Which Response Criteria Best Help Predict Survival of Patients with Hepatocellular Carcinoma Following Chemoembolization? A Validation Study of Old and New Models

Abstract

To identify differences in radiologic assessment methods and determine optimal imaging criteria for response evaluation in hepatocellular carcinoma (HCC) patients treated with chemoembolization. Institutional review board approval was obtained, and patient informed consent was waived. The present study included 332 patients with intermediate stage HCC and Child-Pugh A cirrhosis who underwent serial chemoembolization. All measurable target lesions of 1 cm or larger in diameter were uni- and bidimensionally measured both at baseline and during follow-up. Intermodel agreement among the guidelines of the World Health Organization (WHO), Response Evaluation Criteria in Solid Tumors (RECIST), the European Association for the Study of the Liver (EASL), and modified RECIST (mRECIST) were examined. The most reliable model was selected on the basis of the correlation with survival prediction. The κ values of comparisons among WHO, RECIST, and mRECIST guidelines were less than 0.20, whereas the κ value for the comparison of EASL and mRECIST guidelines was 0.94. In patients with a partial response (PR), stable disease (SD), or progressive disease (PD), compared with patients with a complete response (CR), hazard ratios (HRs) for survival were 2.99 (95% confidence interval [CI]: 2.14, 4.17), 3.49 (95% CI: 1.71, 7.10), and 15.63 (95% CI: 9.51, 25.69), respectively, for EASL criteria. In patients with a PR, SD, or PD, compared with patients with a CR, the HRs were 2.75 (95% CI: 1.96, 3.87), 6.32 (95% CI: 3.67, 10.90), and 16.06 (95% CI: 9.76, 26.43), respectively, for mRECIST guidelines (P<.001). The C index for the multivariate model was 0.76 (95% CI: 0.72, 0.79) for both EASL and mRECIST guidelines, thus exhibiting satisfactory capability to help predict survival. The Cox regression model revealed that both mRECIST and EASL guidelines were independent predictors of overall survival (P<.001 for both). The enhancement models more accurately helped predict long-term survival in HCC patients treated with chemoembolization.