Título: Neighborhood designs in phase-I clinical Trials
Authors: José Antonio Moler
Co-authors: Maider Sada, Nancy Flournoy
Language: en

An experiment that aims to estimate the stimulus (or, in a wide sense, dose) that would trigger a prespecified proportion of positive responses is a dose-finding design. In a toxicity problem, the uncertainty about the behavior of a new drug forces experimenters to be cautious in the design of the clinical tria (experiment). Adaptive designs provide allocation rules that assign a dose to the present patient or cohort of patients considering previous responses and allocations. But they do not converge a.s. to the target dose. To overcome this problem we propose neighborhood designs which allocate patients to the two adjacent doses closest to the current target dose estimate. We present their theoretical properties and a comparative simulation study with relevant designs used in practice.