Abstract
AIM: This study aims to investigate α-fetoprotein (AFP) trajectories for prediction of survival outcomes after hepatic arterial infusion chemotherapy (HAIC) treatment in large hepatocellular carcinoma (HCC). METHODS: From May 2014 to June 2020, 889 eligible patients with large HCC underwent HAIC were retrospectively enrolled from five hospitals. A latent class growth mixed (LCGM) model was applied to distinguish potential AFP level dynamic changing trajectories. Inverse-probability-of-treatment weighted (IPTW) analyses were performed to eliminate unmeasured confounders through marginal structural models. Multivariate Cox proportional hazard regression analyses were used to determine the overall survival (OS) in patients with large HCC. Performance of these serum markers for survival prediction was compared by areas under receiver operating characteristic analysis with the Delong test. RESULTS: The median follow-up time was 23.7 (interquartile range, 3.8-115.3). A total of 1009 patients with large HCC, who underwent HAIC with AFP repeatedly measured 3-10 times, were enrolled in the study. Three distinct trajectories of these serum AFP were identified using the LCGM model: high stable (37.0%; n = 373), low stable (15.7%; n = 159), and sharp-falling (47.3%; n = 477). Multivariate Cox proportional hazard regression analyses found that ALBI stage 2-3, BCLC-C stage and high-stable AFP trajectories were associated with OS. AFP trajectories yield the optimal predictive performance in all risk factors. CONCLUSIONS: The AFP trajectories based on longitudinal AFP change showed outstanding performance for predicting survival outcomes after HAIC treatment in large HCC, which provide a potential monitoring tool for improving clinical decision-making.