Abstract：
OBJECTIVE: This study aimed to develop an ancestry-specific polygenic risk scores (PRSs) for the prediction of breast cancer events in Japanese females and validate it in a longitudinal cohort study.
METHODS: Using publicly available summary statistics of female breast cancer genome-wide association study (GWAS) of Japanese and European ancestries, we, respectively, developed 31 candidate genome-wide PRSs using pruning and thresholding (P + T) and LDpred methods with varying parameters. Among the candidate PRS models, the best model was selected using a case-cohort dataset (63 breast cancer cases and 2213 sub-cohorts of Japanese females during a median follow-up of 11.9 years) according to the maximal predictive ability by Harrell\'s C-statistics. The best-performing PRS for each derivation GWAS was evaluated in another independent case-cohort dataset (260 breast cancer cases and 7845 sub-cohorts of Japanese females during a median follow-up of 16.9 years).
RESULTS: For the best PRS model involving 46,861 single nucleotide polymorphisms (SNPs; P + T method with PT = 0.05 and R2 = 0.2) derived from Japanese-ancestry GWAS, the Harrell\'s C-statistic was 0.598 ± 0.018 in the evaluation dataset. The age-adjusted hazard ratio for breast cancer in females with the highest PRS quintile compared with those in the lowest PRS quintile was 2.47 (95% confidence intervals, 1.64-3.70). The PRS constructed using Japanese-ancestry GWAS demonstrated better predictive performance for breast cancer in Japanese females than that using European-ancestry GWAS (Harrell\'s C-statistics 0.598 versus 0.586).
CONCLUSIONS: This study developed a breast cancer PRS for Japanese females and demonstrated the usefulness of the PRS for breast cancer risk stratification.
METHODS: Using publicly available summary statistics of female breast cancer genome-wide association study (GWAS) of Japanese and European ancestries, we, respectively, developed 31 candidate genome-wide PRSs using pruning and thresholding (P + T) and LDpred methods with varying parameters. Among the candidate PRS models, the best model was selected using a case-cohort dataset (63 breast cancer cases and 2213 sub-cohorts of Japanese females during a median follow-up of 11.9 years) according to the maximal predictive ability by Harrell\'s C-statistics. The best-performing PRS for each derivation GWAS was evaluated in another independent case-cohort dataset (260 breast cancer cases and 7845 sub-cohorts of Japanese females during a median follow-up of 16.9 years).
RESULTS: For the best PRS model involving 46,861 single nucleotide polymorphisms (SNPs; P + T method with PT = 0.05 and R2 = 0.2) derived from Japanese-ancestry GWAS, the Harrell\'s C-statistic was 0.598 ± 0.018 in the evaluation dataset. The age-adjusted hazard ratio for breast cancer in females with the highest PRS quintile compared with those in the lowest PRS quintile was 2.47 (95% confidence intervals, 1.64-3.70). The PRS constructed using Japanese-ancestry GWAS demonstrated better predictive performance for breast cancer in Japanese females than that using European-ancestry GWAS (Harrell\'s C-statistics 0.598 versus 0.586).
CONCLUSIONS: This study developed a breast cancer PRS for Japanese females and demonstrated the usefulness of the PRS for breast cancer risk stratification.
摘要：
目的：本研究旨在开发一种祖先特异性多基因风险评分（PRS），用于预测日本女性乳腺癌事件，并在纵向队列研究中进行验证。
方法：使用日本和欧洲祖先的女性乳腺癌全基因组关联研究（GWAS）的公开汇总统计，我们,分别,使用修剪和阈值（PT）和具有不同参数的LDpred方法开发了31个候选全基因组PRS。在候选的PRS模型中，根据HarrellC-statistics的最大预测能力,使用病例-队列数据集(中位随访时间为11.9年的63例乳腺癌病例和2213个日本女性亚组)选择最佳模型.在另一个独立的病例队列数据集中评估了每个衍生GWAS的最佳PRS（260例乳腺癌病例和7845例日本女性子队列，中位随访时间为16.9年）。
结果：对于涉及源自日本血统GWAS的46,861个单核苷酸多态性（SNP；PT=0.05和R2=0.2的PT方法）的最佳PRS模型，在评估数据集中，Harrell的C统计量为0.598±0.018。PRS五分位数最高的女性与PRS五分位数最低的女性相比，乳腺癌的年龄校正风险比为2.47（95％置信区间，1.64-3.70)。使用日本血统GWAS构建的PRS对日本女性乳腺癌的预测性能优于使用欧洲血统GWAS的PRS（Harrell的C-statistics0.598对0.586）。
结论：这项研究为日本女性开发了乳腺癌PRS，并证明了PRS对乳腺癌风险分层的有用性。
方法：使用日本和欧洲祖先的女性乳腺癌全基因组关联研究（GWAS）的公开汇总统计，我们,分别,使用修剪和阈值（PT）和具有不同参数的LDpred方法开发了31个候选全基因组PRS。在候选的PRS模型中，根据HarrellC-statistics的最大预测能力,使用病例-队列数据集(中位随访时间为11.9年的63例乳腺癌病例和2213个日本女性亚组)选择最佳模型.在另一个独立的病例队列数据集中评估了每个衍生GWAS的最佳PRS（260例乳腺癌病例和7845例日本女性子队列，中位随访时间为16.9年）。
结果：对于涉及源自日本血统GWAS的46,861个单核苷酸多态性（SNP；PT=0.05和R2=0.2的PT方法）的最佳PRS模型，在评估数据集中，Harrell的C统计量为0.598±0.018。PRS五分位数最高的女性与PRS五分位数最低的女性相比，乳腺癌的年龄校正风险比为2.47（95％置信区间，1.64-3.70)。使用日本血统GWAS构建的PRS对日本女性乳腺癌的预测性能优于使用欧洲血统GWAS的PRS（Harrell的C-statistics0.598对0.586）。
结论：这项研究为日本女性开发了乳腺癌PRS，并证明了PRS对乳腺癌风险分层的有用性。
