OBJECTIVES: For many emerging cancer treatments, establishing efficacy by direct observation of overall survival (OS) is considered unethical or impractical, and progression-free survival (PFS) is used as a surrogate endpoint for OS. Though many studies have explored PFS as a surrogate for long-term survival, and confounding factors such as treatment switching or imbalance in subsequent therapies have also been analyzed, to our knowledge no studies have focused on approaches to extrapolating the association when the hazard ratio (HR) for PFS is lower than the range observed in existing trial-level results. This work explored approaches to expanding the potential prediction range for the association between PFS and OS in pretreated non-small cell lung cancer (NSCLC).
METHODS: Randomised trials in pretreated NSCLC (with or without activating alterations) were identified via systematic searches; trials with treatment switching or imbalance in subsequent therapy were excluded. Eighteen trials were included. HRs observed in the trials ranged from 0.48-1.13 for PFS and 0.73-1.14 for OS. To expand the range of HRs, additional data from subgroups in which greater efficacy was observed were included. The association between the HRs for PFS and OS was explored using bivariate meta-regression and models with Wishart prior and product normal formulation (PNF).
RESULTS: Including subgroup data extended the HR range to 0.20-2.87 for PFS and 0.42-2.03 for OS. Bivariate models and models with Wishart prior were more influenced by subgroup data and estimated larger treatment effects for OS for a given effect on PFS than PNF models. Slopes for models including subgroup data (i.e. the change in ln(HR) for OS per unit change in ln(HR) for PFS) were 0.333, 0.342, and 0.236, respectively.
CONCLUSIONS: Inclusion of subgroup data may allow prediction of OS over a wider range of PFS HRs and PNF models may provide a conservative estimate of the association.
