Surrogate properties of survival endpoints in metastatic soft-tissue sarcoma: A meta-analysis.

Abstract

Background: Alternative endpoints to overall survival (OS) such as progresion-free survival (PFS), time-to-progression (TTP) or time-to-treatment failure (TTF) are increasingly used to assess treatment efficacy in randomized controlled trials (RCT) to reduce inclusions and trials’ duration. Their properties in terms of surrogate markers need to be assessed to ensure that they can adequately replace OS. Methods: We conducted a literature review to summarize by cancer type studies evaluating surrogate endpoints for OS. In the absence of data for soft-tissue sarcoma (STS), we assessed surrogate properties for OS of PFS, TTP and TTF in advanced STS. We relied on a meta-analytical framework to estimate individual-level association (association between the candidate surrogate endpoint and OS) and trial-level association (association between the treatment effects on the candidate surrogate and on OS). Statistical methods included weighted linear regression (WLR) and the two-stage method introduced by Burzykowski et al., which relies on the joint modeling of 2 survival endpoints with a copula function and a regression model. Results: Individual data of 2020 patients from 10 European RCTs were analyzed. We censored OS at 2 years and PFS, TTP and TTF at 1 year. Regardless of the method, the highest individual-level association was observed for PFS (R² = 0.62 IC95% [0.27; 0.76]; Kendall's tau = 0.43 IC95% [0.40; 0.46]). Even if WLR suggested that PFS had the highest trial-level association, it was not significant for any of the two methods (WLR: R² = 0.44 IC95% [0.00; 0.69]; two-stage model: R² = 0.01 IC95% [-0.46; 0.48]). Conclusions: Out of the 3 endpoints, PFS had the best surrogate properties. Associations with OS were however moderate and cannot validate PFS as a surrogate for OS. This could be partly explained by a lack of precision due to our small sample size in terms of patients and trials. STS are rare tumors ( < 2% of all cancers) which explains the low number of available trials with usually smaller sample sizes compared to other cancers. We are however collecting additional trials to improve these estimations and complementary analyses are ongoing to validate our regression models on external data.