For phase 1 dose-finding studies, a variety of model-based approaches and study designs with expansion cohorts have been proposed in place of the conventional 3＋3 design, and the benefits of these designs have been reported. However, there are limited reports on the practical details of such statistical work using actual clinical trial data. We present practical procedures for planning, conducting, and decision making in a model-based approach with both dose-escalation and expansion parts using actual clinical trial data, focusing on statistical considerations.

In ex vivo dose-response experiments with multiple animals, it is common to observe variation in the shape of the dose-response curves between animals. This heterogeneity may be explained to some extent by the differences in responsiveness between specimens, which is typically evaluated in two ways：spontaneous reactions with and without the standard stimulant at baseline. If available, these covariates may be used to predict a specimen’s maximal and minimal responses. Meanwhile, the between-substance similarities in dose-response curves are crucial in relative potency estimation. In a parallel-line assay, for example, we need to demonstrate the parallelism between dose-response slopes. When using multiple animals, it is also crucial to confirm the similarity within each animal（Uehara et al. 2016a）．This paper discusses the appropriate use of the above-noted covariates to assess intrasubject parallelism（Uehara 2018）．The influence of the covariates on the dose-response curve is investigated in order to improve the efficiency of the intrasubject parallelism assessment and enhance the precision of the parameter estimates. An example is introduced to demonstrate the proposed method, and simulation study results are reported.