Multi-stage dose expansion cohort (MSDEC) design with Bayesian stopping rule. Issue 4 (4th July 2022)

• Record Type: Journal Article
• Authors: Wang, Shuqi; Tan, Ming
• Is Part Of: Journal of biopharmaceutical statistics. Volume 32:Issue 4(2022)
• Page Start: 600
• Page End: 612
• Publication Date: 2022-07-04
• Subjects: Bayesian stopping rule -- dose expansion cohort -- hybrid frequentist-Bayesian method -- power prior -- MTD; Pharmacy -- Statistical methods -- Periodicals; Drugs -- Testing -- Statistical methods -- Periodicals; Biometry -- Periodicals; Biopharmaceutics -- Periodicals; Pharmacokinetics -- Periodicals; 615.19
• DOI: 10.1080/10543406.2022.2080694
• Languages: English
• ISSNs: 1054-3406

Abstract

ABSTRACT: Phase I trial designs generally fall into three categories: algorithm-based (e.g., the classic 3 + 3 design), model-based (e.g., the continual reassessment method, CRM), and model-assisted designs that combine features of the first two (e.g., the Bayesian Optimal Interval, BOIN, design). The classic '3 + 3' design continues to be the most frequently used design in phase I trials in finding maximum tolerated dose (MTD) due to its simplicity and feasibility, though many other model-based designs such as the Continual Reassessment Method (CRM) have also been proposed and used in various such as immunotherapies trials. The MTD based on three or six patients is not accurate, and dose-expansion cohorts (DEC) are increasingly used to better characterize the toxicity profiles of experimental agents. This article proposes a multi-stage dose-expansion cohort (MSDEC) hybrid frequentist-Bayesian design combining the power prior and the sequential conditional probability ratio test. In this design, results from the dose-escalation part are viewed and treated as historical data, and then are weighted and modeled through power prior. For safety monitoring, the Bayesian stopping rule is developed and the maximum sample size is calculated by a fixed-sample-size test with exact binomial computation. Simulation studies showed that MSDEC reduces the chance that a patient experiences a toxic dose. Power prior provides a reasonable prior for the Bayesian model because the degree of informativeness of the prior can be driven by the ("objective") historical data rather than from expert opinion elicited on parameters in the model.