Statistics and Its Interface

Volume 3 (2010)

Number 2

Maximum-relevance weighted likelihood estimator: application to the continual reassessment method

Pages: 177 – 183

DOI: http://dx.doi.org/10.4310/SII.2010.v3.n2.a5

Authors

Sylvie Chevret (Département de Biostatistique et Informatique Médicale, Université Paris Diderot - Paris 7, Paris, France)

Matthieu Resche-Rigon (Département de Biostatistique et Informatique Médicale, Université Paris Diderot - Paris 7, Paris, France)

Sarah Zohar (Département de Biostatistique et Informatique Médicale, Université Paris Diderot - Paris 7, Paris, France)

Abstract

Typical phase I dose-finding clinical trials, notably in cancer, are characterized by a small number of patients (less than 40), a relatively high number of dose levels (4 to 6) and sequential dose allocation rules. In this setting, the Continual Reassessment Method (CRM) has been recommended as a dose allocation rule that provides a consistent method to converge to the maximal tolerated dose (MTD), possibly based on likelihood (CRML). In this adaptive design setting, we derived a Relevance Weighted Likelihood to propose a robust estimation of the MTD. The main idea is to weight the individual contributions to likelihood using a decreasing function of rank. We compare this method to the CRML throughout simulations.

Keywords

relevance weighted likelihood, phase I, dose-finding clinical trials, continual reassessment method

Full Text (PDF format)