Statistics > Methodology
Title:Dose-Finding for Repeated Presses of a Latched Hall-Call Button: A Continual Reassessment Design with an Efficacy Futility Rule
Abstract:Presses of an elevator hall-call button that is already latched are discarded by the controller, but the practice is common and its dose has never been bounded. We treat the number of redundant presses as a dose and look for the maximum tolerated dose, with a dose-limiting event defined as an audible remark by a bystander that is addressed to the person pressing or to the button. We describe a continual reassessment design in which a Bayesian futility rule on a continuous efficacy endpoint, the error of the controller's arrival estimate, is added to the toxicity model, and we give the stopping boundary in closed form. In simulations over five toxicity scenarios with 2,000 replicates each, the design selected the correct dose in 28% to 90% of replicates, similar to a Bayesian optimal interval design and above the 3+3 design (30% to 42%). With no efficacy the futility rule stopped 46% to 95% of trials, and with an efficacy of 1.0 s per press it stopped fewer than 1%. We then ran the design over 11 weekday mornings. The futility rule stopped the trial after 33 episodes, when the posterior probability that the slope of arrival-estimate error on dose was steeper than s per press fell to 0.018. The estimated maximum tolerated dose was 8 presses (posterior dose-limiting probability 0.27), and the estimated slope was 0.05 s per press (95% CI to ). No dose was recommended for further study.
Submission history
From: Mara Lindqvist[v1] Fri, 20 Nov 2026 21:32:31 UTC (172 KB)