OBJECTIVES: To evaluate ORR and CR as surrogate endpoints for OS by modelling the association between the treatment effects on each surrogate endpoint and OS with aggregate-level data from RCTs investigating first-line therapies for advanced melanoma.

METHODS: A systematic literature review was conducted to identify RCTs reporting hazard ratio of OS (HR_OS) and odds ratios of ORR (OR_ORR) or CR (OR_CR). Bivariate random-effects meta-analysis (BRMA) and weighted linear regression (WLR) models were employed for correlation assessment between OR_ORR/OR_CR and HR_OS. Surrogacy equations from WLR, weighted using RCT sample sizes, were validated internally using leave-one-out cross-validation, and externally against three recently published RCTs (IMspire170, RELATIVITY-047, PIVOT IO-001). The primary analysis included 26 trials. Sensitivity analyses were restricted to these trials: phase III (n=17), crossover-adjusted or crossover-unallowed (n=19), and using RECISTv1.1 in response assessment (n=20).

RESULTS: In the primary analysis for ORR, BRMA and WLR estimated correlation coefficients of -0.59 (95%CI: [-0.80, -0.25]) and -0.61 (95%CI: [-0.82, -0.25]), respectively. For CR, BRMA and WLR estimated correlations of -0.65 (95%CI: [-0.82, -0.37]) and -0.55 (95%CI: [-0.79, -0.16]), respectively. HR_OS was predicted accurately from OR_ORR for 96% of studies (25/26), and from OR_CR for 92% of studies (24/26) in cross-validation. In external validation, the average absolute deviations between published and predicted HR_OS were 0.05 and 0.08 for OR_ORR and OR_CR, respectively. In sensitivity analyses, correlations varied for both ORR (BRMA: -0.53 to -0.63; WLR: -0.56 to -0.64) and CR (BRMA: -0.65 to -0.66; WLR: -0.48 to -0.56), and coverage rate of HR_OS by 95% PIs ranged between 94-100% for ORR and 94-95% for CR during cross-validation. Restricting the evidence base to crossover-adjusted trials strengthened correlations.

CONCLUSIONS: Treatment effects on both response outcomes were moderately correlated with improvements on OS. Cross-validations of surrogacy equations are indicative of the high and early predictive ability of response outcomes for OS benefit.