Background Several MRI-based models for postsurgical prediction of cancer recurrence have been proposed for hepatocellular carcinoma (HCC), but they were predominantly developed using mixed populations including large and multiple tumors, limiting their applicability in patients with small solitary HCC. Purpose To train and test an MRI-based Early Recurrence Individualized Score (MERIS) to help predict early recurrence following resection in patients with solitary HCCs 5 cm or smaller and to compare its performance with a pathologic analysis-based model and existing prognostic systems. Materials and Methods In this retrospective study, patients with treatment-naive, solitary HCCs 5 cm or smaller who underwent curative resection were split into a training set and an external test set. Multivariable Cox regression analysis was conducted to predict time to early recurrence (within 2 years) based on MRI prognostic features and to train MERIS. MERIS performance was assessed using Harrell concordance index (known as the c-index) and compared with a pathologic analysis-based model and existing prognostic systems. Results A total of 325 patients (mean age, 59 years ± 9.7 SD; 253 men) were included (204 in training set; 121 in external test set). MERIS incorporated four features: elevated aspartate aminotransferase (hazard ratio HR, 2.70; P = .006), tumor size (HR, 1.51; P = .01), nonsmooth tumor margin (HR, 2.12; P = .04), and peritumoral hepatobiliary phase hypointensity (HR, 2.97; P = .002). Using a MERIS cutoff of five points, 2-year recurrence-free probabilities were lower in high-risk versus low-risk groups (training set, 69.5% vs 91.4% [P P P Supplemental material is available for this article.
Choi et al. (Sun,) studied this question.