PCN358 COMPARING AND VALIDATING THE OVERALL SURVIVAL EXTRAPOLATIONS FOR IMMUNO-ONCOLOGY TREATMENTS APPLIED IN HTA'S BY NOMA AND TLV

Abstract

To compare the extrapolation of overall survival (OS) in reimbursement submissions of Immuno-Oncology (IO) treatments by the Norwegian Medical Agency (NoMA) and the Dental and Pharmaceutical Benefits Board (TLV), and to validate the OS extrapolation using longer-term trial data that have been published after the health technology appraisals (HTA). This study follow the same overall approach as Bullement et al. (2019), who published a review and validation of OS extrapolation in IO assessments published by NICE. Appraisals of programed death-ligand 1 (PD-L1) treatments published by NoMA and TLV were identified from the agencies’ webpage, and Kaplan-Meier curves and extrapolation methods used in the agency base-case were extracted. A targeted literature search was conducted to identify if longer follow-up on OS had been published. This study only included cases that 1) had a published assessment report from both agencies; and 2) had longer follow-up data on OS published after the HTA assessment. Agency base-cases were recreated by digitizing Kaplan-Meier curves, creating pseudo individual patient level data and applying the base-case extrapolations to these data. Comparison and validation of extrapolations were assessed by visual inspection and by investigating how mean OS estimates were affected by applying the base-case extrapolation to the longer-term follow-up data. Seven cases that fulfilled the inclusion criteria were identified. Graphical representation showed significant variation in long-term OS extrapolation between the agencies that mostly underestimated long-term OS. Applying the base case extrapolation methods to later data-cuts resulted in increased long-term OS estimations. This study demonstrate variation in long-term OS extrapolation of NoMA and TLV in IO HTA’s. Standard parametric survival extrapolations do not seem to capture the potential long-term OS, which suggests that alternative approaches to OS extrapolation of IO treatments, e.g. mixture cure or response based models, are needed.