Supplementary Material for: Tumor radiomic features on pretreatment MRI to predict response to lenvatinib plus an anti–PD-1 antibody in advanced hepatocellular carcinoma: a multicenter study

Introduction: Lenvatinib plus an anti–PD-1 antibody has shown promising anti-tumor effects in patients with advanced hepatocellular carcinoma (HCC), but with clinical benefit limited to a subset of patients. We developed and validated a radiomic-based model to predict objective response to this combination therapy in advanced HCC patients. Methods: Patients (N = 170) who received first-line combination therapy with lenvatinib plus an anti–PD-1 antibody were retrospectively enrolled from 9 Chinese centers; 124 and 46 into the training and validation cohorts, respectively. Radiomic features were extracted from pretreatment contrast-enhanced MRI. After feature selection, clinicopathologic, radiomic, and clinicopathologic-radiomic models were built using a neural network. The performance of models, incremental predictive value of radiomic features compared with clinicopathologic features and relationship between radiomic features and survivals were assessed. Results: The clinicopathologic model modestly predicted objective response with an AUC of 0.748 (95% CI: 0.656–0.840) and 0.702 (95% CI: 0.547–0.884) in the training and validation cohorts, respectively. The radiomic model predicted response with an AUC of 0.886 (95% CI: 0.815–0.957) and 0.820 (95% CI: 0.648–0.984), respectively, with good calibration and clinical utility. The incremental predictive value of radiomic features to clinicopathologic features was confirmed with a net reclassification index of 47.9% (P < 0.001) and 41.5% (P = 0.025) in the training and validation cohorts, respectively. Furthermore, radiomic features were associated with overall survival and progression-free survival both in the training and validation cohorts, but modified albumin-bilirubin grade and neutrophil-to-lymphocyte ratio were not. Conclusion: Radiomic features extracted from pretreatment MRI can predict individualized objective response to combination therapy with lenvatinib plus an anti–PD-1 antibody in patients with unresectable or advanced HCC, and provide incremental predictive value over clinicopathologic features, and are associated with overall survival and progression-free survival after initiation of this combination regimen.

引言：仑伐替尼联合抗PD-1抗体（anti-PD-1 antibody）在晚期肝细胞癌（hepatocellular carcinoma, HCC）患者中展现出颇具前景的抗肿瘤效应，但临床获益仅局限于部分患者群体。本研究开发并验证了一款基于放射组学（radiomic）的模型，用于预测晚期肝细胞癌患者对该联合治疗方案的客观缓解情况。
研究方法：本研究回顾性纳入了来自9家中国中心、接受仑伐替尼联合抗PD-1抗体一线联合治疗的170例患者，其中训练队列124例，验证队列46例。所有患者的治疗前增强磁共振成像（contrast-enhanced MRI）影像均被用于提取放射组学特征（radiomic features）。经特征筛选后，分别构建了临床病理模型、放射组学模型以及临床病理-放射组学联合模型，所用算法为神经网络（neural network）。随后评估了各模型的性能、放射组学特征相较于临床病理特征的增量预测价值，以及放射组学特征与患者生存结局之间的关联。
研究结果：临床病理模型对客观缓解的预测效能中等，训练队列与验证队列的受试者工作特征曲线下面积（area under the receiver operating characteristic curve, AUC）分别为0.748（95%置信区间（confidence interval, CI）：0.656–0.840）与0.702（95%置信区间（CI）：0.547–0.884）。放射组学模型的预测AUC分别为0.886（95%置信区间（CI）：0.815–0.957）与0.820（95%置信区间（CI）：0.648–0.984），且具备良好的校准度与临床实用性。放射组学特征对临床病理模型的增量预测价值得到验证：训练队列与验证队列的净重新分类指数（net reclassification index, NRI）分别为47.9%（P < 0.001）与41.5%（P = 0.025）。此外，无论在训练队列还是验证队列中，放射组学特征均与总生存期（overall survival）及无进展生存期（progression-free survival）相关，而改良白蛋白胆红素分级与中性粒细胞/淋巴细胞比值则未显示出此类关联。
研究结论：从治疗前磁共振成像影像中提取的放射组学特征，可用于预测不可切除或晚期肝细胞癌患者对仑伐替尼联合抗PD-1抗体联合治疗的个体化客观缓解情况，相较临床病理特征具备增量预测价值，且与该联合治疗启动后的总生存期及无进展生存期显著相关。