2026 年第 10 期 第 21 卷

新疆克拉玛依地区人群胃癌和癌前病变情况调查及预测模型构建与验证

Investigation on gastric cancer and precancerous lesions and construction and validation of prediction model in population of Karamay, Xinjiang

作者：刘慧芳1安红梅2崔童1

英文作者：Liu Huifang1 An Hongmei2 Cui Tong1

单位：1新疆维吾尔自治区人民医院克拉玛依医院克拉玛依市中心医院消化内科，克拉玛依834099；2新疆第二医学院，克拉玛依834000

英文单位：1Department of Gastroenterology Karamay Hospital of Xinjiang Uygur Autonomous Region People′s Hospital Karamay Central Hospital Karamay 834099 China; 2Xinjiang Second Medical College Karamay 834000 China

关键词：胃癌；癌前病变；新疆克拉玛依地区；油田公司

英文关键词：Gastriccancer;Precancerouslesions;Karamay,Xinjiang;Oilfieldcompany

• 摘要：

• 目的  探究新疆克拉玛依地区某油田公司员工及家属胃癌和癌前病变的患病情况，构建预测模型并验证其效能。方法 选取2023年1月至2025年6月新疆克拉玛依地区某油田公司员工及家属中参与胃癌筛查的人群作为研究对象。将经胃镜和组织病理学检查确定为胃癌以及癌前病变的患者纳入观察组（256例），另按照1∶1比例匹配无胃癌及癌前病变人群纳入对照组（256例）。测量2组调查对象血清胃蛋白酶原Ⅰ（PG Ⅰ）、PG Ⅱ、胃泌素17水平和测量幽门螺杆菌（Hp）含量，并进行问卷调查。比较2组PG Ⅰ、PG Ⅱ、PG Ⅰ/PG Ⅱ比值（PGR）、胃泌素17、Hp及一般情况、家族遗传史、生活习惯、饮食习惯等。采用二分类条件Logistic回归分析胃癌及癌前病变患者的相关因素，建立风险预测模型并评价。结果 多因素Logistic回归分析发现，年龄、男性、胃癌家族史、吸烟、饮酒、饮食腌制品、饮食熏制品、饮食辛辣食物、饮食烫热、三餐不规律、焦虑、抑郁、Hp阳性、胃泌素17均是胃癌及癌前病变患者的危险因素（均P＜0.05），PGR是保护因素（P＜0.05）。随机森林根据基尼系数减少平均值排序所示，Hp阳性、PGR、饮食腌制品、胃癌家族史、饮酒是胃癌及癌前病变相关因素中重要性占据前五位的指标，对胃癌及癌前病变有重要影响。受试者工作特征曲线分析结果表明Logistic回归模型预测新疆克拉玛依地区某油田公司员工及家属胃癌及癌前病变的敏感度、特异度、曲线下面积分别为91.8%、88.7%、0.938；经Hosmer-Lemeshow检验，Logistic回归模型预测概率与实际概率拟合良好（χ2=3.159，P=0.604）；Bootstrap法内部验证显示该模型预测的敏感度、特异度、曲线下面积分别为90.5%、88.3%、0.931；模型预测概率与实际概率的平均绝对误差为0.031，模型预测胃癌及癌前病变的概率与实际发生率一致性较好。结论 年龄、男性、胃癌家族史、吸烟、饮酒、饮食腌制品、饮食熏制品、饮食辛辣食物、饮食烫热、三餐不规律、焦虑、抑郁、Hp感染、PGR、胃泌素17均与新疆克拉玛依地区某油田公司员工及家属胃癌和癌前病变发生相关，基于此构建的Logistic回归模型可推广用于油田职工及家属胃癌及癌前病变风险的筛查中。

• Objective To investigate the prevalence of gastric cancer and precancerous lesions among employees and their families of an oilfield company in Karamay, Xinjiang, construct a prediction model and verify its efficacy. Methods Participants undergoing gastric cancer screening from employees and their families of an oilfield company in Karamay, Xinjiang from January 2023 to June 2025 were enrolled as research subjects. Patients confirmed with gastric cancer and precancerous lesions by gastroscopy and histopathological examination were included in the observation group (256 cases), and individuals without gastric cancer and precancerous lesions were matched at a ratio of 1∶1 into the control group (256 cases). Serum levels of pepsinogen Ⅰ (PG Ⅰ), PG Ⅱ, gastrin-17 and Helicobacter pylori (Hp) content were detected in the two groups, and questionnaire surveys were conducted. PG Ⅰ, PG Ⅱ, PG Ⅰ/PG Ⅱ ratio (PGR), gastrin-17, Hp status, general information, family genetic history, living habits and dietary habits were compared between the two groups. Binary conditional Logistic regression was used to analyze the related factors of gastric cancer and precancerous lesions, and a risk prediction model was established and evaluated. Results Multivariate Logistic regression analysis showed that age, male gender, family history of gastric cancer, smoking, alcohol consumption, intake of pickled food, smoked food, spicy food, scalding hot food, irregular three meals, anxiety, depression, Hp positivity and gastrin-17 were risk factors for gastric cancer and precancerous lesions (all P＜0.05), while PGR was a protective factor (P＜0.05). According to the mean decrease in Gini coefficient sorted by random forest, Hp positivity, PGR, pickled food intake, family history of gastric cancer and alcohol consumption ranked the top five in importance among related factors, exerting significant impacts on gastric cancer and precancerous lesions. Receiver operating characteristic curve analysis showed that the sensitivity, specificity and area under the curve of the Logistic regression model for predicting gastric cancer and precancerous lesions were 91.8%, 88.7% and 0.938 respectively. Hosmer-Lemeshow test indicated good fitting between predicted probability and actual probability of the model (χ2=3.159, P=0.604). Internal validation by Bootstrap method showed that the sensitivity, specificity and area under the curve were 90.5%, 88.3% and 0.931 respectively; the mean absolute error between predicted probability and actual probability was 0.031, with good consistency between predicted risk and actual incidence of gastric cancer and precancerous lesions. Conclusion Age, male gender, family history of gastric cancer, smoking, alcohol consumption, intake of pickled food, smoked food, spicy food and scalding hot food, irregular three meals, anxiety, depression, Hp infection, PGR and gastrin-17 are associated with the occurrence of gastric cancer and precancerous lesions among employees and their families of the oilfield company in Karamay, Xinjiang. The constructed Logistic regression model can be popularized for risk screening of gastric cancer and precancerous lesions in oilfield staff and their families.