| 刘 建,于 静,张 师,等.基于SEER数据库利用机器学习预测胃多原发恶性肿瘤患者的预后[J].肿瘤学杂志,2026,32(8):656-666. |
| 基于SEER数据库利用机器学习预测胃多原发恶性肿瘤患者的预后 |
| Prognostic Analysis of Patients with Multiple Primary Gastric Cancers Using Machine Learning Based on the SEER Database |
| 投稿时间：2026-06-29 |
| DOI：10.11735/j.issn.1671-170X.2026.08.B005 |
|
 |
| 中文关键词: 胃肿瘤 多原发恶性肿瘤 倾向性评分匹配 机器学习 预后 |
| 英文关键词:gastric neoplasms multiple primary malignant tumor propensity score matching machine learning prognosis |
| 基金项目:河北省医学科学研究课题计划（20240544） |
|
| 摘要点击次数: 12 |
| 全文下载次数: 3 |
| 中文摘要: |
| 摘 要：［目的］ 基于美国SEER数据库回顾性分析胃多原发恶性肿瘤患者的临床特征和预后，并探讨机器学习模型对预后预测的优劣。［方法］ 收集SEER数据库2000—2022年胃癌患者的临床资料，采用Kaplan-Meier法计算生存期，组间比较采用Log-rank检验。利用1∶1倾向性评分匹配（propensity score matching，PSM）分析调整混杂因素，比较PSM前后仅胃癌组和胃多原发恶性肿瘤组的生存差异。采用Cox比例风险回归模型（Cox模型）和10折交叉验证确定影响胃多原发恶性肿瘤患者预后的因素。将数据按照7∶3分为训练集和验证集，利用Cox模型、随机森林（random forest，RF）生存模型（RF模型）和生存树（survival tree，ST）模型（ST模型）对生存情况进行预测并进行模型效果评价。［结果］ 共纳入胃癌患者53 752例，其中有4 174例胃癌患者罹患了第二原发恶性肿瘤。胃癌和第二原发恶性肿瘤的中位诊断年龄分别为68岁和71岁。同时性和异时性第二原发恶性肿瘤分别为1 190例和2 984例，中位发病时间间隔分别为1个月和45个月。结直肠癌和肺癌为胃多原发恶性肿瘤患者最常见的第二原发癌种。Cox多因素分析显示，男性、分化程度高、发病年龄大、分期晚、腺癌、手术史、两癌发病间隔短、第二原发癌种（肺癌）、家庭年收入低以及患者所在地区为小城市等为胃多原发恶性肿瘤患者的独立预后危险因素（P均<0.05）。时间依赖受试者工作特征曲线显示，胃多原发恶性肿瘤患者1年、3年和5年的曲线下面积分别为0.899、0.765和0.739。3种预后模型比较，Cox和RF模型优于ST模型。［结论］ 男性、高龄、分期晚、两癌发病间隔短、第二原发癌种（肺癌）、家庭年收入低以及患者所在地区为小城市等为胃多原发恶性肿瘤患者的预后危险因素，据此建立的生存预测模型效果较好，可为临床医师提升对胃多原发癌患者的早期识别和诊治提供数据支持。 |
| 英文摘要: |
| Abstract： ［Objective］ Based on a retrospective analysis of the United States SEER database， to investigate the clinical characteristics and prognosis of patients with multiple primary malignant gastric cancers， and to explore the advantages and disadvantages of machine learning models in predicting prognosis. ［Methods］ Clinical data of gastric cancer patients from 2000 to 2022 in the SEER database were collected. Overall survival was estimated by the Kaplan-Meier method， and Log-rank test was used for inter-group comparison. Propensity score matching (PSM) at a 1∶1 ratio was performed to adjust for confounding factors， and survival differences between the gastric cancer-only group and the gastric multiple primary cancer group were compared before and after PSM. The Cox proportional hazards regression model was used to identify prognostic factors in patients with gastric multiple primary cancer. Data were randomly divided into training and validation sets at a 7∶3 ratio. The Cox model， random forest survival (RF) model， and survival tree (ST) model were used to predict survival， and model performance was evaluated. ［Results］ A total of 53 752 patients with gastric cancer were included， among whom 4 174 had suffered from a second primary cancer. The median age at diagnosis for gastric cancer and second primary cancer was 68 years old and 71 years old， respectively. Synchronous and metachronous second primary cancer occurred in 1 190 and 2 984 patients， with median time latency of 1 month and 45 months， respectively. Colorectal cancer and lung cancer were the most common second primary cancer types. Cox multivariate analysis revealed that male， high degree of differentiation， older age at onset， advanced stage， adenocarcinoma， a history of surgery， a short interval between the onset of two cancers， a second primary cancer type (lung cancer)， low annual household income， and residing in a small city were independent prognostic risk factors for patients with multiple primary gastric cancers (all P <0.05).The time-dependent receiver operating characteristic (ROC) curve showed that the area under the curve for patients with multiple primary gastric cancers at 1， 3， and 5 years were 0.899， 0.765， and 0.739， respectively. Among the 3 prognostic models， the Cox and RF models were superior to the ST model. ［Conclusion］ Male， high degree of differentiation， older age at onset， advanced stage， adenocarcinoma， a history of surgery， a short interval between the onset of two cancers， a second primary cancer type (lung cancer)， low annual household income， and residing in a small city are independent prognostic risk factors for patients with multiple primary gastric cancers. The survival prediction model established has a good effect and can provide data support for clinicians to improve the early identification and diagnosis and treatment of patients with multiple primary gastric cancers. |
|
在线阅读
查看全文 查看/发表评论 下载PDF阅读器 |