A comparison of priors for variance parameters in Bayesian basket trials

TL;DR

This paper compares different prior choices for variance parameters in Bayesian basket trials, highlighting the advantages of penalized complexity priors over traditional options through simulation studies.

Contribution

It provides a comprehensive comparison of priors, including the recently introduced penalized complexity priors, and demonstrates their performance advantages in certain scenarios.

Findings

PC priors perform similarly to half-t priors in general.

PC priors excel in homogeneous response scenarios.

They offer easier handling of shrinkage degree via a single parameter.

Abstract

Phase II basket trials are popular tools to evaluate efficacy of a new treatment targeting genetic alteration common to a set of different cancer histologies. Efficient designs are obtained by pooling data from the different arms (e.g., cancer histologies) via Bayesian hierarchical modelling, with a variance parameter controlling the strength of shrinkage of each arm treatment effect to the overall treatment effect. One critical aspect of this approach is that prior choice on the variance plays a major role in determining the strength of shrinkage and impacts the operating characteristics of the design. We review the priors most commonly adopted in previous works and compare them with the recently introduced penalized complexity (PC) priors. Our simulation study shows comparable behaviour for the PC prior and the gold standard choice half-t prior, with the former performing better in the homogeneous scenario where all histologies respond similarly to the treatment. We argue that PC priors offer advantages over other priors because they allow the user to handle the degree of shrinkage by means of only one parameter and can be elicited based on clinical opinion when available.