Improved adaptive randomization strategies for a seamless Phase I/II dose-finding design

Donglin Yan, Nolan A. Wages, Emily V. Dressler

Research output: Contribution to journalArticlepeer-review

4 Scopus citations


In this article, we propose and evaluate three alternative randomization strategies to the adaptive randomization (AR) stage used in a seamless Phase I/II dose-finding design. The original design was proposed by Wages and Tait in 2015 for trials of molecularly targeted agents in cancer treatments, where dose-efficacy assumptions are not always monotonically increasing. Our goal is to improve the design’s overall performance regarding the estimation of optimal dose as well as patient allocation to effective treatments. The proposed methods calculate randomization probabilities based on the likelihood of every candidate model as opposed to the original design which selects the best model and then randomizes doses based on estimations from the selected model. Unlike the original method, our proposed adaption does not require an arbitrarily specified sample size for the adaptive randomization stage. Simulations are used to compare the proposed strategies and a final strategy is recommended. Under most scenarios, our recommended method allocates more patients to the optimal dose while improving accuracy in selecting the final optimal dose without increasing the overall risk of toxicity.

Original languageEnglish
Pages (from-to)333-347
Number of pages15
JournalJournal of Biopharmaceutical Statistics
Issue number2
StatePublished - Mar 4 2019

Bibliographical note

Publisher Copyright:
© 2018, © 2018 Taylor & Francis.


  • Seamless I/II adaptive design
  • adaptive randomization
  • continual reassessment method
  • dose finding
  • molecularly targeted agent
  • optimal biological dose

ASJC Scopus subject areas

  • Statistics and Probability
  • Pharmacology
  • Pharmacology (medical)


Dive into the research topics of 'Improved adaptive randomization strategies for a seamless Phase I/II dose-finding design'. Together they form a unique fingerprint.

Cite this