广西医学

目的探讨影响胃肠道间质瘤（GIST）患者术后肿瘤特异性生存（CSS）情况的因素，并构建患者术后CSS情况的列线图预测模型。方法纳入SEER数据库中5 207例GIST术后患者的临床资料。将患者按7 ∶3比例随机分为建模组（n=3 647）和验证组（n=1 560）。采用单因素和多因素COX回归模型分析GIST患者术后CSS情况的影响因素，并基于多因素COX回归分析结果采用R语言软件构建GIST患者术后1年、3年和5年CSS情况的列线图预测模型。使用一致性指数、校准曲线、受试者工作特征（ROC）曲线评估模型性能，采用K折交叉验证法进行内部验证，采用决策曲线评估模型的临床适用性。基于净重新分类改善指数、综合判别改善指数、一致性指数的改善、ROC曲线下面积比较列线图预测模型与改良美国国立卫生研究院（M-NIH）分级的预测性能。采用X-tile软件确定列线图预测模型的最佳截断值，以此对GIST患者重分类后使用Kaplan-Meier法进行生存分析。结果多因素COX回归分析结果显示，年龄、性别、肿瘤大小、肿瘤位置、肝转移情况、美国癌症联合委员会（AJCC）分期、分化程度、综合分期、有丝分裂率、淋巴结检出数量及手术方式为GIST患者术后CSS情况的影响因素（P＜0.05）。基于影响因素（除综合分期外）构建的列线图预测模型具有中等的区分度及良好的校准度、预测效能及泛化能力。列线图预测模型的预测性能、临床获益均优于M-NIH分级。以184.7分和281.0分为最佳截断值将GIST患者重分类为高、中、低风险组，在建模组和验证组中，3个亚组患者的术后CSS情况差异有统计学意义（P＜0.05）。结论年龄、性别、肿瘤大小、肿瘤位置、肝转移情况、AJCC分期、分化程度、综合分期、有丝分裂率、淋巴结检出数量及手术方式是GIST患者术后CSS情况的影响因素。基于相关影响因素构建的列线图预测模型的预测性能优于M-NIH分级，有助于临床医师更好评估GIST患者的术后生存状态，并在术后随访中优化临床决策。

ObjectiveTo explore the factors affecting postoperative cancer-specific survival (CSS) status in patients with gastrointestinal stromal tumor (GIST), and to establish a nomogram prediction model of postoperative CSS status in patients. MethodsThe clinical data of 5207 GIST patients after operation were enrolled from the SEER database. Patients were randomly divided into modeling group (n=3647) or validation group (n=1560) according to a ratio of 7 ∶3. The univariate and multivariate COX regression models were employed to analyze the influencing factors for postoperative CSS status in GIST patients, and a nomogram prediction model of 1-, 3-, and 5-year postoperative CSS status in GIST patients was established by using the R language software based on the results of multivariate COX regression analysis. The model performance was evaluated by using consistency index, calibration curve, receiver operating characteristic (ROC) curve. The K-fold cross-validation was used to perform internal validation, and the decision curve was employed to evaluate clinical applicability of the model. The predictive performance was compared between the nomogram prediction model and Modified National Institutes of Health (M-NIH) classification based on net reclassification improvement index, integrated discrimination improvement index, consistency index improvement, and area under the ROC curve. The X-tile software was used to determine the optimal cut-off value of the nomogram prediction model, so as to perform survival analysis by using Kaplan-Meier method after reclassifying GIST patients. ResultsThe results of multivariate COX regression analysis revealed that age, gender, tumor size, tumor location, liver metastasis status, American Joint Committee on Cancer (AJCC) stage, differentiated degree, comprehensive stage, mitotic rate, number of lymph node detection, and surgical method were the influencing factors for postoperative CSS status in GIST patients (P＜0.05). The nomogram prediction model established based on these influencing factors (except for comprehensive stage) exerted medium discrimination, and favorable calibration, predictive efficiency, generalization ability. The predictive performance, clinical benefit of the nomogram prediction model were superior to M-NIH classification. GIST patients were reclassified as high-, medium-, and low-risk group when taking 184.7 points and 281.0 points as the optimal cut-off value, and in the modeling group and the validation group, there was a statistically significant difference in postoperative CSS status between the 3 subgroups (P＜0.05). ConclusionAge, gender, tumor size, tumor location, liver metastasis status, AJCC stage, differentiated degree, comprehensive stage, mitotic rate, number of lymph node detection, and surgical method are the influencing factors for postoperative CSS status in GIST patients. The predictive performance of the nomogram prediction model established based on relevant influencing factors is superior to M-NIH classification, which is beneficial for clinicians preferably evaluate postoperative survival status in GIST patients, and optimize clinical decision-making during postoperative follow-up.