An information-theoretic approach for selecting arms in clinical trials

TL;DR

This paper introduces an information-theoretic method for selecting the best treatment arms in clinical trials, focusing on accuracy, simplicity, and applicability without relying on parametric assumptions.

Contribution

It proposes a novel, intuitive arm selection criterion based on weighted information measures, applicable to multinomial outcomes in clinical trial design.

Findings

Accurately identifies optimal arms without parametric assumptions

Demonstrates good asymptotic properties and small sample performance

Outperforms some existing methods in simulations

Abstract

The question of selecting the "best" amongst different choices is a common problem in statistics. In drug development, our motivating setting, the question becomes, for example: what is the dose that gives me a pre-specified risk of toxicity or which treatment gives the best response rate. Motivated by a recent development in the weighted information measures theory, we propose an experimental design based on a simple and intuitive criterion which governs arm selection in the experiment with multinomial outcomes. The criterion leads to accurate arm selection without any parametric or monotonicity assumption. The asymptotic properties of the design are studied for different allocation rules and the small sample size behaviour is evaluated in simulations in the context of Phase I and Phase II clinical trials with binary endpoints. We compare the proposed design to currently used alternatives and discuss its practical implementation.