Abstract: To explore a novel prediction model based on the combined use of multiple common tumor markers to improve the accuracy of prognostic prediction for gastric cancer (GC). Methods    This multi-center retrospective analysis collected 5119 GC patients treated with radical resection from January 2015 to December 2022 in 6 hospitals. Based on the inclusion and exclusion criteria, 2947 patients were finally enrolled. Basic information about patients, including sex, age, tumor location, tumor size, differentiation, vascular invasion, neural invasion, depth of invasion(T), lymph node metastasis(N), pathological stage and preoperative serum tumor markers were assessed for analysis. The calibration curves, consistency index (C-index) and decision curve analysis (DCA) were employed to evaluate the precision and credibility of nomogram. Results    R software randomly selected 4/5 patients (2357/2947) to be included in the training cohort and the remaining 1/5 (590/2947) for the validation cohort. CEA, CA72-4, CA19-9 and CA125 were incorporated into risk-score. Patients were divided into high or low risk-score group based on the optimal cut-off value. Age, tumor location, T, N, and risk-score group were enrolled in the nomogram. The C-index of nomogram was 0.782 (95%CI: 0.744-0.820) in training cohort and 0.802 (95%CI: 0.733-0.871) in the validation cohort. The DCA curves showed the proposed predictive model had good clinical applicability. Calibration curves indicated the OS values predicted by nomogram fit well with the actual OS observed in the patients. Conclusions    The novel prognostic prediction model based on preoperative serum tumor markers has potential clinical application value in predicting OS probability of gastric cancer patients undergoing radical surgery. Thus, it can guide the follow-up treatment of GC patients after surgical intervention. 

Key words: gastric cancer, serum tumor markers, LASSO-Cox regression, prognostic model, nomogram

摘要：  目的    探索基于联合使用多种常用的肿瘤标记物的预测模型，提高对胃癌预后预测的准确性。 方法 采用多中心、回顾性研究，收集2015年1月至2022年12月6家医院接受根治性胃切除术治疗的5119例胃癌病人。根据纳入及排除标准，最终纳入2947例病人。评估病人的基本信息，包括性别、年龄、肿瘤位置、肿瘤大小、分化程度、血管侵犯、神经侵犯、浸润深度（T）、淋巴结转移（N）、病理分期和术前血清肿瘤标记物进行分析。采用校准曲线、一致性指数（C-index）和决策曲线分析（DCA）对列线图的可信度进行评估。 结果    R语言随机选择4/5（2357/2947）的病人纳入训练队列，剩余1/5（590/2947）为验证队列。CEA、CA72-4、CA19-9和CA125被纳入风险评分。根据最佳临界值将病人分为高、低风险评分组。模型中纳入年龄、肿瘤位置、T、N和风险评分分组。在训练队列中，列线图的C指数为0.782（95%CI：0.744-0.820），在验证队列中为0.802（95%CI：0.733-0.871）。DCA曲线表明，该预测模型具有良好的临床适用性。校准曲线显示列线图预测的OS值基本与实际相符。 结论    基于术前血清肿瘤标记物的新型预后预测模型在预测胃癌根治术后病人的总生存期方面具有较好的临床应用价值，可指导胃癌病人的术后治疗。

关键词: 胃癌, 血清肿瘤标记物, LASSO-Cox回归, 预后模型, 列线图