Towards Personalized Breast Cancer Risk Management: A Thai Cohort Study on Polygenic Risk Scores

Abstract

Polygenic Risk Scores (PRS) are now playing an important role in predicting overall risk of breast cancer risk by means of adding contribution factors across independent genetic variants influencing the disease. However, PRS models may work better in some ethnic populations compared to others, thus requiring populaion specific validation. This study evaluates the performance of 140 previously published PRS models in a Thai population, an underrepresented ethnic group. To rigorously evaluate the performance of 140 breast PRS models, we employed generalized linear models (GLM) combined with a robust evaluation strategy, including Five fold cross validation and bootstrap analysis in which each model was tested across 1,000 bootstrap iterations to ensure the robustness of our findings and to identify models with consistently strong predictive ability. Among the 140 models evaluated, 38 demonstrated robust predictive ability, identified through > 163 bootstrap iterations (95% CI: 163.88). PGS004688 exhibited the highest performance, achieving an AUROC of 0.5930 (95% CI: 0.5903,0.5957) and a McFadden's pseudo R squared of 0.0146 (95% CI: 0.0139,0.0153). Women in the 90th percentile of PRS had a 1.83 fold increased risk of breast cancer compared to those within the 30th to 70th percentiles (95% CI: 1.04,3.18). This study highlights the importance of local validation for PRS models derived from diverse populations, demonstrating their potential for personalized breast cancer risk assessment. Model PGS004688, with its robust performance and significant risk stratification, warrants further investigation for clinical implementation in breast cancer screening and prevention strategies. Our findings emphasize the need for adapting and utilizing PRS in diverse populations to provide more accessible public health solutions.