Skip to main content

Conference Program

Subpage Hero

Loading

Assessing Breast Cancer Risk: A Comparative Analysis of Ancestry-Adjusted PRS Models in Women of Ashkenazi Jewish Heritage

Cancer Genetics and Therapeutics
  • Primary Categories:
    • Genetic Counseling
  • Secondary Categories:
    • Genetic Counseling
Introduction:
The ability to predict breast cancer (BC) risk has seen promising advances through the use of polygenic risk scores (PRS), which calculate an individual's genetic predisposition by analyzing multiple genetic variants. However, a critical limitation of current PRS models is their lack of adjustments for individuals from diverse ancestral backgrounds, which can reduce their predictive accuracy for populations outside of those of primarily European descent. In this study, we aimed to improve the accuracy of BC risk prediction for women of Ashkenazi Jewish (AJ) heritage by examining the performance of three PRS models.

Methods:
We evaluated the BC risk prediction of three PRS models: the Mavaddat 313-SNP PRS (PRS313), a cross-ancestry PRS model adjusted for five different ancestry groups (caPRS), and an enhanced cross-ancestry model (caPRSx), specifically adding an AJ ancestry group, alongside European, African, South Asian, East Asian, and Admixed American groups. Ancestry-specific principal components (PCs) were calculated using 1000 Genomes Project data, with 100 additional AJ samples for caPRSx. Each ancestry-specific PRS was adjusted by subtracting the linear regression-based PRS predicted from the first five PCs in unaffected individuals, then normalized using the standard deviation (SD) of the corresponding population in the reference set. A cross-ancestry PRS model was constructed by linearly combining the highest-performing PRS for each ancestry, weighted by fractional ancestry.

The association between BC risk and caPRS models was evaluated in AJ women from the UK Biobank and Women’s Health Initiative using multivariable logistic regression adjusted for age, ovarian cancer history, family history of BC, and cohort. Model performance was assessed via PRS skewness, OR per SD, and changes in remaining lifetime risk when integrating each PRS with the Tyrer-Cuzick clinical model.

Results:
Our findings revealed notable differences in skewness across the three PRS models, with skewness values indicating that caPRSx had the least skewed distribution. Specifically, Pearson skewness coefficients for PRS313, caPRS, and caPRSx were 0.050, 0.039, and 0.0022, respectively, suggesting that caPRSx more closely aligned with a normal distribution, which is associated with more accurate predictions.

All three models showed significant associations with BC risk in the validation cohorts, with caPRSx demonstrating the best performance for AJ women. The odds ratios per SD were 1.53 (95% CI, 1.45–1.62) for PRS313, 1.66 (95% CI, 1.57–1.75) for caPRS, and 1.67 (95% CI, 1.58–1.77) for caPRSx. Adjusting the PRS using a reference panel that included AJ women resulted in a lower average lifetime risk estimate, decreasing by 3% across 2,068 AJ women in the validation cohorts. This underscores the critical importance of accurate reference panels in generating reliable PRS calculations. The observed discrepancies are likely due to insufficient correction for genetic background and the absence of AJ women in the 1000 Genomes reference dataset.

Conclusion:
In summary, the caPRSx model outperformed the PRS313 and caPRS models in predicting breast cancer risk among AJ women. These findings underscore the importance of including specific ancestry adjustments in polygenic risk models, as demonstrated by the superior performance of caPRSx when AJ ancestry was incorporated. Given the known association of BRCA1 and BRCA2 mutations in AJ women and the heightened screening in this population, it is essential that all breast cancer PRS models account for populations beyond the 1000 Genomes reference dataset when implemented clinically. Implementing such refined models in clinical practice could support more precise risk assessments and contribute to risk appropriate preventive strategies for breast cancer among women from diverse backgrounds.

Agenda

Sponsors