Abstract: A Predictive Nomogram for the Risk of Tumor Deposits in Stage I-III Gastric Cancer        HUA Ke-lei*，HUO Ming-ke，DONG Zhi-chuang，et al. *Department of General Surgery，Henan Cancer Hospital，Affiliated Tumor Hospital of Zhengzhou University，Zhengzhou 450008，China
Corresponding author：REN Ying-kun，E-mail：18903839515@163.com
Abstract    Objective    To screen the risk factors of tumor deposits (TDs) with stage I-III gastric cancer，and use them to develop a predictive nomogram as a visualization tool assisting clinical prediction of TDs occurrence. Methods  Participants (n=870) were stage Ⅰ-Ⅲ gastric cancer who were retrospectively selected from the Affiliated Hospital of Zhengzhou University from September 2015 to September 2018，and 75% of them were randomly assigned to a training group (n=654) and the other 25%  to a verification group (n=216). Patients’ Clinicopathological data and biochemical data were collected. Independent predictors of TDs were screened by Lasso regression analysis，and further analyzed using multivariate Logistic regression analysis，then the finally determined ones were used to develop a predictive nomogram. The performance of the nomogram was verified in the verification group. Finally，the area under the ROC curve（AUC），calibration curve and decision curve analysis（DCA） were used to evaluate the identification ability，accuracy and clinical applicability of the nomogram. Results    Among the 870 cases，121 had TDs，and the other 749 did not. The overall survival rate and disease-free survival rate of TDs positive patients were significantly worse than those of negative patients. The findings of Lasso regression with multivariate Logistic regression analyses showed that tumor size （OR=1.363，95%CI 1.172-1.584），CEA（OR=1.041，95%CI 1.004-1.078），CA19-9（OR=1.007，95%CI 1.003-1.011），pT（OR=2.229，95%CI 1.397-3.557） and pN（OR=1.639，95%CI 1.271-2.113） were associated with TDs. The predictive nomogram was established by employing the above-mentioned variables. The AUC of the nomogram for identifying TDs in the training group was 0.865（95%CI 0.826-0.904），and in the validation group was 0.858（95%CI 0.783-0.912），the calibration curve displayed a general consistency between the modeling and actual curves. According to the DCA of the modeling and actual groups，the nomogram had a better clinical benefit when the probability threshold was set at 1%-99%，respectively. Conclusion    We successfully established and verified a nomogram （with the above-mentioned five variables ） with a high accuracy，which may be used as a tool facilitating the improvement in screening of TDs in high-risk gastric cancer patients.

Key words: stomach neoplasms, tumor deposits, risk factors, prediction model, nomogram

摘要： 目的    分析Ⅰ~Ⅲ期胃癌病人癌结节发生的危险因素，构建并验证能够辅助临床预测癌结节发生的可视化评价工具。方法    回顾性分析2015年9 月至2018年9月郑州大学附属肿瘤医院普外科收治的870例Ⅰ~Ⅲ期胃癌病人的临床资料，在R软件下按照3∶1 的比例使用简单随机抽样将病人分为训练集（n=654）和验证集（n=216）。通过 Lasso 回归分析筛选独立预测因素，利用多因素 Logistic 回归分析进一步探讨并建立列线图预测模型，并由验证集评估预测癌结节发生列线图预测模型的可行性；分别采用受试者工作特征（ROC）曲线下面积（AUC）、校正曲线和决策曲线分析（DCA）对预测模型的区分度、准确度和临床实用性进行评估。结果    870例病人中，癌结节阳性病人121例，阴性病人749例，癌结节阳性病人在总生存率和无病生存率方面均劣于阴性病人，差异有统计学意义（P<0.05）。Lasso 回归结合多因素Logistic回归分析结果显示，肿瘤大小（OR=1.363，95%CI 1.172-1.584）、CEA（OR=1.041，95%CI 1.004-1.078）、CA19-9（OR=1.007，95%CI 1.003-1.011）、pT分期（OR=2.229，95%CI 1.397-3.557）和pN分期（OR=1.639，95%CI 1.271-2.113）是胃癌病人发生癌结节的独立危险因素（P均<0.05）。利用上述变量构建列线图预测模型在训练集中预测癌结节发生的AUC为0.865（95%CI 0.826-0.904），验证集中为0.858（95%CI 0.783-0.912）。在验证集和训练集中校准曲线均显示出较好的拟合度。训练集和验证集DCA的结果显示阈值概率在1%~99%时使用该模型预测癌结节发生风险的净收益更高。结论    用肿瘤大小、CEA、CA19-9、pT和pN分期构建列线图预测模型，有助于提高发生癌结节高危胃癌病人的识别和筛选能力。

关键词: 胃肿瘤, 癌结节, 危险因素, 预测模型, 列线图