New algorithms that better predict breast cancer risk
Some 90 000 women die of breast cancer every year in the EU despite the relatively high efficacy of first-line treatments. For those with a genetic predisposition, disease can be prevented through the likes of intensified screening, chemoprevention or prophylactic surgery. But even for the most cautious and informed women, these efforts are still very much hit or miss. The trouble is, cancer risks associated with most gene variants are still unknown, or have large uncertainties. To enable better patient counselling and management, practitioners need more reliable evaluation methods. The BRIDGES (Breast Cancer Risk after Diagnostic Gene Sequencing) project was launched in 2015 to allow for more accurate identification of women at high risk of breast cancer. Peter Devilee, coordinator of the project on behalf of the Leiden University Medical Center (LUMC), discusses the project’s breakthroughs and their expected impact on future cancer care.
What are the remaining obstacles to the association of specific genes with breast cancer risks?
Peter Devilee: We have been able to significantly associate a gene with breast cancer risk for a while now, but there is indeed a problem with the confidence intervals of the effect sizes which are generally too wide. Moreover, existing association analyses have been conducted on too small samples, which means that some uncovered associations could be spurious. An important question prior to our project was how to best determine this effect size, because many of the existing approaches introduce statistical biases. The main obstacle is sample size and the need for extensive and accurate descriptions of a patient’s disease history.
How do you propose to overcome these problems?
We proposed two things in our project: The first is to investigate the association of breast cancer with each gene tested by commercial companies on their ‘oncogene panels’ in a very large series of cases and controls. The second objective focuses on the genes that were already solidly associated with breast cancer: We want to assess with the highest possible accuracy their effect size, that is, try to narrow down existing confidence intervals. We previously built a very large case-control data set with DNA samples and extended clinical data, which covers over 120 000 individuals from the general population, which provided us with a head start.
Did you use other data sources as well?
No, but the project required us to generate the DNA sequencing of the genes of all individuals who took part in our studies. Given the large sample-size – unprecedented in 2014, when the project was conceived – we needed to develop a methodology that had a very high throughput at a very low cost per sample. We finally managed to sequence 35 genes at a price under EUR 10 per sample.
What are the project’s most important outcomes?
The project has succeeded in narrowing down confidence intervals for the five ‘major’ breast cancer genes: BRCA1, BRCA2, PALB2, CHEK2 and ATM. Four other genes (BARD1, RAD51C, RAD51D, TP53) were also definitively uncovered as ‘breast cancer genes’. This will have important clinical ramifications for women receiving genetic counselling. We excluded 19 genes from being associated with breast cancer, although there remains a remote possibility that these genes may be associated with a very low risk (the rarity of their occurrence precluded the exclusion of risks <2-fold). For several other genes, the study found suggestive associations with, for example, certain subtypes of breast cancer such as oestrogen receptor-negative breast cancer, which has certain prognostic features. Larger follow-up studies will have to find out whether these associations are real or not.
How do your online tools work exactly? Who can use them?
The results of our study are currently being incorporated into an online tool called CanRisk. The algorithm running behind that tool is called BOADICEA and has been developed by one of the BRIDGES partners, the University of Cambridge. This tool combines various risk factors, both genetic and non-genetic (such as family history, BMI, hormonal status, parity, etc.), into a single risk-score. With this score, women know what their risk of developing breast cancer over their lifetime is. Anyone would be able to use the tool, but at present the entry of variables is so strictly protocolised and demanding that it’s intended for use by healthcare providers, for example genetic counsellors.
What are the concrete benefits for patients?
The tool is intended mostly for healthy women who would like to prevent breast cancer development. This includes, for example, women who suspect they are at risk (for example because their mother had breast cancer) and are considering preventive measures such as prophylaxis, more intensive screening, or lifestyle adaptations. For breast cancer patients, the tool might help predict the odds of developing a second breast cancer, but that feature is still under development.
What are your follow-up plans, if any?
Because the tool has been externally validated and CE-marked as a medical device, it is ready to be introduced in cancer family clinics, and several centres are already exploring its use. We will need to find out when to use it, how to use it, and how individuals will respond to this knowledge of personalised risk. Another potential impact is related to population screening programmes for breast cancer that are being run in many EU countries. These are usually offered to women reaching a certain age, but many have been advocating a risk-based entry into these programmes, which should prove to be more cost-effective. This is not as simple as it sounds though. We need more evidence of our solution’s efficacy, and this implies large population-based efforts in which individuals can exploit genetic data for personal health improvement. This is an issue that receives much attention in the new Horizon Europe Health programme.
Keywords
BRIDGES, breast cancer, algorithms, clinical data, gene variants, prediction