It is time for polygenic risk scores to be used in breast cancer risk assessments
George Busby, Allelica CSO & Co-Founder
Much of the development of polygenic risk scores (PRSs) centers on assessing their ability to predict risk in different populations. This undoubtedly provides assurance of their clinical validity, which is a crucial component of PRS development. As we’ve written about before, when deciding on whether a genetic test should be used, we also need to understand how it can be used to inform clinical decision making.
For some diseases, most notably Atherosclerotic Cardiovascular Disease (ASCVD) and breast cancer (BC), there are already established clinical models that assess risk and guidelines that can be used to provide direction on how to act when risk is increased. Examples include assessment of 10 year ASCVD risk with the Framingham Risk Score, Pooled Cohort Equations, SCORE2 or QRISK3 algorithms, and the Tyrer-Cuzick, Gail, and BOADICEA algorithms for defining 5 year, 10 year or lifetime risk of BC.
Given their demonstrable ability to identify women at high risk, how should we think about combining PRSs within BC risk models to ensure that their potential should be realized?
In this article, we’ll explore how PRSs can be used to augment clinical risk models for BC and will argue that it is no longer acceptable to ignore the key component of disease risk that can be gleaned from a genetic assessment of BC risk.
Where we are now with breast cancer prevention
When it comes to cancer prevention, what we really mean is early detection. While there are some environmental risk factors that influence cancer risk (obesity and poor lifestyle being the main ones), the key driver that lowers cancer mortality is catching the disease early. Breast cancer is one of the most common cancers, and over the last two decades disease incidence has gone up and mortality has gone down. The reason that we see this pattern is because national scale screening programs have been implemented at scale across most of the world. Typically using mammograms to identify subclinical disease, these screening programs provide an opportunity for most women to get checked regularly so that BC can be detected early.
These national screening programs largely run using a one-size fits all approach. National guidelines dictate when to start screening and the cadence at which to have repeat surveillance. The main criteria for when to start screening is a woman’s age, and this tends to be the same for the whole population.
At this point, it’s important to highlight that the vast majority of women will not get breast cancer. Across western populations, there is an approximately 1 in 8 chance of getting breast cancer, which means that 7 out of 8 (87%) of women won’t get BC. It is also important to note that this 1 in 8 risk is not equal across the population: some women will have higher (or lower) risk than others due to measurable differences in risk factors.
If women at higher risk can be accurately identified, then screening programs will stand to benefit by targeting such women with earlier and more frequent surveillance.
So why don’t all women have an equal risk of disease? A range of risk factors are known to increase BC risk, including exposure to estrogen (for example because of earlier menarche, later menopause, or long term use of hormone replacement therapy), physiological factors such as high BMI and breast density, and family history of disease.
A woman’s genetics also influences her breast cancer risk. This includes risk from high impact, but exceedingly rare, genetic mutations in known cancer susceptibility genes such as BRCA1/2, TP53, PALB2, CHEK2 and ATM as well as polygenic risk based on much more common variants. Currently, being a carrier of a rare variant can be ascertained through a genetic test which is typically offered if a family member is a known carrier. This can be used to identify women at higher than average population risk, at which point more intensive strategies can be implemented to mitigate this risk.
However, despite their proven ability to identify women at high breast cancer risk, PRSs are not currently part of the toolbox available to clinicians wishing to use all available information to classify a woman’s risk.
Acknowledging both that BC is common and its causes multifaceted, national guidelines, such as those released by the NCCN and which are utilized by physicians across the US, promote the quantification of risk so that, for example, those identified at a greater than 1.7% 5 year risk or greater than 20% residual lifetime risk should be eligible for increased risk screening.
Measuring breast cancer risk
It follows that to deploy risk-based stratification of screening and intervention resources, you need to be able to measure risk. This is enabled through risk assessments such as those mentioned above. These can be used to identify women at high risk, who can then receive tailored, guideline-informed risk mitigation strategies to counteract this risk. These assessments can be as basic as understanding whether there is a family history of early onset breast cancer, to more detailed assessments of a range of clinical risk factors that produce an overall estimate of a woman’s risk.
The most commonly used BC risk assessment tool in the US is the Tyrer-Cuzick model. This uses information on a range of different risk factors to estimate a woman’s 10 year and residual lifetime risk of disease, and national guidelines explicitly define thresholds above which certain actions should be taken.
Genetic testing for the presence of known pathogenic variants in known breast cancer susceptibility genes can be used to identify women with ‘hereditary cancer syndrome’. This is initiated when there is some evidence that a woman might be carrying a mutation, either because a close relative is known to carry a mutation, or because a close relative suffered from early onset BC.
Either way, the implicit understanding is that risk can be measured and those at higher risk should be offered more opportunity to proactively mitigate that high risk.
The role of polygenic risk in cancer assessments
So where do PRSs for breast cancer play a role in risk assessments? As we alluded to above, there are is now compelling evidence from multiple groups, based on analyses across different populations (e.g. UK, USA, Finland, Japan), that PRSs identify a component of disease risk that is not captured by other risk factors, including family history, and that this risk is far more common in the general population than genetic risk conferred by rare pathogenic variants in known BC susceptibility genes such at BRCA1/2, ATM, PALB2 and CHEK2.
We recently added to this evidence in a new preprint that describes our work building and validating ancestry-specific PRSs. Importantly, we demonstrated that our PRSs have good and largely equivalent predictive power across diverse populations. It is important to demonstrate that PRSs can be shown to predict risk across different groups, as this is often held up as a reason to hold back on clinical implementation. The lack of transferability of PRSs is no longer an argument against their broad-scale use.
Models with PRS are better than those without
Given the accepted importance of other risk factors we wanted to go further and explore the use of PRS within clinical risk models. To this end, we integrated our PRSs into the Tyrer-Cuzick risk model and demonstrated that by applying these PRSs to populations comprising individuals of diverse genetic ancestries we can improve the performance of the Tyrer-Cuzick risk model in identifying women at various high risk disease thresholds.
So models integrating PRSs lead to better overall risk prediction, and there’s also evidence that PRSs identify women whose disease occurs earlier on in life. As we’ve discussed, cancer prevention is really about early detection, so we want to improve our ability to identify women at higher risk so that they can be checked more often and from an earlier age.
The effect of having a high breast cancer PRS relative to single gene variants
In addition to improvements in clinical risk models, we also assessed the level of risk conferred by having a high PRS on its own, and compared this to the risk conferred from carrying a rare pathogenic variant. We found that our PRSs can identify women at equivalent risk of breast cancer to carriers of pathogenic variants in key breast cancer susceptibility genes, but who are much more common in the population.
PRSs are just as good at identifying risk as single gene variants in all but the most penetrant BC susceptibility genes, and potentially better, given the proportion of the population identified at high risk from PRSs is much greater.
Additional use cases for PRSs in breast cancer
Our research has concentrated on demonstrating the value and utility of integrating PRSs in clinical risk models, but identifying women at high risk of BC has at least two other potential uses.
Breast density is an important risk factor for breast cancer. Unfortunately, while dense breasts lead to increased risk, they also cause mammograms to be less effective. In such cases, Magnetic Resonance Imaging (MRI) is a better, but much more expensive, option for screening. PRSs offer a way of identifying women with dense breasts who are at greater risk and therefore who should be prioritized for MRI over mammographic surveillance.
Multi-cancer early detection (MCED) tools leverage the fact that tumours, even very small ones, release DNA fragments into the bloodstream, which can be picked up through a genetic test. These tests are exciting, but currently suffer from low positive predictive power. That is, there are a lot of false positives, or people who get a positive test result who do not actually have cancer. PRSs can be used to identify those who are at high risk of breast (and other) cancers to improve the predictive value of these tests.
Driving down breast cancer mortality with PRSs
Around one third of new breast cancers in the United States are diagnosed at a late-stage. As we note above, earlier diagnosis leads to better outcomes. In the US, five year survival from late stage diagnosis is much lower (25%) than when BC is diagnosed earlier (88%). As approximately 300,000 women a year are diagnosed with breast cancer, this means that there are 100,000 women every year who are being diagnosed with breast cancer later than they could be.
Allelica’s breast cancer PRSs can be used to improve risk classification of clinical risk models. Using a lifetime high risk threshold of 20%, our latest research found a 5% improvement in the classification of risk using our PRS-integrated version of the Tyrer-Cuzick model compared to the standard version that does not include PRS.
If PRS-integrated risk models were used to identify more of these late stage diagnoses at an earlier stage, then, as many as 5,000 breast cancer cases a year might be identified earlier, this could potentially translate to up to 3,150 or 1.5% fewer annual BC deaths based on the difference in 5 year survival between early and late stage diagnosis.
If you’re interested in providing PRS tests to your patients or developing clinical PRS tests in your lab please get in touch with us today.
