Accurate breast cancer risk prediction could improve risk-reduction paradigms if thoughtfully used in clinical practice. Identification of at-risk women is the first step in tailoring risk screening and risk-reduction protocols to women's needs. Using the UK Biobank, we validated a simple risk model to predict breast cancer risk in the general population. Our simple breast cancer risk (BRISK) model integrates a combination of impactful breast cancer-associated risk factors including extended family history and polygenic risk allowing for the removal of moderate factors currently found in comprehensive traditional models. Using two versions of BRISK, differing by 77-single-nucleotide polymorphisms (SNP) versus 313-SNP polygenic risk score integration, we found improved discrimination and risk categorization of both BRISK models compared with one of the most well-known models, the Breast Cancer Risk Assessment Tool (BRCAT). Over a 5-year period, at-risk women classified ≥3% 5-year risk by BRISK had a 1.829 (95% CI = 1.710–1.956) times increased incidence of breast cancer compared with the population, which was higher than the 1.413 (95% CI = 1.217–1.640) times increased incidence for women classified ≥3% by BCRAT.

Prevention Relevance:

In this prospective population-based cohort study, we show the improved performance of a new risk assessment model compared with a gold-standard model (BCRAT). The classification of at-risk women using this new model highlights the opportunity to improve risk stratification and implement existing clinical risk-reduction interventions.

Breast cancer is a multifactorial and heterogeneous disease for which numerous risk factors have been identified (1–3). Accordingly, many models have been developed over the years to predict breast cancer risk, thus enabling increased screening and risk-reduction options for at-risk women. Each risk model incorporates slightly different risk factors (4, 5); however, some risk factors are stronger than others (6). These include high-penetrance hereditary risk, family history, breast architecture (which includes everything from mammographic density to biopsy status), and polygenic risk. Polygenic risk is a promising epidemiologic risk factor (7, 8) with a robust amount of evidence derived from multiple large international data sets but its value is greater when integrated with other risk factors (9, 10).

Breast cancer risk assessment is valuable in determining which women might benefit from risk-reducing interventions such as supplemental screening measures and risk-reducing medication. Much focus over the past two decades has been on high-penetrance hereditary genetic risk of breast cancer. Although hereditary breast and ovarian cancer represents a significant risk to a carrier, at a population level, these women represent a fraction of the total number of women who are diagnosed with breast cancer. The large CARRIERS and BRIDGES studies looked at 12 and 13 susceptibility genes, respectively, and found pathogenic variants in around 5% of breast cancer cases and around 1.6% of controls (11–13). The population prevalence is less than 1% depending on the combination of susceptibility genes assessed (14, 15).

The United States Preventive Services Taskforce, American Society of Clinical Oncology, American Cancer Society, and the National Comprehensive Cancer Network have identified two actionable thresholds of risk to help define clinical outcomes. The lifetime risk threshold where magnetic resonance imaging surveillance is offered is 20% (16). In the United States, there are two separate 5-year risk thresholds; the 3% threshold is where the benefits of risk-reducing medication begin to outweigh the risks, although the lower threshold of 1.67% should not be dismissed, particularly in younger women (17–19). These thresholds are not used independently, but in the context of a woman's full medical history where weighing the benefits and risk of the intervention are considered; use of risk–benefit tables can help (20). Although not addressed specifically in the analysis of these data, guidelines vary by country. For example, UK National Institute for Health and Care Excellence guidelines use 10-year and lifetime risk to categorize by average (<3%, <17%, respectively), moderate (≥3% to <8%, ≥17 to <30%), and high risk (≥8%, ≥30%) to guide breast cancer screening and risk-reduction recommendations (21).

The Breast Cancer Risk Assessment Tool (BRCAT) was originally developed to estimate 5-year breast cancer risk in women participating in mammography screening trials (22). Since that time, BCRAT has been used to determine eligibility for risk-reducing medication in breast cancer prevention trials (23) and remains a standard tool in estimating 5-year breast cancer risk in the general population. Despite the efficacy of risk-reducing medication, clinician recommendation and patient uptake are both dismal, although both vary greatly depending on the clinical scenario (24). Recently, efforts have been made to improve risk prediction to increase the uptake of prophylactic endocrine therapy in women at increased risk (25–27).

Robust risk assessment models that use comprehensive pedigree data—such as IBIS (28), BOADICEA (29), and BRCAPRO (30)—exist, but in a real-world setting, practical opportunities to screen the general population with these models may not be available (31), although efforts are being made to continuously improve the models (32). However, IBIS is used in several large-scale, population-level studies in the UK (PROCAS and BC-Predict) that suggest both clinical feasibility and reciprocal interest from patients (33–35).

These comprehensive models were originally built to assess risk in a familial and pathogenic variant hereditary risk setting and integrate a large pedigree analysis (29, 36–40) that provide lifetime risk scores and a short-term risk score (5 or 10 year; refs. 38, 41). Other models such as BCRAT (42) and BCSC (43) were built without an extensive pedigree analysis are targeted toward general population risk assessment, and focus on short-term (5-year) risk scores. All models are underused in the clinic for a variety of reasons including time constraints associated with ease of use.

We and others have previously investigated the improvement in breast cancer risk prediction from combining a polygenic risk score (PRS) based on a panel of single-nucleotide polymorphisms (SNP) with established risk prediction models that incorporate clinical data (9, 10, 37, 44–49). Unlike previously published data where polygenic risk is combined with an existing comprehensive clinical model, we created a simple clinical model—which we call BRISK—by identifying the major risk factors for breast cancer and using them in a format that would be fast and easy to collect in a clinical setting. The goal of our risk model is to simplify data collection while still providing improved or equivalent performance compared with the comprehensive risk prediction models (50).

Herein, we compare the performance of two versions of our clinical model against the performance of the commonly used BCRAT in the UK Biobank. For clarity, the single variable that differs between the BRISK versions is the integrated PRS SNP panel and throughout the manuscript we use the following names to distinguish the two BRISK models: 77-SNP BRISK and 313-SNP BRISK (7, 8). A general population cohort such as the UK Biobank is of interest because it represents the large number of women who are typically considered as being at average risk of developing breast cancer. Most women who develop breast cancer fall within this so-called average-risk category, and it is these women who could benefit greatly from accurate risk prediction.

Calculation of risk scores

The BRISK model was constructed by bringing together evidence for the major risk factors for breast cancer. The input variables are age, number of affected female first-degree relatives, age of youngest affected first-degree relative, number of affected female second-degree relatives, percent mammographic density (or BI-RADS category), body mass index, menopausal status, and a PRS (50). In this paper, the BRISK model risk factors that we had access to include age, number of affected female first-degree relatives, body mass index, menopausal status, and a PRS. We did not have access to age of affected first-degree relative, second-degree family history or breast density data within the cohort and therefore these were left out of the calculation. The family history risks were based on those in the Collaborative Group on Hormonal Factors in Breast Cancer analysis (51). These risks were smoothed and centered to have a population average risk of 1 (Supplementary Table S1). The estimates for body mass index for pre- and post-menopausal women were taken from Hopper and colleagues (52). The 77-SNP PRS was from Mavaddat and colleagues (8) and the 313-SNP PRS was from Mavaddat and colleagues (7).

We calculated the 5-year and full-lifetime BRISK scores using the method in Supplementary Methods S1. We calculated the 5-year BCRAT risk (22, 53) using the BCRAT calculation function in the BCRA package (BCRA: Breast Cancer Risk Assessment. [cited 29 Jul 2022]. Available from https://dceg.cancer.gov/tools/risk-assessment/bcra) in R (RRID: SCR_001905; version 4.0.3). The risk factors in the BCRAT model include: age, number of affected female first-degree relatives, age at menarche, age at first live birth, benign breast biopsy status/number, atypical hyperplasia diagnosis.

UK Biobank

The UK Biobank (RRID: SCR_012815) is a population-based cohort of more than 500,000 middle-aged (40–69 years) adults who were recruited from 2006 to 2010 (54, 55). Baseline assessment information collected at study entry included comprehensive clinical, lifestyle and demographic information as well as specimens, such as DNA for genotyping analysis. This cohort does show evidence of healthy volunteer bias when comparing lifestyle habits as well as all-cause mortality against the general population (54, 56).

We identified 200,195 active (i.e., had not withdrawn their participation as at February 22, 2022) UK Biobank members who were female, were genetically Caucasian, were ages 40 to 69 years at baseline assessment date, had SNP data available, were unaffected at enrolment into the UK Biobank, were genetically unrelated, and had not died or been diagnosed with breast cancer within the first six weeks of follow-up (Supplementary Table S2). We used the ukb_gen_samples_to_remove function of the R (version 4.0.3) package ukbtools (57) to limit the final data set to unrelated women (i.e., no pairs with closer than third-degree relatedness).

Statistical analysis

We used age and date of baseline assessment to determine the date at which 5 years of follow-up was completed for all women. We used both self-reported and linked cancer registry data (ICD9 code 174 or ICD10 code C50) to determine the earliest diagnosis of invasive breast cancer for each affected woman. Follow-up began at the date of the baseline assessment and ended at whichever came first of date of breast cancer diagnosis, or date at which 5 years of follow-up was completed. For the calculations of standardized incidence ratios (SIR), follow-up of unaffected women was censored at age of death for those who died before completing 5 years of follow-up. We first assessed the SIR of the number of breast cancers expected using age-specific and calendar year-specific population incidence rates for women in England compared with the number observed during the 5 years of follow-up, overall and by 10-year age group for the years 2006–2017 (Office of National Statistics, Cancer Registration Statistics. [cited 2020 Apr 15]. Available from https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/datasets/cancerregistrationstatisticscancerregistrationstatisticsengland).

We assessed association using Cox regression to estimate the hazard ratio (HR) per quintile of risk. Harrell's C-index was used to assess the ability of the models to distinguish between affected and unaffected women (i.e., the discrimination of the models). We used logistic regression to evaluate calibration (the overall fit of the models). To assess dispersion, we tested whether the coefficients from a logistic regression of the log odds of each of the 5-year risk scores were equal to 1 (<1 is overdispersion and >1 is underdispersion). We then tested overall calibration by constraining the logistic regression models to have a slope of 1 and tested whether the intercept was equal to 0 (<0 is overestimation of risk and >0 is underestimation of risk; refs. 58, 59). These results were then illustrated using the PMCALPLOT module in Stata (RRID: SCR_012763).

We calculated SIRs stratified by commonly used clinical cutoffs of the number of breast cancers observed during the 5 years of follow-up to the number expected by each of the models. The cutoffs were derived from the clinical 5-year risk thresholds indicating increased risk of breast cancer according to the NCCN (>1.67%; ref. 18) and ASCO (17) and the USPSTF (≥3%; ref. 19). To investigate the ability of the models to stratify breast cancer risk in the population, we calculated the SIR of the number of breast cancer cases expected using sex- and age-specific population incidence rates for England for the years 2006–2017 (Office of National Statistics, Cancer Registration Statistics. [cited 2020 Apr 15]. Available from https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/datasets/cancerregistrationstatisticscancerregistrationstatisticsengland) and the number observed during the 5 years of follow-up for deciles of risk for each model. We used Stata (RRID: SCR_012763; version 17.0) for analyses; all statistical tests were two sided and P values <0.05 were considered nominally statistically significant.

Ethics approval

The UK Biobank operates under the UK Biobank Ethics and Governance Framework. [cited 2023 Feb 23]. Available from https://www.ukbiobank.ac.uk/media/0xsbmfmw/egf.pdf. The UK Biobank has Research Tissue Bank approval (REC #11/NW/0382) that covers analysis of data by approved researchers. All participants provided written informed consent to the UK Biobank before data collection began. This research was conducted using the UK Biobank resource under Application Number 47401.

Data availability

The raw data for this study were provided by the UK Biobank and we do not have permission to share the data. Researchers wishing to access the data used in this study can apply directly to the UK Biobank at https://www.ukbiobank.ac.uk/register-apply. Stata 17.0 code for the data management and analysis is available from the corresponding author for noncommercial purposes.

Of the 200,195 women in the analysis data set, the 197,057 unaffected women had a mean age of 57.0 years (standard deviation [SD] = 7.9) at their baseline assessment. The 3,138 affected women who were diagnosed with invasive breast cancer during the 5 years of follow-up had a mean age of 58.2 (SD = 7.5) years women at their baseline assessment and a mean age at diagnosis of 60.8 years (SD = 7.6). In unaffected women, 195,635 (99%) completed 5 years of follow-up and 1,422 (1%) died before completing the follow-up period.

All SNPs were available in the UK Biobank for the 77-SNP PRS (8), whereas 8 (3%) SNPs for the 313-SNP PRS (7) were not available in the UK Biobank. For the 77-SNP PRS, 55% of women had all SNP genotypes and 34% were missing only one SNP, whereas for the 313-SNP PRS, 61% had at least 98% of the 305 available SNP genotypes. Summary statistics for the risk factors used in the BRISK and BCRAT models are provided in Supplementary Table S3. Affected women had mean 5-year risks of 2.35% (SD = 1.72%) for 77-SNP BRISK, 2.14% (SD = 1.94%) for 313-SNP BRISK, and 1.62% (SD = 0.07%) for BCRAT. In unaffected women the mean 5-year risks were 1.75% (SD = 1.20%) for 77-SNP BRISK, 1.44% (SD = 1.30%) for 313-SNP BRISK, and 1.46% (SD = 0.62% for BCRAT). Distribution of 5-year risk scores between affected and unaffected using each model is shown in Supplementary Fig. S1.

Overall, there were 7% fewer breast cancers observed than the number expected using population incidence rates for the 5 years of follow-up (SIR = 0.936; 95% CI = 0.904–0.969). This was evident in the 50–59 years age group (SIR = 0.908; 95% CI = 0.853–0.966) and the 60–69 years age group (SIR = 0.903; 95% CI = 0.859–0.949), but not in the 40–49 years age group, which saw more breast cancers than expected (SIR = 1.108; 95% CI = 1.021–1.202).

For the BRISK model, the 77-SNP BRISK had an HR per quintile of risk of 1.38 (95% CI = 1.34–1.42, P < 0.001) and the 313-SNP BRISK had an HR per quintile of risk of 1.45 (95% CI = 1.40–1.49, P < 0.001). BCRAT had a HR per quintile of risk of 1.12 (95% CI = 1.08–1.16, P < 0.001).

Harrell's C-index was 0.628 (95% CI = 0.618–0.638) for the 77-SNP BRISK, 0.649 (95% CI = 0.640–0.695) for the 313-SNP BRISK and 0.567 (95% CI = 0.556–0.577) for BCRAT. The 313-SNP BRISK was a statistically significant improvement in discrimination over both the 77-SNP BRISK (z = 6.22, P < 0.001) and BCRAT (z = 11.71, P < 0.001).

The 5-year risks for 77-SNP BRISK (α = −0.119; 95% CI = −0.154, −0.083; P < 0.001) and 313-SNP BRISK (α = −0.088; 95% CI = −0.123, −0.052; P < 0.001) overestimated risk overall, but as shown in Fig. 1 and Supplementary Table S4, overestimation of risk was only evident in the top two deciles of risk for the 313-SNP BRISK and the 77-SNP BRISK. For BCRAT, the 5-year risk underestimated risk overall (α = 0.068; 95% CI = 0.033, 0.103; P < 0.001), with Fig. 1 showing that, as with the other models, the top decile overestimated risk.

Next, an SIR analysis comparing the number of cancers observed during the 5 years of follow-up to the number expected by each of the models, overall and by 10-year age group, was performed using categories based on the clinically actionable 5-year risk thresholds of 1.67% and 3.0% (Table 1 and Fig. 2). BCRAT had an SIR of 1.413 (95% CI = 1.217–1.640) when classifying women at increased risk (≥3% 5-year risk score), representing 172 (5.5%) of the 3,138 affected women. In comparison, the 77-SNP BRISK had an SIR of 1.667 (95% CI = 1.549–1.795) for the high-risk women, representing 710 (22.7%) of 3,138 affected women, and the 313-SNP BRISK model had an SIR of 1.829 (95% CI = 1.710–1.956) representing 850 (27.1%) of 3,138 affected women. In Table 1 and Fig. 2, we also show this analysis broken down by 10-year age groups.

We conducted a reclassification analysis to assess the ability of the 313-SNP BRISK model to improve upon the classifications of BCRAT (Table 2) and found an overall net reclassification index of 0.278 (95% CI = 0.255–0.300). The 313-SNP BRISK classified more affected women into a higher category of 5-year risk and more unaffected women into a lower category of risk than BCRAT.

Current supplemental breast cancer screening and risk-reduction recommendations are based on family history, having a mutation in a high-penetrance gene and risk assessment models. BCRAT is a well-known and well-established tool, but it is underused in the general population. Other models such as IBIS, BOADICEA, and BCRAPRO were originally developed for family history-rich settings and are time-consuming to complete and thus more difficult to implement at a general population level. In an effort to improve upon current risk models, we and others have constructed risk models that integrate polygenic risk with established clinical models such as the BCRAT, IBIS, BOADICEA, BRCAPRO, and BCSC (9, 10, 37, 46, 60–62). These data support the value of PRS as a component of risk prediction models. Further improvements can be made by streamlining the risk models to facilitate clinician adoption. The BRISK model was designed with clinician time constraints in mind and includes that strongest risk factors for breast cancer. As expected (7), we found that the performance of 313-SNP BRISK was better than that of 77-SNP BRISK, and both were better than BCRAT.

Assessing model calibration, we show that both the BRISK models and the BCRAT were generally well calibrated within the UK Biobank population (Fig. 1). Both 77-SNP BRISK and 313-SNP BRISK overestimated risk in the top decile of risk and to a lesser extent in the ninth decile. In the top decile of risk, the predicted risk was 3.1% for 77-SNP BRISK and 4.9% for 313-SNP BRISK, whereas the observed risks were 3.1% and 3.5%, respectively. These observed risks were both over the threshold of 3%, which indicates increased risk and where discussions around risk-reducing medication and supplemental screening options are available to clinicians. Of note, BRCAT modestly overestimated risk in the top decile (similar to the overestimation of both BRISK models in the ninth decile) but neither the predicted risk (2.8%) nor the observed risk (2.4%) surpassed the 3% 5-year risk threshold. Because of overestimation of risk in the top two deciles for BRISK, there is a risk of overscreening and unnecessary recommendation of risk-reducing medication for a small number of women; however, in both these circumstances, clinical decision-making is made in the context of the patient's full medical history and not based solely on a risk score.

Using discrimination and net reclassification performance, we show an improvement by both BRISK models over BCRAT, and 313-SNP BRISK performs better than 77-SNP BRISK. The 313-SNP BRISK model's ability to classify women into at-risk categories show significant improvement compared with the BCRAT (net reclassification improvement 0.278; 95% CI = 0.255–0.300; P < 0.001). But, more importantly, the 313-SNP BRISK model was able to categorize 4.9 times as many women into the at-risk category (5-year risk score ≥3%) compared with BCRAT, which has significant implications for clinical risk reduction. These data suggest that, based on 5-year risk scores of ≥3% in women ages 40–69 years, approximately 13% of the general population should be having discussions regarding the risks and benefits of risk-reducing medication, which strongly outweighs the risks for this group of women (17). This percentage is consistent with published estimates (33).

Using 313-SNP BRISK, risk prediction is improved across all age ranges, but the actionable benefit can be observed in the extremes of the risk categories: <1% or ≥3% 5-year risk scores. Taking advantage of the large sample size, we were able to look at the SIRs stratified by 10-year age groups (Table 1). Risk models perform differently across the age groups, highlighting the value and specific use-cases of the upper and lower 5-year risk scores: 3% and 1.67%.

The 5-year risk threshold of 1.67% can be achieved by age alone; women around the age of 60 years approach this level of risk. Using the 3% 5-year risk threshold in these women is necessary for improved risk prediction. Comparing the percentages of affected women ages 60–69 years (column 2 of Table 1 divided by the total number of affected) with 5-year risks ≥3%, 313-SNP BRISK identifies 3.5 times more affected women than BCRAT, and once classified, they are more likely to develop breast cancer than women identified by BCRAT. This latter statement is supported by the SIR observed in the ≥3% category of women which is higher in BRISK compared with BCRAT (Fig. 2). Women in their 60s, can be more effectively identified as being at increased risk above 3%. This improved stratification over the BCRAT only improves as we move down by decade. For women in their 50s, 13 times more affected women are classified in this at-risk category ≥3% using 313-SNP BRISK compared with BRCAT. For women in their 40s, none of the affected women are classified as ≥3% 5-year risk by BCRAT, but 7.3% of affected women are classified by 313-SNP BRISK.

However, for women in their early 40s, a 1.67% 5-year risk score represents a 2-fold increase in risk compared with the average woman. There is a 5.2-times increase in affected women classified by the 313-SNP BRISK model over the BCRAT in 40–49-year-olds using this lower threshold of 1.67%; and 313-SNP BRISK classifies 6% of affected women over the 3% threshold whereas the BCRAT does not put any affected women over this threshold. Importantly, 313-SNP BRISK identified approximately a third of the affected women in this age range, suggesting the importance of risk stratification in decision-making discussions around breast cancer risk-reducing behaviors including screening compliance.

Interestingly, we observed an increase in breast cancer incidence among women aged 40–49 years within the UK Biobank participants in comparison with the UK population incidence rates. This is opposite from the “healthy volunteer” bias that we observe in older decades. In the 50–69-year-olds, one might assume they were having regular mammograms at higher rates than the general population. This same theory might apply to the 40–49-year-olds where women enrolled in the study were more vigilant about their healthcare and thus were diagnosed earlier than expected population rates. Or because these women enrolled in the study initially because they had some concerns about their health; or these women may have been pathogenic variant carriers in a known breast cancer susceptibility gene (63). But recent evidence suggests that pathogenic variant carriers are uncommon enough that at a population level, their presence does not change discrimination potential of risk assessment (64). We did not observe a difference in BMI between UK Biobank participants and the general population; however, there are data to suggest that UK Biobank participants are less likely to live in socioeconomically deprived areas than nonparticipants in the general population, which has been associated with “healthy volunteer” bias (56).

Overall, the 313-SNP BRISK classified more women into an increased-risk category (a 5-year risk ≥3%) compared with BCRAT. Despite the known overestimation of risk for 313-SNP BRISK, women classified as being at increased risk were at substantially increased risk of breast cancer (Fig. 2): 1.8 times population risk, which was higher than the 1.4 times population risk seen for women identified by BCRAT.

Some limitations exist in this study. The BRISK models used here did not incorporate second-degree family history, age of affected first-degree relatives or mammographic density information due to the lack of existing data within the UK Biobank. However, in real-world scenarios, often times clinical information including biopsy status, family history and breast density information is not readily available. Despite federal legislation efforts, there are still states in the United States that have no breast density notification laws in place.

Herein, we focused on the 5-year risk score capabilities for comparative purposes to the BCRAT model. We show the improved risk-stratification potential of 313-SNP BRISK over BRCAT. The driving factor in the improvement over BCRAT is the incorporation of the PRS. The BRCAT model does not predict lifetime risk score because of the limited family history component, so we do not attempt a comparison with lifetime risk. However, the 313-SNP BRISK model is an appropriate tool to assess lifetime risk as it is “largely dependent on family history,” as defined by the NCCN breast cancer screening practice recommendations (16). We previously cross-validated BRISK in a nested case–control analysis of the Nurses’ Health Study (50). Consistent with our previous analyses, 313-SNP BRISK categorizes around 13% of women with a full-lifetime risk score greater than 20% and around 13% of women with 5-year risk scores ≥3% (Supplementary Table S5).

Keeping the limitations of clinical practice in mind, we have developed a simple breast cancer risk prediction model by integrating the most important polygenic and clinical risk factors. It is becoming widely accepted that a PRS adds significant value to risk stratification, and standard risk models including BOADICEA (CanRisk) and IBIS (version 8) now both allow for the incorporation of PRS into their risk scores. Using the UK Biobank, we have shown that our simple breast cancer risk model has improved discrimination and has a net reclassification improvement compared with BCRAT for 5-year risk prediction. Although both BRISK models were well calibrated for the majority of women, we found an overestimation of risk in the top decile of risk; however, the observed risks for these women remained over the 3% 5-year risk threshold. Women with a 5-year risk of over 3% according to 313-SNP BRISK had substantially increased SIRs compared with population incidence rates and these were higher than the SIRs for women identified as over 3% by BCRAT. These performance metrics have significant implications to existing risk-reduction paradigms in place for at-risk women.

E.L. Spaeth is employed by Phenogen Sciences, the US subsidiary of Genetic Technologies. G.S. Dite reports a patent 2022279367 pending to Genetic Technologies Limited and a patent 2021903955 pending to Genetic Technologies Limited; and is employed by Genetic Technologies Limited. The company had no role in the conceptualization, design, data analysis, decision to publish, or preparation of the manuscript. R. Allman reports a patent for PCT/AU2022/051459 pending. No disclosures were reported by the other authors.

E.L. Spaeth: Conceptualization, supervision, validation, investigation, writing–original draft, project administration, writing–review and editing. G.S. Dite: Conceptualization, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. J.L. Hopper: Conceptualization, methodology, writing–review and editing. R. Allman: Conceptualization, supervision, writing–review and editing.

We wish to thank Mr. Lawrence Whiting for his invaluable expertise in the management of the large data files from the UK Biobank. We also wish to thank Dr. James Dowty for his expert guidance in the use of competing mortality in the calculation of absolute risks. This study was fully funded by Genetic Technologies Limited, which had no role in the conceptualization, design, data analysis, decision to publish, or preparation of the manuscript.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Cancer Prevention Research Online (http://cancerprevres.aacrjournals.org/).

1.
Nelson
HD
,
Zakher
B
,
Cantor
A
,
Fu
R
,
Griffin
J
,
O'Meara
ES
, et al
.
Risk factors for breast cancer for women aged 40 to 49 years: a systematic review and meta-analysis
.
Ann Intern Med
2012
;
156
:
635
48
.
2.
Thun
M
,
Linet
MS
,
Cerhan
JR
,
Haiman
CA
,
Schottenfeld
D
.
Cancer epidemiology and prevention
.
New York
:
Oxford University Press
;
2017
.
1328
p
.
3.
Hartmann
LC
,
Sellers
TA
,
Frost
MH
,
Lingle
WL
,
Degnim
AC
,
Ghosh
K
, et al
.
Benign breast disease and the risk of breast cancer
.
N Engl J Med
2005
;
353
:
229
37
.
4.
Brentnall
AR
,
Harkness
EF
,
Astley
SM
,
Donnelly
LS
,
Stavrinos
P
,
Sampson
S
, et al
.
Mammographic density adds accuracy to both the Tyrer-Cuzick and Gail breast cancer risk models in a prospective UK screening cohort
.
Breast Cancer Res
2015
;
17
:
147
.
5.
Vachon
CM
,
Pankratz
VS
,
Scott
CG
,
Haeberle
L
,
Ziv
E
,
Jensen
MR
, et al
.
The contributions of breast density and common genetic variation to breast cancer risk
.
J Natl Cancer Inst
2015
;
107
:
dju397
.
6.
Hopper
JL
.
Odds per adjusted standard deviation: comparing strengths of associations for risk factors measured on different scales and across diseases and populations
.
Am J Epidemiol
2015
;
182
:
863
7
.
7.
Mavaddat
N
,
Michailidou
K
,
Dennis
J
,
Lush
M
,
Fachal
L
,
Lee
A
, et al
.
Polygenic risk scores for prediction of breast cancer and breast cancer subtypes
.
Am J Hum Genet
2019
;
104
:
21
34
.
8.
Mavaddat
N
,
Pharoah
PD
,
Michailidou
K
,
Tyrer
J
,
Brook
MN
,
Bolla
MK
, et al
.
Prediction of breast cancer risk based on profiling with common genetic variants
.
J Natl Cancer Inst
2015
;
107
:
djv036
.
9.
Allman
R
,
Dite
GS
,
Hopper
JL
,
Gordon
O
,
Starlard-Davenport
A
,
Chlebowski
R
, et al
.
SNPs and breast cancer risk prediction for African American and Hispanic women
.
Breast Cancer Res Treat
2015
;
154
:
583
9
.
10.
Dite
GS
,
MacInnis
RJ
,
Bickerstaffe
A
,
Dowty
JG
,
Allman
R
,
Apicella
C
, et al
.
Breast cancer risk prediction using clinical models and 77 independent risk-associated SNPs for women aged under 50 years: Australian Breast Cancer Family Registry
.
Cancer Epidemiol Biomarkers Prev
2016
;
25
:
359
65
.
11.
Hu
C
,
Hart
SN
,
Gnanaolivu
R
,
Huang
H
,
Lee
KY
,
Na
J
, et al
.
A population-based study of genes previously implicated in breast cancer
.
N Engl J Med
2021
;
384
:
440
51
.
12.
Domchek
SM
,
Yao
S
,
Chen
F
,
Hu
C
,
Hart
SN
,
Goldgar
DE
, et al
.
Comparison of the prevalence of pathogenic variants in cancer susceptibility genes in black women and non-hispanic white women with breast cancer in the United States
.
JAMA Oncol
2021
;
7
:
1045
50
.
13.
Breast Cancer Association Consortium
.
Breast cancer risk genes — association analysis in more than 113,000 women
.
N Engl J Med
2021
;
384
:
428
39
.
14.
Maxwell
KN
,
Domchek
SM
,
Nathanson
KL
,
Robson
ME
.
Population frequency of germline BRCA1/2 mutations
.
J Clin Oncol
2016
;
34
:
4183
5
.
15.
van der Groep
P
,
van der Wall
E
,
van Diest
PJ
.
Pathology of hereditary breast cancer
.
Cell Oncol (Dordr)
2011
;
34
:
71
88
.
16.
National Comprehensive Cancer Network
.
NCCN Cinical Practice Guidelines in Oncology
.
Breast Cancer Screening and Diagnosis V1.2022
. [
cited 2022 June 1
]
Available from
: <https://www.nccn.org/professionals/physician_gls/pdf/breast-screening.pdf>.
Web Page
.
17.
Visvanathan
K
,
Fabian
CJ
,
Bantug
E
,
Brewster
AM
,
Davidson
NE
,
DeCensi
A
, et al
.
Use of endocrine therapy for breast cancer risk reduction: ASCO clinical practice guideline update
.
J Clin Oncol
2019
;
37
:
3152
65
.
18.
National Comprehensive Cancer Network
.
NCCN Cinical Practice Guidelines in Oncology
.
Breast Cancer Risk Reduction V1.2022
[
cited 2022 June 1
]
Available from
: <https://www.nccn.org/professionals/physician_gls/pdf/breast_risk.pdf>.
Web Page
.
19.
Force USPST
,
Owens
DK
,
Davidson
KW
,
Krist
AH
,
Barry
MJ
,
Cabana
M
, et al
.
Medication use to reduce risk of breast cancer: US preventive services task force recommendation statement
.
JAMA
2019
;
322
:
857
67
.
20.
Freedman
AN
,
Yu
B
,
Gail
MH
,
Costantino
JP
,
Graubard
BI
,
Vogel
VG
, et al
.
Benefit/risk assessment for breast cancer chemoprevention with raloxifene or tamoxifen for women age 50 years or older
.
J Clin Oncol
2011
;
29
:
2327
33
.
21.
Excellence NIfHaC
.
Familial breast cancer: classification, care and managing breast cancer and related risks in people with a family history of breast cancer CG164
.
NICE
2013
, [
cited 2023 February 7
]
Available from
: <https://www.nice.org.uk/guidance/cg164/chapter/recommendations>
22.
Costantino
JP
,
Gail
MH
,
Pee
D
,
Anderson
S
,
Redmond
CK
,
Benichou
J
, et al
.
Validation studies for models projecting the risk of invasive and total breast cancer incidence
.
J Natl Cancer Inst
1999
;
91
:
1541
8
.
23.
Vogel
VG
.
The NSABP study of tamoxifen and raloxifene (STAR) trial
.
Expert Rev Anticancer Ther
2009
;
9
:
51
60
.
24.
Ball
S
,
Arevalo
M
,
Juarez
E
,
Payne
JD
,
Jones
C
.
Breast cancer chemoprevention: an update on current practice and opportunities for primary care physicians
.
Prev Med
2019
;
129
:
105834
.
25.
Keane
H
,
Huilgol
YS
,
Shieh
Y
,
Tice
JA
,
Belkora
J
,
Sepucha
K
, et al
.
Development and pilot of an online, personalized risk assessment tool for a breast cancer precision medicine trial
.
NPJ Breast Cancer
2021
;
7
:
78
.
26.
Huilgol
YS
,
Keane
H
,
Shieh
Y
,
Hiatt
RA
,
Tice
JA
,
Madlensky
L
, et al
.
Elevated risk thresholds predict endocrine risk-reducing medication use in the Athena screening registry
.
NPJ Breast Cancer
2021
;
7
:
102
.
27.
Kim
JO
,
Schaid
DJ
,
Vachon
CM
,
Cooke
A
,
Couch
FJ
,
Kim
CA
, et al
.
Impact of personalized genetic breast cancer risk estimation with polygenic risk scores on preventive endocrine therapy intention and uptake
.
Cancer Prev Res (Phila)
2021
;
14
:
175
84
.
28.
Tyrer
J
,
Duffy
SW
,
Cuzick
J
.
A breast cancer prediction model incorporating familial and personal risk factors
.
Stat Med
2004
;
23
:
1111
30
.
29.
Antoniou
AC
,
Cunningham
AP
,
Peto
J
,
Evans
DG
,
Lalloo
F
,
Narod
SA
, et al
.
The BOADICEA model of genetic susceptibility to breast and ovarian cancers: updates and extensions
.
Br J Cancer
2008
;
98
:
1457
66
.
30.
Parmigiani
G
,
Berry
D
,
Aguilar
O
.
Determining carrier probabilities for breast cancer-susceptibility genes BRCA1 and BRCA2
.
Am J Hum Genet
1998
;
62
:
145
58
.
31.
Archer
S
,
Babb de Villiers
C
,
Scheibl
F
,
Carver
T
,
Hartley
S
,
Lee
A
, et al
.
Evaluating clinician acceptability of the prototype CanRisk tool for predicting risk of breast and ovarian cancer: a multi-methods study
.
PLoS One
2020
;
15
:
e0229999
.
32.
Carver
T
,
Hartley
S
,
Lee
A
,
Cunningham
AP
,
Archer
S
,
Babb de Villiers
C
, et al
.
CanRisk tool-a web interface for the prediction of breast and ovarian cancer risk and the likelihood of carrying genetic pathogenic variants
.
Cancer Epidemiol Biomarkers Prev
2021
;
30
:
469
73
.
33.
Evans
DGAS
,
Stavrinos
P
, et al
.
Improvement in risk prediction, early detection and prevention of breast cancer in the NHS breast screening programme and family history clinics: a dual cohort study
.
Chapter 4, PROCAS: Predicting Risk of Breast Cancer at Screening
.
Southampton (UK)
:
NIHR Journals Library
;
2016
.
34.
Hawkins
R
,
McWilliams
L
,
Ulph
F
,
Evans
DG
,
French
DP
.
Healthcare professionals' views following implementation of risk stratification into a national breast cancer screening programme
.
BMC Cancer
2022
;
22
:
1058
.
35.
French
DP
,
Astley
S
,
Brentnall
AR
,
Cuzick
J
,
Dobrashian
R
,
Duffy
SW
, et al
.
What are the benefits and harms of risk stratified screening as part of the NHS breast screening programme? Study protocol for a multi-site non-randomised comparison of BC-predict versus usual screening (NCT04359420)
.
BMC Cancer
2020
;
20
:
570
.
36.
Anderson
H
,
Bladström
A
,
Olsson
H
,
Möller
TR
.
Familial breast and ovarian cancer: a Swedish population-based register study
.
Am J Epidemiol
2000
;
152
:
1154
63
.
37.
Pal Choudhury
P
,
Brook
MN
,
Hurson
AN
,
Lee
A
,
Mulder
CV
,
Coulson
P
, et al
.
Comparative validation of the BOADICEA and Tyrer-Cuzick breast cancer risk models incorporating classical risk factors and polygenic risk in a population-based prospective cohort of women of European ancestry
.
Breast Cancer Res
2021
;
23
:
22
.
38.
Glynn
RJ
,
Colditz
GA
,
Tamimi
RM
,
Chen
WY
,
Hankinson
SE
,
Willett
WW
, et al
.
Comparison of questionnaire-based breast cancer prediction models in the nurses' health study
.
Cancer Epidemiol Biomarkers Prev
2019
;
28
:
1187
94
.
39.
Terry
MB
,
Liao
Y
,
Whittemore
AS
,
Leoce
N
,
Buchsbaum
R
,
Zeinomar
N
, et al
.
10-year performance of four models of breast cancer risk: a validation study
.
Lancet Oncol
2019
;
20
:
504
17
.
40.
Quante
AS
,
Whittemore
AS
,
Shriver
T
,
Strauch
K
,
Terry
MB
.
Breast cancer risk assessment across the risk continuum: genetic and nongenetic risk factors contributing to differential model performance
.
Breast Cancer Res
2012
;
14
:
R144
.
41.
Quante
AS
,
Whittemore
AS
,
Shriver
T
,
Hopper
JL
,
Strauch
K
,
Terry
MB
.
Practical problems with clinical guidelines for breast cancer prevention based on remaining lifetime risk
.
J Natl Cancer Inst
2015
;
107
:
djv124
.
42.
Rockhill
B
,
Spiegelman
D
,
Byrne
C
,
Hunter
DJ
,
Colditz
GA
.
Validation of the Gail et al. model of breast cancer risk prediction and implications for chemoprevention
.
J Natl Cancer Inst
2001
;
93
:
358
66
.
43.
Tice
JA
,
Bissell
MCS
,
Miglioretti
DL
,
Gard
CC
,
Rauscher
GH
,
Dabbous
FM
, et al
.
Validation of the breast cancer surveillance consortium model of breast cancer risk
.
Breast Cancer Res Treat
2019
;
175
:
519
23
.
44.
Allman
R
,
Spaeth
E
,
Lai
J
,
Gross
SJ
,
Hopper
JL
.
A streamlined model for use in clinical breast cancer risk assessment maintains predictive power and is further improved with inclusion of a polygenic risk score
.
PLoS One
2021
;
16
:
e0245375
.
45.
Starlard-Davenport
A
,
Allman
R
,
Dite
GS
,
Hopper
JL
,
Spaeth Tuff
E
,
Macleod
S
, et al
.
Validation of a genetic risk score for Arkansas women of color
.
PLoS One
2018
;
13
:
e0204834
.
46.
Shieh
Y
,
Hu
D
,
Ma
L
,
Huntsman
S
,
Gard
CC
,
Leung
JW
, et al
.
Breast cancer risk prediction using a clinical risk model and polygenic risk score
.
Breast Cancer Res Treat
2016
;
159
:
513
25
.
47.
Kurian
AW
,
Hughes
E
,
Simmons
T
,
Bernhisel
R
,
Probst
B
,
Meek
S
, et al
.
Performance of the IBIS/Tyrer-Cuzick model of breast cancer risk by race and ethnicity in the women's health initiative
.
Cancer
2021
;
127
:
3742
50
.
48.
Lakeman
IMM
,
Rodríguez-Girondo
M
,
Lee
A
,
Ruiter
R
,
Stricker
BH
,
Wijnant
SRA
, et al
.
Validation of the BOADICEA model and a 313-variant polygenic risk score for breast cancer risk prediction in a Dutch prospective cohort
.
Genet Med
2020
;
22
:
1803
11
.
49.
Yang
X
,
Eriksson
M
,
Czene
K
,
Lee
A
,
Leslie
G
,
Lush
M
, et al
.
Prospective validation of the BOADICEA multifactorial breast cancer risk prediction model in a large prospective cohort study
.
J Med Genet
2022
;
59
:
1196
205
.
50.
Allman
R
,
Mu
Y
,
Dite
GS
,
Spaeth
E
,
Hopper
JL
,
Rosner
BA
.
Validation of a breast cancer risk prediction model based on the key risk factors: family history, mammographic density and polygenic risk
.
Breast Cancer Res Treat
2023
; doi: .
51.
Collaborative Group on Hormonal Factors in Breast Cancer
.
Familial breast cancer: collaborative reanalysis of individual data from 52 epidemiological studies including 58,209 women with breast cancer and 101,986 women without the disease
.
Lancet
2001
;
358
:
1389
99
.
52.
Hopper
JL
,
Dite
GS
,
MacInnis
RJ
,
Liao
Y
,
Zeinomar
N
,
Knight
JA
, et al
.
Age-specific breast cancer risk by body mass index and familial risk: prospective family study cohort (ProF-SC)
.
Breast Cancer Res
2018
;
20
:
132
.
53.
Gail
MH
,
Brinton
LA
,
Byar
DP
,
Corle
DK
,
Green
SB
,
Schairer
C
, et al
.
Projecting individualized probabilities of developing breast cancer for white females who are being examined annually
.
J Natl Cancer Inst
1989
;
81
:
1879
86
.
54.
Sudlow
C
,
Gallacher
J
,
Allen
N
,
Beral
V
,
Burton
P
,
Danesh
J
, et al
.
UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age
.
PLoS Med
2015
;
12
:
e1001779
.
55.
Bycroft
C
,
Freeman
C
,
Petkova
D
,
Band
G
,
Elliott
LT
,
Sharp
K
, et al
.
The UK biobank resource with deep phenotyping and genomic data
.
Nature
2018
;
562
:
203
9
.
56.
Fry
A
,
Littlejohns
TJ
,
Sudlow
C
,
Doherty
N
,
Adamska
L
,
Sprosen
T
, et al
.
Comparison of sociodemographic and health-related characteristics of UK biobank participants with those of the general population
.
Am J Epidemiol
2017
;
186
:
1026
34
.
57.
Hanscombe
KB
,
Coleman
JRI
,
Traylor
M
,
Lewis
CM
.
ukbtools: an R package to manage and query UK Biobank data
.
PLoS One
2019
;
14
:
e0214311
.
58.
Van Calster
B
,
McLernon
DJ
,
van Smeden
M
,
Wynants
L
,
Steyerberg
EW
.
Calibration: the Achilles heel of predictive analytics
.
BMC Med
2019
;
17
:
230
.
59.
Huang
Y
,
Li
W
,
Macheret
F
,
Gabriel
RA
,
Ohno-Machado
L
.
A tutorial on calibration measurements and calibration models for clinical prediction models
.
J Am Med Inform Assoc
2020
;
27
:
621
33
.
60.
Hurson
AN
,
Pal Choudhury
P
,
Gao
C
,
Hüsing
A
,
Eriksson
M
,
Shi
M
, et al
.
Prospective evaluation of a breast-cancer risk model integrating classical risk factors and polygenic risk in 15 cohorts from six countries
.
Int J Epidemiol
2022
;
50
:
1897
911
.
61.
Lee
A
,
Mavaddat
N
,
Wilcox
AN
,
Cunningham
AP
,
Carver
T
,
Hartley
S
, et al
.
BOADICEA: a comprehensive breast cancer risk prediction model incorporating genetic and nongenetic risk factors
.
Genet Med
2019
;
21
:
1708
18
.
62.
Li
H
,
Feng
B
,
Miron
A
,
Chen
X
,
Beesley
J
,
Bimeh
E
, et al
.
Breast cancer risk prediction using a polygenic risk score in the familial setting: a prospective study from the breast cancer family registry and kConFab
.
Genet Med
2017
;
19
:
30
5
.
63.
Patel
AP
,
Wang
M
,
Fahed
AC
,
Mason-Suares
H
,
Brockman
D
,
Pelletier
R
, et al
.
Association of rare pathogenic DNA variants for familial hypercholesterolemia, hereditary breast and ovarian cancer syndrome, and Lynch syndrome with disease risk in adults according to family history
.
JAMA Netw Open
2020
;
3
:
e203959
.
64.
Evans
DGR
,
van Veen
EM
,
Harkness
EF
,
Brentnall
AR
,
Astley
SM
,
Byers
H
, et al
.
Breast cancer risk stratification in women of screening age: Incremental effects of adding mammographic density, polygenic risk, and a gene panel
.
Genet Med
2022
;
24
:
1485
94
.
This open access article is distributed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license.