可解释机器学习模型的开发与验证:用于治疗前预测肝细胞癌靶向联合免疫治疗的应答
这项回顾性多中心研究纳入两家机构接受靶向联合免疫治疗的413例肝细胞癌患者,基于临床和CT特征构建并验证逻辑回归、支持向量机、随机森林和XGBoost模型预测客观应答和总生存。XGBoost模型表现最佳,训练集和验证集AUC分别为0.802(95%CI 0.744-0.860)和0.805(95%CI 0.741-0.868);BCLC分期、肿瘤数目、肿瘤边缘、瘤周强化和大血管侵犯为显著预测因素;XGBoost模型划分的高风险评分与更短总生存相关(HR 0.740,95%CI 0.665-0.823,P<0.001)。
要点
- XGBoost模型预测靶向联合免疫治疗应答的验证集AUC为0.805(95%CI 0.741-0.868)。
- 显著预测因素包括BCLC分期、肿瘤数目、肿瘤边缘、瘤周强化和大血管侵犯。
- XGBoost模型划分的高风险评分与更短总生存相关(HR 0.740,P<0.001)。
核心结论
- XGBoost模型能有效预测接受靶向联合免疫治疗的肝细胞癌患者的治疗应答和预后,提供无创工具以指导治疗决策和优化临床结局。
研究设计与样本
- 回顾性多中心研究,纳入两家机构接受靶向联合免疫治疗的413例肝细胞癌患者,收集临床和CT特征,开发并验证逻辑回归、支持向量机、随机森林和XGBoost模型预测治疗应答和总生存,采用AUC和Kaplan-Meier分析评估模型性能,并用SHAP分析解释模型预测。
关键影像表现/征象
- 显著预测因素包括BCLC分期、肿瘤数目、肿瘤边缘、瘤周强化和大血管侵犯。
- XGBoost模型划分的高风险评分与更短总生存相关(HR 0.740,95%CI 0.665-0.823,P<0.001)。
诊断标准与关键数值
- XGBoost模型在训练队列的AUC为0.802(95%CI 0.744-0.860),在验证队列的AUC为0.805(95%CI 0.741-0.868)。
- XGBoost模型划分的高风险评分与更短总生存相关(HR 0.740,95%CI 0.665-0.823,P<0.001)。
临床应用与阅片要点
- 原文未据此给出推荐。
局限性
- 原文未报告局限性。
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)。
英文原文(摘要)· 点击展开
BACKGROUND: Targeted and immune therapies are used for unresectable hepatocellular carcinoma (HCC), but their efficacy is limited by tumor heterogeneity. Identifying patients who will benefit from targeted and immune therapy is crucial for optimizing treatment strategies and prognosis. We aimed to develop and validate interpretable machine learning (ML) models integrating clinical and CT imaging characteristics for pretreatment prediction of objective response and prognosis to targeted and immune therapy in HCC.
METHODS: This retrospective multicenter study included 413 patients from two institutions who received targeted and immune therapy. Clinical and CT characteristics were collected. Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) models were developed and validated to predict treatment response and overall survival (OS). Model performance was assessed by area under the curve (AUC) and Kaplan-Meier analysis. Additionally, SHAP (SHapley Additive exPlanations) analysis explained model predictions by aggregating the attribution values for each input feature.
RESULTS: The XGBoost model demonstrated the best performance with AUCs of 0.802 (95 % CI: 0.744-0.860) and 0.805 (95 % CI: 0.741-0.868) in training and validation cohorts. Significant predictors included Barcelona Clinic Liver Cancer (BCLC) stage, tumor number, tumor margin, peritumoral enhancement and macrovascular invasion. Kaplan-Meier analysis showed that high-risk scores stratified by XGBoost model were associated with shorter OS (HR: 0.740, 95 % CI: 0.665-0.823, P < 0.001).
CONCLUSION: XGBoost model effectively predicted treatment response and prognosis in HCC patients undergoing targeted and immune therapy, offering a noninvasive tool to guide treatment decisions and optimize clinical outcomes.
引用
- Huang R, Liu K, Yan L, et al. Development and validation of interpretable machine learning model for pre-treatment predicting the response to targeted and immune therapy in hepatocellular carcinoma. Eur J Radiol. 2025. PMID: 40818216. doi:10.1016/j.ejrad.2025.112353
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。