Multimode Ultrasound Model for Predicting the Early Treatment Response of Anti-VEGF Agents Plus Anti-PD-1 Antibody in Patients with Unresectable Hepatocellular Carcinoma

The purpose of the study described was to establish prediction models to initially screen the beneficiary patients with unresectable hepatocellular carcinoma (HCC) in the treatment of anti-vascular endothelial growth factor (VEGF) agents plus anti-programmed cell death-1 (PD-1) antibody. A total of 62 patients were enrolled in this study. All patients underwent ultrasound (US), color ddoppler flowing imaging (CDFI), contrast-enhanced ultrasound (CEUS) and laboratory examinations within 2 wk before the treatment. Tumor response was assessed according to mRECIST criteria. Univariate and multivariate analyses were used to select the independent predictors. US + CDFI, CEUS and FULL models were established. Three models were displayed by nomography. Receiver operating characteristic (ROC) and calibration curves were drawn to evaluate the predictive ability of models. Decision curve analysis (DCA) was used to assess the clinical utility of models. On univariate and multivariate analysis, the US boundary (p = 0.037), halo (p = 0.002) and CDFI (p = 0.024) were included in the US + CDFI model. CEUS boundary (p = 0.001) and washout time (p < 0.001) were included in the CEUS model. The number of lesions (p = 0.104), halo on US (p = 0.014), CDFI (p = 0.057) and washout time on CEUS (p = 0.015) were incorporated into the FULL model. The C indices of the US + CDFI, CEUS and FULL models were 0.918, 0.920 and 0.973. CEUS and FULL models yielded a good net benefit for almost all threshold probabilities. Nomograms based on US, CDFI, CEUS and clinical characteristics could help to non-invasively predict the response to treatment with anti-PD-1 antibodies plus anti-VEGF agents.