Using a model based approach to inform dose escalation in a Ph I Study by combining emerging clinical and prior preclinical information: an example in oncology
S.Y. Amy Cheung, James W.T. Yates, Peter Lawrence, Marcelo Marotti, Barry Davies, Paul Elvin, Christine Stephens, Paul Stockman and Andrew Foxley
AstraZeneca R & D, Macclesfield, UK
Objectives: Oncology Phase I studies are typically small, open-label, sequential studies enrolling 3-6 patients per dose escalation [1]. Deriving a recommended dose, schedule and potential combination option is one of the main goals.
Rule based methods are used to identify the recommended dose [1] based on clinical data generated during the study. The disadvantages of these methods are that they are unable to use all previous information on the study and cannot easily provide extrapolation to untested schedules. Model based approaches for human studies [2] allow utilization of all available data, and the relationship between dose, exposure and effect to be determined.
This example of a first time in man trial demonstrates the benefits of incorporating model based approaches to inform dose escalation.
Methods: Prior to this study, pre-clinical information was reviewed to identify and prioritize key data for analysis that would provide useful signals for tolerability.
The predicted human PK profile was used as prior information to enable analysis of the sparse datasets emerging from the first few cohorts. Pharmacodynamic models developed from pre-clinical data were reapplied to the clinical data.
The models were updated with data from each successive cohort of patients and then used to simulate the endpoints for a range of proposed dose escalations to inform the clinical team with predicted outcomes. These models were also used to explore options for further arms of the study to investigate alternative schedules.
Results: Even from small data sets the models developed were robust to inform escalation. This was demonstrated in part by the ability to predict to untested doses and schedules. The simulations of continuous variables allowed for dose increments and the starting dose for alternative schedules to be determined using a quantitative basis. This included the instigation and escalation of intermittent dosing arms that proceeded to identify the recommended dose more quickly than would have been the case with a classical approach.
Conclusions: The utilization of pre-clinical, clinical PK, safety and PD data in model based dose escalation allows rapid learning in early phase clinical development. This real-time approach using simulation of scenarios based on the available information has enabled a development program to identify the RD for a range of schedules efficiently thereby improving trial outcome.
References:
[1] Le Tourneau C. et al. JNCI Vol. 101, Issue 10, 2009
[2] Goodwin R. et al. Euro J of Can 48, 170-178, 2012
