www.nature.com/bjc ARTICLE Epidemiology A systematic review and quality assessment of individualised breast cancer risk prediction models Javier Louro1,2,3, Margarita Posso1,2, Michele Hilton Boon4, Marta Román1,2, Laia Domingo1,2, Xavier Castells 1,2 and María Sala1,2 BACKGROUND: Individualised breast cancer risk prediction models may be key for planning risk-based screening approaches. Our aim was to conduct a systematic review and quality assessment of these models addressed to women in the general population. METHODS: We followed the Cochrane Collaboration methods searching in Medline, EMBASE and The Cochrane Library databases up to February 2018. We included studies reporting a model to estimate the individualised risk of breast cancer in women in the general population. Study quality was assessed by two independent reviewers. Results are narratively summarised. RESULTS: We included 24 studies out of the 2976 citations initially retrieved. Twenty studies were based on four models, the Breast Cancer Risk Assessment Tool (BCRAT), the Breast Cancer Surveillance Consortium (BCSC), the Rosner & Colditz model, and the International Breast Cancer Intervention Study (IBIS), whereas four studies addressed other original models. Four of the studies included genetic information. The quality of the studies was moderate with some limitations in the discriminative power and data inputs. A maximum AUROC value of 0.71 was reported in the study conducted in a screening context. CONCLUSION: Individualised risk prediction models are promising tools for implementing risk-based screening policies. However, it is a challenge to recommend any of them since they need further improvement in their quality and discriminatory capacity. British Journal of Cancer (2019) 121:76–85; https://doi.org/10.1038/s41416-019-0476-8 BACKGROUND organised screening programmes do not use these models Mammography screening has been associated with a reduction in routinely. One reason for not including these models in screening breast cancer mortality and therefore organised breast cancer context is the high uncertainty with regards to its applicability in screening programmes using mammography have been well screening settings. Also, the emergence of new risk prediction established worldwide.1–4 Although there is not a single consensus, factors such as the expression of single nucleotide polymorphisms current screening programmes generally recommend biennial or (SNPs) needs to be appropriately summarised before recommend- triennial screening in Europe and annual or biennial screening in ing one of the models into screening practice. the US with variations in the recommended targeted age.2–5 These Like any other source of information, risk prediction models recommendations usually consider age as the sole risk factor have limitations that should be evaluated before using them. A leading women to be invited for screening from age 40–50 until rigorous risk of bias assessment of the existing individualised risk age 70–74, depending on the programmes. models is needed to clarify the overall quality and applicability of The likelihood that a woman will benefit from screening each model. Therefore, the aim of this systematic review is to mammography depends on her risk for developing clinically update the existing evidence, conduct a critical appraisal and risk significant breast cancer in her lifetime. Taking individual risk of bias assessment and summarise the results of the individualised factors beyond age into account should enable the classification risk models which are used to estimate the risk of breast cancer in of women into groups at varying risk of breast cancer. women in the general population. Personalised risk-based screening going beyond the current ‘one-size fits all' recommendation may increase the effectiveness and benefit-harm balance of breast cancer screening. Individua- METHODS lised risk prediction models for breast cancer are a key element to Data sources and searches develop risk-based screening approaches since they are designed We performed a systematic review of the literature following the to quantify the risk that can predict whether an individual woman standard Cochrane Collaboration methods8 and adhering to the would develop breast cancer in a defined period.6 PRISMA statement reporting recommendations.9 A predetermined A number of risk prediction models that include classical review protocol was registered (CRD42018089842) in the PROS- risk factors are commonly used in clinical contexts.7 However, PERO database (date of registration 1 March 2018). The Patient, 1 Department of Epidemiology and Evaluation, IMIM (Hospital del Mar Medical Research Institute), Barcelona, Spain; 2Research Network on Health Services in Chronic Diseases (REDISSEC), Barcelona, Spain; 3European Higher Education Area (EHEA) Doctoral Programme in Methodology of Biomedical Research and Public Health in Department of Pediatrics, Obstetrics and Gynecology, Preventive Medicine and Public Health, Universitat Autónoma de Barcelona (UAB), Bellaterra, Barcelona, Spain and 4MRC/CSO Social and Public Health Sciences Unit, University of Glasgow, Glasgow, UK Correspondence: Margarita Posso (mposso@parcdesalutmar.cat) Received: 10 January 2019 Accepted: 25 April 2019 Published online: 22 May 2019 © The Author(s) 2019 Published by Springer Nature on behalf of Cancer Research UK A systematic review and quality assessment of individualised breast. . . J Louro et al. 77 Intervention, Comparison, Outcomes (PICO) question of this systematic review is the following: Should individualised breast MEDLINE EMBASE The Cochrane library cancer risk prediction models vs. no risk prediction models be (N = 2032) (N = 1718) (N = 216) used to develop risk-based screening approaches for women in the general population? We retrieved relevant literature by using a combination of controlled vocabulary and keyword search terms in the following databases: (i) Medline (accessed through PubMed); (ii) The Records identified through database searching Cochrane Library; and (iii) EMBASE (accessed through Ovid). (N = 3966) Terms related to breast cancer recurrence were excluded in order to avoid retrieving citations out of the scope of this systematic review. We adapted the search algorithms to the requirements of Duplicated records each database and used validated filters to retrieve systematic (N = 922) reviews and primary studies as needed. We reviewed references of included studies that could potentially fulfil our eligibility criteria. The detailed search strategy is reported in Supplementary table 1. Records screened by title and abstract (N = 2974) We searched primary studies of individualised breast cancer risk models searching each database from its inception up to February 2018. Excluded records (N = 2895) Study selection Eligible studies were those published in English that reported a model to estimate the individualised risk of breast cancer in Full text articles assessed for eligibility (N = 79) women in the general population. We included models that assessed more than one risk factor and reported the quantitative 1234567890();,: characteristics of the risk prediction model. If multiple publications were based on the same individualised risk model, the most Full text articles excluded extensive report of the model in terms of risk factors reported was (N = 57) chosen. We excluded external validation studies that replicated (Did not assess a new risk prediction model, N = 23) (Secondary publication, N = 18) previous models without adding any additional information such (Wrong setting, population, N = 8) as a new design for collecting the inputs data, modifications on (Not risk prediction model, N = 6) (Wrong study outcomes, N = 2) the risk factors or the risk model method. Articles identified from the search were loaded into EndNote X7.7.1 for Windows (2008, Version 12.0.4) and duplicates were Additional records removed. identified through other sources (N = 2) Data extraction and quality assessment One reviewer screened the search results based on title and abstract, and a second reviewer performed a quality check of the Studies included in the synthesis (N = 24) study screening by reviewing 20% of the references. Two reviewers independently confirmed eligibility based on the full Fig. 1 PRISMA flowchart text of the relevant articles. In case of disagreement between researchers, the inclusion of studies was determined by con- sensus. We reported the result of this process with a PRISMA operating characteristic curve (AUROC), the net reclassification flowchart (Fig. 1). index (NRI) and the expected observed (E/O) ratio. The NRI was We used a predefined form to extract the following information not included in the tables because it was only reported in 2 out of from included studies: author, publication date, country, study 24 articles. The characteristics of the included models and the risk design, the name of the model if available, sample characteristics, prediction outcomes reported preclude the possibility to pool sample size, type of breast cancer, the method of analysis, and data across studies. Therefore, a narrative synthesis has been validation of the model. Data abstraction was conducted by one conducted. Key study characteristics, validation and accuracy of reviewer and checked by another. individual risk models, and methodological quality are described Two reviewers carried out the assessment of the risk of bias in tables and summarised in a narrative manner. Results are independently and final quality assessment was based on presented according to the original model that they reported. consensus. We used the ISPOR-AMCP-NPC Questionnaire10 to assess the relevance and credibility of each risk prediction study and the following sources of limitations: (i) internal and external RESULTS validation; (ii) bias due to the study design for risk estimates; (iii) Study inclusion limitations in data inputs; (iv) appropriateness of the model The database searches for primary studies retrieved 2974 citations, analysis; (v) reporting bias; (vi) interpretation bias; and (vii) conflict of which 79 were considered potentially relevant. These 79 studies of interest. The risk of bias for each domain was rated as low, high were screened in full text. We found a systematic review of or unclear. For systematic reviews we used the AMSTAR 2 critical Anothaisintawee et al.,7 which we used as a source of primary appraisal tool.11 studies. In addition, two studies were included after a manual inspection of papers’ references.12,13 After the full text was Data synthesis and analysis checked, 24 studies12–35 met the inclusion criteria and were We evaluated the model validation by assessing both the considered in the evidence synthesis. Details about study discriminative power and the calibration accuracy estimated for inclusion with reasons for exclusion are described in the flow- the women in the general population. When available in the chart (Fig. 1), and a list of references to excluded studies is included publication, we extracted the area under the receiver provided in Supplementary table 2. A systematic review and quality assessment of individualised breast. . . J Louro et al. 78 Characteristics of the included studies a. BCRAT model. The first BCRAT model publication did not The included studies can be grouped according to the risk report the AUROC, however, later publications of this model model that they reported, the Breast Cancer Risk Assessment reported a range that varied from 0.56 to 0.68. The three Tool (BCRAT), the Breast Cancer Surveillance Consortium (BCSC), publications that included the original risk factors, age, the Rosner & Colditz model, the International Breast family history of breast cancer, age at first birth, menarche, Cancer Intervention Study (IBIS), and other original models. and previous biopsies, reported low AUROC values, 0.56 to The study by Zhang et al.13 is included in two of the 0.62.14,20,23 Similarly, the AUROC reported by Boyle et al.16 groups (BCRAT and Rosner & Colditz models) because it and Matsuno et al.25 were 0.60 and 0.61, although these provides information of both models and presents its results authors added BMI, HRT, alcohol, physical activity and diet, separately. A brief summary of the 24 included studies is and ethnicity into the model. Zhang et al.13 with the new presented in Table 1 and the extended characteristics in variables reach an AUROC of 0.65 and Tice et al.30 reported Supplementary table 3. in 2005 a higher AUROC value of 0.68 which was obtained just adding breast density to the original five risk factors a. Breast Cancer Risk Assessment Tool ‘BCRAT’ model. This (Table 1). Zhang et al.13 also reported the NRI to validate model was first published in the United States in 1989 that his model improved the previous ones with a assessing age, family history of breast cancer, age at first result of 8%. birth, menarche, and previous biopsies as risk factors for b. BCSC model. The published value of the AUROC for the predicting individualised breast cancer risk.22 After this first BCSC model was moderate, ranging from 0.64 to 0.69. Tice publication, eight studies were identified that were based et al. included age, family history, previous biopsies, breast on BRCAT model but modified the data collection design, density reported by the Breast Imaging Reporting and Data assessed additional risk factors or changed the statistical System (BI-RADS), and ethnicity into the model in 2008 and method. In addition to the five risk factors proposed in 1989, obtained a value of 0.66 for the AUROC.31 Instead of other variables such as body mass index (BMI), weight, BI-RADS, Kerlikowske et al. assessed changes in breast hormone replacement therapy (HRT), alcohol consumption, density obtaining a similar result, 0.64.24 Using previous physical activity, diet, breast density, atypical hyperplasia, benign breast disease, Tice et al. obtained a slightly higher breast inflammatory disease, parity, a polygenic risk score or AUROC value of 0.67 in 2015.32 More recently, in 2015 and hormones information have been included in updated 2016, Vachon et al.12 added to the model a polygenic risk versions (Table 1).13,14,16,17,20,23,25,26,30 score and Shieh et al.29 a combination between a b. Breast Cancer Surveillance Consortium ‘BCSC’ model. One polygenic risk score and BMI reporting a value of 0.69 relevant variation of the BCRAT model opens the path to and 0.65 for the AUROC respectively (Table 1). Vachon the emergence of the BCSC model first published by Tice et al.12 also demonstrated the improvement of discrimina- et al. in 2008 in the United States.31 In this study, Tice et al. tory accuracy estimating the NRI with a positive result used data from a cohort to create an individualised risk of 11%. prediction model that combines age, family history, c. Rosner & Colditz model. The discriminatory accuracy of this previous biopsies, breast density, and ethnicity. The BCSC model varied from 0.61 to 0.68. The authors assessed age, model has been further evaluated by other authors12,24,29,32 family history, age at first birth, menarche, BMI, benign and it currently includes previous benign breast diseases breast disease, menopause, HRT, age at subsequent births, and polygenetic risk score using SNPs as risk factors alcohol, and weight. They obtained an AUROC of 0.64 and (Table 1). 0.61 for ER + /PR + and ER-/PR- tumours, respectively.19 c. Rosner & Colditz model. Parallel to the BCSC model, another The addition of oestradiol levels to the model was tested model based on the ‘Nurses' Health Study’ cohort devel- by Rosner et al. who obtained a 0.65 AUROC value in oped by Rosner & Colditz in 1996 was also developed in the 2008.28 Finally the addition of a polygenic risk score, United States. This model currently includes 11 risk factors: mammographic density and endogenous hormones by age, menarche, menopause, age at first birth, age at Zhang et al.13 reached a 0.68 AUROC value (Table 1) and subsequent births, previous benign breast disease, HRT, obtained an improvement of the discriminative accuracy family history, weight, BMI, alcohol consumption, and also reflected in a NRI of a 9.5%. oestradiol levels.18,19,27,28 In the same way as in the BCRAT, d. IBIS model. The IBIS model original paper33 does not Zhang et al.13 analysed this model adding breast density, a include any validation and does not present the AUROC. polygenic risk score and endogenous hormones as risk Nevertheless, it has been externally validated showing an factors. AUROC of 0.57 which increases to 0.61 when adding d. International Breast Cancer Intervention Study ‘IBIS’ model. mammographic density.36 The IBIS model33 includes genetic information adding the e. Other models. Overall, the AUROC values of these models BRCA genes and a hypothetical susceptibility gene. were not higher than those shown by the above- e. Other models. Four studies reporting different models were mentioned models, varying from 0.62 to 0.64, although also identified.15,21,34,35 Apart from the above-mentioned they included a large number of risk factors. However, the risk factors, the models also assessed other variables such as model reported by Eriksson et al.21 did show an AUROC of abortion, breastfeeding, height, and previous mammogra- 0.71 that was the highest AUROC value identified in this phy results. Particularly relevant is the Eriksson model21 systematic review (Table 1). This model, in addition, is the since it was the only one targeted to the screening only one that estimates a 2-year risk, while the rest of population. In this study, the authors included risk factors models estimate the risk at a longer time horizon. This that were available at mammography screening examina- could explain the difference in AUROC values since it tion: age, BMI, HRT, family history, menopause, breast becomes more difficult to predict risk as the time horizon density, and presence of microcalcifications and/or masses increases. in the screen-mammogram. Discriminatory accuracy Calibration accuracy Fifteen out of the 24 studies reported the discriminatory accuracy Nine out of the 24 studies reported the calibration accuracy as the as the AUROC (Table 1 and Fig. 2). E/O ratio (Table 1). Table 1. Summary of included studies Study ID Targeted population Risk factors (number of categories) Discriminatory accuracy Calibration (E/O ratio)a (AUROC)a BCRAT model Banegas 2017 Hispanic, 25–79 years Age (2), Menarche (3), Previous biopsies (2), Age at first birth (3), First degree breast cancer (2) US-born: 0.56 US-born: 0.93b Foreign-born: 0.62 Foreign-born: 1.52b Boyle 2004 Caucasian, 20–74 years Age (n), Menarche (3), Age at first birth (3), First degree breast cancer (2), BMI (3), Alcohol (3), 0.6 1.03b Physical activity (3), HRT (2), Diet beta-carotene/vitE (5), Diet fruits/vegetables (5) Chen 2006 Caucasian, 35–74 years Age (2), Weight (6), Breast density (5), Menarche (3), Previous biopsies (3), Age at first birth (4), None None First degree breast cancer (3), Atypical hyperplasia (2) Decarli 2006 Caucasian, 20–74 years Age (2), Menarche (3), Previous biopsies (3), Age at first birth (4), First degree breast cancer (3) 0.59 0.96 Gail 1989 Caucasian, 20–79 years Age (2), Menarche (3), Previous biopsies (3), Age at first birth (4), First degree breast cancer (3) None None Gail 2007 African-American, 35–64 years Age (2), Menarche (3), Previous biopsies (3), Age at first birth (4), First degree breast cancer (3) 0.56 0.93b Matsuno 2011 Asian, 20–55 years Age (2), Menarche (3), Previous biopsies (3), Age at first birth (4), First degree breast cancer 0.61 0.85b (n), Ethnicity (6) Novotny 2006 Multiple ethnicities, 23–84 years Age (2), Menarche (3), Previous biopsies (3), Age at first birth (4), First degree breast cancer None None (3), First degree family history of cancer (5), Parity (n), Breast inflammatory disease (2) Tice 2005 Multiple ethnicities, older than Age (2), Menarche (3), Previous biopsies (3), Age at first birth (4), First degree breast cancer 0.68 None 35 years (3), Breast density (4) Zhang 2018 Caucasian 30–64 years Age (2), Menarche (3), Previous biopsies (3), Age at first birth (4), First degree breast cancer 0.65 None (2), Polygenic Risk Score(n), Mammographic density (n), Estrone Sulphate (n), Testosterone (n), Prolactin (n) Breast Cancer Surveillance Consortium ‘BCSC’ model Kerlikowske 2015 Multiple ethnicities, 35–74 years Age (n), Ethnicity (6), First degree breast cancer (2), Previous biopsies (2), Changes in breast 0.64 5-years: 0.98 density (16) 10-years: 0.95 Shieh 2016 Multiple ethnicities. Age was not Age (n), Ethnicity (6), First degree breast cancer (2), Previous biopsies (2), Breast density (4), 0.65 None specified Polygenetic risk score (n), BMI (n) Tice 2008 Multiple ethnicities, 35–84 years Age (n), Ethnicity (4), First degree breast cancer (2), Previous biopsies (2), Breast density (BI- 0.66 1.03 RADS) Tice 2015 Multiple ethnicities, 35–74 years Age (n), Ethnicity (4), First degree breast cancer (2), Breast density (4), Benign breast 0.67 5-years: 1.04 J Louro et al. disease (6) 10-years: 1.05 Vachon 2015 Multiple ethnicities. Age was not Age (n), Ethnicity (6), First degree breast cancer (2), Previous biopsies (2), Breast density (4), 0.69 None specified Polygenetic risk score (n) Rosner & Colditz model based on the ‘Nurses' Health Study’ Colditz 2000 Caucasian, 30–64 years Age (n), Menarche (n), Age at first birth (n), Menopause (n), Age at subsequent births (n), None None Benign breast disease (2), HRT (n), First degree breast cancer (2), Weight (n), BMI (n), Alcohol (n) Colditz 2004 Caucasian, 30–64 years Age (n), Menarche (n), Age at first birth (n), Menopause (n), Age at subsequent births (n), ER+/PR+: 0.64 None Benign breast disease (2), HRT (n), First degree breast cancer (2), Weight (n), BMI (n), Alcohol ER−/PR−: 0.61 (n) Rosner 1996 Caucasian, 30–64 years Age (n), Menarche (n), Age at first birth (n), Menopause (n), Age at subsequent births (n) None None Rosner 2008 Caucasian, 30–64 years Age (n), Menarche (n), Age at first birth (n), Menopause (n), Age at subsequent births (n), 0.65 None Benign breast disease (2), HRT (n), First degree breast cancer (2), Weight (n), BMI (n), Alcohol (n), Estradiol levels (n) Zhang 2018 Caucasian 30–64 years Age (n), Menarche (n), Age at first birth (n), Menopause (n), Age at subsequent births (n), 0.68 None Benign breast disease (2), HRT (n), First degree breast cancer (2), Weight (n), BMI (n), Alcohol (n), Polygenic Risk Score(n), Mammographic density (n), Estrone Sulphate (n), Testosterone (n), Prolactin (n) A systematic review and quality assessment of individualised breast. . . 79 A systematic review and quality assessment of individualised breast. . . J Louro et al. 80 a. BCRAT model. Of the 10 studies derived of the The original publication reported the Observed/Expected ratio. E/O ratios were calculated based on the original information. AUROC area under the receiver operating characteristic curve. E/O expected/ Calibration (E/O ratio)a Post menopause: 1.01 BCRAT model, five reported the calibration accuracy. Pre menopause: 1.00 observed, BMI body mass index, HRT hormone replace treatment, ER oestrogen receptor, PR progesterone receptor, BI-RADS Breast Imaging Reporting and Data System. (n): continuous variable in model Banegas et al.14 presented heterogeneous results depend- ing on the provenance of the population, reporting an E/O ratio of 0.93 for US-born and 1.52 for foreign-born women. Although Matsuno et al.25 added new variables to the original BCRAT model, the E/O ratio was 0.85, which was None None None None None the lowest of the group, whereas the other studies published E/O ratios that varied from 0.93 to 1.0316,20,23 (Table 1). Discriminatory accuracy b. BCSC model. Tice et al. published in 2008 a value of 1.03 Post menopause: 0.62 Pre menopause: 0.63 for the E/O ratio when looking at 5-year risk.31 Using Discriminatory and calibration accuracy values represents the statistics published in the original articles for the general population Subgroup values are not reported here previous benign breast disease, they obtained a similar result in 2015, with an E/O ratio of 1.04 for 5-year risk and 1.05 for 10-year risk.32 When Kerlikowske et al. (AUROC)a assessed changes in breast density the ratio decreased None None None obtaining a 0.98 for 5-year risk and 0.95 for 10-year risk.24 0.71 0.64 The studies of Vachon et al. and Shieh et al. did not present validation regarding the calibration accuracy of Age (7), Menarche (2), Previous biopsies (2), Age at first birth (2), First degree breast cancer Age (7), BMI (n), HRT (2), Breast cancer family history (2), Menopause (2), Breast density (4), birth (5), Menopause (n), Atypical Hyperplasia (2), Lobular carcinoma in situ (2), Height (3), birth (5), Menopause (n), Atypical Hyperplasia (2), Lobular carcinoma in situ (2), Height (3), Age (n), Gen phenotype (6), Family history (n, relationship, age), Menarche (n), Age at first Age (n), Gen phenotype (6), Family history (n, relationship, age), Menarche (n), Age at first the model (Table 1). Age (9), Age at first birth (4), Ethnicity (6), Menopause (2), First degree breast cancer (5), c. Rosner & Colditz model. Of the five studies based on the Previous biopsies (4), Breast density (5), HRT (3), BMI (5), Previous false positive or true Rosner & Colditz model,13,18,19,27,28 none of them reported Age (n), Menarche (3), Age at first birth (5), BMI (2), Breast cancer family history (2) calibration accuracy statistics of their models for the women in the general population. d. IBIS model. The IBIS model original paper33 does not report any calibration statistic. Nevertheless, other articles have validated it showing an E/O ratio of 1.67.36 e. Other models. The study Barlow et al.15 was the only one that reported calibration accuracy and presented the closest E/O ratio to one of all the studies included in this review taking values of 1.00 and 1.01 for pre and post- menopausal status respectively (Table 1). negative screen result (2), Menopausal status (3) Quality assessment The quality of the included studies was moderate due to some limitations in the discriminative power, study design, and data inputs. The studies did not show important limitations with Risk factors (number of categories) (2), Breastfeeding (2), Abortion (2) Microcalcifications (5), Mases (n) regards to the validation, appropriateness of the model analysis, reporting or interpretation of the results (Fig. 3). A summary of the risk of bias assessment per each source of limitation is presented here and the detailed appraisal and judgements in Supplementary table 4. Internal and external validation Ten studies14–17,20,23,25,26,30,31 validated their models by compar- ing the results with those published by Gail et al.,22 three BMI (5) BMI (5) studies24,29,32 compared with Tice et al.,31 one21 compared with both Gail et al.22 and Tyrer et al.,33 one13 compared with both Gail et al.22 and the results of a Rosner & Colditz model external Multiple ethnicities. Age was not Multiple ethnicities. Age was not validation37 and three studies did not report the model validation in the primary articles.19,22,34 Six studies assessed Asian women. Age was not Multiple ethnicities, 35 to internal validation with a sample of the population that Caucasian, 40–74 years generated data for the model,15,16,24,29,31,32 and four with an Targeted population external population.14,20,23,25 Despite not having reported the Asian, 35–70 years external validation in the primary articles, the Rosner & Colditz model18,19,27,28 reported external validation in a subsequent article mentioned before.37 Nine studies used the expected/ specified. specified specified 84 years observed event ratio to measure the calibration accuracy of the model.14–16,20,23–25,29,31 Other original models Table 1 continued Bias due to the study design Thirteen studies used a case-control design to obtain breast Eriksson 2017 cancer risk estimates,12–14,16,17,20–23,25,26,29,34 five studies used Barlow 2006 Wang 2014 Ueda 2003 Tyrer 2004 Tyrer 2004 prospective cohorts,15,18,19,27,28 and four models used retro- IBIS model Study ID spective cohorts.24,30–32 The study of Wang et al.35 and the study of Tyrer et al.33 used risk estimates obtained from a systematic b a review of the literature. A systematic review and quality assessment of individualised breast. . . J Louro et al. 81 a AUC values reported by the BRCAT model b AUC values reported by the BCSC model studies. 0.80 0.80 0.75 0.75 0.70 0.70 0.65 0.65 0.60 0.60 0.55 0.55 0.50 0.50 0.45 0.45 0.40 0.40 Banegas Banegas Boyle Decarli Gail Matsuno Tice Zhang Tice 2008 Kerlikowske Tice 2015* Shieh 2016 Vachon 2015 2017 US- 2017 2004* 2006 2007 2011 2005 2018 2015* born foreign-born *Boyle 2004 did not report confidence intervals. *Did not present confidence intervals. c AUC values reported by the Rosner and Colditz models studies. d AUC values reported by other original models. 0.80 0.80 0.75 0.75 0.70 0.70 0.65 0.65 0.60 0.60 0.55 0.55 0.50 0.50 0.45 0.45 0.40 0.40 Colditz 2004 Colditz 2004 Rosner 2008 Rosner 2008 Zhang 2018 Eriksson 2017 Barlow 2006 Barlow 2006 Wang 2014 ER+/PR+ ER–/PR– without estradiol with estradiol premenopausal posmenopausal Fig. 2 Area under the ROC curve (AUROC) and Confidence Intervals reported by the included studies. a AUROC values reported by the BCRAT model studies. b AUROC values reported by the BCSC model studies. c AUROC values reported by the Rosner & Colditz model studies. d AUROC values reported by other original models Limitations of data inputs biopsies. Breast density, benign breast disease, and polygenetic Sixteen studies obtained most of the input parameters from self- score were predominant in the BCSC model. Although during the reported questionnaires.13–20,22,23,25–28,30,34 The study of Matsuno last decade the models have shown improvements in their et al.25 also imputed ethnicity for women with missing data. discriminatory accuracy, it remains at best moderate with a maximum AUROC value of 0.71 reported by Eriksson et al.21 The Appropriateness of the model analysis calibration accuracy was very heterogeneous ranging from 0.85 to Thirteen studies12–17,20,22,23,25,26,29,34 used logistic regression to 1.52. Furthermore, the quality of the studies was not high due to estimate the risk of having breast cancer according to the limitations in the discriminative accuracy, study design, and data assessed risk factors, five used proportional hazard Cox inputs. models,21,24,30–32 four used Poisson regression models,18,19,27,28 and the other two studies used risk estimates obtained from a Agreements and disagreements with other reviews systematic review of the literature.33,35 In this systematic review, we found that the number of individualised breast cancer risk prediction models has increased Reporting bias steadily over the past three decades. This finding is in agreement Twenty one studies reported all relevant and necessary informa- with the narrative overview published by Cintolo-Gonzalez et al. in tion for the model creation.12–23,25–29,31,33–35 Conversely, a critical 2017,38 and it updates the results of a previous systematic review lack of information was found in the other three studies.24,30,32 published by Anothaisintawee et al. in 2012.7 In contrast to these reviews, however, our aim was to provide innovative information regarding the quality of the identified prediction models. Thus, we DISCUSSION have identified and rigorously analysed the strengths and Summary of main results limitations of 24 individualised models in order to adjust our This systematic review included 24 studies that aimed to estimate conclusions to the quality of the evidence. the individual risk of developing breast cancer in women in the We have identified two new trends with regards to the use and general population. Twenty studies were based on four specific development of the models, which are the increased use of the risk models (the BCRAT, the BCSC, the Rosner & Colditz and the BCSC model and the inclusion of common genetic variation in the IBIS model),16–20,22–33 whereas four studies used other original prediction models. As compared to the information published in models.15,21,34,35 The most extensively used were the BCRAT, IBIS the review of Anothaisintawee et al.,7 we found that in contrast to and the BCSC models. The number of risk factors included in the the BCRAT and Rosner & Colditz models that were the most models ranged from five to 18. Other than age, which was the frequently cited models up to 20107 the BCSC model has only risk factor present in all models, the BCRAT model also concentrated the attention of several authors during the last five included family history, age at first birth, menarche, and previous years, although its discriminatory accuracy has not dramatically A systematic review and quality assessment of individualised breast. . . J Louro et al. 82 a Validation Design Data Analysis Reporting Interpretation Conflict of interest 0% 25% 50% 75% 100% Yes (low risk) Unclear No (high risk) Review authors’ judgments presented as percentages across all included studies. Conflict of interest b c Conflict of interest Interpretation Interpretation Reporting Validation Reporting Validation Analysis Analysis Design Design Data Data + – – – + + + Kerlikowsky 2015 + + + + – + + Banegas 2017 Shieh 2016 + + + – + + + Boyle 2004 + – – – + + – Tice 2008 + + + + + + + Chen 2006 + – – – + + – Tice 2015 + + + + – + + Decarli 2006 + – – – + + – Vachon 2015 + + + – + + – Gail 1989 + – – – + – – Risk of bias summary for the studies that Gail 2007 + – – – + + – reported the BCSC model. Matsuno 2011 + – – – + + – d Conflict of interest Novotny 2006 – – – – + + – Interpretation Reporting Validation Tice 2005 + + – + – + – Analysis Design Data Zhang 2018 + + – – + + + Risk of bias summary for the studies that Tyrer 2004 – – – + + + – reported the BCRAT model. Risk of bias summary for the studies that reported the IBIS model. e Conflict of interest Interpretation f Conflict of interest Reporting Validation Analysis Interpretation Design Data Reporting Validation Analysis Design Data Colditz 2000 – + – + + + + Colditz 2004 + + – + + + – Barlow 2006 + + – – + + + Rosner 1996 + + – + + + – Eriksson 2017 + – + + + + + Rosner 2008 + + – + + + + Ueda 2003 – – – – + + + Zhang 2018 + + – – + + + Wang 2014 + – – – + + + Risk of bias summary for the studies that Risk of bias summary for the studies that reported the Rosner and Colditz models. reported the other original models. Fig. 3 Risk of bias summary: review authors’ judgments about each risk of bias item for each included study. a Review authors’ judgments presented as percentages across all included studies. b Risk of bias summary for the studies that reported the BCRAT model studies. c Risk of bias summary for the studies that reported the BCSC model studies. d Risk of bias summary for the studies that reported the Rosner & Colditz model studies. e Risk of bias summary for the studies that reported other original models improved. Second, none of the models in the review of would be correctly categorised as higher risk compared to a Anothaisintawee et al.7 included genetic information as a risk randomly chosen woman without disease. The discriminatory factor. By contrast, we have identified four models including accuracy estimate does not express whether the model is more or genetic information: the IBIS model33 that includes genetic less accurate in predicting the risk of specific individuals but phenotype in their updated version, the BCSC model that includes measures the capacity of the model to determine which women a polygenetic score in both 201512 and 201629 publications, as are at higher/lower risk for developing breast cancer. Thus, both well as the article by Zhang et al. that added a polygenic risk score calibration accuracy and discriminatory accuracy should be to both the BCRAT and the Rosner & Colditz models.13 assessed. Contrary to what is expected, we found that authors Most of the included studies reported the AUROC to determine reported the E/O ratio only in less than half of the included the probability that a randomly chosen woman with disease studies. In addition to the AUROC value, the studies of Zhang et al. A systematic review and quality assessment of individualised breast. . . J Louro et al. 83 and Vachon et al.12,13 also reported an improvement in the net publications reporting external model validation in different reclassification index (NRI) of the BCRAT, and Rosner & Colditz settings and countries. These studies may help to understand models, as well as in the BCSC model, respectively. the performance of a model in a specific context, but this issue Overall, the information provided by the AUROC and the E/O was out of the scope of our review and, therefore, we have not ratio was consistent suggesting that the included models have included external validation studies. As an example of the moderate discriminatory accuracy and calibration accuracy when relevance of these studies, we can inform that the BCRAT model applied to the women in the general population. Nevertheless, it has more than 50 articles informing the external validation of must be taken into account that despite the great importance of these models in different countries.46 The Rosner-Colditz model validation in terms of AUROC and E/O ratio, the presence of low has also been validated in several studies, one of the most values of AUROC or clearly different from 1 values of the E/O ratio complete validations being the one performed in 2013 by the does not mean that these models are useless. On the contrary, authors themselves.37 On the other hand, we found that although models are clinically useful even with moderate AUROC since they the Eriksson et al.19 model reports the highest AUC (0.71), this can reclassify individuals at the extremes of risk.39 Thus, the model has not been externally validated, which increases the verdict on risk models should not be based solely on these uncertainty about its applicability. estimators. Instead, they need to be prospectively evaluated in Also, there were limitations in data inputs, mostly due to the clinical trials. In fact, there are currently two very large randomised fact that in several models the information was provided by self- trials assessing risk-based screening strategies. Both of them are reported questionnaires that may affect the accuracy of the using individualised models. Both the IBIS and the BCSC models results. Finally, there is a limitation when comparing the AUROC or are being tested in the European trial MyPeBS (My Personalised E/O ratio across the models given that there is great heterogeneity Breast Screening).40 Also, the BCSC model is being tested in the US amongst them. The models were targeted to different popula- WISDOM trial (Women Informed to Screen Depending On tions, included different sets of risk factors, and often used Measures of risk).41 different methodologies. We have taken into account all these variations and presented the results by model categories. Applicability and completeness of evidence The distribution of risk factors in such different populations may Potential biases in the review process affect the applicability of the models to different contexts. The fact This systematic review was limited to studies published in English that different subtypes of breast cancer may have different and did not involve an active search for grey literature, which is genetic markers is widely accepted.42 These differences, the literature that is not formally published in sources such as books nature of breast cancer itself and its low incidence may condition or journal articles. Therefore, some models may not have been a low discriminatory accuracy of a model. In other words, in the identified. However, since we have conducted a comprehensive general population, there is a low probability of having breast literature search in Medline, EMBASE and The Cochrane Library, cancer (even in the highest risk group). This low probability may we estimate that the loss of information due to the study selection mean that the discriminatory power of a breast cancer risk model criteria is low. Some key genetically oriented models, such as won’t be as high as a risk model targeted to other common BOADICEA47 and BRACAPRO48 were not included in this review diseases such as cardiovascular events, for instance. Another because they are aimed at high risk women and not useful for potential limitation in the applicability in the screening context is women in the general population in the screening context. Full- the completeness and the number of included risk factors, which text screening and data abstraction process were performed by ranged from five to 18. Nevertheless, some potentially relevant two researchers, which increase the quality of the review process. risk factors such as genetic markers have been only included in Moreover, as far as we know, this is the first review assessing the few models. Recent studies43,44 have shown that adding genetic risk of bias of the identified risk prediction models. information as a risk factor can increase the discriminative accuracy of the different models which opens the line for further evaluation. An evaluation that should first assess the calibration of CONCLUSIONS these models in prospective cohort studies. The development of individualised breast cancer risk prediction Overall, women are usually screened using mammography. models has increased over the last three decades, but the Particularly in Europe, most programmes invite women for improvements in both the discriminatory power and calibration screening every 2 years.2 The presence of some mammographic accuracy are still limited. Despite the time that has passed since features in these screening mammograms may be related to the the first model was published and a large number of available risk of developing breast cancer, as has been recently pointed out publications, only one model addressed to women attending a by some authors.21,45 Only one of the 24 models identified in this population-based screening programme21 was identified. Cur- systematic review included microcalcifications and masses found rently, it is still a challenge to recommend any of the models as at mammography as risk factors in the model.21 Time-changing the standard for predicting individual risk in screening context. variables such as radiological variables may not be as stable as However, the models have been updated by adding new personal history. However, in a screening context, this information variables, such as common genetic variation or radiologic is especially relevant because it is easily available from previous variables and have shown improvements in their quality as well screening examinations. as in their discriminative accuracy. These new variables need further evaluation to confirm its promising impact in the Quality of the evidence prediction capacity to propose personalised strategies for breast We found variability in the design of the studies that were used to cancer screening. obtain the cancer risk estimates. Notably, the study design used in the BCSC model was a cohort, which is a robust epidemiology design that allows developing and validating prediction models. ACKNOWLEDGEMENTS Another frequently used design was the case-control study, The authors thank Ms. Lorea Galnares-Cordero for her contributions to the design of nested or not. Contrary to the cohort study, time-changing the search strategy and the initial retrieval of the citations. The authors also thank Ms. variables may not be well obtained in case-control studies. Julieta Politi and Mr. José María Montero-Moraga for their contribution in the screening process. Javier Louro is a Ph.D. candidate at the Methodology of Regarding the external validation, the models showed some lim- Biomedical Research and Public Health program, Universitat Autònoma de Barcelona itations given that few of them were further evaluated in different (UAB), Barcelona, Spain. contexts. As far as we know, there are numerous scientific A systematic review and quality assessment of individualised breast. . . J Louro et al. 84 AUTHOR CONTRIBUTIONS 12. Vachon, C. M., Pankratz, V. S., Scott, C. G., Haeberle, L., Ziv, E., Jensen, M. R., et al. J.L., M.P., M.S. and X.C. designed the study, and J.L. and M.P. wrote the manuscript. J. The contributions of breast density and common genetic variation to breast L., M.P. and M.H.B. performed the screening, data abstraction and quality assessment cancer risk. J. Natl. Cancer Inst. 107, dju397 (2015). of included studies. M.S., L.D. and M.R. contributed to the analyses and interpreted 13. Zhang, X., Rice, M., Tworoger, S. S., Rosner, B. A., Eliassen, A. H., Tamimi, R. M. et al. the data. M.R., M.H.B. and X.C. collaborated in drafting the manuscript and revising it Addition of a polygenic risk score, mammographic density, and endogenous critically for important intellectual content. All authors read and approved the final hormones to existing breast cancer risk prediction models: a nested case-control manuscript. study. PLoS Med. 15, e1002644 (2018). 14. Banegas, M. P., John, E. M., Slattery, M. L., Gomez, S. L., Yu, M., LaCroix, A. Z., et al. Projecting Individualized Absolute Invasive Breast Cancer Risk in US Hispanic ADDITIONAL INFORMATION Women. J. Natl. Cancer Inst. 109, djw215 (2017). Supplementary information is available for this paper at https://doi.org/10.1038/ 15. Barlow, W. E., White, E., Ballard-Barbash, R., Vacek, P. M., Titus-Ernstoff, L., Carney, P. s41416-019-0476-8. A. et al. Prospective breast cancer risk prediction model for women undergoing screening mammography. J. Natl. Cancer Inst. 98, 1204–1214 (2006). Competing interests: The authors declare no competing interests. 16. Boyle, P., Mezzetti, M., La Vecchia, C., Franceschi, S., Decarli, A. & Robertson, C. Contribution of three components to individual cancer risk predicting breast cancer risk in Italy. Eur. J. Cancer Prev. 13, 183–191 (2004). Ethics approval and consent to participate: All procedures performed in this study 17. Chen, J., Pee, D., Ayyagari, R., Graubard, B., Schairer, C., Byrne, C. et al. Projecting were in accordance with the ethical standards of the ethics committee of Parc de absolute invasive breast cancer risk in white women with a model that includes Salut Mar (CEIC Parc de Salut Mar) and with the 1964 Helsinki declaration and its later mammographic density. J. Natl. Cancer Inst. 98, 1215–1226 (2006). amendments or comparable ethical standards. Neither specific patient consent nor 18. Colditz, G. A. & Rosner, B. Cumulative risk of breast cancer to age 70 years ethics committee’s approval were required because we used published articles that according to risk factor status: data from the Nurses’ Health Study. Am. J. Epi- were obtained from open access databases. demiol. 152, 950–964 (2000). 19. Colditz, G. A., Rosner, B. A., Chen, W. Y., Holmes, M. D. & Hankinson, S. E. Risk Data availability: The datasets analysed during the current study are publicly factors for breast cancer according to estrogen and progesterone receptor status. available from the corresponding author. J. Natl. Cancer Inst. 96, 218–228 (2004). 20. Decarli, A., Calza, S., Masala, G., Specchia, C., Palli, D. & Gail, M. H. Gail model for Consent for publication: Not applicable. prediction of absolute risk of invasive breast cancer: independent evaluation in the Florence-European Prospective Investigation Into Cancer and Nutrition cohort. J. Natl. Cancer Inst. 98, 1686–1693 (2006). Funding: This work was partially supported by Agència de Qualitat i Avaluació 21. Eriksson, M., Czene, K., Pawitan, Y., Leifland, K., Darabi, H. & Hall, P. A clinical Sanitàries de Catalunya (AQuAS) and by grants from Instituto de Salud Carlos III model for identifying the short-term risk of breast cancer. Breast Cancer Res. 19, FEDER (grant numbers: PI15/00098 and PI17/00047). JL is core funded by the 29 (2017). Research Network on Health Services in Chronic Diseases (RD12/0001/0015). MHB is 22. Gail, M. H., Brinton, L. A., Byar, D. P., Corle, D. K., Green, S. B., Schairer, C. et al. core funded by the UK Medical Research Council (funding code: MC_UU_12017/15) Projecting individualized probabilities of developing breast cancer for white and the Scottish Government Chief Scientist Office (funding code: SPHSU15). None of females who are being examined annually. J. Natl. Cancer Inst. 81, 1879–1886 the funders participated in the design of the study, collection, analysis, or (1989). interpretation of data, or in writing the manuscript. 23. Gail, M. H., Costantino, J. P., Pee, D., Bondy, M., Newman, L., Selvan, M. et al. Projecting individualized absolute invasive breast cancer risk in African American Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims women. J. Natl. Cancer Inst. 99, 1782–1792 (2007). in published maps and institutional affiliations. 24. Kerlikowske, K., Gard, C. C., Sprague, B. L., Tice, J. A. & Miglioretti, D. L. One versus two breast density measures to predict 5 and 10-year breast cancer risk. Cancer Epidemiol. Biomark. Prev. 24, 889–897 (2015). REFERENCES 25. Matsuno, R. K., Costantino, J. P., Ziegler, R. G., Anderson, G. L., Li, H., Pee, D. et al. 1. The Independent UK Panel on Breast Cancer Screening. The benefits and harms of Projecting individualized absolute invasive breast cancer risk in Asian and Pacific breast cancer screening: an independent review. Lancet. 380, 1778–1786 (2012). Islander American women. J. Natl. Cancer Inst. 103, 951–961 (2011). 2. The European Commission Initiative on Breast Cancer (ECIBC). Recommenda- 26. Novotny, J., Pecen, L., Petruzelka, L., Svobodnik, A., Dusek, L., Danes, J. et al. Breast tions from European Breast Guidelines. 2016. https://ecibc.jrc.ec.europa.eu/ cancer risk assessment in the Czech female population—an adjustment of the recommendations/. original Gail model. Breast Cancer Res Treat. 95, 29–35 (2006). 3. U.S. Preventive Services Task Force. Screening for breast cancer: U.S. Preventive 27. Rosner, B. & Colditz, G. A. Nurses’ health study: log-incidence mathematical Services Task Force recommendation statement. Ann. Intern. Med. 151, 716–726 model of breast cancer incidence. J. Natl. Cancer Inst. 88, 359–364 (1996). (2009). W-236. 28. Rosner, B., Colditz, G. A., Iglehart, J. D. & Hankinson, S. E. Risk prediction models 4. Oeffinger, K. C., Fontham, E. T., Etzioni, R., Herzig, A., Michaelson, J. S., Shih, Y. C. with incomplete data with application to prediction of estrogen receptor-positive et al. Breast cancer screening for women at average risk: 2015 guideline update breast cancer: prospective data from the Nurses’ Health Study. Breast Cancer Res. from the American Cancer Society. JAMA 314, 1599–1614 (2015). 10, R55 (2008). 5. Mandelblatt, J. S., Stout, N. K., Schechter, C. B., van den Broek, J. J., Miglioretti, 29. Shieh, Y., Hu, D., Ma, L., Huntsman, S., Gard, C. C., Leung, J. W. et al. Breast cancer D. L., Krapcho, M. et al. Collaborative modeling of the benefits and harms asso- risk prediction using a clinical risk model and polygenic risk score. Breast Cancer ciated with different U.S. Breast Cancer Screening Strategies. Ann. Intern Med. Res Treat. 159, 513–525 (2016). 164, 215–225 (2016). 30. Tice, J. A., Cummings, S. R., Ziv, E. & Kerlikowske, K. Mammographic breast density 6. Steyerberg, E. W. Clinical Prediction Models. A Practical Approach to Development, and the Gail model for breast cancer risk prediction in a screening population. Validation, and Updating. (Springer Science, New York, 2009). Breast Cancer Res Treat. 94, 115–122 (2005). 7. Anothaisintawee, T., Teerawattananon, Y., Wiratkapun, C., Kasamesup, V. & 31. Tice, J. A., Cummings, S. R., Smith-Bindman, R., Ichikawa, L., Barlow, W. E. & Thakkinstian, A. Risk prediction models of breast cancer: a systematic review of Kerlikowske, K. Using clinical factors and mammographic breast density to esti- model performances. Breast Cancer Res Treat. 133, 1–10 (2012). mate breast cancer risk: development and validation of a new predictive model. 8. Higgins, J. P. T., Green, S. (eds). Cochrane Handbook for Systematic Reviews of Ann. Intern Med. 148, 337–347 (2008). Interventions Version 5.1.0. 2011. http://handbook-5-1.cochrane.org/. 32. Tice, J. A., Miglioretti, D. L., Li, C. S., Vachon, C. M., Gard, C. C. & Kerlikowske, K. 9. Moher, D., Shamseer, L., Clarke, M., Ghersi, D., Liberati, A., Petticrew, M. et al. Breast density and benign breast disease: risk assessment to identify women at Preferred reporting items for systematic review and meta-analysis protocols high risk of breast cancer. J. Clin. Oncol. 33, 3137–3143 (2015). (PRISMA-P) 2015 statement. Syst. Rev. 4, 1 (2015). 33. Tyrer, J., Duffy, S. W. & Cuzick, J. A breast cancer prediction model incorporating 10. Jaime Caro, J., Eddy, D. M., Kan, H., Kaltz, C., Patel, B., Eldessouki, R. et al. Ques- familial and personal risk factors. Stat. Med. 23, 1111–1130 (2004). tionnaire to assess relevance and credibility of modeling studies for informing 34. Ueda, K., Tsukuma, H., Tanaka, H., Ajiki, W. & Oshima, A. Estimation of indivi- health care decision making: an ISPOR-AMCP-NPC Good Practice Task Force dualized probabilities of developing breast cancer for Japanese women. Breast report. Value Health 17, 174–182 (2014). Cancer 10, 54–62 (2003). 11. Shea, B. J., Reeves, B. C., Wells, G., Thuku, M., Hamel, C., Moran, J. et al. AMSTAR 2: 35. Wang, Y., Gao, Y., Battsend, M., Chen, K., Lu, W. & Wang, Y. Development of a risk a critical appraisal tool for systematic reviews that include randomised or non- assessment tool for projecting individualized probabilities of developing breast randomised studies of healthcare interventions, or both. BMJ 358, j4008 (2017). cancer for Chinese women. Tumour Biol. 35, 10861–10869 (2014). A systematic review and quality assessment of individualised breast. . . J Louro et al. 85 36. Brentnall, A. R., Harkness, E. F., Astley, S. M., Donnelly, L. S., Stavrinos, P., Sampson, 45. Castells, X., Tora-Rocamora, I., Posso, M., Roman, M., Vernet-Tomas, M., Rodri- S. et al. Mammographic density adds accuracy to both the Tyrer-Cuzick and Gail guez-Arana, A. et al. Risk of breast cancer in women with false-positive breast cancer risk models in a prospective UK screening cohort. Breast Cancer Res. results according to mammographic features. Radiology 280, 379–386 (2016). 17, 147 (2015). 46. Wang, X., Huang, Y., Li, L., Dai, H., Song, F. & Chen, K. Assessment of performance 37. Rosner, B. A., Colditz, G. A., Hankinson, S. E., Sullivan-Halley, J., Lacey, J. V. Jr. & of the Gail model for predicting breast cancer risk: a systematic review and meta- Bernstein, L. Validation of Rosner-Colditz breast cancer incidence model using an analysis with trial sequential analysis. Breast Cancer Res. 20, 18 (2018). independent data set, the California Teachers Study. Breast Cancer Res Treat. 142, 47. Antoniou, A. C., Pharoah, P. P., Smith, P. & Easton, D. F. The BOADICEA model of 187–202 (2013). genetic susceptibility to breast and ovarian cancer. Br. J. Cancer 91, 1580–1590 38. Cintolo-Gonzalez, J. A., Braun, D., Blackford, A. L., Mazzola, E., Acar, A., Plichta, J. K. (2004). et al. Breast cancer risk models: a comprehensive overview of existing models, 48. Berry, D. A., Parmigiani, G., Sanchez, J., Schildkraut, J. & Winer, E. Probability validation, and clinical applications. Breast Cancer Res Treat. 164, 263–284 (2017). of carrying a mutation of breast-ovarian cancer gene BRCA1 based on family 39. Khera, A. V., Chaffin, M., Aragam, K. G., Haas, M. E., Roselli, C., Choi, S. H. et al. history. J. Natl. Cancer Inst. 89, 227–238 (1997). Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nat. Genet. 50, 1219–1224 (2018). 40. MyPeBS. Randomized comparison of risk-stratified versus standard breast cancer Open Access This article is licensed under a Creative Commons screening in European women aged 40–70 (MyPeBS). 2017. http://www. Attribution 4.0 International License, which permits use, sharing, brumammo.be/documents/docs/bmm-my-pebs-clinical-trial-protocol.pdf. adaptation, distribution and reproduction in any medium or format, as long as you give 41. Esserman, L. J., Study, W. & Athena, I. The WISDOM Study: breaking the deadlock appropriate credit to the original author(s) and the source, provide a link to the Creative in the breast cancer screening debate. NPJ Breast Cancer 3, 34 (2017). Commons license, and indicate if changes were made. The images or other third party 42. Sorlie, T., Perou, C. M., Tibshirani, R., Aas, T., Geisler, S., Johnsen, H. et al. Gene material in this article are included in the article’s Creative Commons license, unless expression patterns of breast carcinomas distinguish tumor subclasses with indicated otherwise in a credit line to the material. If material is not included in the clinical implications. Proc. Natl. Acad. Sci. USA. 98, 10869–10874 (2001). article’s Creative Commons license and your intended use is not permitted by statutory 43. Garcia-Closas, M., Gunsoy, N. B., Chatterjee, N. Combined associations of genetic regulation or exceeds the permitted use, you will need to obtain permission directly and environmental risk factors: implications for prevention of breast cancer. J. from the copyright holder. To view a copy of this license, visit http://creativecommons. Natl. Cancer Inst. 2014;106. org/licenses/by/4.0/. 44. Maas, P., Barrdahl, M., Joshi, A. D., Auer, P. L., Gaudet, M. M., Milne, R. L. et al. Breast cancer risk from modifiable and nonmodifiable risk factors among white women in the United States. JAMA Oncol. 2, 1295–1302 (2016). © The Author(s) 2019