机器学习预测胰腺癌1年生存率、转移状态及治疗依赖性生存
该研究基于1040例胰腺导管腺癌(PDAC)患者的国际单中心数据库,训练机器学习(ML)模型预测1年生存率、影像学M0/M1转移状态及不同治疗方案的生存时间。1年生存率预测准确率内部/外部队列为72%/70%,M0/M1状态为79%/72%;针对辅助和姑息化疗方案分别建模,C-index在切除、局部晚期和转移亚组分别为0.60、0.71、0.71。模型有望支持肿瘤委员会决策并生成合成对照臂。
要点
- ML模型预测PDAC 1年生存率准确率内部/外部队列72%/70%。
- 影像学M0/M1状态预测准确率内部/外部队列79%/72%。
- 姑息化疗方案生存模型C-index优于辅助方案(0.71 vs 0.60)。
核心结论
- ML模型可有效预测PDAC患者1年生存率、转移状态及治疗依赖性生存,支持个体化治疗决策。
研究设计与样本
- 基于1040例PDAC患者的国际单中心数据库,纳入超过23个临床变量,训练并验证ML模型。
主要发现
- 1年生存率预测准确率内部/外部队列分别为72%和70%。
- 影像学转移M0/M1状态预测准确率内部/外部队列分别为79%和72%。
- 针对mFOLFIRINOX、吉西他滨联合或不联合白蛋白紫杉醇方案建模,切除亚组C-index=0.60,局部晚期亚组C-index=0.71,转移亚组C-index=0.71。
- 姑息化疗方案的预测性能最佳。
关键数值与推荐
- 1年生存率预测准确率内部队列72%、外部队列70%。
- M0/M1状态预测准确率内部队列79%、外部队列72%。
- 生存模型C-index:切除亚组0.60,局部晚期亚组0.71,转移亚组0.71。
临床怎么用
- 模型可辅助肿瘤委员会制定个体化治疗策略。
- ML模型可用于未来临床试验中生成合成对照臂。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: Pancreatic adenocarcinoma (PDAC) remains one of the most lethal types of cancer, characterized by its unspecific symptoms, aggressive nature and late-stage diagnosis. PATIENTS AND METHODS: In this study, Machine Learning (ML) models were trained and applied on the largest international, monocentric database, comprising over 23 clinical variables from 1040 PDAC patients. In this study, the potential of ML models was assessed for the 1-year survival rate, metastatic M0/M1 status and therapy-dependent survival time generating predictive and prognostic ML-based biomarkers. RESULTS: Consistent predictive performance was achieved for the 1-year survival rate (accuracy, internal/external cohort 72 % / 70 %) and radiological metastatic M0/M1 status (Accuracy, internal/external cohort 79 % / 72 %). Moreover, distinct ML-based survival models were trained for approved first-line adjuvant and palliative chemotherapy regimen (mFOLFIRINOX, gemcitabine with or without nab-paclitaxel) across resected (C-index = 0.60), locally advanced (C-index = 0.71) and metastatic (C-index = 0.71) PDAC subgroups, demonstrating that predictive performance is best for palliative regimens. CONCLUSIONS: The models developed in this study may serve as a foundation for supporting tumor board decisions and demonstrate that personalized, ML-based therapy concepts are feasible in the near future. In addition, the models proposed in this study are helpful for generating ML-based synthetic control arms in future clinical trials.
引用
- Schneider F, Chen H, Pelzer U, et al. The basis for future personalized therapy approaches - Machine learning-generated 1-year survival rate, metastatic status and therapy-dependent survival in pancreatic cancer patients. Eur J Cancer. 2025. PMID: 41468770. doi:10.1016/j.ejca.2025.116189
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。