Extrapolation of Time-to-Event Survival Outcomes of Histology-Independent Therapies Using a Bayesian Hierarchical Model

Abstract

Introduction

Health technology assessment of histology-independent therapies (HITs) requires statistical methods that can capture heterogeneity in outcomes while allowing borrowing of information between tumor sites to inform cost-effectiveness analysis. In this study, we extend previous work on binary outcomes to the application of Bayesian hierarchical models (BHMs) for extrapolation of overall survival from pembrolizumab-treated patients with microsatellite instability-high/deficient mismatch repair solid tumors.

Methods

We considered BHMs based on 1- or 2-parameter distributions for extrapolation of survival outcomes. The scale or rate parameter of each model was assumed exchangeable among tumor types, and the shape parameter was assumed the same for all tumor types in the 2-parameter models. We compared overall survival (OS) and estimated mean survival time for each BHM with the corresponding nonhierarchical model.

Results

Extrapolated OS showed similar results between the BHM and standard models for colorectal, endometrial, and gastric cancers. Small intestine and biliary cancers showed higher OS estimates with a BHM than the standard models due to a combination of smaller sample sizes, information sharing in the BHM, and the use of a common shape parameter. Estimated mean survival times were similar between the BHM and equivalent standard model. However, the BHM showed reduced uncertainty in all cases.

Conclusions

We have demonstrated that BHMs provide a suitable framework to extrapolate time-to-event outcomes for HITs. The results provide extrapolated curves for OS that vary by tumor site, thus capturing and quantifying the inherent heterogeneity within the patient population. BHMs offer advantages in terms of reduced uncertainty around parameters that are often key drivers in cost-effectiveness analyses, such as estimated OS, through the borrowing of information between tumor sites.

Highlights

• Bayesian hierarchical models (BHMs) reduced uncertainty in extrapolation of time-to-event outcomes for histology-independent treatments compared with nonhierarchical models fit to each tumor site.

• Reduced uncertainty around the mean survival time is a key factor of cost-effectiveness analyses of histology-independent treatments.

• BHMs provide a suitable framework for extrapolating histology-independent survival outcomes, effectively integrating prior knowledge and explicitly capturing heterogeneity between different tumor sites.