Surveillance Strategy after Complete Ablation of Initial Recurrent Hepatocellular Carcinoma: A Risk-Based Machine Learning Study

Abstract

PurposeTo investigate surveillance strategies for initial recurrent hepatocellular carcinoma (irHCC) after ablation to support clinical decision making, as there is no consensus regarding the monitoring strategy for irHCC after ablation. Materials and MethodsData from patients with irHCC who received ablation were retrospectively collected at 2 medical centers. The risk of tumor relapse in each month was calculated through random survival forest methodology, and follow-up schedules were arranged thereafter to maximize the capability of relapse detection at each visit. ResultsThe cumulative 0.5-, 1-, 1.5-, and 2-year risk-adjusted probabilities in the training/validation cohorts were 26.2%/21.5%, 42.3%/39.4%, 55.5%/52.6%, and 61.3%/63.2%, respectively, with the highest recurrence rate occurring in the second month (maximum, 7.9%/7.4%). The surveillance regime primarily concentrated on the first year after treatment, especially the initial 6 months. The delay in detecting tumor recurrence gradually decreased when the total number of follow-up visits increased from 4 to 8. Compared with the control strategies, this schedule (follow-up visits at 2, 4, 6, 9, 12, and 18 months) reduced the delay in detection. The benefits of this surveillance regime were evident when the patients were followed up 6 times. The proposed 6-visit surveillance strategy significantly decreased the delay in detection compared with the control 7-visit approach (1.32 months vs 1.82 months, respectively; P < .001). ConclusionsThe proposed new surveillance schedule minimized the delay in detecting recurrence in patients with irHCC after ablation. The risk-related machine learning method described in this study could be applied to develop follow-up strategies for other patients with hepatocellular carcinoma.