Polygenic breast cancer risk scores strive to overcome racial bias



The potential of polygenic risk scores (PRSs) to become key components in the assessment of individual risk for disease in the clinical setting is inching closer to fruition; however, the technology is plagued by one glaring omission of most existing PRSs – the lack of applicability to those of non-European ancestry.

Polygenic risk scores predict an individual’s risk of disease based on common genetic variants identified in large genomewide association studies (GWASs). They have gained ground in research, as well as in the unregulated realm of the direct-to-consumer market where they are sold as add-ons to DNA ancestry kits such as 23andMe and MyHeritage.com.

While the risk scores show strong validation in estimating risk among people of European descent, their striking caveat is the lack of applicability to other ancestries, particularly African, and their use in practice outside of clinical trials is discouraged in National Comprehensive Cancer Network guidelines.

Study underscores need for ethnically diverse datasets

In a recent study published in JAMA Network Open, researchers evaluated the use of polygenic risk scores’ models in a clinical setting. Researchers tested 7 PRSs models for breast cancer risk against the medical records data of 39,591 women of European, African, and Latinx ancestry.

The PRSs models – all used only for research purposes – included three models involving European ancestry cohorts, two from Latinx cohorts, and two from women African descent.

After adjusting for factors including age, breast cancer family history, and ancestry, the PRSs from women with European ancestry highly corresponded to breast cancer risk, with a mean odds ratio of 1.46 per standard deviation increase in the score.

PRSs were also generalized relatively well among women of Latinx ancestry with a mean OR of 1.31. The authors noted that association is likely caused by Latinx individuals in the United States having a greater proportion of European ancestry than individuals with African ancestry. Importantly, however, the effect size was lower for women of African ancestry with a highest OR of 1.19 per standard deviation.

In the highest percentiles of breast cancer risk, women of European descent had odds ratio as high as 2.19-2.48, suggesting a statistically significant association with overall breast cancer risk. No statistically significant associations were found among women of Latinx and African-ancestry.

The PRSs models were smaller for women of non-European ancestry and included fewer genetic variants for women of non-European ancestry were notably smaller and hence reflected fewer genetic variants. Of the two risk scores involving African ancestry, the Women’s Health Initiative for Women with African ancestry risk score had just 75 variants, while the African diaspora study (ROOT) had 34 variants, compared with 3,820 and 5,218 in the two largest European ancestry PRSs, the Breast Cancer Association Consortium and the UK Biobank, respectively.

“These results highlight the need to improve representation of diverse population groups, particularly women with African ancestry, in genomic research cohorts,” the authors wrote.

First author, Cong Liu, PhD, of Columbia University Irving Medical Center, New York, said that efforts are underway to improve the inclusivity in the Electronic Medical Records and Genomics network data set used in this study.

“Until well-developed and validated PRSs for women with non-European ancestry become available, the current PRSs based on cohorts with European ancestry could be adapted for Latinx women, but not women with African ancestry until additional data sets become available in this important and high-risk group,” Dr. Liu and colleagues wrote.

In a commentary published with the study, Payal D. Shah, MD, of the Basser Center for BRCA at the University of Pennsylvania, Philadelphia, said that PRSs are “disproportionately applicable to patients with European ancestry and are insufficiently vetted and developed in other populations. If an instrument exists that has clinical utility in informing effective cancer risk mitigation strategies, then we must strive to ensure that it is available and applicable to all.”


Recommended Reading