Each week, we will be publishing layman’s abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.
Bayesian methods for the analysis of early‐phase oncology basket trials with information borrowing across cancer types. Statistics in Medicine. 2020; 39: 3459–3475. https://doi.org/10.1002/sim.8675, , , .
A new type of clinical trial called basket trial has recently drawn increasing attention, which was motivated by the change in the focus of oncology research, from histological properties of tumors in each organ, to the genetic aberration potentially shared by multiple cancer types. The basket trial tests treatment efficacy simultaneously on multiple cancer types with common aberration. Assuming homogeneous effects, various methods have been recently developed to increase the trial power by borrowing information across cancer types, which, however, tend to give highly inflated type I error rate that can be unacceptable even for early-phase trials.
This paper first investigates several representative information borrowing methods for the analysis of early-phase basket trials, including the conventional Bayesian hierarchical model, the exchangeability-nonexchangeability (EXNEX) approach, and Liu’s two-stage approach. A novel method named the Bayesian hierarchical model with a correlated prior (CBHM) is then proposed, which conducts more flexible borrowing across cancer types according to sample similarity. Simulation results show the advantage of all information borrowing approaches compared to independent analysis in terms of trial power, when a large proportion of the cancer types truly respond to the treatment. But when the treatment is effective only on a small subset of the cancer types, the existing information borrowing approaches may have lowered power and inflated type I error rate. The proposed CBHM approach has the most robust performance, with a power similar to EXNEX or Liu’s approach, and substantially lowered inflation in type I error control under all considered simulating settings, which can be quite meaningful in clinical application. Further, the general instructions and open-source code provided will allow researchers to conduct similar analysis in a broader range of clinical trials.