betaFeedback

WorksMathematics

A General Theory of Nonparametric Response-Adaptive Randomization

Abstract

Response-Adaptive Randomization (RAR) assigns incoming patients to treatments with probabilities that depend on previously observed responses. While most existing RAR procedures rely on finite-dimensional summaries of treatment performance (such as rates, means, or model parameters), we develop a nonparametric (NP) distribution-based framework in which the treatment outcome distributions are learned sequentially and adaptation is driven by their functionals. The framework naturally accommodates multiple treatments and is particularly suited to endpoints for which average effects may conceal relevant differences across the outcome distribution. We introduce an ethical NP-RAR rule driven by a pre-specified distributional score and a nonparametric extension of the multi-arm efficient randomized adaptive design (NP-ERADE) able to approach generic distribution-dependent targets selected according to ethical or inferential optimality criteria. We establish strong consistency and joint functional central limit theorems for the empirical treatment distributions, together with asymptotic normality of functional scores, estimated targets and allocation proportions. We further show that the NP-ERADE allocation proportions attain the semiparametric efficiency bound for the target. We develop post-trial inference for score contrasts and distributional features, including crossing locations, and investigate finite-sample properties, robustness and inferential performance via simulations.

Paper details

vlnc:W_7sv7e5sb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…