尽管已经进行了一些GWAS研究来调查膀胱肿瘤发生的基因变异，但对可能影响膀胱癌（BC）风险的基因相互作用了解甚少。通过利用来自英国生物库的大规模参与者，我们建立了一个由4000名高加索参与者（2000例和2000非病例）组成的发现数据库，一个由1648名高加索参与者（824例和824非病例）和856名非高加索参与者（428例和428非病例）组成的验证数据库。然后，我们基于机器学习方法（即GenEpi）进行了与BC风险相关的全基因组SNP-SNP相互作用调查。此外，我们使用选定的相互作用来构建一个基于Cox比例风险模型的集成相互作用增强的多基因风险评分（iPRS）的BC筛查模型。经过Bonferroni校正，我们在17条染色体上确定了10个具有统计学显著性的SNP对。其中，有4个SNP-SNP相互作用与高加索参与者的BC风险呈正相关（OR为1.57-2.03），而有6个SNP-SNP相互作用与BC风险呈负相关（OR为0.54-0.65）。只有4个SNP-SNP相互作用在非高加索参与者中被一致地确定，并位于ST7L-ADSS2、FHIT-CHDH、LARP4B-LHPP和RBFOX3-MPRIP位置。此外，与最低分位数相比，iPRS显示了1.81的风险比（95%CI：1.46-2.09），并具有比PRS（AUC：0.86；95% CI：0.76-0.95；P-DeLong检验= 2.2×10-4）更高的增强AUC（0.91；95% CI：0.85-0.97）。总之，本研究发现了几个与BC风险相关的重要SNP-SNP相互作用，并开发了一种BC筛查的iPRS模型，可以帮助在早期表现之前识别处于高风险状态的BC患者。© 2023 The Authors. International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC.

Although GWASs have been conducted to investigate genetic variation of bladder tumorigenesis, little is known about genetic interactions that may influence bladder cancer (BC) risk. By leveraging large-scale participants from UK Biobank, we established a discovery database with 4000 Caucasian participants (2000 cases vs 2000 non-cases), a database with 1648 Caucasian participants (824 cases vs 824 non-cases) and 856 non-Caucasian participants (428 cases vs 428 non-cases) as validation. We then performed a genome-wide SNP-SNP interaction investigation related to BC risk based a machine learning approach (ie, GenEpi). Moreover, we used the selected interactions to build a BC screening model with an integrated interaction-empowered polygenic risk score (iPRS) based on Cox proportional hazard model. With Bonferroni correction, we identified 10 statistically significant pairs of SNPs, which located in 17 chromosomes. Of these, four SNP-SNP interactions were found to be positively associated with BC risk among Caucasian participants (ORs 1.57-2.03), while six SNP-SNP interactions showed negatively associated with BC risk (ORs 0.54-0.65). Only four of the SNP-SNP interactions were consistently identified in non-Caucasian participants located in ST7L-ADSS2, FHIT-CHDH, LARP4B-LHPP and RBFOX3-MPRIP. In addition, the iPRS showed a HR of 1.81 (95% CI: 1.46-2.09) compared the highest tertile to the lowest tertile, with an enhanced AUC (0.91; 95% CI:0.85-0.97) than PRS (AUC: 0.86; 95% CI:0.76-0.95; P-DeLong test  = 2.2 × 10-4 ). In summary, this study identified several important SNP-SNP interactions for BC risk, and developed an iPRS model for BC screening, which may help to identify the people at high-risk state of BC before early manifestation.© 2023 The Authors. International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC.