Abstract
BACKGROUND: Early detection and screening of gastric cancer (GC) remain a critical clinical priority. This study aimed to develop and externally validate a simple, questionnaire-based prediction risk score (PRS) to identify individuals at elevated risk of GC prior to gastroscopy in outpatient settings. METHODS: This multicenter study developed the PRS using regression coefficients derived from 4800 participants in the Guangzhou cohort (training cohort) and conducted internal validation in the remaining 2058 participants (testing cohort). External validation was subsequently performed in an independent cohort of 1811 participants from Shenzhen cohort and in 424,897 participants from the China Kadoorie Biobank (CKB cohort). Participants were then stratified into average-risk and high-risk groups according to comparable levels of predicted risk. RESULTS: The PRS included five predictors: age, sex, body mass index (BMI), family history of GC in first-degree relatives, and consumption of pickled food, with a total score ranging from 0 to 11. In the training cohort, the prevalence of GC was 1.4% in the average-risk group (0–4 points) and 8.3% in the high-risk group (5–11 points) (P < 0.001). Applying gastroscopy to high-risk participants allowed detection of 85.2% (178/209) of GC cases, including 36.5% (65/178) of early cases, while 55.1% of procedures could be avoided in the average-risk group. The PRS demonstrated good discrimination across the training, testing, Shenzhen cohort, and CKB cohorts (area under the receiver operating characteristic curve [AUC]: 0.780, 0.789, 0.732, and 0.723, respectively) and showed satisfactory calibration (P > 0.05). CONCLUSIONS: The PRS has good performance in predicting the risk of GC and reducing unnecessary gastroscopy screening, thus improves cost-effectiveness and saves medical resources. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12885-025-15294-3.