Background
Multiple randomized controlled trials have shown that screening mammography reduces mortality from breast cancer for women who are over 50 years old [
1]. Screening programs have been set up in different countries since the 1970s with recommendations on the screening interval and the age to start screening [
2]. There is a strong consensus that women above 50 should attend routine screening. Recommendations for this group of women have remained largely unchanged for over 40 years [
1,
2].
The benefit of screening mammography for younger women aged 40 to 49 years is less clear [
1]. Recommendations for screening mammography across different countries and time periods are inconsistent and subject to change [
2,
3]. In many cases, a woman is to make a personal choice based on risk factors such as personal and family history of the disease, often with the help of professional advice from a doctor [
1].
With advances made in disease prediction, the approach to breast screening is now leaning towards a tailored, individual risk-based approach. For example, mammographic breast density, which refers to the proportion of fibroglandular breast tissue compared to fat seen on mammography, is a risk factor for breast cancer and is increasingly communicated to screening participants [
4,
5]. Women with dense breasts are informed of their increased risk of breast cancer development and reduced sensitivity of mammography to detect breast cancer so that they can make a better-informed decision as to whether they should undergo supplemental imaging screening adjunct to mammography [
5].
The risk of breast cancer is multifactorial. Apart from mammographic density, other known conventional risk factors include family history, menarche age, menopause age, height, body mass index, age at first childbirth, menopausal hormone therapy, and benign breast disease [
6]. Many of these factors have been incorporated into prediction models to estimate the personal risk of developing breast cancer [
7].
Breast cancer has a significant genetic component. It has been estimated that 27–31% of breast cancer risk may be explained by heritable factors [
8,
9]. Frequently described breast cancer predisposition genes that are highly penetrant include
ATM,
BRCA1,
BRCA2,
CHEK2,
PALB2,
BARD1,
RAD51C,
RAD51D, and
TP53 [
10]. However, pathogenic mutations in these genes are rare in the population. Polygenic risk scores (PRS) computed from another class of genetic variants that are of smaller effect sizes individually but more common in the population have shown promise to add information to better stratify individuals with different breast cancer risks as compared to age-based screening programs [
11‐
13].
Non-genetic risk prediction models are attractive as they are non-invasive and are easier to implement in a general population or primary care screening setting. Currently, breast cancer risk prediction is predominantly based on information on age, family history, lifestyle, and reproductive factors. These data can be collected at a low cost. There is evidence that genetics contribute to risk prediction but the data generation will incur additional costs to individuals or the health system. Hence, there is a need to evaluate how much information genetics can add to the identification of high-risk individuals over non-genetic risk factors. In this case-only analysis involving 7600 Asian breast cancer patients, we look at the overlap of individuals with a family history of breast cancer and those identified to be at high risk based on family history, the Gail model, breast cancer predisposition genes, and polygenic risk score.
Discussion
Currently, in many countries, population-based mammography screening is recommended based on age alone. However, not every woman is at the same level of risk of developing breast cancer. In practice, family history of the disease is widely used as a risk assessment tool. Breast cancer risk of women with a sister or a mother with breast cancer is reported to be approximately twice as high as those who do not have first-degree family members diagnosed with the disease [
26]. In addition, family history information of high quality is reported to be highly correlated to the carriership of actionable genomic variants [
21]. Prediction models using breast cancer risk factor information collected using questionnaires, such as the Gail model, are also widely used [
27]. On the individual level, these risk estimates are encouraged to be included in conversations with clinicians to help make informed decisions about potential interventions, including chemoprevention with tamoxifen [
27,
28].
While family history and conventional breast cancer risk factors may change over time and thus require updates and reassessments, an individual’s genetic risk based on either established breast cancer predisposition genes or PRS may be determined at birth. However, the implementation of genetic tests in population-wide screening is highly debatable. Pathogenic variants in high-penetrance breast cancer genes are rare; hence, most women in the general population will not benefit and may develop a false sense of security [
29]. Previously, the evidence that common genetic variants (used in the calculation of PRS) provide superior risk stratification over conventional breast cancer risk factors is lacking [
30,
31]. There was also no consensus on which variants to include in the PRS calculation. However, recent international mega-consortia studies examining over a hundred thousand women show that the tail ends of PRS enable more precise risk differentiation [
11‐
13].
With the latest developments in genetic risk prediction, it is timely to consider whether every woman in the general population should be genetically screened for high-risk genes and the use of PRS in a screening program. Our findings show that both genetic and conventional risk stratification tools have their own merits and are able to identify unique individuals at risk. Each risk assessment tool is a partial predictor at best. The inclusion of multiple predictive tools can pick up additional high-risk individuals who are missed out from using any one tool alone. In our study, family history and genetic risk perform better for women below age 50, as compared to the Gail model. This is noteworthy as the entry age for subsidized breast screening in many countries is 50 years. Genetic risk profiles will help younger women in making informed decisions on whether they should start screening at an earlier age. High-risk individuals may benefit from specific recommendations or interventions based on their personal breast cancer risk profiles.
In countries where breast screening uptake is low, breast cancer risk assessment tools function more than just predictive scores. The knowledge of breast cancer risk on an individual level may serve as a tool to motivate behavioral change. For example, a Finnish study studied the impact of genetic and non-genetic personal risk scores for cardiovascular diseases on health behavior in over 7000 participants. The results show that risk-reducing behavior is observed in participants across all risk strata, although more individuals at high risk made a health behavioral change (42.6% vs 33.5% of individuals not at high risk) [
32]. The contributions of genetic and non-genetic risk profile feedback were reported to be independent of each other [
32], further supporting the inclusion of both genetic and non-genetic risk factors for stratification in screening programs.
In terms of mammography screening, PRS has been reported to perform well at identifying the women who are most likely to benefit from this mode of detection [
33]. The association between PRS and tumor characteristics in our study confirms this observation. Not all tumors grow at the same rate. Despite the advances in technology, routine mammography screening on average fails to detect ~10–30% of all breast cancers [
5,
34]. Some of these missed tumors are interval cancers that are diagnosed between two screening episodes [
35]. Women at high risk based on PRS will thus benefit from increased screening frequency and compliance to screening.
The main strength of our study is that this is one of the largest and most well-characterized breast cancer cohorts of Asian women. However, many of the breast cancer risk assessment tools and the PRS are established based on European populations and their utility in Asian breast cancer populations remains unclear. Approximately half of the breast cancer population were identified as high risk, suggesting that other factors not considered in the risk prediction models (e.g., mammographic density, physical activity, alcohol, smoking) studied may be responsible. We classified breast cancer patients into risk categories based on 5-year absolute risks at the age of breast cancer diagnosis; this may not be representative of women without breast cancer. While we are likely to overestimate the 5-year absolute risk, results from the prospective cohort (SCHS) support the use of genetic factors on top of family history and the Gail model. As this is a case-only cohort, the proportions from the risk classification analysis are not representative of the general population, where most women will be classified as low risk. Nonetheless, this will not affect the comparison of how different criteria identify women at high risk.
While this study’s main focus was to highlight the lack of an overlap between high-risk women identified by genetic and non-genetic risk factors, it is worth noting that other works in the field have studied the potential improvements in risk prediction by combining different risk factors. For instance, Choudhury et al. explored the value of adding mammographic density and PRS to classical risk factors in a population of women of European ancestry [
36]. In a more recent study, Yang et al. assessed the performance of breast cancer risk prediction models incorporating genetic and non-genetic risk factors in 20,444 breast cancer cases and 106,450 controls from the Asia Breast Cancer Consortium [
37]. These developments are complementary to the findings of this study and will help pave the way for more patient-centric, data-driven healthcare systems in the future.
Acknowledgements
The study was made possible with the corporation of the participants and the assistance of research team members; from SGBCC – Jenny Liu, Siew Li Tan, Siok Hoon Yeo, Ting Ting Koh, Amanda Ong, Michelle Jia Qi Mo, Ying Jia Chew, Jin Yee Lee, Jing Jing Hong, Hui Min Lau, Ganga Devi D/O Chandrasegran, and Nur Khaliesah Binte Mohamed Riza. We thank the Singapore Cancer Registry for assistance in the identification of cancer cases in the SCHS cohort. MyBrCa and MyMammo thanks study participants and all research staff at Cancer Research Malaysia, University Malaya, and Sime Darby Medical Centre who assisted in recruitment and interviews (particularly Ernie Azwa Yusop, Hanani Che Halim, Faizah Harun, Farhana Fadzli, Leelavathy Krishnan, Shivaani Mariapun, Maheswari Jaganathan, Meow Keong Thong, Daphne S. C Lee, Sheau-Yee Lee, Sze-Yee Phuah, Kah-Nyin Lai, Shao-Yan Lau, Pui-Yoke Kwan, Pei-Sze Ng, Sook-Yee Yoon, Siti Norhidayu Hasan, Siu-Wan Wong, and Heamanthaa Padmanabhan) for their contributions and commitment to this study.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.