OBJECTIVES: To evaluate RFS as a surrogate endpoint for OS in early-stage HCC by assessing treatment-level correlations using published literature.

METHODS: A systematic literature review was conducted by searching Embase, MEDLINE®, and CENTRAL from database inception to 07-20-2022 to obtain RCTs reporting HRs for RFS and OS for adults with early-stage HCC who had underwent curative ablation or resection. The correlation estimates between HR_RFS and HR_OS were computed using bivariate random-effects meta-analysis (BRMA) and weighted linear regression (WLR) where trials were weighted by their sample size. Surrogacy equations were obtained from WLR. Validity of the surrogacy equations were assessed via leave-one-out cross-validation (LOOCV). Primary analysis included all trials (number of trials [n]=47). Several sensitivity analyses were conducted including (i) a sensitivity of trials that explicitly reported death in the definition of RFS (n=17) and (ii) a sensitivity of trials that explicitly reported death and were in the adjuvant setting (n=12).

RESULTS: In the primary analysis, the estimated correlation coefficient was 0.67 from BRMA (95% CI: 0.52, 0.79) and 0.73 (0.61, 0.85) from WLR. The resulting surrogate threshold effect was 0.77, suggesting that observing an HR_RFS less than 0.77 is required to predict a nonzero positive effect on the HR_OS. The surrogacy equation was log(HR_OS)=‑0.12+0.97×log(HR_RFS). In sensitivity analyses (i) and (ii), correlation estimates were even stronger (0.88 to 0.91). In LOOCV, alignment between observed and predicted HR_OS was 95.8% for the primary analysis and ranged from 91.7% to 96.2% for the sensitivity analyses.

CONCLUSIONS: Meaningful correlations between the treatment effects on RFS and OS in early stage were observed across all analyses. The highly accurate surrogacy equation between the treatment effects may enable earlier assessments of OS benefit from the RFS benefit for early-stage HCC. Restricting the evidence base to trials explicitly reporting death in the definition of RFS led to stronger correlation estimates.