Prognostic value of PIVKA-II and routine perioperative laboratory indices for recurrence prediction after curative resection of hepatocellular carcinoma

BACKGROUND Although α-fetoprotein (AFP) is widely used in hepatocellular carcinoma (HCC), its prognostic value is limited and approximately 30-40% of patients have non-elevated AFP at diagnosis. Protein induced by vitamin K absence-II (PIVKA-II) and routinely available perioperative laboratory parameters represent potentially underutilized prognostic resources, but their independent contribution to recurrence prediction following curative resection has not been systematically characterised within a unified hierarchical modelling framework with formal statistical comparison. METHODS This single-centre retrospective cohort study included 517 consecutive patients undergoing curative hepatic resection for HCC between January 2018 and December 2020. Three hierarchically nested multivariable Cox proportional hazards models were constructed: M1 included clinical and pathological covariates; M2 added log-transformed AFP and PIVKA-II; M3 further incorporated ten preoperative hematological/biochemical indices. Model performance was evaluated via time-dependent area under the receiver operating characteristic curve (AUC) with bootstrap 95% confidence intervals, calibration, decision curve analysis, and SHapley Additive exPlanations (SHAP) interpretation, separately for progression-free survival (PFS) and distant metastasis-free survival (DMFS). Internal validation incorporating the entire model construction pipeline was performed using 200 bootstrap resamples. RESULTS Among 517 patients, 302 PFS events and 62 distant metastasis events were observed during a maximum follow-up of 96 months. PIVKA-II emerged as the dominant prognostic predictor (M3 PFS: HR 1.295 [1.218, 1.376], Wald |z| = 8.20). The addition of AFP and PIVKA-II to clinical variables (M2 vs. M1) provided statistically significant discriminative gain for PFS (60-month AUC difference + 0.080 [0.051, 0.110], p < 0.001), whereas the further addition of ten hematologic indices (M3 vs. M2) produced only marginal and non-significant gain (60-month AUC difference + 0.006 [-0.004, 0.016], p = 0.232). For DMFS, M2 and M3 performed equivalently. Optimism-corrected C-index for M3 PFS, accounting for variable selection, was 0.766 [0.745, 0.788], with calibration slope 0.835 [0.562, 1.001] indicating mild overfitting. PIVKA-II demonstrated time-varying behaviour (pH p = 0.0001), with hazard ratio rising from 1.221 [1.132, 1.317] in the first 24 months to 1.573 [1.401, 1.766] thereafter; reported overall HR therefore represents a time-averaged estimate. CONCLUSIONS PIVKA-II carries substantial independent prognostic information for postoperative HCC recurrence prediction, with effect magnitude exceeding all other examined predictors. The addition of broader hematological/biochemical panels yields small and statistically non-significant incremental gains over a clinically practical model using PIVKA-II alongside clinical variables. These findings, derived from a single-centre HBV-predominant cohort with internal validation only, are exploratory and hypothesis-generating; external validation in independent cohorts with diverse aetiologies is required before clinical implementation.