Dose Estimation

This paper presents state-of-the-art statistical methods for dose-finding experiments in first-in-man clinical studies of drugs for the treatment of malignancies. Most early-phase clinical trials are not hypothesis driven, which might be the reason why statistical considerations have been largely ignored in dose-finding studies. The standard experimental design for dose-finding clinical studies employs a rule-based, dose-escalation scheme in which escalation depends on the number of patients at a dose level who experience dose-limiting toxicity. The standard design is widely used because of its algorithm-based simplicity for clinical investigators.

In the last two decades, new approaches for dose-finding have been proposed, all aiming to (i) model the toxicity of a new treatment as a percentile of the dose-toxicity relationship; (ii) minimize the number of patients treated at unacceptably high toxic dose levels; and (iii) minimize the number of patients needed to complete the study. In this paper, we describe some of these methodologies in simple terms for nonstatisticians.