British Journal of Cancer www.nature.com/bjc ARTICLE OPEN Epidemiology Benefits and harms of annual, biennial, or triennial breast cancer mammography screening for women at average risk of breast cancer: a systematic review for the European Commission Initiative on Breast Cancer (ECIBC) ✉ 5✉ Carlos Canelo-Aybar 1,2,3 , Margarita Posso1,4, Nadia Montero1, Ivan Solà1,2, Zuleika Saz-Parkinson , Stephen W. Duffy 6 , Markus Follmann , Axel Gräwingholt8, Paolo Giorgi Rossi9 and Pablo Alonso-Coello1,2 7 © European Union 2021 BACKGROUND: Although mammography screening is recommended in most European countries, the balance between the benefits and harms of different screening intervals is still a matter of debate. This review informed the European Commission Initiative on Breast Cancer (BC) recommendations. METHODS: We searched PubMed, EMBASE, and the Cochrane Library to identify RCTs, observational or modelling studies, comparing desirable (BC deaths averted, QALYs, BC stage, interval cancer) and undesirable (overdiagnosis, false positive related, radiation related) effects from annual, biennial, or triennial mammography screening in women of average risk for BC. We assessed the certainty of the evidence using the GRADE approach. RESULTS: We included one RCT, 13 observational, and 11 modelling studies. In women 50–69, annual compared to biennial screening may have small additional benefits but an important increase in false positive results; triennial compared to biennial screening may have smaller benefits while avoiding some harms. In younger women (aged 45–49), annual compared to biennial screening had a smaller gain in benefits and larger harms, showing a less favourable balance in this age group than in women 50–69. In women 70–74, there were fewer additional harms and similar benefits with shorter screening intervals. The overall certainty of the evidence for each of these comparisons was very low. CONCLUSIONS: In women of average BC risk, screening intervals have different trade-offs for each age group. The balance probably favours biennial screening in women 50–69. In younger women, annual screening may have a less favourable balance, while in women aged 70–74 years longer screening intervals may be more favourable. British Journal of Cancer (2022) 126:673–688; https://doi.org/10.1038/s41416-021-01521-8 INTRODUCTION programmes, including which is the optimal mammography Breast cancer (BC) is the second most prevalent cancer in the screening interval. world and the most frequent among women [1]. In the European Published recommendations on mammography screening Union, 404,920 women were diagnosed with BC and 98,755 frequencies vary among organisations. The National Health women died during 2018 [2]. Despite these high rates, the Service Breast Screening Program (NHSBSP) of the United King- mortality risk of BC has decreased over the last decades due to dom, recommends screening every 3 years to women aged 50–70 improvements in treatment, services quality, and to early (47–73 in England) [4]. The United States Prevention Services Task diagnosis linked to the implementation of population-based Force (USPSTF) recommends biennial mammography for women screening programmes [3]. However, there is still ongoing aged 50–74 and making a case by case decision for women in research and debate on how to best implement BC screening their 40s [5]. The American Cancer Society recommends annual 1 Iberoamerican Cochrane Centre - Department of Clinical Epidemiology and Public Health, Biomedical Research Institute Sant Pau (IIB Sant Pau), Barcelona, Spain. 2CIBER de Epidemiología y Salud Pública (CIBERESP), Madrid, Spain. 3Department of Paediatrics, Obstetrics and Gynaecology, Preventive Medicine, and Public Health. PhD Programme in Methodology of Biomedical Research and Public Health, Universitat Autònoma de Barcelona, Bellaterra, Barcelona, Spain. 4Department of Epidemiology and Evaluation, IMIM (Hospital del Mar Medical Research Institute), Barcelona, Spain. 5European Commission, Joint Research Centre (JRC), Ispra, Italy. 6Wolfson Institute of Preventive Medicine, Queen Mary University of London, London, UK. 7German Cancer Society, Berlin, Germany. 8Radiologie am Theater, Paderborn, Germany. 9Epidemiology Unit, Azienda Unità Sanitaria Locale – IRCCS di Reggio Emilia, Reggio Emilia, Italy. ✉email: Carlos.canelo.ay@gmail.com; Zuleika.SAZ-PARKINSON@ec.europa.eu Received: 9 October 2020 Revised: 20 June 2021 Accepted: 30 July 2021 Published online: 26 November 2021 Published on Behalf of CRUK C. Canelo-Aybar et al. 674 screening between the ages of 45 and 54 (with the option of time trend (before-after), or analysis of population surveillance registries, starting annual screening between 40 and 44), and screening and (iii) decision analytic models (hereafter referred to as modelling every two years from age 55 or continue annually if the woman is studies) (Supplementary Tables S3a and 3b). All studies included at least in good health and expected to live ten more years [6]. two screening intervals in one of the age groups of interest; screening Previous studies have suggested that the balance between intervals from observational studies should had been defined based on at least two examinations prior to diagnosis; modelling studies should have benefits and harms for different screening intervals might vary assumed 100% adherence to the screening programmes and applied no depending on the age subgroup. A modelling study found that for discounting to the effects. Due to sparse empirical evidence in the 45–49 every 1000 women aged 50–74, biennial screening avoided seven age subgroup, we included RCTs and observational studies that recruited BC deaths, while annual screening had similar benefits but caused women from 40 to 49. more harms [7]. Observational data from the US Breast Cancer We excluded studies of women at high risk for breast cancer, i.e. having Surveillance Consortium (BCSC) registries, observed that preme- known susceptibility gene mutations (BRCA1/BRCA2), a history of previous nopausal women undergoing biennial screening had more BC breast cancer or lobular neoplasia, exposure to chest irradiation (other lesions with less favourable prognostic characteristics compared than diagnostic imaging over that anatomical area) or having a direct family member with breast cancer. to those having annual screening [8]. Pairs of reviewers (CCA, MP), after calibration, assessed eligibility and In 2015, the European Commission Initiative on Breast Cancer reviewed the full text of the selected references. Discrepancies were (ECIBC) was launched to develop the European Guidelines on Breast resolved either by consensus or with the help of a third reviewer. Cancer Screening and Diagnosis. This article describes the systematic review that informed the recommendations about mammography screening intervals for women of average breast cancer risk in three Data extraction and risk of bias assessment CCA and MP independently extracted details of the study design, patient separate age subgroups [9, 10]. During the guideline development population, setting, screening method, follow-up, mammography inter- process [9], the Guidelines Development Group (GDG) made detailed vals and results. If needed, we requested additional data from the considerations about the balance between desirable and undesirable authors. We assessed the risk of bias (or credibility for modelling studies) effects [9], values and preferences, equity, acceptability and feasibility; with the following tools: (I) for RCTs with the Cochrane Risk of Bias these considerations are described in the published methodology and Assessment tool [12] (ii) for observational studies with the Risk of Bias in summary of recommendations [9, 10] (https://healthcare-quality.jrc.ec. Non-randomised Studies of Intervention (ROBINS-I) [13], (iii) for 1234567890();,: europa.eu/european-breast-cancer-guidelines/screening-ages-and- modelling studies with the Questionnaire to Assess Relevance and frequencies). Credibility of Modelling Studies (the ISPOR-AMCP-NPC Good Practice Task Force) [14]. METHODS Data analysis Structured question and outcome prioritisation We prioritised observational studies reporting the longest observation time The clinical question ‘Should an annual, biennial or triennial screening when different studies used the same surveillance or clinical registries from frequency be used for screening asymptomatic women?’ was prioritised by an identical population covering overlapping time periods. We prioritised the GDG (Box 1: Structured clinical question). This review focused on the the more direct evidence for a European population of average risk when three age subgroups for which the European Guidelines previously issued data was stratified by women´s characteristics (i.e. white women instead of recommendations for screening (45–49, 50–69, and 70–74 years old). The other ethnic groups). GDG prioritised the outcomes using a 1–9 scale (7–9 critical; 4–6 important; Modelling studies reported the incremental number of events for each 1–3 of limited importance) [11]. screening interval compared to a non-screening scenario. For some studies, we calculated the number of events by subtracting overlapping age groups (i.e. to obtain events in annual screening in women 45– 49 Data sources and searches years old, we subtracted the estimates in women 50–69 from the larger We initially searched MEDLINE (via PubMed, October 2016), EMBASE (via group of 45–69). We used the estimates for women with scattered Ovid, October 2016) and CENTRAL (via The Cochrane Library, October fibroglandular breast density when they were only reported by breast 2016) databases using pre-defined algorithms for individual studies. We density categories. Across the different studies, we presented the range of updated our initial search in MEDLINE (via PubMed) and EMBASE (via Ovid) the absolute difference of events per each pairwise screening interval in April 2020 (Supplementary Table S1: Protocol Systematic Review, comparison. Supplementary Table S2: Search strategy). We did not attempt to conduct a meta-analysis of relative risks (RR) or odds ratios (OR) from empirical studies because there were not enough Study selection studies across age groups to be meaningful or because several We included studies published in English of the following designs: (I) publications reported the same population data at overlapping time randomised clinical trials (RCTs), (ii) observational studies such as cohorts, periods. Box 1. PICO structured clinical question Population Intervention Comparison Outcomes Women who are at average risk Annual, biennial or triennial mammography Another interval (annual, 1. Breast cancer mortality of breast cancer: screening (film or digital) biennial or triennial) 2. Incidence of interval cancer • 45–49 years 3. Stage of breast cancer • 50–69 years 4. Radiation induced breast • 70–74 years cancers 5. Deaths due to radiation induced breast cancers 6. Quality of life 7. False positive related adverse effects 8. Overdiagnosis British Journal of Cancer (2022) 126:673 – 688 C. Canelo-Aybar et al. 675 3920 unique records identified from databases (2860 from original search, and 1060 from updating) 3846 records excluded after title/abstract screening 74 full text articles assessed for eligibility: Excluded at full text (n=48) -8 SRs did not haveall outcomes of interest assessed -17 individual studies did not have a comparison of screening intervals 1 additional study referred by the expert -3 individual studies without panel acceptable definition of intervals -4 individual studies did not include age groups or intervals for 27 individual publications included: comparison -2 studies had updated -1 RCT (2 publications) publications -13 observational studies (14 publications) -11 modelling studies -13 excluded or other reasons Fig. 1 PRISMA flow diagram of literature search and selection. Certainty of the evidence RESULTS We rated the certainty of the evidence, as high, moderate, low or very low, Search results for each outcome based on the standard GRADE approach for RCTs and We included 22 studies from 2860 unique citations in our initial observational studies [15, 16]. To apply the GRADE approach to modelling evidence synthesis in October 2016 which was used to develop studies, we considered the certainty would depart from the lowest the ECIBC recommendations. After the updated search in April, certainty of the bodies of evidence that informed the main inputs in the model. We used the credibility and relevance items from the ISPOR-AMCP- 2020, we included 3 additional studies comprising a total of NPC tool to inform the judgments for the risk of bias and indirectness 25 studies (from 27 publications) during both periods: one RCT domains. [18, 19], 11 modelling studies [7, 20–29] and 13 observational As is customary in systematic reviews, we adopted a partially studies (Fig. 1) [8, 30–42]. The list of excluded studies and reasons contextualised approach to rate the certainty of evidence, this means for exclusion are described in Supplementary Table S4. that for a point (or range) estimate of a single outcome we assessed our certainty that the true effects lie within the boundaries of what we Studies’ characteristics consider a trivial, small, medium or large effect without considering the We provide here a summary of the study design, and the main evidence from other outcomes [17]. During the development of results for only the three age groups of interest. When there is recommendations, guideline panel members might consider our results using a contextualised approach which means considering the evidence empirical data (from observational or RCTs) we rely primarily on from other critical outcomes (i.e. whether the benefits are consistent across those estimates instead of simulated number of events from outcomes) when rating the certainty for a single outcome [17]. modelling studies. To interpret the modelling estimated events, British Journal of Cancer (2022) 126:673 – 688 C. Canelo-Aybar et al. 676 we must consider that they represent the estimated events for a microsimulation study projected adverse events related to cohort of individuals from the time of screening until death or radiation exposure from mammography exams in women 50–74 during the individual´s lifetime (or other given time point). The years of age (Table 2) [21]. One additional study adapted a estimated 10-year probability of false positive or false biopsy microsimulation Markov model to the German context to assess recommendation in the observational studies were estimated annual, biennial, and triennial routine screening in women aged using a previously described statistical model [43]. A detailed 50–69 [29]. reporting of the results from studies covering larger age groups The remaining four modelling studies implemented non- (i.e. 66–74 years) can be found in Supplementary Table S3a, b. individual models. One transition model evaluated annual versus The only available RCT was conducted between 1989 and 1996, biennial screening intervals in Japan [23]. One Markov model the United Kingdom Co-ordinating Committee on Cancer assessed breast cancer deaths averted and overdiagnosis due to Research (UKCCCR) trial of Breast Screening Frequency and screening for women in the United Kingdom [20], and another randomly allocated 99,389 women aged 50–62 to either annual study applied the model developed by Preston to estimate or triennial screening [19]. Of the women originally invited to radiation related events [25]. We obtained non-publicly available either of the screening arms, 38,492 (77%) attended triennial data of a transition modelling study for a Spanish cohort described screening and 37,530 (76%) attended annual screening. The elsewhere (Table 2) [27, 52]. primary end point was predicted mortality based on two validated risk-models. However, as the UKCCR published observed data for Benefits and harms in women aged 45–49 (Tables 3/4) survival up to the end of 2006, we reported these estimates in our Observational studies. A Finish study suggested an increase in assessment [18]. the risk of BC mortality in annual versus triennial screening Nine studies performed analysis from surveillance systems data (incidence RR 1.14; 95%CI 0.59–2.19) although the estimate was of the United States which differed in the time periods covered very uncertain [39]. The odds of advanced breast cancer stage and the age group of the women included. Eight studies used (IIB–IV) may be higher in women with a history of biennial national Breast Cancer Surveillance Consortium (BCSC) mammo- screening compared to annual screening (OR 1.17; 95%CI graphy registries which were linked to the Surveillance, Epide- 0.93–1.46) among incident breast cancers from US registries [37]. miology, and End Results (SEER) pathology registries In women of normal weight, the 10-year probability of false [8, 30, 32, 34, 37, 38, 40]. One study used the Vermont Breast positive results was 11.2% (95%CI 9.8–12.8%) with annual screen- Cancer Surveillance System (VBCSS) from the state of Vermont ing and 6.0% (95%CI 5.4–6.6%) with biennial screening [32]. The [33]. The studies included two types of analysis: first a case series probability of a false positive biopsy recommendation was 11.4% of invasive BC that were used to evaluate the association between (95%CI 10.5–12.4%) with annual screening, 5.9% (95%CI 5.6–6.2%) screening intervals and adverse tumour characteristics, and with biennial screening, and 3.9% (95%CI 3.7–4.1%) with triennial secondly, they estimated the 10-year cumulative probabilities of screening among white women [38]. false positive results and false positive biopsy recommendations Moreover, indirect evidence from the wider age group of (Table 1) [43]. women (40–79) suggested that the incidence of interval cancers A quasi-experimental study included women aged 40–49 who may be lower among annually screened (0.07%) compared to were invited to attend a screening programme in Finland. Those biennially screened (0.15%) women, but it was very uncertain women born in an even calendar year were invited for given the small number of events [35]. mammography screening every year, while those born in an odd calendar year were invited to screening every 3 years [39]. Modelling studies. One study estimated, across six microsimula- One study conducted a comparative analysis of two time periods tion models, a median of 30 more deaths averted per 100,000 in British Columbia-Canada, before and after 1997, year when the women undergoing annual screening compared to biennial Screening Mammography Program of British Columbia (SMPBC) screening in the US population [7], while the median number of changed its policy from annual to biennial mammography for additional QALYs gained with annual screening was 480 more women aged 50–79 [31]. compared to biennial screening [7]. In the same modelling study, Two studies included women from screening programmes at the overdiagnosis estimation was higher with annual screening medical centers from the US. The first performed a retrospective compared to biennial screening [7]. One modelling study assessed analysis of data from women who chose to attend either annual or the risk of radiation induced adverse events in this age group and biennial mammography examinations in a screening programme found that annual screening yielded 14 more induced BC and 2 of the University of California San Francisco Medical Center [35]. more deaths per 100,000 screened women compared to biennial The second study was a retrospective cohort of women without screening [21]. previous diagnosis of BC who attended a routine screening examination at Columbia University Medical Center in New York; Benefits and harms in women aged 50–69 (Tables 3/4) the screening interval was defined using the time elapsed since Randomised clinical trials. Duffy et al. reported in the UKCCR their previous exam according to their electronic clinical records study, over a median of 162 months of follow-up, that annual [42] (Table 1). screening may decrease the risk of BC mortality compared to Six studies used microsimulation models developed within the triennial screening among attenders to the prevalent screening Cancer Intervention and Surveillance Modelling Network (CISNET) (RR = 0.89, 95% CI 0.73−1.07) [18]. Moreover, there was a small collaboration: Model D (Dana-Farber) [44], Model E (Erasmus) [45], difference in the size of the tumour at diagnosis, with a major Model GE (Georgetown-Einstein) [46], Model M (MD Anderson) proportion of them being 10 mm or smaller in the annual [47], Model S (Stanford) [48], and Model W (Wisconsin-Harvard) screening group compared to the triennial group (25% vs. 19%) [49]. Each of these models has its own characteristics which are [18, 19]. described elsewhere [50], they vary in the model structures and assumptions such as factors conditioning screen detection, Observational studies. One study in a province of Canada individual risk factors or allowing spontaneous regression of comparing the period before and after mammography screening ductal carcinoma in-situ (DCIS) lesions [51]. Four studies assessed changed from annual to biennial found there may be little to no mammography screening intervals for the U.S. population difference in mortality (MR 1.06; 95%CI 0.76, 1.46) or interval cancer reporting the median estimates from two to six models (RR 0.98; 95%CI 0.90–1.06) between the two-time periods [31]. [7, 21, 22, 24]. Two studies simulated screening for a Canadian Miglioretti et al. found there may be no difference in the risk of population based on an adaptation of Model W [26, 28]. One advanced BC stage (IIB–IV) in the age groups 50–69 (adjusted RR 0.98; British Journal of Cancer (2022) 126:673 – 688 Table 1. Characteristics of the clinical trials, and observational studies identified in the literature search. Author, year Country, period Design/screening intervals No enrolled Inclusion/exclusion criteria Age ranges, Outcomes of interest (years) Randomized clinical trial BSFTG, 2002 [19], United Kingdom Attender women to prevalent -Annual: 37,530. I: women attending prevalent 50–62 -BC mortality. Duffy, 2008 [18] 1989–2006 screening at NHS breast screening -Triennial: 38,492. screening. -Interval cancer. program invited to conventional (3 E: women with BC diagnosed prior years) interval or to three annual to the trial. screenings. Observational studies Braithwaite, 2012 United States Data from five BCSC registries (those -Annual (BC I: women with at least two 66–89 Reported by co-morbidity 1999–2006 matched to Medicare claims) linked to cases):1227. mammograms. score categories: SEER programs/ tumor registries. -Biennial (BC -Stage of BC (IIB+). British Journal of Cancer (2022) 126:673 – 688 -BC-case series: women with incident cases): 453. -FP results. DCIS or invasive BC; interval groups -FP-Cohort: 137,949. -FP recommendations. defined as 1 (9–18 months) or 2 (19–30 months) years, based on the two most recent mammograms prior diagnosis. -FP-cohort: all first and subsequent screening mammography from 1999 to 2006, without BC diagnosis after 1 year of last examination. Coldman, 2008 [31] Canada Pre-post screening policy changing -Annual (before july E: women with a prior BC diagnosis 50–79 -BC mortality. evaluation. Data from the SMPBC 1996): 152,226 not eligible to attend SMPBC. -Interval cancers. linked to VSA registries over two time -Biennial (July 1996 periods. From 1988 to 1997 women or after): 184,764 50–79 years recommended annual screening, 1998–2005 changed to a biennial recommendation. Dittus, 2013 [32] United States Data from seven BCSC registries linked -Annual (BC I: women with at least two 40–74 Reported by BMI categories: 1994–2008 to SEER programs/tumor registries: cases): 2766. mammograms before BC diagnosis. -Stage of BC (IIB+). -BC-case series: women with incident -Biennial (BC E: history of BC diagnosis, reporting -FP results. C. Canelo-Aybar et al. BC; intervals groups defined as 1 cases): 1666. hormone therapy use. -FP recommendations. (9–18 months) or 2 (19–30 months) -FP-Cohort: 555,343 years based on the time between two most recent mammograms prior diagnosis. -FP-cohort: all screening mammography from 1994 to 2008, without BC diagnosis after 1 year of last examination. Goel, 2007 [33] United States Data from the VBCSS which collects -Annual (BC I: women with at least two >40 years 1994–2002 information from patients, radiologists cases): 1236 mammograms before BC diagnosis. and hospital pathology. -Biennial (BC E: intervals of less than 273 days -BC case series: women with incident cases): 439 between mammograms. History of BC, interval groups defined as 1 BC diagnosis. (0.75–1.49 years), 2 (1.5–2.49 years) years based on the time between two most recent mammograms prior diagnosis. 677 Table 1 continued 678 Author, year Country, period Design/screening intervals No enrolled Inclusion/exclusion criteria Age ranges, Outcomes of interest (years) Hubbard, 2011 [34] United States, Data from seven BCSC registries linked -Annual (BC I: women with at least two 40–59 -BC stage (IIB+). 1994–2006 to SEER programs/tumor registries. cases): 36,445 mammograms. -FP results. -BC case series: women with incident -Biennial (BC E: women with BC at or after -Bx recommendations. invasive BC; interval groups defined as cases): 27,775 60 years. 1 (9–18 months) or 2 (19–30 months) -FP-Cohort: 169,456 years based on the time between two most recent mammograms prior diagnosis. -FP-cohort: screening mammograms from 1994 to 2004 or 2007 (depending on the registry). Hunt, 1999 [35] United States Retrospective analysis from -Annual: 19,905 I: previous normal screening 40–79 -Interval cancers 1985–1997 prospectively collected data of women -Biennial: 4306 mammography, asymptomatic that choose annual or biennial physician referred women from six screening mammography, performed contiguous counties. by University of California San E: – Francisco Medical Center at screening mammography mobile van. Kerlikowske, 2013 [8] United States Data from BCSC registries linked to -Annual (BC I: women with diagnosis of incident 40–74 Reported by breast density 1994–2008 SEER programs/tumor registries. cases): 7039 invasive or DCIS BC and at least 2 categories: -BC case series: women with incident -Biennial (BC prior mammograms. -BC stage(IIB+). DCIS or invasive BC; interval groups cases): 3476 E: History of BC diagnosis. -FP results. C. Canelo-Aybar et al. defined as 1 (0.75–1.49 years) or 2 -Triennial (BC -Bx recommendations. (1.5–2.49 years) years based on the cases): 959 time between two most recent -FP-Cohort: 922,624 mammograms prior diagnosis. -FP-cohort: all first and subsequent screening mammography from 1994 to 2008, without BC diagnosis after 1 year of last examination. Parvinen, 2011 [39] Finland Population based, quasi-experimental. -Triennial: 6,926 I: – 40–49 -BC mortality. Klemi, 1997 [36] 1987–2003 Mailed screening invitation to women -Annual: 7839 E: – -Interval cancer. aged 50 or more at biennial interval. Women less than 50 years, and born even-numbered year invited to annual screening and those born odd- numbered year were invited to triennial screening. McGuinnes, 2018 [42] United States Retrospective cohort. Women were -Annual: 1399 I: no previous diagnosis of BC, age <50 years -FP results. 2014–2015 approached during routine screening -Biennial: 335 ≥18 years ≥50 years mammography at Columbia University Medical Center in New York. Annual interval from 9 to 18 months, biennial from 18 to 30 months. More than 913 days (3+ years) were considered non-compliant. Less than 274 days was considered recall imaging. Miglioretti, 2015 [37] United States Data from seven BCSC registries linked -Annual (BC I: women with at least two 40–85 Reported by menopausal 1996–2012 to SEER programs/tumor registries. cases): 12,070 mammograms. and HT use categories: -BC-case series: women with incident British Journal of Cancer (2022) 126:673 – 688 Table 1 continued Author, year Country, period Design/screening intervals No enrolled Inclusion/exclusion criteria Age ranges, Outcomes of interest (years) DCIS or invasive BC; intervals groups -Biennial (BC -BC stage(IIB+). defined as 1 (11–14 months) or 2 cases): 3370 -Interval cancer. (23–26 months) years based on the time between two most recent mammograms prior diagnosis. O’Meara, 2013 [38] Unitesd States Data from seven BCSC registries linked -Annual (BC I: women with at least two 40–74 Reported by breast density 1994–2008 to SEER programs/tumor registries. cases): 8876 mammograms. and ethnic categories: -BC-case series: women with incident -Biennial (BC -BC stage(IIB+). DCIS or invasive BC; interval groups cases): 4265 -Interval cancer. defined as 1 (9–18 months) or 2 -Triennial (BC -FP results. (>18–30 months), or 3 cases): 1255 -Bx recommendations. (>30–42 months) years based on the -FP-cohort: 1,276,312 British Journal of Cancer (2022) 126:673 – 688 time between two most recent mammograms prior diagnosis. -FP-cohort: all screening mammography from 1994 to 2008, without BC diagnosis after 1 year of last examination. Sanderson, 2015 [41] Unites States Data from Medicare claims and SEER. -Irregular (BC I: non-Hispanic white or black 69–84 Reported by ethnicity 1995–2000 -BC-case series: women with incident cases): 29,712 ethnicity, complete Medicare category (non-Hispanic BC; groups defined based on -Biennial (BC coverage during 4-year before BC white or black women): mammography screening periodicity cases): 11,227 diagnosis; primary BC diagnosed -Mortality among BC cases over the 4 years prior to diagnosis as -Annual (BC cases): between 69 to 84 years. (according to screening (a) no or irregular mammography 23,355. E: BC diagnosed by autopsy or death interval history). screening, (b) biennial mammography, certificate, stage IV cancer. and (c) annual mammography. White, 2004 [40] Unites States Data from seven BCSC registries linked -Annual (BC I: women with diagnosis of invasive 40–89 Reported by breast density: 1996–2001 to SEER programs/tumor registries. cases): 5400. or DCIS BC and at least two prior -BC stage(IIB+). -BC series: women with incident DCIS -Biennial (BC mammograms. -Interval cancer. or invasive BC; intervals groups cases): 2440. E: history of BC. defined as 1 (9–18 months) or 2 (>18–30 months) years based on the C. Canelo-Aybar et al. time between two most recent mammograms prior diagnosis. SMPBC Screening Mammography Program of British Columbia, VSA Vital Statistics Agency, SEER Surveillance Epidemiology and End Results, BCSC Breast Cancer Surveillance Consortium, VBCSS Vermont Breast Cancer Surveillance System, DCIS ductal carcinoma in situ; BC breast cancer, NHS National Health Services, FP false positive, Bx biopsy, I inclusion criteria, E exclusion criteria. 679 680 Table 2. Characteristics of decision analysis studies identified. Author, year Modelled Design/screening intervals Strategies Parameters Years of Outcomes of interest population screening Non-individual based models Gunsoy, 2014 [20] United Kingdom Markov-cohort simulation model. -Healthy, Six strategies defined by: -(F): up to 85 years old 40–73 -Breast cancer deaths preclinical non-progressive in situ, preclinical -Starting age: 40, 47, 50 years. -(SA): on uptake, sensitivity, and averted progressive in situ, preclinical invasive, -Interval: triennial, annual, and hybrid sojourn time. -Overdiagnosis- diagnosed in situ, and diagnosed invasive (annual for 40–47, triennial thereafter) -(O): difference in the cumulative breast cancer by NPI category, death from BC, -Screening scenarios stopped at 70 or incidence of invasive in situ cancer. and death from other causes. 73 years. -(MA): NHS breast cancer program. Tsunematsu, 2015 [23] Japan, United States. Transition cohort model to simulate impact Twelve strategies defined by: -(F): NR 40–79 -Breast cancer deaths of screening for the Japanese and US -Starting age: 40, 50 years -(O): NR averted population. The source of stage distributions -Intervals: annual, biennial -(SA:): mortality rate of undetected BC -FP results were data from the Japanese Breast Cancer -Screening stopping at 69, 74, and -(MA): SE: 81.5%; SP 90.4–94.7% (Japan) Society and the BCSC and National Cancer 79 years. Data BASE respectively. Vilaprinyo, 2014 [27]* Spain Stochastic transition model. Extension of the Twenty strategies defined by: -(C): born from 1948 to 1952 40–79 -Breast cancer deaths Lee and Zelen model to estimate incidence -Starting ages: 40, 45, 50. -(F): time horizon was 40–79 years averted and prevalence, -Intervals: annual, biennial -(O): NR -Overdiagnosis -Screening stopping at 69, 70, 74 (SA): not provided by authors -QALYS and 79. -(MA): SE: 0.55 for 40–45 years, 0.70 for -FP results 45–50 years, 0.75 for 50–70 years and -Benign breast bx 0.80 for >70 year -(Q): includes anxiety, and FP results Yaffe, 2011 [25] Canada Model by Preston (excess absolute risk of Six strategies defined by: -(F): screening began up to 109 years 40–74 - Radiation induced BC radiation induced BC). Applied to Canadian -Starting ages: 40, 50 -(SA): using relative model instead of - Radiation induced population of 2002. Digital mammography. -Intervals: annual, biennial (hybrid absolute model, latency years, BC deaths annually in 40 s, biennial thereafter). survival rates. -Screening stopping at 49, 59 years. Individual based models C. Canelo-Aybar et al. Arnold, 2019 [29] Germany A microsimulation-Markov model included 6 Three regular screening strategies and -(F): from age of 50, until the end of life -QALY N= health states: healthy (no breast cancer); additional strategies based on or 100 years. -Biopsy after false positive 3,000,000 women ductal carcinoma in situ (DCIS); localized, individual risk assessment (not shown) -(T): specific treatment based on screening regional, or distant invasive breast cancer; -Interval: annual, biennial, triennial hormone receptors and death. -Starting age: 50 years -(MA): digital mammography sensitivity -Screening stopping: 69 years based in BCSC. -(O): -(SA): univariate and probabilistic sensitivity analysis (e.g. DCIS incidence, invasive cancer incidence, invasive cancer morality) Mandelblatt, 2016 [7] United States Six micro simulation models developed Eight strategies defined by: -(C): born in 1970 of average-risk and 40–74 Reported as median across N = 1000 women within the CISNET collaboration: model D, -Starting age: 40, 45, or 50 years average breast density. models: model E, model GE, model M, model S and -Interval: annual, biennial and hybrid -(F): from age 25 years until death or -BC deaths averted model W. (annual in 40 s, biennial thereafter). -All age 100 -Overdiagnosis Updating of models include 1) portrayal of strategies stop screening at 74. -(O): models assume proportions of DCIS -QALYS (includes molecular subtypes based on ER and non-progressive; models M and W overdiagnosis) HER2 status, current population incidence, assumed some non-progressive invasive -FP results digital screening, and update therapies. cancers -Benign breast bx -(T): specific treatment based on hormone receptors -(MA): digital mammography sensitivity based in BCSC. -(ME): RR 0.72 (95%CI 0.65–0.75) Miglioretti, 2016 [21] United States Two micro simulation modeling approaches Eight strategies defined by: -(MA): digital mammography sensitivity 40–74 - BC deaths averted N = 100,000 women for digital mammography. MISCAN-Fadia -Starting age: 40, 45, or 50 years based in BCSC - Radiation induced BC model and a new model for radiation -Interval: annual, biennial and hybrid -(BS): views and compressed thickness - Radiation induced exposure (which accounts for repeated (annual in 40 s, biennial thereafter). from DMIST BC deaths mammography or radiation exposure and -All strategies stop screening at 74. -(RD): product of half the number of BS). Excess of radiation induced BC using the views of both breast by dose per view. results from Preston. Mittmann, 2018 [28]** Canada One modified microsimulation, from the Eleven screening scenarios: -(T): specific treatment based on 40–74 -QALYs N= perspective of the Ontario public health care -Annual, biennial, triennial, and hybrid hormone receptors 2,000,000 women system model W developed within the of these -(C): all women born in 1960 validated CISNET collaboration. Discrete event, -Starting age: 40 or 50 years. against US data and modified against stochastic simulation based on the US Canadian data. British Journal of Cancer (2022) 126:673 – 688 Table 2 continued Author, year Modelled Design/screening intervals Strategies Parameters Years of Outcomes of interest population screening population. -Screening stopping at: 49, 69 or -(F): lifetime horizon Model simulated the lives of women at 74 years. -(SA): input cost varied for key resources 6-month intervals. in one-way analysis -(MA): digital mammography sensitivity based in BCSC. Trentham-Dietz, 2016 [22] United States Three micro simulation models developed Six screening scenarios: -(C): born in 1970 40–74 Reported as median across N = 1000 within the CISNET collaboration: model GE, -Annual, biennial, or triennial digital -(F): from age 25 years until death or models (stratified by BD): model W, model E. Model applied to mammography. age 100 -BC deaths averted. population subgroups based on 4 breast -Starting age: 50 or 65 (received -(O): models assume proportions of DCIS -Overdiagnosis. density levels and 4 exemplar relative risk biennial from 50 to 64). non-progressive; models W assumed -QALYS (includes levels: average, postmenopausal obesity, -Stopping age: 74. some non-progressive invasive cancers overdiagnosis). history of benign breast biopsy result, history -(T): specific treatment based in -FP results. of lobular carcinoma in situ. hormone receptors -Benign breast bx. -(MA): digital mammography accuracy based in BCSC. British Journal of Cancer (2022) 126:673 – 688 Van Ravestein, 2012 [24] United States Four micro simulation models developed Five screening scenarios: -(T): specific treatment based in 40–49 -BC deaths averted. N = 1000 women within the CISNET collaboration: model D, -Interval: annual and biennial hormone receptors -FP results. model E, model GE and model W. The models -Film or digital mammography -(C): born in 1960 of average-risk. include biennial screening for women 50 to -Age group: 40–49 years -(MA): digital and film mammography 74 years extended with 4 screening scenarios All scenarios estimated incremental accuracy based in BCSC. varying by screening interval (annual and effects compared to 50–74 screening. biennial) and screening method (film and digital). Yaffe, 2015 [26] Canadian, One model from the CISNET collaboration Eleven screening scenarios: -(F): from age 40 years until death or 40–74 -BC deaths averted. N= (model W), adapted to the Canadian context. -Interval: annual, biennial, triennial age 99. -FP results. 2,000,000 women Treatment effectiveness was implemented on (and two hybrid scenarios). -(C): born in 1960 of average-risk. a cure/no cure model. The model allowed -starting age: 40 or 50 years. different proportion of hormone receptors -stopping age: 69 or 74 years. subgroups. Model D: Dana-Farber Cancer Institute Boston Massachusetts; Model E: Erasmus Miscan-Fadia; University Medical Center Rotterdam, the Netherlands; Model GE: Georgetown University Medical Centre, Washington, DC, and Albert Einstein College of Medicine, Bronx, New York; Model M: MD Anderson Cancer Center, Houston, Texas; Model W: University of Wisconsin; CISNET: Cancer Intervention and Surveillance Modeling Network; BCSC: Breast Cancer Surveillance Consortium; ER: oestrogen receptor; HER2: human epidermal growth factor receptor 2; NPI: Nottingham prognostic index; QALY: quality adjusted life-years; (SA): sensitivity analysis; (C): women born cohort; (F): time of follow-up, horizon time; (MA): mammography accuracy; (ME): mammography effectiveness; (O): overdiagnosis assumptions; (T): tailored treatment; (RD): radiation dose; BC: breast cancer; FP: false positive. *Unpublished data were provided by the authors. **A previous study by the same authors and using the same model and population was excluded (Mittmann 2015) as the updated study provided a more detailed description of the outcomes. C. Canelo-Aybar et al. 681 682 Table 3. Summary of the desirable and undesirable effects from RCTs and observational studies for different screening intervals and age groups.* Age group Annual vs. Biennial Triennial vs. Biennial Annual vs. Triennial Certainty of evidence N° of studies, Relative Absolute N° of studies, Relative Absolute N° of studies, Relative Absolute countries effect (95% reduction (95%CI) countries effect reduction (95% countries effect (95% reduction (95%CI) CI) (95%CI) CI) CI) Breast cancer (BC) mortality 40–49 – – – – – – 1 (Finland) RR 1.14 (0.59 3 more (7 fewer to Very low [39] to 2.19)** 21 more) 50–69 1 (Canada) IRR: 0.94 4 fewer (22 fewer – – – 1 (UK) [18]*** RR 0.93 (0.76 42 fewer (144 Very low for annual vs biennial [31] (0.68 to 1.31) to 21 more) to 1.12) fewer to 72 more) and moderate vs triennial BC stage (IIB–IV) 40–49 1 (US) [37] OR 0.85 (0.75 NE 1 (US) [38] OR 0.78 NE – – – Very low for all comparisons to 0.96) (0.54 to 1.11) 50–69 1 (US) [37] OR: 0.93 NE 1 (US) [38] OR: 0.83 NE – – – Very low for all comparisons (0.81 to 1.09) (0.65 to 1.07) 70–74 1 (US) [37] OR 0.98 (0.76 NE – – – – – – Very low to 1.27) Interval cancer 40–49 1 (US) [35] RR 0.46 (0.16 81 fewer (126 – – – – – – Very low to 1.36) fewer to 54 more) 50–69 1 (US) [37] A: 22% (21% to 30%) of BC cases 1 (US) [7] T: 44% (41% to 48%) of 1 (US) [7] A: 30% (29% to 31%) of BC cases Very low for all comparisons C. Canelo-Aybar et al. B: 27% (26% to 29%) of BC cases BC cases T: 44% (41% to 48%) of BC cases B: 41% (39% to 42%) of BC cases 70–74 1 (US) [30] A: 23% (22% to 25%) – – – – – – Very low B: 33% (30% to 36%) False positive results—10 year cumulative probability per woman 40–49 1 (US) [32] A: 67% (65% to 68%) 1 (US) [32] T: 30% (29% to 30%) 1 (US) [38] A: 65% (63% to 65%) Very low for all comparisons B: 45% (44% to 46%) B: 41% (41% to 42%) T: 29% (29% to 30%) 50–69 1 (US) [32] A: 54% (53% to 55%) 1 (US) [38] T: 25% (25% to 25%) 1 (US) [38] A: 55% (55% to 56%) T: 25% (25% Very low for all comparisons B: 34% (34% to 35%) B: 35% (35% to 36%) to 25%) 70–74 1 (US) [30] A: 47% (45% to 50%) – – – – – – Very low B: 27% (26% to 28%) False positive biopsy recommendation—10 year cumulative probability per woman 40–49 1 (US) [32] A: 11% (10% to 13%) 1 (US) [32] T: 4% (4% to 4%) 1 (US) [38] A: 11% (11% to 12%) Very low for all comparisons B: 6% (5% to 7%) B: 6% (6% to 6%) T: 4% (4% to 4%) 50–69 1 (US) [32] A: 8% (7% to 9%) 1 (US) [38] T: 4% (4% to 4%) 1 (US) [38] A: 10% (9% to 10%) Very low for all comparisons B. 5% (4% to 5%) B: 5% (5% to 6%) T: 4% (4% to 4%) 70–74 1 (US) [30] A: 9% (8% to 11%) – – – – – – Very low B: 4% (4% to 5%) To review the reference for each study and the reasons for downgrading the certainty of the evidence see Supplementary file Table S4–S12. *Only the study with the longest time of observation was included when there were several publications with overlapping time periods. When studies provided results stratified by women´s characteristics, we extracted data from subgroups more similar to European context (i.e. white women instead of other ethnic groups). **We calculated the confidence interval from the raw data reported in the publication as the original interval was not consistent with the main effect and lower interval bound. ***Randomized clinical trial study. British Journal of Cancer (2022) 126:673 – 688 Table 4. Summary of desirable and undesirable effects (number of events) from modelling studies for different screening intervals and age groups (per 100,000 screened women).** Age group Annual vs. Biennial Triennial vs. Biennial Annual vs. Triennial Certainty of evidence across comparisons N° of N° of events Absolute N° of studies, N° of events Absolute N° of studies, N° of events Absolute studies, per arm reduction countries per arm reduction countries per arm reduction countries Breast cancer deaths averted (BC) 45–49* 2 (US) A: 70–90 30 more to 1 (Spain) [27] T: 47 5 fewer 1 (Spain) [27]# A: 33 14 fewer Very low for all [7, 21]## B: 39–40 51 more B: 52 T: 47 comparisons 50–69 3 (Canada, A: 631–870 165 more to 2 (Canada, Spain) T: 397–400 120 fewer to 2 (Canada, Spain) A: 631–740 234 more to Very low for all Japan, B: 426–705 205 more [26, 27] B: 426–520 29 fewer [26, 27] T: 397–400 340 more comparisons Spain) [23, 26, 27] 70–74* 2 (Canada, A: 100–142 3 fewer to 2 (Canada, Spain) T: 80–136 10 fewer to 2 (Canada, Spain) A: 100–142 6 more to Very low for all Spain) B: 90–145 10 more [26, 27] B: 90–145 9 fewer [26, 27] T: 80–136 20 more comparisons [26, 27] British Journal of Cancer (2022) 126:673 – 688 Overdiagnosis 45–49* 2 (Spain, A: 143–200 24 more to 1 (Spain) [27] T: 88 31 fewer 1 (Spain) [27] A: 143 55 more Very low for all US) [7, 27] B: 0–119 200 more B: 119 T: 88 comparisons 50–69 1 (Spain) A: 904 295 more 1 (Spain) [27] T: 500 109 fewer 1 (Spain) [27] A: 904 404 more Very low for all [27] B: 609 B: 609 T: 500 comparisons 70–74* 1 (Spain) A: 269 33 more 1 (Spain) [27] T: 193 43 fewer 1 (Spain) [27] A: 269 76 more Very low for all [27] B: 236 B: 236 T: 193 comparisons QALYs 45–49* 2 (Spain, A: 727–1540 62 more to 1 (Spain) [27] T: 653 12 fewer 1 (Spain) [27] A: 727 74 more Very low for all US) [7, 27] B: 665–1060 480 more B: 665 T: 653 comparisons 50–69 3 (Canada, A: 4400–7100 500 to 3 (Canada, Germany, T: 3300–4386 1200–328 fewer 3 (Canada, Germany, A: 4,400–7,100 1,100 to Very low for all Germany, B: 3900–5000 2100 more Spain) [27–29] B: 3900–5000 Spain) [27–29] T: 3,300–4,386 3,100 more comparisons Spain) [27– 29] 70–74* 2 (Canada, A: 336–600 91 fewer to 2 (Canada, Spain) T: 300–398 200 fewer to 2 (Canada, Spain) A: 336–600 62 fewer to Very low for all Spain) B: 427–500 100 more [26, 27] B: 427–500 29 fewer [26, 27] T: 300–398 300 more comparisons [26, 27] False positive results 45–49* 2 (Spain, A: 9150–56,700 2849 more to 1 (Spain) [27] T: 4831 1470 fewer 1 (Spain) [27] A: 9,150 4,319 more Very low for all US) [7, 27] B:6301–26,700 30,000 more B: 6301 T: 4,831 comparisons C. Canelo-Aybar et al. 50–69 3 (Canada, A: 13,567 more to 2 (Canada, Spain) T: 19,600 fewer to 2 (Canada, Spain) A: 18,354 to Very low for all Japan, 42,606–152,800 63,300 more [26, 27] 24,252–69,900 4787 fewer [26, 27] 42,606–152,800 82,900 more comparisons Spain) B: B: T: [23, 27, 29] 29,039–89,500 29,039–89,500 24,252–69,900 70–74* 2 (Canada, A: 5766–24,500 2307 more to 2 (Canada, Spain) T: 2295–12,700 4700 fewer to 2 (Canada, Spain) A: 3,471 more to Very low for all Spain) B: 3459–17,400 7100 more [26, 27] B: 3459–17,400 1164 fewer [26, 27] 5,766–24,500 11,800 more comparisons [26, 27] T: 2,295–12,700 Benign biopsy recommendations 45–49* 2 (Spain, A:409–5600 201 more to 1 (Spain) [27] T: 108 100 fewer 1 (Spain) [27] A: 409 301 more Very low for all US) [7, 27] B: 208–3000 2600 more B: 208 T: 108 comparisons 50–69 3 (Canada, A: 904–16,300 295 more to 3 (Canada, Germany, T: 1300 fewer to 3 (Canada, Germany, A: 1,289 fewer to Very low for all Germany, B: 609–14,400 4900 more Spain) [26, 27, 29] 2166–14,100 300 fewer Spain) [26, 27, 29] 3,455–16,300 5,900 fewer comparisons Spain) B: T: 2,166–14,100 [26, 27, 29] 2487–14,400 70–74* 2 (Canada, A: 428–3200 No difference to 2 (Canada, Spain) T: 171–3200 300 fewer to 2 (Canada, Spain) A: 428–3,200 0 more to Very low for all Spain) B: 287–3500 142 more [26, 27] B: 287–3500 116 fewer [26, 27] T: 171–3,200 257 more comparisons [26, 27] Radiation induced BC 45–49* 1(US) [21] A: 32 14 more – – – – – – Very low for all B: 18 comparisons 683 C. Canelo-Aybar et al. 684 95%CI 0.80–1.21) and 60–69 (adjusted RR 0.99; 95%CI 0.79–1.24) with *Number of events was not directly reported for this age group. We made an ad-hoc calculation subtracting the events from overlapping age groups (e.g. number of QALYS in women 45 to 69 years minus the Unpublished data from one study (Vilaprinyo 2014) reported 19 fewer BC deaths averted with annual compared to biennial screening. This result was inconsistent with the other studies and, therefore, is not To review the reference for each study and the reasons the certainty of the evidence was downgraded see: Supplementary file Table S4 to S12. When more than one study informing an outcome, the number Certainty of evidence annual versus biennial screening [37]. Another study in the US found across comparisons uncertain evidence that triennial screening compared to biennial # Only one study providing unpublished data informed this comparison. The result was in a different direction than the other bodies of evidence and thus cautious interpretation is recommended. screening in white women might be associated to lower odds of Very low for all Very low for all Very low for all comparisons comparisons comparisons stage IIB-IV (OR 0.83; 95%CI 0.65–1.07) but it was not consistent with the observed difference in large tumour size (>20 mm) (OR 1.15; 95% CI 0.93–1.41), or presence of lymph nodes (OR 0.98; 95%CI 0.80–1.21) at BC diagnosis [38]. From a US study using mammography and tumour registries, the 10-year probability of a false positive result was 55.2% (95%CI 54.8–55.7%) with annual screening, 35.4% (95%CI 35.0–35.7%) with reduction Absolute biennial screening, and 24.8% (95%CI 24.5–25.2%) with triennial screening [38]. The cumulative 10-year probability of having a false – – – positive biopsy recommendation was 9.7% (95%CI 9.3–10.1%) with annual screening, 5.4% (95%CI 5.2–5.6%) with biennial screening, and 3.7% (95%CI 3.6–3.9%) with triennial screening [38]. These findings N° of events were consistent with the risk of false positive results observed in a per arm retrospective cohort of a screening programme of New York [42]. – – – Modelling studies. In a Canadian modelling study, the number of BC deaths averted per 100,000 women aged 50–69 screened Annual vs. Triennial annually, biennially or triennially compared to no screening was **The certainty of evidence departed from low as the input parameters that inform the modelling studies were of low to very low certainty. 740, 520 and 400, respectively [26]. In another study, including N° of studies, three models tailored to the US population, the number of BC deaths averted per 100,000 screened women aged 50–74, with countries scattered fibroglandular breast density, was 690, 520 and 400 for annual, biennial and triennial screening [22] and the number of – – – QALYs gained was 6000, 4700 and 3600, respectively [22]. A microsimulation model for the German population found a median of 4400, 3900 and 3330 additional QALYs with annual, reduction biennial and triennial screening [29]. Absolute The estimated overdiagnosis was greater with more frequent screening intervals. In women with scattered fibroglandular density aged 50– 74, a microsimulation model study estimated – – – 2900, 2000 and 1600 for annual, biennial and triennial screening N° of events compared to no screening per 100,000 women [22]. A similar trend was reported in a study using non-individual models for a per arm Spanish cohort of women aged 50–69 [27]. A microsimulation model estimated the risk of radiation – – – induced adverse events in 100,000 women aged 50–74 to be of 27 induced BC cases with biennial screening and 49 with annual Triennial vs. Biennial screening [21]. The attributed number of radiation related deaths simulated was 4 with biennial screening and 7 with annual N° of studies, screening for the same age group [21]. A similar difference countries between biennial and annual screening intervals was observed from an excess absolute risk model of radiation induced BC [25]. represents the range of point estimates reported across studies. – – – Benefits and harms in women aged 70–74 (Tables 3/4) Observational studies. Three studies provided estimations of advanced BC stage (IIB–IV) in older women, using population 13 more to to 3 more reduction Absolute registries but for different age ranges (i.e. 66–89 [30], 70–85 [37] 22 more 2 more 1 more and 70–89 years [40]). In the age group of 70–85, the proportion of tumours at stage IIB–IV were no different among newly diagnosed BC with a history of biennial or annual screening (OR 0.98 95%CI 0.76–1.27) [37]. N° of events One study estimated that the 10-year cumulative probability of A: 27–49 B: 14–27 per arm false positive results for women between the ages of 75 and 89 A: 3–7 B: 2–4 Annual vs. Biennial may be higher with annual screening (47%, 95%CI 44.9–49.5%) estimates from 50 to 69 years). A: 6 B: 4 compared to biennial screening (26.6%, 95%CI 25.7–27.5%) [30]. Death by radiation induced BC The cumulative probability of false positive biopsy recommenda- 2 (Canada, 2 (Canada, 1 (US) [21] countries tions may also be higher for annual screening (9%, 95%CI 8–11%) included in the table. [21, 25]‡ [21, 25]‡ studies, Table 4 continued compared to biennial screening (4%, 95%CI 4–5%) [30]. N° of US) US) Modelling studies. The estimated difference for BC deaths Age group between the different intervals might be small. A microsimulation model estimated the number of BC deaths averted for annual, 45–49* 50–69 50–69 biennial and triennial screening to be 100, 90 and 80, respectively, ## compared to no screening per 100,000 screened Canadian women British Journal of Cancer (2022) 126:673 – 688 C. Canelo-Aybar et al. 685 [26]. This result was consistent with the one reported in a non- available evidence was mostly of very low certainty and precludes individual model for a Spanish cohort which showed almost us from reaching firm conclusions. In women 50–69 years old, similar benefits for the three screening intervals (unpublished annual compared to biennial screening may have small additional data) [27], and a small number of QALYs gained since life benefits but an important increase in false positive results. expectancy is lower in this age group. Triennial compared to biennial screening suggests the latter Only one non-individual based model estimated overdiagnosis provides more benefits but also some additional harms. In for this age group and it showed a small increasing trend with younger women (45–49), the more frequent screening intervals shorter screening intervals from 193 for triennial screening to 269 (going from biennial to annual screening) provides smaller for annual screening [52]. incremental benefits (i.e. number of BC deaths averted), nearly similar incremental estimates of overdiagnosis and slightly more Risk of bias and certainty of the evidence incremental harms (i.e. false positive results and false positive Overall, the certainty of the evidence was very low, and therefore biopsies recommendations from observational studies) than in the differences observed between the possible combinations of women 50–69 years of age. Thus the overall balance between screening intervals and age groups are uncertain. The exemption benefits and harms is more favourable in the latter age group. was the evidence from the only RCT included in this systematic Finally, among women aged 70–74, the smaller incremental harms review which was downgraded to moderate certainty due to and similar benefits with shorter screening intervals suggests that imprecision [19]. longer intervals probably have a more favourable overall balance, The evidence from observational studies was limited among but the difference may be small. other factors by indirectness as for the age group of 45–49 we We observed sparse data, especially in older women and for only identified studies including a broader age range from 40 to critical outcomes, such as BC mortality or disease stage at 49 years of age at the time of invitation to screening, and from diagnosis. The only included RCT showed that annual screening, some studies we had to extract results from specific subgroups of compared to triennial screening, probably reduces BC mortality in women (e.g. normal weight or white women). All secondary women 50–62 years of age. Observational evidence consisted of analysis from surveillance registries were also subject to mis- population registries from different time periods with high classification bias of the interventions as the periodicity of uncertainty. We considered modelling evidence when empirical screening was assigned based on different time ranges that evidence was not available. However, its certainty was very low elapsed between the two latest mammographies prior to due to indirectness, since data for input parameters mostly come diagnosis. Additionally, US studies used opportunistic screening, from opportunistic screening settings,. Model studies suggested thus women might have anticipated or delayed the mammo- that in women aged 50–69 the benefits with annual screening graphy due to preferences or indications given by radiologists. may be a bit larger but may also be associated to relevant harms, We decided that for modelling studies, our GRADE assessment including the possibility of a small increase of new BC lesions departed from low certainty after considering methodological induced by radiation exposure; thus, biennial screening may limitations of key input evidence (i.e. mammography sensitivity provide a more favourable balance, while in other age groups the estimated from BCSC registries including women from wider age potential benefits gains with more frequent screening intervals groups than our clinical question and with a clinical follow-up may be smaller. restricted to only one year [53], or no formal assessment of risk of bias in the individual-patient-data meta-analysis used to inform Our results in the context of previous research treatment effectiveness [54]) and that credibility assessment of Our results are broadly consistent but more comprehensive than model development was limited due to suboptimal reporting. previous reviews. The USPFTF based their assessment on one There was also limited reporting of formal sensitivity analysis to modelling study (included in our review), concluding that when assess the impact of input data assumptions on the simulated moving from biennial to annual mammography, regardless of the events [21, 24, 25]. We had concerns about indirectness given that starting age, there is a small increase in averted deaths but with a most models used observational data from the US to inform their large increase of harms [7]. A systematic review conducted by the input parameters (i.e. radiation induced BC), and because in one American Cancer Society included an indirect comparison modelling study data was only available by different levels of between RCTs and a model study from the CISNET collaboration, breast density (i.e. scattered fibroglandular density) [22]. Finally, concluded that beginning screening with more frequent intervals one study providing unpublished data (Vilaprinyo 2014) [27] likely results in a greater mortality reduction but the magnitude is reported fewer BC deaths averted with annual compared to uncertain [55]. biennial or triennial screening in the age group of 45–49 years. The modelling estimates of harms due to overdiagnosis This result was not internally consistent (i.e. annual screening had remains a matter of debate as there is no consensus on the the largest number of BC deaths averted from 45 to 69) and methods to quantify this outcome [56], and many assumptions differed from other studies or bodies of evidence; thus we are made, including the clinical impact of DCIS and the included this result cautiously only if other studies were not probability of some cancers to spontaneously regress [50]. It is available (Table 4). worth noting that there is also considerable uncertainty in the The detailed risk of bias assessment per study is available under evidence coming from RCTs. For example, a review including request. The evidence profiles for all age groups and intervals only studies that did not invite women of the control group to comparisons describing the reasons for downgrading the certainty screening at the end of the trial period, reported a relevant of evidence are available from Supplementary Tables S5–S13. In proportion of overdiagnosis [57]. However, the UK age trial the evidence profiles we prioritised the reporting of evidence from showed that the cumulative incidence of invasive cancers was observational/randomised studies over modelling studies (i.e. false similar, if not higher, in women who underwent only one positive results). mammogram after the age of 50 compared to women who underwent annual mammography from 40 to 49, and then entered a triennial screening programme [58]. DISCUSSION The cost-effectiveness of implementing different screening Main findings intervals has been studied in few microsimulation models. One Our systematic review shows that in women of average breast study assessed the impact of extending the Dutch screening cancer risk, screening intervals may have different trade-offs programme in women under 50, showing that biennial strategies between benefits and harms for each age group. However, the were cost-effective while other alternatives, such as annual British Journal of Cancer (2022) 126:673 – 688 C. Canelo-Aybar et al. 686 screening starting at 45, resulted in less favourable incremental screening. For example, previous research has highlighted that breast cost-effectiveness ratios (ICERs) [59]. However, the study used an density influences both mammography accuracy and risk of 80% adherence to screening [59], which might have influenced developing breast cancer [64, 65]. For further information on the the relative trade-offs between different screening intervals, as complete recommendations formulated in the European Guidelines previously described [22]. In women from the US between 50 and on Breast Cancer Screening and Diagnosis, please visit the ECIBC 74 years of age, with different breast densities and individual risk website (https://healthcare-quality.jrc.ec.europa.eu/european-breast- level of developing BC, triennial strategies were considered cost- cancer-guidelines/screening-ages-and-frequencies). effective (at a threshold of $100 000 per QALY) for subgroups with average risk and low breast density, while biennial strategies were cost-effective for other breast density subgroups at an average or DISCLAIMER intermediate risk [22]. All views expressed in this article are strictly those of the authors. Limitations and strengths Although we included only English language articles, the risk of DATA AVAILABILITY selection bias is probably small as we also screened previous All data sources used during this study are described in this published article and its systematic reviews and consulted the GDG experts, not identifying additional information files. The datasets analysed are available from the additional studies. Some results are not directly transferable to the corresponding author on reasonable request. European context; for example the cumulative 10-year false positive rates from US studies are higher than those reported in organised REFERENCES European screening programmes. However, we assumed that the 1. Ferlay J, Ervik M, Lam F, Colombet M, Mery L, Piñeros M et al. Global Cancer difference between intervals would be more comparable across Observatory: Cancer Today. Lyon, France: International Agency for Research on different settings. The scarce available empirical evidence to evaluate Cancer. Available from: https://gco.iarc.fr/today, accessed [25-07-2019]. 2018. the trade-offs between benefits and harms limited our conclusions. 2. ECIS. European Cancer Information System From https://ecis.jrc.ec.europa.eu, We therefore included modelling evidence to complement the gaps accessed on 20-July-2019. 2018. in the evidence, an approach that is recommended for interventions 3. Duffy SW, Tabar L, Yen AM, Dean PB, Smith RA, Jonsson H, et al. Mammography such as population screening [60]. screening reduces rates of advanced and fatal breast cancers: Results in 549,091 women. Cancer. 2020;126:2971–9. Implications for practice and research 4. Borrelli C, Cohen S., Duncan A, Given-Wilson R, Jenkins J, Kearins O, et al. NHS Our findings may have different implications for practice Breast Screening Programme. Clinical guidance for breast cancer screening assessment. Fourth edition Nov 2016. depending on the age group, the balance between benefits and 5. Siu AL, Force, U. S. P. S. T. Screening for breast cancer: U.S. Preventive services harms, available resources for public health services, and how task force recommendation statement. Ann Intern Med. 2016;164:279–96. women value the different outcomes. In the case of women 6. Oeffinger KC, Fontham ET, Etzioni R, Herzig A, Michaelson JS, Shih YC, et al. Breast invited to an opportunistic screening programme (or considering cancer screening for women at average risk: 2015 guideline update from the screening) a shared decision-making process to carefully explain American Cancer Society. JAMA. 2015;314:1599–614. the pros and cons of each decision is warranted. Similarly, given 7. Mandelblatt JS, Stout NK, Schechter CB, van den Broek JJ, Miglioretti DL, Krapcho the low certainty of evidence and the variability and uncertainty of M, et al. Collaborative modeling of the benefits and harms associated with dif- how women value outcomes at stake, guideline panellists are ferent U.S. Breast Cancer Screening Strategies. Ann Intern Med. 2016;164:215–25. likely to formulate conditional recommendations, as opposed to 8. Kerlikowske K, Zhu W, Hubbard RA, Geller B, Dittus K, Braithwaite D, et al. Out- comes of screening mammography by frequency, breast density, and post- strong ones. The scope of this review is determined by the menopausal hormone therapy. JAMA Intern Med. 2013;173:807–16. European Breast Guidelines screening recommendations; [10] 9. Schunemann HJ, Lerda D, Dimitrova N, Alonso-Coello P, Grawingholt A, Quinn C, thus, policy makers should note that we did not include modelling et al. Methods for development of the European Commission Initiative on Breast estimates for women between the ages of 40 and 44 as screening Cancer Guidelines: recommendations in the era of guideline transparency. Ann is not suggested in this age group [10]. Also, readers should be Intern Med. 2019;171:273–80. careful when interpreting the effects of screening intervals across 10. Schunemann HJ, Lerda D, Quinn C, Follmann M, Alonso-Coello P, Rossi PG, et al. the different age groups, as comparisons are limited by the small Breast cancer screening and diagnosis: a synopsis of the European Breast number of screening rounds in the 45 to 49 and 70 to 74 age Guidelines. Ann Intern Med. 2019;172:46–56. groups, compared to the 50–69 age group. 11. Guyatt GH, Oxman AD, Kunz R, Atkins D, Brozek J, Vist G. et al. GRADE guidelines: 2. Framing the question and deciding on important outcomes. J Clin Recommendations about mammography screening intervals Epidemiol.2011;64:395–400. will also depend on the magnitude and relative importance of 12. Higgins JP, Altman DG, Gotzsche PC, Juni P, Moher D, Oxman AD, et al. The potential harms. Narayan, et al. assessed to what extent harms Cochrane Collaboration’s tool for assessing risk of bias in randomised trials. BMJ. should decrease in order to make a screening interval with an 2011;343:d5928. unfavourable balance of benefits and harms acceptable [61]. They 13. Sterne JA, Hernan MA, Reeves BC, Savovic J, Berkman ND, Viswanathan M, et al. found that for annual screening a reduction of 31% false positive ROBINS-I: a tool for assessing risk of bias in non-randomised studies of inter- results would be required to support a recommendation in favour ventions. BMJ. 2016;355:i4919. of starting at 50, although this was in the context of false positive 14. Jaime Caro J, Eddy DM, Kan H, Kaltz C, Patel B, Eldessouki R, et al. Questionnaire rates prevailing in the US [61]. Policy makers should probably to assess relevance and credibility of modeling studies for informing health care decision making: an ISPOR-AMCP-NPC Good Practice Task Force report. Value consider implementing interventions to improve mammography Health. 2014;17:174–82. performance, mitigating concerns about potential harms. For 15. Guyatt GH, Oxman AD, Schunemann HJ, Tugwell P, Knottnerus A. GRADE example previous studies suggest that comparing mammograms guidelines: a new series of articles in the Journal of Clinical Epidemiology. J Clin. with prior exams can significantly reduce the recall rate while Epidemiol. 2011;64:380–2. maintaining the same detection rate [62, 63]. 16. Guyatt GH, Oxman AD, Vist GE, Kunz R, Falck-Ytter Y, Alonso-Coello P, et al. Several research priorities were identified during this review, with GRADE: an emerging consensus on rating quality of evidence and strength of feedback from the GDG experts, such as need for: (i) empirical recommendations. BMJ. 2008;336:924–6. research on the effectiveness of the different screening intervals due 17. Hultcrantz M, Rind D, Akl EA, Treweek S, Mustafa RA, Iorio A, et al. The GRADE to the current very low certainty of evidence; ii) cost-effectiveness Working Group clarifies the construct of certainty of evidence. J Clin Epidemio. 2017;87:4–13. studies using unitary costs from different settings, and in particular for 18. Duffy SW, B. R. Long term mortality results from the UK screening frequency trial. women aged 45 to 49, iii) assessment of alternative imaging EJC Supplements. 2008;6:48. modalities, iv) tailored screening according to risk vs population British Journal of Cancer (2022) 126:673 – 688 C. Canelo-Aybar et al. 687 19. Breast Screening Frequency Trial, G. The frequency of breast cancer screening: 43. Hubbard RA, Miglioretti DL, Smith RA. Modelling the cumulative risk of a false- results from the UKCCCR Randomised Trial. United Kingdom Co-ordinating positive screening test. Stat Methods Med Res. 2010;19:429–49. Committee on Cancer Research. Eur J Cancer. 2002;38:1458–64. 44. Lee SJ, Li X, Huang H, Zelen M. The Dana-Farber CISNET model for breast cancer 20. Gunsoy NB, Garcia-Closas M, Moss SM. Estimating breast cancer mortality screening strategies: an update. Med Decis Making. 2018;38:44S–53S. reduction and overdiagnosis due to screening for different strategies in the 45. van den Broek JJ, van Ravesteyn NT, Heijnsdijk EA, de Koning HJ. Simulating the United Kingdom. Br J Cancer. 2014;110:2412–9. Impact of Risk-Based Screening and Treatment on Breast Cancer Outcomes with 21. Miglioretti DL, Lange J, van den Broek JJ, Lee CI, van Ravesteyn NT, Ritley D, et al. MISCAN-Fadia. Med Decis Making. 2018;38:54S–65S. Radiation-induced breast cancer incidence and mortality from digital mammo- 46. Schechter CB, Near AM, Jayasekera J, Chandler Y, Mandelblatt JS. Structure, graphy screening: a modeling study. Ann Intern Med. 2016;164:205–14. function, and applications of the Georgetown-Einstein (GE) Breast Cancer 22. Trentham-Dietz A, Kerlikowske K, Stout NK, Miglioretti DL, Schechter CB, Ergun Simulation Model. Med Decis Making. 2018;38:66S–77S. MA, et al. Tailoring Breast Cancer Screening Intervals by Breast Density and Risk 47. Huang X, Li Y, Song J, Berry DA. A Bayesian Simulation Model for Breast Cancer for Women Aged 50 Years or Older: Collaborative Modeling of Screening Out- Screening, Incidence, Treatment, and Mortality. Med Decis Making. 2018 Apr;38 comes. Ann intern Med. 2016;165:700–12. (1_suppl):78S–88S. 23. Tsunematsu M, Kakehashi M. An analysis of mass screening strategies using a 48. Plevritis SK, Sigal BM, Salzman P, Rosenberg J, Glynn P. A stochastic simulation mathematical model: comparison of breast cancer screening in Japan and the model of U.S. breast cancer mortality trends from 1975 to 2000. J Natl Cancer Inst United States. J Epidemiol. 2015;25:162–71. Monogr. 2006; 86–95, https://doi.org/10.1093/jncimonographs/lgj012. 24. van Ravesteyn NT, Miglioretti DL, Stout NK, Lee SJ, Schechter CB, Buist DS, et al. 49. Alagoz O, Ergun MA, Cevik M, Sprague BL, Fryback DG, Gangnon RE, et al. The Tipping the balance of benefits and harms to favor screening mammography University of Wisconsin breast cancer epidemiology simulation Model: an update. starting at age 40 years: a comparative modeling study of risk. Ann Intern Med. Med Decis Making. 2018;38:99S–111S. 2012;156:609–17. 50. Mandelblatt JS, Near AM, Miglioretti DL, Munoz D, Sprague BL, Trentham-Dietz A, 25. Yaffe MJ, Mainprize JG. Risk of radiation-induced breast cancer from mammo- et al. Common model inputs used in CISNET collaborative breast cancer mod- graphic screening. Radiology. 2011;258:98–105. eling. Med Decis Making. 2018;38:9S–23S. 26. Yaffe MJ, Mittmann N, Lee P, Tosteson AN, Trentham-Dietz A, Alagoz O, et al. 51. van den Broek JJ, van Ravesteyn NT, Mandelblatt JS, Cevik M, Schechter CB, Lee Clinical outcomes of modelling mammography screening strategies. Health Rep. SJ, et al. Comparing CISNET breast cancer models using the maximum clinical 2015;26:9–15. incidence reduction methodology. Med Decis Making. 2018;38:112S–125S. 27. Vilaprinyo E, Forne C, Carles M, Sala M, Pla R, Castells X, et al. Cost-effectiveness 52. Carles M, Vilaprinyo E, Cots F, Gregori A, Pla R, Roman R, et al. Cost-effectiveness and harm-benefit analyses of risk-based screening strategies for breast cancer. of early detection of breast cancer in Catalonia (Spain). BMC Cancer. 2011;11:192. PLoS One. 2014;9:e86858. 53. Lehman CD, Arao RF, Sprague BL, Lee JM, Buist DS, Kerlikowske K, et al. National 28. Mittmann N, Stout NK, Tosteson ANA, Trentham-Dietz A, Alagoz O, Yaffe MJ. Cost- performance benchmarks for modern screening digital mammography: update effectiveness of mammography from a publicly funded health care system per- from the breast cancer surveillance consortium. Radiology. 2017;283:49–58. spective. CMAJ Open. 2018;6:E77–E86. 54. Early Breast Cancer Trialists’ Collaborative, G, Peto R, Davies C, Godwin J, Gray R, 29. Arnold M, Pfeifer K, Quante AS. Is risk-stratified breast cancer screening eco- Pan HC, et al. Comparisons between different polychemotherapy regimens for nomically efficient in Germany? PLoS ONE. 2019;14:e0217213. early breast cancer: meta-analyses of long-term outcome among 100,000 women 30. Braithwaite D, Zhu W, Hubbard RA, O’Meara ES, Miglioretti DL, Geller B, et al. in 123 randomised trials. Lancet. 2012;379:432–44. Screening outcomes in older US women undergoing multiple mammograms in 55. Myers ER, Moorman P, Gierisch JM, Havrilesky LJ, Grimm LJ, Ghate S, et al. community practice: does interval, age, or comorbidity score affect tumor char- Benefits and harms of breast cancer screening: a systematic review. JAMA. acteristics or false positive rates? J Natl Cancer Inst. 2013;105:334–41. 2015;314:1615–34. 31. Coldman AJ, Phillips N, Olivotto IA, Gordon P, Warren L, Kan L. Impact of chan- 56. Ripping TM, Ten Haaf K, Verbeek ALM, van Ravesteyn NT & Broeders MJM. ging from annual to biennial mammographic screening on breast cancer out- Quantifying overdiagnosis in cancer screening: a systematic review to evaluate comes in women aged 50-79 in British Columbia. J Med Scree. 2008;15:182–7. the methodology. J Natl Cancer Inst. 2017;109:djx060. 32. Dittus K, Geller B, Weaver DL, Kerlikowske K, Zhu W, Hubbard R, et al. Impact of 57. Canelo-Aybar C, Ferreira DS, Ballesteros M, Posso M, Montero N, Sola I et al. mammography screening interval on breast cancer diagnosis by menopausal Benefits and harms of breast cancer mammography screening for women at status and BMI. J Gen Intern Med. 2013;28:1454–62. average risk of breast cancer: a systematic review for the European Commission 33. Goel A, Littenberg B, Burack RC. The association between the pre-diagnosis Initiative on Breast Cancer. J. Med Screen. 2021; https://doi.org/10.1177/ mammography screening interval and advanced breast cancer. Breast Cancer Res 0969141321993866. Treat. 2007;102:339–45. 58. Moss SM, Wale C, Smith R, Evans A, Cuckle H, Duffy SW. Effect of mammographic 34. Hubbard RA, Kerlikowske K, Flowers CI, Yankaskas BC, Zhu W, Miglioretti DL. screening from age 40 years on breast cancer mortality in the UK Age trial at 17 Cumulative probability of false-positive recall or biopsy recommendation after 10 years’ follow-up: a randomised controlled trial. Lancet. Oncol. 2015;16:1123–32. years of screening mammography: a cohort study. Ann Intern Med. 59. Sankatsing VD, Heijnsdijk EA, van Luijt PA, van Ravesteyn NT, Fracheboud J, de 2011;155:481–92. Koning HJ. Cost-effectiveness of digital mammography screening before the age 35. Hunt KA, Rosen EL, Sickles EA. Outcome analysis for women undergoing annual of 50 in The Netherlands. Int J Cancer. 2015;137:1990–9. versus biennial screening mammography: a review of 24,211 examinations. AJR 60. Habbema JD, Wilt TJ, Etzioni R, Nelson HD, Schechter CB, Lawrence WF, et al. Am J Roentgenol. 1999;173:285–9. Models in the development of clinical practice guidelines. Ann Intern Med. 36. Klemi PJ, Toikkanen S, Rasanen O, Parvinen I, Joensuu H. Mammography 2014;161:812–8. screening interval and the frequency of interval cancers in a population-based 61. Narayan AK, Elkin EB, Lehman CD, Morris EA. Quantifying performance thresholds screening. Br J Cancer. 1997;75:762–6. for recommending screening mammography: a revealed preference analysis of 37. Miglioretti DL, Zhu W, Kerlikowske K, Sprague BL, Onega T, Buist DS, et al. Breast USPSTF guidelines. Breast Cancer Res Treatment. 2018;172:463–8. tumor prognostic characteristics and biennial vs annual mammography, age, and 62. Hayward JH, Ray KM, Wisner DJ, Kornak J, Lin W, Joe BN, et al. Improving menopausal status. JAMA Oncol. 2015;1:1069–77. screening mammography outcomes through comparison with multiple prior 38. O’Meara ES, Zhu W, Hubbard RA, Braithwaite D, Kerlikowske K, Dittus KL, et al. mammograms. AJR Am J Roentgenol. 2016;207:918–24. Mammographic screening interval in relation to tumor characteristics and false- 63. Roelofs AA, Karssemeijer N, Wedekind N, Beck C, van Woudenberg S, Snoeren PR, positive risk by race/ethnicity and age. Cancer. 2013;119:3959–67. et al. Importance of comparison of current and prior mammograms in breast 39. Parvinen I, Chiu S, Pylkkanen L, Klemi P, Immonen-Raiha P, Kauhava L, et al. cancer screening. Radiology. 2007;242:70–77. Effects of annual vs triennial mammography interval on breast cancer incidence 64. McCormack VA, dos Santos Silva I. Breast density and parenchymal patterns as and mortality in ages 40-49 in Finland. Br J Cancer. 2011;105:1388–91. markers of breast cancer risk: a meta-analysis. Cancer Epidemiol Biomarkers Prev. 40. White E, Miglioretti DL, Yankaskas BC, Geller BM, Rosenberg RD, Kerlikowske K, 2006;15:1159–69. et al. Biennial versus annual mammography and the risk of late-stage breast 65. Boyd NF, Guo H, Martin LJ, Sun L, Stone J, Fishell E, et al. Mammographic density cancer. J Natl Cancer Inst. 2004;96:1832–9. and the risk and detection of breast cancer. N Engl J Med. 2007;356:227–36. 41. Sanderson M, Levine RS, Fadden MK, Kilbourne B, Pisu M, Cain V, et al. Mam- mography screening among the elderly: a research challenge. Am J Med. 2015;128:1362 e1367–1314. ACKNOWLEDGEMENTS 42. McGuinness JE, Ueng W, Trivedi MS, Yi HS, David R, Vanegas A, et al. Factors It is with genuine gratitude and warm regard that we would like to dedicate this associated with false positive results on screening mammography in a popula- publication to our dearest colleague Sue Warman who passed away on 21–06–2021. The tion of predominantly Hispanic women. Cancer Epidemiol Biomarkers Prev. authors would like to sincerely thank all members of the Guidelines Development Group 2018;27:446–53. of the European Commission Initiative on Breast Cancer for their participation in the British Journal of Cancer (2022) 126:673 – 688 C. Canelo-Aybar et al. 688 discussions generated by this systematic review which led to the different recommenda- Margarita Posso, Ivan Solà and Pablo Alonso-Coello are employees of the tions they developed in the European Guidelines on Breast Cancer Screening and Iberoamerican Cochrane Collaboration. Stephen W. Duffy, Markus Follmann, Axel diagnosis (https://healthcare-quality.jrc.ec.europa.eu/european-breast-cancer-guidelines). Gräwingholt and Paolo Giorgi Rossi are GDG members of the ECIBC guidelines. AUTHOR CONTRIBUTIONS ADDITIONAL INFORMATION Carlos Canelo-Aybar, Margarita Posso, Nadia Montero, Ivan Sol and Pablo Alonso-Coello Supplementary information The online version contains supplementary material were responsible for conducting the systematic review. Carlos Canelo-Aybar and available at https://doi.org/10.1038/s41416-021-01521-8. Margarita Posso conducted the search, data extraction, and analysis. Zuleika Saz- Parkinson, Paolo Giorgi Rossi, Stephen W. Duffy, Markus Follmann, Stephen W. Duffy, Correspondence and requests for materials should be addressed to Carlos Canelo- Markus Follmann, Axel Gräwingholt and Paolo Giorgi Rossi contributed to the definition of Aybar or Zuleika Saz-Parkinson the research protocol, and provided comments to the preliminary results of the systematic review. Carlos Canelo-Aybar drafted the first version of the article. All authors Reprints and permission information is available at http://www.nature.com/ contributed to the interpretation and reporting of the results and provided comments on reprints subsequent versions of the article. All authors read and approved the final manuscript prior submission. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. FUNDING INFORMATION The systematic review was carried out by Iberoamerican Cochrane Collaboration under the Framework contract 443094 for procurement of services between the European Commission’s Joint Research Centre and Asociación Colaboración Open Access This article is licensed under a Creative Commons Cochrane Iberoamericana. Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party ETHICS APPROVAL AND CONSENT TO PARTICIPATE material in this article are included in the article’s Creative Commons license, unless Not applicable. indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory CONSENT TO PUBLISH regulation or exceeds the permitted use, you will need to obtain permission directly Not applicable. from the copyright holder. To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/. COMPETING INTERESTS Zuleika Saz Parkinson was an employee of the Joint Research Centre, European © European Union 2021 Commission when this systematic review was carried out. Carlos Canelo-Aybar, British Journal of Cancer (2022) 126:673 – 688