Challenges and opportunities of predicting overall survival benefit from improvements to recurrence-free survival in stage II/III melanoma: a correlation meta-analysis

Abstract

Background

We evaluated the association between treatment effects on recurrence-free survival (RFS) and overall survival (OS) in randomized controlled trials (RCTs) studying resected stage II/III melanoma.

Methods

Hazard ratios (HRs) of RFS and OS were obtained from a literature review. Bivariate random-effects meta-analysis (BRMA) and weighted linear regression (WLR) models estimated correlations [95% confidence interval (CI)] between HR_RFS and HR_OS. Slopes and intercepts of surrogacy equations were estimated. Surrogate threshold effect was derived from WLR for various sample sizes. Validity and predictive performance of WLR were assessed using leave-one-out cross-validation. Sensitivity analyses evaluated impact of RCTs violating proportional hazards assumption, publication year, treatments’ mechanism of action, and cancer stage.

Results

Across 30 RCTs, treatments included interferon-α (n = 17), other immunotherapy-containing regimens (n = 10), immune checkpoint inhibitors (n = 3), and targeted therapies (n = 2). BRMA (0.68, 95% CI 0.45-0.82) and WLR (0.71, 95% CI 0.42-0.87) estimated moderate correlation between HR_RFS and HR_OS. Surrogate threshold effect was 0.66/0.68 for studies with 800/1000 patients. Slope coefficients were statistically significant in both models (95% CI 0.09-0.61 BRMA; 95% CI 0.41-0.92 WLR). The 95% prediction intervals around the HR_OS predicted by WLR accurately contained 29/31 (93.5%) of observed HR_OS. Across sensitivity analyses correlations ranged between 0.69 and 0.84 (BRMA) and 0.55 and 0.77 (WLR).

Conclusions

Statistically meaningful correlation between HR_RFS and HR_OS can assist earlier predictions of OS benefit from improvements in RFS for RCTs in resected stage II/III melanoma and provide insights for the earlier evaluation of emerging therapies. Primary model predictions should be approached with caution as nearly half of the evidence base comprised interferon-α trials.