Advancements in Efficient Randomized Adaptive Design in Clinical Trials
Open Access DepositedAdvancements in Efficient Randomized Adaptive Design in Clinical TrialsIn randomized controlled experiments, response-adaptive randomization (RAR) aims to allocate more patients to superior treatments while maintaining valid and efficient statistical inference. The Efficient Randomized Adaptive Design (ERADE) is particularly distinguished in this class for achieving the Cramér–Rao lower bound for the allocation proportion, indicating optimal statistical efficiency. However, existing results for ERADE are limited to two-arm clinical trials with im- mediate responses. Several important practical features of real trials are not addressed, including multiple treatment arms, delayed and missing responses, and model misspecification. These gaps motivate methodological developments that extend ERADE to more practical trial settings while preserving its desirable asymptotic features. Chapter 2 extends the Efficient Randomized Adaptive Design (ERADE), originally proposed by Hu et al. [32] for two-arm clinical trials, to settings with multiple treatment options, which are increasingly common in modern clinical research. We develop a multi-arm version of ERADE that preserves the simplicity and statistical efficiency of the original procedure while allowing flexible allocation across more than two treatment arms. Theoretical analyses show that the proposed design retains the key asymptotic properties established in the two-arm case, including the consistency of allocation proportions and asymptotic normality. Simulation studies further demonstrate that the multi-arm ERADE performs well in finite samples across a range of trial scenarios. Finally, we illustrate the practical applicability and effectiveness of the generalized design through a real-world trial redesign. Chapter 3 develops a grouped-updating version of the Efficient Randomized Adaptive Design, where allocation is updated after groups of responses rather than after each patient. Grouped-updating structures have been previously investigated for Doubly Biased Coin Design (DBCD) by Zhai et al. [81]
here, we develop the corresponding framework for ERADE. We establish that the proposed Group ERADE preserves the key asymptotic properties of the classical ERADE. The framework is further extended to incorporate delayed and missing responses, and the associated asymptotic properties are derived. Simulation studies and a real clinical trial redesign demonstrate that the Group ERADE achieves high efficiency and stable allocation behavior under various target allocation schemes, and we compare its performance with that of the Group DBCD. Chapter 4 investigates the robustness of the Efficient Randomized Adaptive Design under misspec- ification of both the design and analysis models. Recently, Ye et al. [78] established robustness properties under model misspecification for the Doubly Biased Coin Design (DBCD). Much like other response-adaptive procedures, ERADE is typically studied under the assumption that the response model is correctly specified. However, real clinical data rarely conform to the idealized models assumed during trial design or analysis, creating uncertainty about ERADE’s behavior under model misspecification. We evaluate robustness under two types of model misspecification. The first is design-model misspecification, where the assumed response distribution used to guide allocation updates does not match the true data-generating distribution. The second is analysis-model misspec- ification, where the regression model used to estimate the treatment effect is incorrectly specified. We show that, even when the design model is misspecified, the ERADE allocation proportions remain consistent and asymptotically normal, highlighting the stability of the procedure. On the inference side, we study three commonly used treatment-effect estimators—difference-in-means, ANCOVA I, and ANCOVA II—and allow each regression model to be misspecified. For all three estimators, we derive consistency and asymptotic normality of the treatment effect estimators. Simulation studies further illustrate how ERADE behaves under a variety of model-misspecification scenarios, confirming the robustness predicted by the theoretical results.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Alkhnefr_gwu_0075A_17861.pdf | 2026-06-24 | Open Access |
|