Advertisement

P51 Blended Survival Curves: A New Approach to Extrapolation for Time-to-Event Clinical Trial Data in Health Technology Assessment

      Objectives

      Survival extrapolation is generally required in the cost-effectiveness analysis to estimate the survival benefit of a new intervention, due to the limited duration of randomized controlled trials (RCTs). Current techniques of extrapolation often assume constant treatment effect beyond the observed period in the RCT, which is implausible and highly influential in survival estimates for resource allocation decisions. The objective of this study is to develop a novel methodology based on “blending” survival curves as a possible solution.

      Methods

      We mixed a flexible Cox semi-parametric model conducted in Bayesian setting to fit the observed data and a parametric model either by prior assumptions or external data on the long-term expected behavior of the underlying survival curves. The two are “blended” into a single survival curve that is equivalent to the Cox model over the follow-up intervals and gradually approaching to the parametric model over the extrapolation period based on a weight function. The weight function and mixing area of the blended curve control the way the internal and external data sources influence the estimated survival.

      Results

      A 4-year follow-up RCT of rituximab in combination with fludarabine and cyclophosphamide (RFC) v. fludarabine and cyclophosphamide alone (FC) for the first-line treatment of chronic lymphocytic leukemia is used to illustrate the method. Two kinds of prior information, registry data and summary of clinical knowledge were respectively used for the long-term estimate.

      Conclusions

      Long-term extrapolation with various assumptions of treatment effect may give significantly different estimated mean survival gains. The blending process allows a consideration of plausible scenarios, abandoning the over-optimistic constant treatment effect and provides sufficient flexibility. Not only internal but also external validity could be carefully considered since a wide range of external evidence can be used to inform the long-term estimate, including hard data from real world and clinical expert opinion.