基于胃镜及超声内镜下表现预测胃间质瘤恶性潜能评分系统的建立和临床验证

Establishment and clinical validation of a predictive scoring system for malignant gastric stromal tumors based on endoscopic and endoscopic ultrasound findings

摘要:

目的 建立一个术前基于胃镜及超声内镜下表现预测胃间质瘤恶性潜能的评分系统并进行验证。方法 将2017年1月1日至2023年12月31日在江苏省中医院接受治疗并经术后病理证实为胃间质瘤的286例患者作为建模组，根据美国国立卫生研究院分类系统将227例极低/低风险患者纳入低恶性潜能（low malignant potential，LMP）组，将59例中/高风险患者纳入高恶性潜能（high malignant potential，HMP）组。采用LASSO回归分析筛选变量，探究HMP胃间质瘤的预测因素，并构建列线图预测模型。采用Bootstrap法在建模组中进行内部验证。使用2021年1月1日至2023年12月31日在泰州市人民医院接受治疗并术后病理证实为胃间质瘤的85例患者资料进行外部验证。在建模组和外部验证组中绘制受试者工作特征曲线、校准曲线及临床决策曲线。结果 肿瘤长径（coef=0.755）、肿瘤形状（coef=0.015）、肿瘤位置（coef=0.008）、生长方式（coef=-0.026）、囊变性（coef =0.685）、肿瘤表面溃疡改变（coef =-0.545）是HMP胃间质瘤的独立预测因素。基于上述6个因素构建的预测模型列线图在建模组和外部验证组的受试者工作特征曲线下面积分别为0.959（95%CI：0.898~0.903）和0.959（95%CI：0.857~1.000）。采用Bootstrap法对模型进行内部验证，其准确度为0.917，Kappa=0.737。根据校准曲线及临床决策曲线所示，模型在建模组和外部验证组中均具有较好的校准度及较高的净获益度。结论 肿瘤长径、肿瘤形状、肿瘤位置、生长方式、囊变性、肿瘤表面溃疡改变是HMP胃间质瘤的独立预测因素。基于上述因素构建的列线图模型具有较好的区分度及校准度，为临床医师术前预测胃间质瘤恶性潜能提供了有效便捷的可视化工具。

Abstract:

Objective To establish a scoring system for preoperative prediction of the malignant potential of gastric stromal tumors based on gastroscopic and endoscopic ultrasound features, along with validation. Methods A total of 286 patients who were treated in Jiangsu Province Hospital of Chinese Medicine from January 1, 2017 to December 31, 2023 and diagnosed as having gastric stromal tumors by postoperative pathology were enrolled in the modeling group. According to National Institutes of Health classification system, 227 very-low/low-risk patients were classified into the low malignant potential (LMP) group, and the 59 intermediate/high-risk patients into the high malignant potential (HMP) group. LASSO regression analysis was performed to identify predictive factors for HMP gastric stromal tumors, and a nomogram prediction model was developed. Internal validation using the Bootstrap method was performed on the modeling group, and external validation was performed on data from 85 patients who were treated and diagnosed as having gastric stromal tumors by postoperative pathology in Taizhou People''s Hospital from January 1, 2021 to December 31, 2023. The receiver operator characteristic (ROC) curves, calibration curves, and decision curve analyses were employed in both the modeling and external validation groups. Results Tumor size (coef=0.755), tumor shape (coef=0.015), tumor location (coef=0.008), growth pattern (coef=-0.026), cystic change (coef=0.685), and surface unceration change (coef=-0.545) were the independent predictive factors for HMP gastric stromal tumors. The nomogram-based prediction model constructed using these factors achieved an area under the ROC curve of 0.959 (95%CI: 0.898-0.903) in the modeling group and 0.959 (95%CI: 0.857-1.000) in the external validation group. The model demonstrated good accuracy (0.917) and a Kappa value of 0.737 in internal validation. Calibration curve and decision curve analyses indicated strong calibration and high net benefit in both the modeling and the external validation groups. Conclusion Tumor size, tumor shape, tumor location, growth pattern, cystic change, and surface ulceration change are independent predictive factors for HMP gastric stromal tumors. The nomogram model developed based on these factors offers effective and convenient visualization for clinicians to predict the malignant potential of gastric stromal tumors preoperatively.