Feifang Hu

Optimality, Variability, Power: Evaluating Response-Adaptive Randomization Procedures for Treatment Comparisons

We provide a theoretical template for the comparison of response-adaptive randomization procedures for clinical trials. Using a Taylor expansion of the noncentrality parameter of the usual chi-squared test for binary responses, we show explicitly the relationship among the target allocation proportion, the bias of the randomization procedure from that target, and the variability induced by the randomization procedure. We also generalize this relationship for more than two treatments under various multivariate alternatives. This formulation allows us to directly evaluate and compare different response-adaptive randomization procedures and different target allocations in terms of power and expected treatment failure rate without relying on simulation. For K = 2 treatments, we compare four response-adaptive randomization procedures and three target allocations based on multiple objective optimality criteria. We conclude that the drop-the-loser rule and the doubly adaptive biased coin design are clearly superior to sequential maximum likelihood estimation or the randomized play-the-winner rule in terms of decreased variability, but the latter is preferable because it can target any desired allocation. We discuss how the template developed in this article is useful in the design and evaluation of clinical trials using response-adaptive randomization.

Markov Chain Marginal Bootstrap

Markov chain marginal bootstrap (MCMB) is a new method for constructing confidence intervals or regions for maximum likelihood estimators of certain parametric models and for a wide class of M estimators of linear regression. The MCMB method distinguishes itself from the usual bootstrap methods in two important aspects: it involves solving only one-dimensional equations for parameters of any dimension and produces a Markov chain rather than a (conditionally) independent sequence. It is designed to alleviate computational burdens often associated with bootstrap in high-dimensional problems. The validity of MCMB is established through asymptotic analyses and illustrated with empirical and simulation studies for linear regression and generalized linear models.

Implementing optimal allocation in sequential binary response experiments

For sequential experiments with K treatments, we establish two formal optimization criteria to find optimal allocation strategies. Both criteria involve the sample sizes on each treatment and a concave noncentrality parameter from a multivariate test. We show that these two criteria are equivalent. We apply this result to specific questions: (1) How do we maximize power of a multivariate test of homogeneity with binary response?, and (2) for fixed power, how do we minimize expected treatment failures? Because the solutions depend on unknown parameters, we describe a response-adaptive randomization procedure that “targets” the optimal allocation and provides increases in power along the lines of 2–4% over complete randomization for equal allocation. The increase in power contradicts the conclusions of other authors who have explored other randomization procedures for K = 2 and have found that the variability induced by randomization negates any benefit of targeting an optimal allocation.

Asymptotics in randomized urn models

This paper studies a very general urn model stimulated by designs in clinical trials, where the number of balls of different types added to the urn at trial n depends on a random outcome directed by the composition at trials 1,2,...,n-1. Patient treatments are allocated according to types of balls. We establish the strong consistency and asymptotic normality for both the urn composition and the patient allocation under general assumptions on random generating matrices which determine how balls are added to the urn. Also we obtain explicit forms of the asymptotic variance-covariance matrices of both the urn composition and the patient allocation. The conditions on the nonhomogeneity of generating matrices are mild and widely satisfied in applications. Several applications are also discussed.

EFFICIENT RANDOMIZED-ADAPTIVE DESIGNS

Response-adaptive randomization has recently attracted a lot of attention in the literature. In this paper, we propose a new and simple family of response-adaptive randomization procedures that attain the Cramer-Rao lower bounds on the allocation variances for any allocation proportions, including optimal allocation proportions. The allocation probability functions of proposed procedures are discontinuous. The existing large sample theory for adaptive designs relies on Taylor expansions of the allocation probability functions, which do not apply to nondifferentiable cases. In the present paper, we study stopping times of stochastic processes to establish the asymptotic efficiency results. Furthermore, we demonstrate our proposal through examples, simulations and a discussion on the relationship with earlier works, including Efrons biased coin design.

Asymptotic properties of covariate-adaptive randomization

Balancing treatment allocation for influential covariates is critical in clinical trials. This has become increasingly important as more and more biomarkers are found to be associated with different diseases in translational research (genomics, proteomics and metabolomics). Stratified permuted block randomization and minimization methods [Pocock and Simon Biometrics 31 (1975) 103-115, etc.] are the two most popular approaches in practice. However, stratified permuted block randomization fails to achieve good overall balance when the number of strata is large, whereas traditional minimization methods also suffer from the potential drawback of large within-stratum imbalances. Moreover, the theoretical bases of minimization methods remain largely elusive. In this paper, we propose a new covariate-adaptive design that is able to control various types of imbalances. We show that the joint process of within-stratum imbalances is a positive recurrent Markov chain under certain conditions. Therefore, this new procedure yields more balanced allocation. The advantages of the proposed procedure are also demonstrated by extensive simulation studies. Our work provides a theoretical tool for future research in this area.

Balancing continuous covariates based on Kernel densities

The balance of important baseline covariates is essential for convincing treatment comparisons. Stratified permuted block design and minimization are the two most commonly used balancing strategies, both of which require the covariates to be discrete. Continuous covariates are typically discretized in order to be included in the randomization scheme. But breakdown of continuous covariates into subcategories often changes the nature of the covariates and makes distributional balance unattainable. In this article, we propose to balance continuous covariates based on Kernel density estimations, which keeps the continuity of the covariates. Simulation studies show that the proposed Kernel-Minimization can achieve distributional balance of both continuous and categorical covariates, while also keeping the group size well balanced. It is also shown that the Kernel-Minimization is less predictable than stratified permuted block design and minimization. Finally, we apply the proposed method to redesign the NINDS trial, which has been a source of controversy due to imbalance of continuous baseline covariates. Simulation shows that imbalances such as those observed in the NINDS trial can be generally avoided through the implementation of the new method.

A Unified Family of Covariate-Adjusted Response-Adaptive Designs Based on Efficiency and Ethics

Response-adaptive designs have recently attracted more and more attention in the literature because of its advantages in efficiency and medical ethics. To develop personalized medicine, covariate information plays an important role in both design and analysis of clinical trials. A challenge is how to incorporate covariate information in response-adaptive designs while considering issues of both efficiency and medical ethics. To address this problem, we propose a new and unified family of covariate-adjusted response-adaptive (CARA) designs based on two general measurements of efficiency and ethics. Important properties (including asymptotic properties) of the proposed procedures are studied under categorical covariates. This new family of designs not only introduces new desirable CARA designs, but also unifies several important designs in the literature. We demonstrate the proposed procedures through examples, simulations, and a discussion of related earlier work.

Generalized multidimensional dynamic allocation method.

Dynamic allocation has received considerable attention since it was first proposed in the 1970s as an alternative means of allocating treatments in clinical trials which helps to secure the balance of prognostic factors across treatment groups. The purpose of this paper is to present a generalized multidimensional dynamic allocation method that simultaneously balances treatment assignments at three key levels: within the overall study, within each level of each prognostic factor, and within each stratum, that is, combination of levels of different factors Further it offers capabilities for unbalanced and adaptive designs for trials. The treatment balancing performance of the proposed method is investigated through simulations which compare multidimensional dynamic allocation with traditional stratified block randomization and the Pocock-Simon method. On the basis of these results, we conclude that this generalized multidimensional dynamic allocation method is an improvement over conventional dynamic allocation methods and is flexible enough to be applied for most trial settings including Phases I, II and III trials.

Outcome-adaptive randomization for a delayed outcome with a short-term predictor: imputation-based designs.

Delay in the outcome variable is challenging for outcome-adaptive randomization, as it creates a lag between the number of subjects accrued and the information known at the time of the analysis. Motivated by a real-life pediatric ulcerative colitis trial, we consider a case where a short-term predictor is available for the delayed outcome. When a short-term predictor is not considered, studies have shown that the asymptotic properties of many outcome-adaptive randomization designs are little affected unless the lag is unreasonably large relative to the accrual process. These theoretical results assumed independent identical delays, however, whereas delays in the presence of a short-term predictor may only be conditionally homogeneous. We consider delayed outcomes as missing and propose mitigating the delay effect by imputing them. We apply this approach to the doubly adaptive biased coin design (DBCD) for motivating pediatric ulcerative colitis trial. We provide theoretical results that if the delays, although non-homogeneous, are reasonably short relative to the accrual process similarly as in the iid delay case, the lag is also asymptotically ignorable in the sense that a standard DBCD that utilizes only observed outcomes attains target allocation ratios in the limit. Empirical studies, however, indicate that imputation-based DBCDs performed more reliably in finite samples with smaller root mean square errors. The empirical studies assumed a common clinical setting where a delayed outcome is positively correlated with a short-term predictor similarly between treatment arm groups. We varied the strength of the correlation and considered fast and slow accrual settings.