Plasma five-protein risk score for prognostic stratification in patients with advanced hepatocellular carcinoma treated with atezolizumab plus bevacizumab

We aimed to characterize longitudinal plasma proteomic changes and develop a non-invasive prognostic protein signature for patients with advanced hepatocellular carcinoma (HCC) treated with atezolizumab plus bevacizumab. Paired plasma samples from 71 patients at baseline and after three treatment cycles were analyzed using Olink proximity extension assays. A baseline proteomic risk score was derived using elastic-net Cox regression and stability selection, with progression-free survival (PFS) as the primary endpoint. Longitudinal risk trajectories were evaluated using a Cycle 3 landmark analysis. Cross-platform translation was evaluated using ELISA in an assay-bridging subset and a non-overlapping evaluation cohort. Treatment induced broad immune-inflammatory and angiogenic shifts. A five-protein risk score (5PRS; MUC16, MPO, TNFSF14, GZMA, and MB) achieved time-dependent AUCs of 0.79 and 0.93 for 6- and 12-month PFS, respectively. Median PFS was significantly longer in the low-risk group than in the high-risk group (16.1 vs. 4.0 months, P  < 0.001), and the 5PRS remained independently associated with PFS after adjustment for clinical factors. Among 52 patients who remained progression-free at Cycle 3, longitudinal 5PRS transitions further stratified subsequent PFS and provided prognostic information beyond early radiological response. ELISA-based translation showed good concordance with the Olink-derived classification and retained an association with PFS in the evaluation cohort. The 5PRS stratified PFS in patients with HCC treated with atezolizumab plus bevacizumab, while longitudinal tracking provided additional prognostic information beyond early radiological response. The feasibility of ELISA-based translation supports further prospective validation of the 5PRS as a non-invasive prognostic tool.