Evaluation of Surrogate Endpoints (SES) for Previously Untreated Unresectable or Metastatic Melanoma (MMEL): Analyses from a Longitudinal Electronic Health Record Database in the United States (US)

OBJECTIVES: To evaluate progression-free survival (PFS), time-to-progression (TTP), time-to-next-treatment-or-death (TNTD), and time-to-treatment discontinuation (TTD) as SEs for overall survival (OS) in previously untreated unresectable or mMel in real-world settings.

METHODS: Individual patient data was acquired from the US Flatiron Health database comprising >265 cancer clinics. Individual-level correlation between each SE and OS was estimated using Spearman’s rank correlation. To calculate treatment effect correlations, in absence of randomized settings, patients were allocated to 12 non-overlapping clusters with synthetic treatment and control arms using propensity score matching to balance baseline age, race, sex, ECOG, and BRAF mutation status. Pearson’s correlation coefficient, estimated by weighted linear regression (WLR), was used to measure the treatment-effect correlation for each SE. Predictive performance of WLR was assessed via cross-validation. Correlation measures are reported as (estimate; 95% CI).

RESULTS: 489 patients were eligible for the individual-level analysis and 428 for the cluster-level analysis, with a median of 36 patients per cluster (range: 18-46). Patients were all diagnosed post-2011, mostly stage IV (58.5%) and BRAF wild-type (53.2%). Among patients assigned to clusters, most received either single or combinations of immunotherapy, or BRAF-MEK inhibitors (84.8%). Individual-level correlations were estimated as (0.62; 0.54-0.72) for PFS, (0.67; 0.56-0.76) for TNTD, (0.60; 0.43-0.74) for TTD, and (0.47; 0.36-0.58) for TTP. Correlations between treatment-effects were (0.78; 0.53-0.97) for PFS, (0.77; 0.36-0.98) for TNTD, (0.87; 0.69-0.97) for TTD and (0.67; 0.36-0.97) for TTP. In cross validation, the 95% prediction intervals of OS hazard ratios included their reported counterparts in >83% of the clusters for each SE.

CONCLUSIONS: Correlations between the candidate SEs and OS were moderate in this study. Predictions from the derived surrogacy equations may enable earlier assessment of real-world treatment effect on OS for previously untreated unresectable or mMel; however, classification of real-world data in small clusters may result in wide uncertainty around estimates.