Performance evaluation of four prediction models for risk stratification in gastric cancer screening among a high-risk population in China.

OBJECTIVE

Early detection of gastric cancer (GC) is a critical step for decreasing mortality. The aim of this study was to evaluate the performance of four prediction models for risk stratification in the screening of GC and precancerous lesions among a large, high-risk population in China.

DESIGN

This study was a retrospective analysis of data from the Provincial Gastric Cancer Screening Program (Zhejiang, China) spanning the period between October 2016 and April 2019, in which 97,541 individuals from the urban areas of 10 cities in Zhejiang province, China participated in this program. Demographic and clinical characteristics data were collected, and serum pepsinogens I and II, gastrin-17, and anti-H. pylori IgG antibody were detected. Participants were asked to voluntarily undergo gastroscopy. The performance of the ABC method, new ABC method, Tu's prediction model, and Li's prediction model, which stratified participants into low-, medium- and high-risk subgroups, were evaluated using the area under the receiver-operating characteristic (ROC) curve (AUC) and Youden index.

RESULTS

Among the participants, 6005 (3447 males and 2558 females, mean age of 58.35 years), voluntarily underwent gastroscopy. Overall, 72 (1.20%) GC cases (30 early and 42 advanced) and 2006 cases with precancerous lesions (270 atrophic gastritis, 1634 intestinal metaplasia, and 102 dysplasia/intraepithelial neoplasia) were identified. Notably, Li's prediction model achieved the greatest AUC and Youden index values (0.708 and 0.319, respectively) for predicting GC, and exhibited the greatest ability to detect precancerous lesions, especially intestinal metaplasia.

CONCLUSION

Li's prediction model performs the best for risk stratification in the screening, detection, and diagnosis of GC and precancerous lesions, whereas the overall performance of the other three models is similar ( www.chictr.org.cn , ChiCTR2100043363).