EURO 2025 Leeds
Abstract Submission

1099. Simulation Optimisation for Clinical Trial Design

Invited abstract in session TA-4: Journal of Simulation. Computer Modelling and Simulation, stream OR Journals.

Tuesday, 8:30-10:00
Room: Rupert Beckett LT

Authors (first author is the speaker)

1. Luke Rhodes-Leader
Management Science, Lancaster University
2. Matthew Darlington
STOR-i, Lancaster University
3. Tom Parke
Berry Consultants
4. Peter Jacko
Ma, Lancaster University

Abstract

The design of complex clinical trials, particularly those that involve multi-stage analyses, can be a complicated task. It includes considerations on several operating characteristics such as the number of participants (trial size) and statistical properties such as the type-I and type-II error rates. The trial configurations involve multiple parameters, including trial sizes, number of interim analyses, each analysis’ significance level, and stopping thresholds for success/futility. To evaluate these metrics and explore the configurations, a stochastic simulation is often needed. To get an accurate evaluation of a configuration (needed to ensure a trial meets regulatory standards) requires the simulation to be run many times (approximately 100,000), which may take several hours, for each simulation scenario. By exploring and altering the parameters, one can find a trial configuration with good performance.
Current practice can be trial and error, learning from the designer’s experience and the output of the simulation model. This is time consuming. This talk will discuss a project aiming to build a simulation optimisation approach to make the searching for optimal designs more systematic and efficient. In particular, Bayesian Optimisation was applied, with some adaptations to handle application specific considerations, such as stochastic constraints on the type-I error. The results suggest able to identify good solutions much more quickly than the usual approach.

Keywords

Status: accepted

Back to the list of papers