Development and validation of a CT-based online calculator for prognosis and postoperative adjuvant transcatheter arterial chemoembolization benefit prediction in resected hepatocellular carcinoma: a multicenter retrospective cohort study

Background: Selecting hepatocellular carcinoma (HCC) patients who may benefit from postoperative adjuvant transarterial chemoembolization (PA-TACE) after surgical resection (SR) remains challenging. This study aimed to create a CT-based online tool to predict overall survival (OS) and PA-TACE benefit. Methods: A total of 1,770 patients who underwent preoperative contrast-enhanced CT treated with SR only or PA-TACE for HCC were retrospectively enrolled from three distinct institutions. The patients included 1,040 in the development cohort, 448 in the internal validation cohort, 236 in the external validation cohort and 46 in the RNA-sequencing cohort. SR-only and PA-TACE nomograms were built by integrating CT imaging features and clinicopathological variables, enabling personalized predictions of 5-year OS. Virtual-twin analyses were performed to estimate the degree of improvement or detriment of PA-TACE. Enrichment analyses were performed to investigate the mechanisms underlying recommendation groups. Results: SR-only and PA-TACE nomograms showed better predictive performance (C-index ≥ 0.694), better calibration, and lower prediction error (integrated Brier score ≤ 0.176) than widely adopted staging systems in all cohorts. An online calculator was generated for the individualized quantification of the survival improvement after PA-TACE and made available for clinicians at http://120.78.200.107:8312/PA-TACE.html. We identified 611/1,724 (35.4%) patients who were recommended to undergo PA-TACE whose average probability of 5-year OS and restricted mean survival time would increase by 19.4% and 22.5 months, respectively, compared to resection without PA-TACE. Approximately 43.6% of patients may necessitate adjustments to their postoperative treatment protocols. Patients in the recommended group exhibited a greater likelihood of possessing high-risk characteristics, which were associated with an immunosuppressive tumor microenvironment (all P<0.01). Conclusion: Our online prognostic prediction calculator integrates preoperative CT imaging features and clinicopathological variables to predict the survival probability of patients with or without PA-TACE, as well as the potential benefits of PA-TACE in HCC patients following SR.