目的 构建2型糖尿病（T2MD）患者注射司美格鲁肽所致胃肠系统相关不良反应的风险预测模型。方法 选取2021年4月—2024年4月如皋市人民医院收治的141例T2MD患者作为研究对象，根据注射司美格鲁肽后是否发生胃肠系统相关不良反应，将患者分为发生不良反应组（n＝37）和未发生不良反应组（n＝104）。比较两组患者的临床资料和胃肠系统相关指标。采用单因素和多因素Logistic回归分析筛选出发生不良反应的独立影响因素。采用贝叶斯多变量联合模型分析相关指标纵向变化过程对胃肠不良反应的影响。绘制受试者工作特征（ROC）曲线评价预测模型的准确性。采用拟合优度检验（H-L）验证模型的预测能力。采用决策曲线分析（DCA）评价模型的临床实用性。结果 两组患者的体质量指数（BMI）存在统计学差异（P＜0.05）。与治疗前相比，两组患者治疗2、3和4个月后的肠鸣音次数、胃窦收缩频率（ACF）和胃窦运动指数（MI）呈现下降趋势，且发生不良反应组的下降幅度高于未发生不良反应组；胃潴留量、胃动素（MTL）、胃泌素（GAS）、胃排空时间（GET）、洛桑肠衰竭评价（LIFE）评分呈现上升趋势，且发生不良反应组的上升幅度高于未发生不良反应组（P＜0.05）。Logistic回归分析结果表明，BMI、肠鸣音次数、胃潴留量、MTL、GAS、GET是发生不良反应的独立影响因素。贝叶斯多变量联合模型结果显示，BMI每纵向升高1 kg·m-2，不良反应的发生风险升高2.291倍；肠鸣音次数每纵向升高1次·min^-1，不良反应的发生风险升高0.787倍；胃潴留量每纵向升高1 mL，不良反应的发生风险升高1.301倍；MTL每纵向升高1 mL，不良反应的发生风险升高1.149倍； GAS每纵向升高1 pg·mL^-1，不良反应的发生风险升高2.403倍； GET每纵向升高1 min，不良反应的发生风险升高1.773倍（P＜0.05）。使用BMI、肠鸣音次数、胃潴留量、MTL、GAS、GET构建预测模型，模型ROC曲线的AUC为0.786（95% CI： 0.743～0.911，P＜0.05）。H-L验证结果显示，校正曲线与实际曲线基本重合。DCA曲线验证结果显示临床实用性较好。结论 肠鸣音次数、胃潴留量、MTL、GAS、GET是T2MD患者注射司美格鲁肽所致胃肠系统相关不良反应的独立影响因素，构建的预测模型预测效果较好，有助于临床决策并降低不良反应发生风险。

Objective To construct a risk prediction model for semaglutide induced gastrointestinal system adverse reactions in patients with type 2 diabetes mellitus (T2MD). Methods A total of 141 patients with T2MD admitted to our hospital from April 2021 to April 2024 were selected as the research objects. According to whether gastrointestinal system related adverse reactions occurred after injection of semaglutide, the patients were divided into adverse reaction group (n ＝37) and non-adverse reaction group (n ＝104). The clinical data and gastrointestinal indexes of the two groups were compared. Univariate and multivariate Logistic regression analysis were used to screen out the independent influencing factors of adverse reactions. Bayesian multivariate joint model was used to analyze the influence of longitudinal changes of related indicators on gastrointestinal adverse events. The receiver operating characteristic (ROC) curve was drawn to evaluate the accuracy of the prediction model. The goodness-of-fit test (H-L) was used to verify the predictive ability of the model. Decision curve analysis (DCA) was used to evaluate the clinical practicability of the model. Results There was a significant difference in body mass index (BMI) between the two groups (P < 0.05). Compared with before treatment, the number of bowel sounds, gastric antral contraction frequency (ACF) and gastric antral motility index (MI) of the two groups showed a downward trend after 2, 3 and 4 months of treatment, and the decline in the adverse reaction group was higher than that in the nonadverse reaction group. Gastric retention, motilin (MTL), gastrin (GAS), gastric emptying time (GET) and Lausanne intestinal failure assessment (LIFE) scores showed an upward trend, and the increase in the adverse reaction group was higher than that in the nonadverse reaction group (P < 0.05). Logistic regression analysis showed that BMI, frequency of bowel sounds, gastric retention, MTL, GAS and GET were independent influencing factors for adverse reactions. The results of Bayesian multivariate joint model showed that the risk of adverse reactions increased by 0.291 times for each 1 kg·m-2 longitudinal increase in BMI. The risk of adverse reactions increased by 0.787 times for each longitudinal increase in the frequency of bowel sounds·min^-1. For each 1 mL longitudinal increase in gastric retention, the risk of adverse reactions increased by 1.301 times. For each 1 mL increase in MTL, the risk of adverse reactions increased by 1.149 times. For each 1 pg·mL^-1 increase in GAS, the risk of adverse reactions increased by 2.403 times. The risk of adverse reactions increased by 0.777 times for every 1-min longitudinal increase in GET (P < 0.05). The prediction model was constructed using BMI, number of bowel sounds, gastric retention, MTL, GAS and GET, and the AUC of the ROC curve of the model was 0.786 (95%CI: 0.743-0.911) (P < 0.05). The results of H-L verification showed that the calibration curve was basically consistent with the actual curve. The DCA curve verification results show that the clinical practicability is good. Conclusion The frequency of bowel sounds, gastric retention volume, MTL, GAS, and GET are independent influencing factors for gastrointestinal-related adverse reactions in T2MD patients caused by subcutaneous injection of semaglutide. The constructed prediction model has a good predictive effect and is helpful for clinical decision-making and reducing the risk of adverse reactions.

R977