A simulation and case study to evaluate the extrapolation performance of flexible Bayesian survival models when incorporating real-world data

TL;DR

This paper evaluates the accuracy of flexible Bayesian survival models, specifically survextrap, in extrapolating long-term survival data by incorporating real-world data, demonstrating improved reliability over traditional methods.

Contribution

The study demonstrates that Bayesian survival models effectively integrate trial and real-world data for accurate long-term survival extrapolation, including uncertainty quantification.

Findings

Survextrap provides accurate long-term survival estimates with real-world data.

Including external data improves extrapolation accuracy over trial data alone.

The model reliably estimates differences in survival between treatments with appropriate assumptions.

Abstract

Background: Assessment of long-term survival for health technology assessment often necessitates extrapolation beyond the duration of a clinical trial. Without robust methods and external data, extrapolations are unreliable. Flexible Bayesian survival models that incorporate longer-term data sources, including registry data and population mortality, have been proposed as an alternative to using standard parametric models with trial data alone. Methods: The accuracy and uncertainty of extrapolations from the survextrap Bayesian survival model and R package were evaluated. In case studies and simulations, we assessed the accuracy of estimates with and without long-term data, under different assumptions about the long-term hazard rate and how it differs between datasets, and about treatment effects. Results: The survextrap model gives accurate extrapolations of long-term survival when long-term data on the patients of interest are included. Even using moderately biased external data gives improvements over using the short-term trial data alone. Furthermore, the model gives accurate extrapolations of differences in survival between treatment groups, provided that a reasonably accurate assumption is made about how the treatment effect will change over time. If no long-term data are available, then the model can quantify structural uncertainty about potential future changes in hazard rates. Conclusions: This analysis shows that Bayesian modelling can give accurate and reliable survival extrapolations by making the most of all available trial and real-world data. This work improves confidence in the use of a powerful tool for evidence-based healthcare decision-making.