Background: 
Gastric cancer (GC) remains a major public health burden in China, and the limited predictive accuracy of existing questionnaire-based risk models highlights the need for improved risk stratification strategies. 
Objectives: 
To determine whether integrating a GC polygenic risk score (PRS112) with the non-genetic Gastric Cancer Risk Score (GCRS) prediction model improves risk prediction and high-risk identification for GC in population-based screening. 
Methods: 
In this two-stage study, we first calculated PRS112 and GCRS in 86,898 participants with complete data for genetics and questionnaire risk factors from the China Kadoorie Biobank (CKB). Then, we developed a new GC integrated risk prediction tool (GC-IRT) that combined PRS112 with GCRS for the 10-year GC risk estimation. Model discrimination, calibration, and net reclassification improvement (NRI) for the prediction of incident GC events were assessed compared with the GCRS model alone. Finally, the clinical suitability of the GC-IRT was evaluated in 5,488 subjects from an endoscopy screening program in Yangzhou. 
Results: 
The integration of PRS112 significantly increased the prediction accuracy, and the change in the C statistic when adding to the GCRS was 0.010 (95% CI: 0.004-0.016) in the CKB cohort. Compared with the GCRS model, the GC-IRT led to a better separation between high- and low-risk groups when using a 10-year risk threshold of 2.1%. Overall, 6.29% (40/636) of incident GC cases were misclassified as low risk by GCRS and correctly classified as high risk by the GC-IRT, compared with 1.89% (12/636) misclassified by the GC-IRT and correctly classified by GCRS. The GC-IRT significantly improved categorical NRI (4.16%, 95% CI: 1.51%-6.57%). The cumulative incidence curves showed that individuals up-classified by the GC-IRT had a 10-year GC risk comparable to those at shared high risk (2.72% vs. 2.60%), and higher than those down-classified by the GC-IRT (2.72% vs. 1.18%). Likewise, all the screened participants in Yangzhou demonstrated similar findings, and those up-classified by GC-IRT had a higher detection rate of GC. 
Conclusions: 
Incorporating a GC PRS into an established clinical risk model improves the precision of risk assessment and identification of high-risk individuals. These findings offer important real-world evidence supporting the potential value of implementing a PRS into clinical and screening practices.