OBJECTIVES: To ensure timely access to innovative therapies, health technology assessments are often informed by survival data with limited follow-up. To facilitate cost-effectiveness analyses, lifetime survival benefits are estimated using statistical extrapolation methods based on trial data. Here we retrospectively validate survival extrapolations based on the second interim analysis (IA2) with additional 5-year follow-up data from the phase 3 KN-177 study.

METHODS: In KN-177, 307 MSI-H/dMMR mCRC patients were randomized 1:1 to receive 1L pembrolizumab for up to 2 years or mFOLFOX6 or FOLFIRI Q2W ± bevacizumab or cetuximab (referred to as standard of care [SoC]). Crossover to pembrolizumab after disease progression was permitted. OS extrapolations based on IA2 data (February, 2020) were compared with additional follow-up from the 5-year data cut-off (July, 2023) using landmark survival times and restricted mean survival time.

RESULTS: OS from IA2 was extrapolated by fitting piecewise parametric survival models to the OS Kaplan–Meier (KM) data. While published guidelines were followed to select the best model, only the exponential curve was considered clinically plausible and so was selected as the base case for cost-effectiveness analyses. The Weibull distribution was explored in scenario analyses. The exponential distribution underpredicted the 7-year survival rate by 10.7% in the pembrolizumab arm and 14.1% in the SoC arm. The Weibull distribution also underpredicted the observed data, but to a lesser extent.

CONCLUSIONS: The 5-year follow-up data demonstrates that extrapolations based on IA2 for both pembrolizumab and SoC were conservative. For SoC, this may have been influenced by the impact of crossover on long-term survival. Alternative distributions were a better fit to the data but resulted in clinically implausible results. Real-world evidence and/or more flexible statistical models could provide a better fit to the observed data while providing clinically plausible long-term extrapolations.