微信公众号

官网二维码
中国癌症防治杂志

中国癌症防治杂志 ›› 2025, Vol. 17 ›› Issue (1): 95-102.doi: 10.3969/j.issn.1674-5671.2025.01.13

• 临床研究 • 上一篇    下一篇

基于SEER 数据库年龄≥55 岁N1b 期甲状腺乳头状癌患者疾病特异生存期预测模型构建及风险分层

  

  1. 安徽医科大学附属阜阳人民医院/阜阳市人民医院甲状腺乳腺外科
  • 出版日期:2025-02-25 发布日期:2025-03-06
  • 通讯作者: 焦大海 E-mail：dahaijiao1008@163.com
  • 基金资助:
    安徽省阜阳市自筹经费科技计划项目（FK202081015）

Development of cancer-specific survival prediction model and risk stratification for patients aged ≥55 years with N1b stage papillary thyroid carcinoma based on the SEER database

  • Online:2025-02-25 Published:2025-03-06

PDF (PC)

76

赞

摘要： 目的 探讨年龄≥55岁N1b期甲状腺乳头状癌（papillary carcinoma of the thyroid，PTC）患者癌症特异性生存期（cancer⁃specific survival，CSS）的影响因素，并构建随机生存森林（random survival forest，RSF）模型和进行风险分层，评估不同治疗策略的生存获益。方法 收集2004—2019年美国国家癌症研究所的监测、流行病学和最终结果（Surveillance，Epidemiology，and End Results，SEER）数据库中符合条件的2 867例PTC患者的临床资料，按7∶3比例分为训练组（n=2 008）和验证组（n=859）。采用Cox比例风险回归模型和Kaplan⁃Meier（K⁃M）生存分析法识别CSS的潜在危险因素。采用Lasso回归筛选关键变量，构建Lasso⁃Cox和RSF预测模型，并通过一致性C指数（C⁃index）、时间依赖受试者工作特征（time⁃dependent receiver operating characteristic，tROC）曲线及Brier评分评估模型的效能，进行风险分层，分析不同治疗策略的生存获益。结果  Cox比例风险回归模型分析显示，年龄较大、远处转移、肿瘤直径较大、腺外侵犯程度高、已婚、不放疗、甲状腺腺叶切除及化疗是CSS的独立危险因素（均P<0.05）。Lasso回归筛选出年龄、肿瘤直径、腺外侵犯和远处转移4个关键变量。Lasso⁃Cox模型和RSF模型的区分度及校准度均较高，在训练组中RSF模型整体表现优于Lasso⁃Cox模型，在验证组中两者差异不明显。Log⁃rank分析显示，高风险组的3年、5年、10年CSS率分别为68.15%、58.63%、37.52%，而低风险组分别为96.86%、94.38%、88.87%。手术方式及放疗对低风险组生存无显著影响，而在高风险组中显示出显著的生存差异。结论 本研究构建的RSF模型对年龄≥55岁N1b期PTC患者的CSS具有较好的预测能力，有助于其风险分层，为临床治疗决策提供依据。

关键词: 甲状腺乳头状癌, 淋巴结转移, 随机生存森林, 生存分析

Abstract: Objective To investigate the influencing factors of cancer⁃specific survival (CSS) in patients aged ≥55 years with N1b stage papillary thyroid carcinoma (PTC), develop a random survival forest (RSF) model and used for risk stratification, and assess the survival benefits of different treatment options. Methods Clinical data of 2, 864 eligible patients with PTC in Surveillance, Epidemiology, and End Results(SEER) database from 2004 to 2019 were collected and randomly divided into the training group (n=2, 008) and the validation group (n=856) at a ratio of 7∶3. The potential risk factors of CSS were determined using Cox proportional hazards regression model and Kaplan⁃Meier (K⁃M) survival analyses. Key variables were selected through Lasso regression , and both Lasso⁃Cox and RSF prediction models were developed. Model performance was evaluated using the concordance index (C⁃index), time⁃dependent receiver operating characteristic (tROC) curves, and Brier scores. Risk stratification was conducted, and survival benefits associated with different treatment strategies were evaluated. Results Cox proportional hazards regression model indicated that older age, distant metastasis, larger tumor size, extensive extrathyroidal extension, marital status, lack of radiotherapy, thyroid lobectomy and receipt of chemotherapy were independent risk factors for CSS (all P<0.05). Lasso regression screened out four key variables: age, tumor size, extrathyroidal extension, and distant metastasis. Both models showed high discrimination and calibration, with the RSF model superior in the training group and comparable in validation group. Log⁃rank analysis revealed that the 3-year, 5-year and 10-year CSS rates in the high⁃risk group were 68.15%, 58.63% and 37.52%, compared with 96.86%, 94.38% and 88.87% in the low⁃risk group, respectively. neither the surgical approach nor radiotherapy significantly affected survival in the low⁃risk group, whereas significant survival differences were observed in the high⁃risk group. Conclusions The RSF model developed in this study demonstrated strong predictive performance for CSS in patients aged ≥55 years with N1b stage PTC. This model facilitates risk stratification and provides a basis for clinical treatment decision⁃making.

Key words: Papillary thyroid cancer, Lymph node metastasis, Random survival forest, Survival analysis

中图分类号: 

  • R736.1