Bayesian probability of success for multivariate generalized linear models

Lead Investigator: Ethan Alt, University of North Carolina at Chapel Hill
Title of Proposal Research: Bayesian probability of success for multivariate generalized linear models
Vivli Data Request: 6521
Funding Source: None
Potential Conflicts of Interest: None

Summary of the Proposed Research:

The project will enable clinicians to power studies with possibly mixed outcome types (e.g., binary, continuous, and count). Because trials typically have primary and multiple secondary endpoints, methods to determine sample size to achieve clinical success across multiple outcomes are necessary. However, such methods are scarce. In earlier work, we have developed a Bayesian approach to multiplicity that is uniformly more powerful than frequentist adjustment procedures, such as the Holm-Bonferroni method, for multiple continuous outcomes (see In this work, we aim to extend this previous work to the case of the multivariate generalized linear model.

Statistical Analysis Plan:

This particular data set was selected because it was a two-arm phase 2 trial that led to a phase 3 trial and had multiple outcomes of mixed types.

We will obtain posterior samples of the regression models and the covariates in order to simulate future study data (to determine a phase 3 sample size based on our method). We will also use the data to inform simulation parameters.

Because we are using the data to illustrate use of our method only, no adjustments will be made to the covariates, missing data, etc. All data will be analyzed as it was in the study protocol as listed on A complete cases analysis will be conducted if the missingness is miniscule. Otherwise, we will utilize a simple imputation method.

Bayesian expected power will be conducted to achieve at least 80% probability of success for the future phase 3 clinical trial.

Requested Studies:

A Multicenter, Randomized, Double-Blind, Placebo-Controlled Exploratory Study to Assess the Effect of Treatment With Prolonged-Release Fampridine (BIIB041) 10 mg Twice Daily on Walking Ability and Balance in Subjects With Multiple Sclerosis
Sponsor: Biogen
Study ID: NCT01597297
Sponsor ID: 218MS205

Public Disclosures:

Alt, E. M., Psioda, M. A., & Ibrahim, J. G. (2023). A Bayesian approach to study design and analysis with type I error rate control for response variables of mixed types. Statistics in Medicine. doi: 10.1002/sim.9696