Towards risk-stratified breast cancer screening: considerations for changes in screening guidelines

Population-wide breast cancer screening programs have been established in many countries to reduce breast cancer mortality through early detection and treatment. These programs offer routine mammographic screening to women at average risk starting at ages 40–50 years, with frequency of 1–3 years, and enhanced screening starting at earlier ages for women at elevated risk due to family history or genetic susceptibility from high-penetrance genetic mutations, e.g., in BRCA1/2 or TP53.¹ Screening guidelines recommend different approaches to identify women at elevated risk, including the use of risk prediction models with pedigree-level family history information and genetic testing.

In this issue of JAMA Oncology, Mukama et al.² call for a change in current screening guidelines for women with a family history of breast cancer. Mukama et al. define “risk-adapted” starting age of screening based on empirical estimates of the 10-year cumulative risk of developing invasive breast cancer for women with different combinations of family history of breast cancer. These estimates were obtained using a nationwide dataset including over 5 million Swedish women born after 1931 and followed through 2015. The risk-adapted starting age of screening, defined as the age by which women with a particular family history profile attained a 10-year cumulative risk comparable to the average risk for women at the recommended age of screening initiation in the general population, varied substantially depending on the number of first- and second-degree relatives diagnosed with breast cancer and the age at diagnosis of the first-degree relatives. For example, when screening is recommended at age 50 years for the general population (corresponding to a 2.2% 10-year risk), women with one affected first-degree relative attained a similar risk level at age 36 years or 41 years, depending on whether the relative was diagnosed before age 40 years or after age 50 years, respectively.

The main novelty of this paper is the use of empirical risk estimates obtained from a very large population to determine starting age of mammography screening for relatives of breast cancer patients. Other strengths include the detailed and dynamic assessment of family history at the population level and internal validation of findings using two-fold cross validation. The authors used some examples to demonstrate that the ages at screening initiation using their approach can be quite different compared to guidelines aimed at “average risk” women. However, differences were smaller when compared to the American College of Radiology guidelines that recommend using family history-based models to determine when women with a family history should start screening. Mukama et al. propose an approach to integrate information from their estimates of risk-adapted starting ages with risk prediction tools to supplement current clinical guidelines; however, as the authors recognize, assessment of the validity and clinical utility of this approach is necessary prior to considering its use in clinical practice. Though Mukama et al. performed internal validation of their risk estimates within their large Swedish population, it is also essential to validate the robustness of findings across diverse, independent populations. The Swedish data are a very valuable resource that could be used in the future to compare observed familial risks in this cohort to risks predicted by existing family-based risk models, such as BOADICEA or IBIS.^5,6

Family history of breast cancer is one of numerous risk factors that may be considered for tailoring screening according to individual risk, and several models have been developed to incorporate complex risk factor information, including breast density and reproductive/hormonal risk factors, in addition to pedigree-level family history and genetic testing.⁴ Furthermore, as breast cancer risk factor profiles and incidence rates vary geographically, tailored breast cancer risk assessment tools can integrate information on population-specific cancer rates and risk factor distributions from nationally representative databases that reflect the target population at risk.^7,8 Although estimating the starting age of screening based on models with family history information alone might be simpler and easier to implement compared to those using more comprehensive risk factor information, this approach would likely result in lower risk discrimination at the population level and produce less accurate individualized risk estimates. Thus, the tradeoffs between simplicity and accuracy need to be balanced for specific implementation settings.

Recently developed polygenic risk scores (PRS) can be used to stratify women in the general population with or without family history of breast cancer according to the genetic risk due to common variants.⁹ Such PRS, e.g., the latest one using 313 common variants associated with breast cancer⁹, can also be used to determine starting age of screening based on genetic risk for such women. For instance, we recently showed that women in the UK at the highest 1% of genetic risk of breast cancer based on the 313-variant PRS would reach the same 10-year risk as women in the UK general population aged 47 years (the recommended age to start screening in the UK) by age 30 years.⁹ Since 85% of women in the Swedish cohort do not have a family history of breast cancer and estimates based on family history only would not apply to them, incorporating additional information from common variants and classical risk factors is important to determine “risk-adapted” starting age of screening. We have recently used the Individualized Coherent Absolute Risk Estimation (iCARE) tool¹⁰ to develop comprehensive breast cancer risk prediction models and demonstrated that integrating information on PRS along with classical risk factors and mammographic breast density can result in substantial improvements in risk stratification.^7,8 We projected further improvements through ongoing efforts to better characterize the genetic architecture of breast cancer.⁸ The BOADICEA and IBIS risk models have also been extended to include information on polygenic risk, in addition to complex family history information, and other risk factors for breast cancer.^5,6 Robust, prospective validation of these comprehensive risk models across populations¹¹ will be critical prior to their application to determine the initiation and frequency of risk-stratified screening.

Mukama et al. provide estimates for risk of developing primary invasive breast cancer. However, because breast cancer is a heterogenous disease with subtypes that differ in their etiology, response to preventive therapies, detection and prognoses, future work is also needed to evaluate subtype-specific risk prediction.¹¹ For instance, estrogen receptor (ER)-positive predictions could improve the identification of women who would benefit from endocrine risk-reducing therapies, while identifying women at high risk of aggressive forms of ER-negative breast cancer could be relevant to determine the best modality for breast cancer screening (e.g., more frequent screening for fast growing tumors). Risk assessment should also identify those at increased risk of interval breast cancer who could benefit most from supplemental imaging or alternative methods for early detection currently under development (e.g., liquid biopsy).¹ In the Mukama et al. study, a substantial proportion of women with a family history of breast cancer had “risk-adapted” starting ages of screening before 50 years, when the performance of screening mammography tends to be lower due to the elevated breast density that tends to occur in younger women. Though this report focused on mammographic breast screening, alternate breast imaging modalities (e.g., MRI, ultrasonography) may be better suited for younger women with dense breasts.¹² Recent work from the U.S. Breast Cancer Surveillance Consortium suggests that an integrated approach encompassing breast density along with other risk factors may be most effective in identifying those at high risk of advanced breast cancer who may benefit most from supplemental breast imaging.¹³

There are many challenges to integrating findings from the Mukama et al. report within the evolving landscape of personalized breast cancer screening. Identification of a subset of the population who could benefit from initiating screening earlier needs to be balanced with potential costs due to false-positive results, particularly since the number of additional cases that would be detectable by screening would still be relatively low in younger women. Population-wide risk-stratified screening programs have the potential to improve the cost-effectiveness of mammographic screening and to reduce potential harms due to overdiagnosis and false-positive findings, but this might require reducing screening for those at low risk, in addition to screening more of those at high risk.¹⁴ Ongoing screening trials (e.g., WISDOM and MyPebs),¹⁵ are currently evaluating the clinical acceptability and utility of risk-stratified screening programs in the general population, and should provide valuable information for the possible implementation of such programs.

In summary, while we agree with Mukama et al. in their call for a change in current screening guidelines to move towards improved strategies for risk-stratified breast cancer screening, more research is needed to determine what are the best approaches, particularly within the context of other existing and emerging risk prediction tools for breast cancer and its biologically heterogeneous subtypes. Other considerations include the cost-effectiveness of such approaches, current limitations of mammographic screening, and challenges in implementation. Ultimately, we hope to improve breast cancer screening and prevention through accurate personalized risk estimates not only for women with a family history of breast cancer, but also in the general population.