction of 47 calcified lesions, something that could improve the accuracy false-­positive results, including related invasive procedures, of the index tests but of an unclear impact in the comparison. and most likely a small increase of 4 additional true BC le- There were other limitations in the study designs, such as not sions detected per 1000 women recalled for assessment. Our having a prespecified threshold9,21 or not including a clinical results were consistent in women with calcified lesions, but | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License CANELO-­AYBAR et al.    2201 FIGURE 5 QUADAS-­2 risk of bias assessment. with a larger imprecision in the sensitivity differences. These women and did not distinguish between the screening and results were taken into account by the GDG, together with diagnostic setting in the analysis.26 other considerations (ie, feasibility) to issue a conditional In contrast to the screening setting, the body of evidence recommendation in favor of DBT, available on the ECIBC on DBT in the assessment of screen-­recalled findings is website (https://healt​hcare​-­quali​ty.jrc.ec.europa.eu/europ​ scarcer. Li et al. published a systematic review in the diagnos- ean-­breas​t-­cance​r-­guide​lines/​diagn​osis/DBT) tic setting, which includes six studies suggesting that DBT may improve specificity but have no effect on sensitivity;11 as well as one of the included studies showing that DBT may 4.2 | Our results in the context of reduce the frequency of ultrasound requests at assessment.17 previous research Our findings are consistent with the observations from Li et al11 regarding specificity, however, we found a greater sen- Previous studies assessing the role of DBT have focused sitivity for DBT. Additionally, we performed a more up-­to-­ on the screening setting. A prospective study from Europe date search identifying five more studies and excluded one reported a 40% improvement in the detection of invasive that did not compare DBT to additional views of DM but cancers and a 15% reduction in the number of false-­positive instead simulated the application of DBT as a triage tool.27 results.25 A systematic review in the screening setting for It is worth noting that our findings are consistent with a asymptomatic women, included 17 studies and found that to- recent study suggesting the use of DBT as an add-­on test, mosynthesis improves BC detection rate and reduces recall, although this study used DBT in a different diagnostic path- with greater improvement in detection rate among European way.28 Sharma et al. compared triple assessment (further 2D studies.10 Alabousi et al published a broader systematic re- mammography views, clinical examination, and ultrasound) view that identified 38 studies and reported that DBT alone versus triple assessment plus DBT among women recalled or combined with DM had a higher sensitivity than DM at screening within the UK National Health Service Breast alone (and also a higher specificity though smaller in magni- Screening Programme.28 The addition of DBT resulted in tude). However, the authors included studies in symptomatic similar sensitivity but higher specificity (38.2% vs 77.5%), | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2202    CANELO-­AYBAR et al. which translated into a reduction in the number of biopsies the evidence, considering the comparative framework of our from 571 of 827 (69.0%) to 298 of 827 (36.0%).28 question of interest. In most of the included studies in our review, the readers interpreted the DBT images having the screening mammog- raphy available to them. One study used synthetic mam- 4.4 | Implications for practice and research mography (SM) in combination with DBT showing results consistent with other studies. This may suggest a potential Our findings may have different implications for practice de- use of SM in the assessment of abnormalities at recall.16 In pending on the stakeholder group. In the case of women, it a different population, Mumin et al, compared the accuracy would depend on how they value the balance between potential of SM+DBT to that of DM+DBT, in women referred for as- benefits and harms derived from the replacement of diagnostic sessment due to the presence of symptoms, finding a high mammography projections with DBT in the assessment setting. agreement between the use of DM or SM in combination Thus, in a society where women are more concerned about with DBT.29 false-­positive results or invasive procedures derived from false-­ positive results, in the context of a shared decision-­making pro- cess, women might be more in favor of DBT. In the case of 4.3 | Limitations and strengths guideline panels and policymakers, they might consider other aspects such as the use of resources or feasibility issues around Our systematic review has some limitations. We included implementing this technology which could influence their deci- only articles in English, but the risk of selection bias is prob- sion to formulate a strong or conditional recommendation. ably small because we screened previous systematic reviews, The available body of evidence for the use of DBT was of and the GDG includes several international experts in the moderate certainty due to risk of bias, thus there is a need to field. Most studies had limitations: (1) In many studies the produce further evidence from implementation data of this radiologist readers had lower experience in DBT compared technology in the assessment context. Among the research to DM, (2) the work-­up in the included studies was not lim- priorities identified during this review, with input from the ited to mammography or DBT but also used other imaging GDG experts, are the following: i) given that ultrasound is modalities (ie, breast ultrasound), which might have hidden often included in the management of assessment after a pos- true differences between the index tests, and (3) some studies itive finding in screening mammography, further research did not implement explicit blinding of readers during the as- should be conducted exploring which subgroups could avoid sessment and used imperfect reference standards that might ultrasound after DBT-­additional projections, as well as which include or not clinical follow-­up. lesions (ie, masses) could be assessed with ultrasound instead Additionally, the included studies did not use the same of additional mammography projections or DBT; ii) Use of thresholds to interpret the results of the index tests or did DBT in women with high mammographic breast density, not clearly define the threshold, referring only to the standard particularly for the assessment of abnormalities at screening clinical practice. However, we considered that it was appro- examination, and iii) whether or not to use one or two views priate to provide pooled estimates because (1) the reported for tomosynthesis in the assessment. definitions were to some degree clinically equivalent (ie, UK-­ MRS 3 to 5 vs BI-­RADS 3 to 5) and (2) our interest was ACKNOWLEDGMENTS not to determine the accuracy of each testing strategy but to The authors would like to sincerely thank all members assess the difference between them, and all studies included of the Guidelines Development Group of the European a paired design. Commission Initiative on Breast Cancer for their participa- Our review has several strengths. The review was devel- tion in the discussions generated by this systematic review oped as part of a clinical guideline development process and which led to the different recommendations they developed the GDG also evaluated DBT as a primary screening test in the European Guidelines on Breast Cancer Screening and guaranteeing a broader view of the evidence. The clinical diagnosis (https://healt​hcare​-­quali​ty.jrc.ec.europa.eu/europ​ question was framed considering the complete screening and ean-­breas​t-­cance​r-­guide​lines). diagnostic pathway, thus placing it in context with the whole work-­up process and the clinical utility of the results. We in- CONFLICT OF INTEREST cluded studies with a paired design to directly compare both Zuleika Saz-­Parkinson and Elena Parmelli were or are cur- index tests. We identified studies that recruited participants rent employees of the Joint Research Centre, European from routine screening programs which makes our results ro- Commission. Carlos Canelo-­ Aybar, Jessica Beltrán, David bust in terms of applicability to the clinical practice. We also Rigau, Margarita Posso, and Pablo Alonso-­Coello, are employ- used recommended statistical methods to pool accuracy stud- ees of the Iberoamerican Cochrane Collaboration. Dr. Giorgi ies and included the GRADE approach to rate the certainty of Rossi as former-­PI of an independent study on HPV-­based | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License CANELO-­AYBAR et al.    2203 5. Schünemann HJ, Lerda D, Dimitrova N, et al. Methods for devel- cervical cancer screening, funded by the Italian Ministry of opment of the European commission initiative on breast cancer Health, data owner, conducted negotiations with Roche diag- guidelines: recommendations in the era of guideline transparency. nostics, Hologic-­Genprobe, Becton-­Dickinson to obtain rea- Ann Intern Med. 2019;171(4):273–­280. gents at reduced price or for free; the reagents obtained were 6. Schünemann HJ, Lerda D, Quinn C, et al. Breast cancer screening not used in his institution. Dr. Lebeau reports grants and reim- and diagnosis: a synopsis of the european breast guidelines. Ann bursement for travel-­related expenses related to consultancy Intern Med. 2020;172(1):46–­56. from Roche Pharma AG, reimbursement for travel-­ related 7. Domingo L, Hofvind S, Hubbard RA, et al. Cross-­national com- parison of screening mammography accuracy measures in U.S., expenses related to consultancy from Novartis Oncology, and Norway, and Spain. Eur Radiol. 2016;26(8):2520–­2528. grants from BioNTech Diagnostics GmbH outside the submit- 8. Brett J, Bankhead C, Henderson B, Watson E, Austoker J. The ted work. Dr. Gräwingholt is the responsible radiologist for psychological impact of mammographic screening. A systematic screening unit Paderborn, Germany, consultant radiologist for review. Psycho-­oncology. 2005;14(11):917–­938. screening programs in Switzerland, and consultant radiolo- 9. Michell MJ, Iqbal A, Wasan RK, et al. A comparison of the ac- gist for Hellenic School of Senology. Annette Lebeau, Axel curacy of film-­ screen mammography, full-­ field digital mam- Gräwingholt, Xavier Castells, Miranda Langendam, Elsa mography, and digital breast tomosynthesis. Clin Radiol. 2012;67(10):976–­981. Pérez, Paolo Giorgi Rossi, and Ruben Van Engen are mem- 10. Marinovich ML, Hunter KE, Macaskill P, Houssami N. Breast bers of the ECIBC Guidelines Development Group. cancer screening using tomosynthesis or mammography: a meta-­analysis of cancer detection and recall. J Natl Cancer Inst. AUTHORS' CONTRIBUTIONS 2018;110(9):942–­949. Carlos Canelo-­ Aybar, Lourdes Acosta, and Pablo Alonso-­ 11. Li T, Marinovich ML, Houssami N. Digital breast tomosynthesis Coello were responsible for conducting the systematic review (3D mammography) for breast cancer screening and for assess- conducted the search and data extraction. Paolo Giorgi Rossi, ment of screen-­recalled findings: review of the evidence. Expert Zuleika Saz-­Parkinson, and Axel Gräwingholt contributed to the Rev Anticancer Ther. 2018;18(8):785–­791. 12. Whiting PF, Rutjes AW, Westwood ME, et al. QUADAS-­2: a re- definition of the research protocol and provided comments to vised tool for the quality assessment of diagnostic accuracy stud- the preliminary results of the systematic review. Carlos Canelo-­ ies. Ann Intern Med. 2011;155(8):529–­536. Aybar drafted the first version of the article. All authors contrib- 13. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, uted to the interpretation and reporting of the results and provided Zwinderman AH. Bivariate analysis of sensitivity and specificity comments on subsequent versions of the article. All authors read produces informative summary measures in diagnostic reviews. J and approved the final manuscript prior submission. Clin Epidemiol. 2005;58(10):982–­990. 14. Hoyer A, Kuss O. Meta-­analysis for the comparison of two di- DATA AVAILABILIT Y STATEMENT agnostic tests to a common gold standard: a generalized linear mixed model approach. Stat Methods Med Res. 2018;27(5):​ All data sources used during this study are described in this 1410–­1421. published article and its additional information files. The 15. Schünemann HJ, Oxman AD, Brozek J, et al. Grading quality of datasets analyzed are available from the corresponding au- evidence and strength of recommendations for diagnostic tests and thor on reasonable request. strategies. BMJ. 2008;336(7653):1106–­1110. 16. Gilbert FJ, Tucker L, Gillan MGC, et al. Accuracy of digital ORCID breast tomosynthesis for depicting breast cancer subgroups in a Axel Gräwingholt https://orcid.org/0000-0003-4778-8731 UK retrospective reading study (TOMMY Trial). Radiology. 2015;277(3):697–­706. Paolo Giorgi Rossi https://orcid.org/0000-0001-9703-2460 17. Brandt KR, Craig DA, Hoskins TL, et al. Can digital breast tomo- Zuleika Saz-­Parkinson https://orcid. synthesis replace conventional diagnostic mammography views for org/0000-0002-3518-5899 screening recalls without calcifications? A comparison study in a sim- ulated clinical setting. AJR Am J Roentgenol. 2013;200(2):291–­298. R E F E R E NC E S 18. Heywang-­Kobrunner S, Jaensch A, Hacker A, Wulz-­Horber S, 1. ECIS. European Cancer Information System. 2018. https://ecis.jrc. Mertelmeier T, Holzel D. Value of digital breast tomosynthesis ec.europa.eu. Accessed 20/July/2019. versus additional views for the assessment of screen-­detected ab- 2. Duffy SW, Tabár L, Yen A-­F, et al. Mammography screening re- normalities -­a first analysis. Breast care. 2017;12(2):92–­97. duces rates of advanced and fatal breast cancers: results in 549,091 19. Tagliafico A, Astengo D, Cavagnetto F, et al. One-­to-­one com- women. Cancer. 2020;126(13):2971–­2979. parison between digital spot compression view and digital breast 3. Saadatmand S, Bretveld R, Siesling S, Tilanus-­Linthorst MM. tomosynthesis. Eur Radiol. 2012;22(3):539–­544. Influence of tumour stage at breast cancer detection on survival in 20. Cornford EJ, Turnbull AE, James JJ, et al. Accuracy of GE digital modern times: population based study in 173,797 patients. BMJ. breast tomosynthesis vs supplementary mammographic views for 2015;351:h4901. diagnosis of screen-­detected soft-­tissue breast lesions. The British 4. Marmot MG, Altman DG, Cameron DA, Dewar JA, Thompson journal of radiology. 2016;89(1058):20150735. SG, Wilcox M. The benefits and harms of breast cancer screening: 21. Whelehan P, Heywang-­ Köbrunner SH, Vinnicombe SJ, an independent review. Br J Cancer. 2013;108(11):2205–­2240. et al. Clinical performance of Siemens digital breast tomosynthesis | 20457634, 2021, 7, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cam4.3803 by Readcube (Labtiva Inc.), Wiley Online Library on [10/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2204    CANELO-­AYBAR et al. versus standard supplementary mammography for the assessment 28. Sharma N, McMahon M, Haigh I, Chen Y, Dall BJG. The potential of screen-­detected soft-­tissue abnormalities: a multi-­reader study. impact of digital breast tomosynthesis on the benign biopsy rate Clin Radiol. 2017;72(1):95 e9–­e15. in women recalled within the UK breast screening programme. 22. Waldherr C, Cerny P, Altermatt HJ, et al. Value of one-­view breast Radiology. 2019;291(2):310–­317. tomosynthesis versus two-­view mammography in diagnostic workup 29. Mumin NA, Rahmat K, Fadzli F, et al. Diagnostic efficacy of syn- of women with clinical signs and symptoms and in women recalled thesized 2D digital breast tomosynthesis in multi-­ethnic malaysian from screening. AJR Am J Roentgenol. 2013;200(1):226–­231. population. Sci Rep. 2019;9(1):1459. 23. Poplack SP, Tosteson TD, Kogel CA, Nagy HM. Digital breast tomosynthesis: initial experience in 98 women with abnor- SUPPORTING INFORMATION mal digital screening mammography. AJR Am J Roentgenol. 2007;189(3):616–­623. Additional supporting information may be found online in 24. Heywang-­Kobrunner SH, Hacker A, Jansch A, Kates R, Wulz-­ the Supporting Information section. Horber S, German RT. Use of single-­view digital breast tomosyn- thesis (DBT) and ultrasound vs. additional views and ultrasound for the assessment of screen-­ detected abnormalities: German How to cite this article: Canelo-­Aybar C, Carrera L, multi-­reader study. Acta Radiol. 2018;59(7):782–­788. Beltrán J, et al. Digital breast tomosynthesis compared 25. Skaane P, Bandos AI, Gullien R, et al. Comparison of digital to diagnostic mammographic projections (including mammography alone and digital mammography plus tomo- magnification) among women recalled at screening synthesis in a population-­based screening program. Radiology. mammography: a systematic review for the European 2013;267(1):47–­56. Commission Initiative on Breast Cancer (ECIBC). 26. Alabousi M, Zha N, Salameh J-­P, et al. Digital breast tomo- synthesis for breast cancer detection: a diagnostic test ac- Cancer Med. 2021;10:2191–2204. https://doi. curacy systematic review and meta-­ analysis. Eur Radiol. org/10.1002/cam4.3803 2020;30(4):2058–­2071. 27. Bernardi D, Ciatto S, Pellegrini M, et al. Prospective study of breast tomosynthesis as a triage to assessment in screening. Breast Cancer Res Treat. 2012;133(1):267–­271.