Improved study design in phase IIb by the use of optimal design, focusing on the precision of dose finding
M Kågedal(1), Joakim Nyberg(2), A Dominicus(1), MO Karlsson(2) A Hooker(2)
(1) AstraZeneca R&D Södertälje, Sweden (2) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden
Objectives: The objective of dose finding studies is to identify the relationship between dose and efficacy to guide selection of doses to be further studied in phase III. Not all parts of the dose exposure response curve are of equal importance and it may be difficult to predict how the precision in parameter estimates translates in to precision in dose-selection. Thus if optimal design methods are used it is important to have a parameter in the model that has a direct relevance for dose selection and to optimize the study design w.r.t. the precision of that parameter.
The objective of this work was to investigate if the design of a dose finding study could be improved by the use of a re-parameterized Emax model by applying the Ds and D-optimal criteria.
Methods: An alternative parameterisation of the sigmoid Emax model, described by Groth, was used [1] where one of the parameters (D*) is the dose corresponding to a particular treatment effect (E*). In the optimization, the optimal design tool PopED [2] was used. The D-optimal criteria was used as well as the Ds criteria with D* as the parameter of interest. Assumptions with regards to steady state pharmacokinetics and the shape of the exposure response curve was based on a drug intended for the treatment of neuropathic pain where pain is assessed using a numerical rating scale graded 0-10. The Ds and D-optimal designs were subsequently evaluated by means of simulation to estimate the probability of correctly estimating the dose corresponding to a treatment effect of 1 versus placebo. Correct estimation of dose was defined as being in the interval 6-24 mg and the true dose was 12 mg.
Results: The optimal design based on the Ds-optimal design was 0, 0.1, 9 and 18 mg and the D-optimal design was 0, 1.5, 7.5 and 18 mg. The probability of correct estimation of dose was 61% based on the Ds-optimal design and 54% based on the D-optimal design.
Conclusions: The re-parameterized sigmoid Emax model provides a means of optimizing dose finding studies for correct dose identification, since the model has a parameter with direct relevance for dose selection. Ds-optimal design can be useful in improving design performance for dose finding by focusing on the parameter of interest (D*).
References:
[1] AV Groth, Alternative parameterisations of saturable (Emax) models allowing for nesting of non-saturable models PAGE 17 (2008) Abstr 1371 [www.page-meeting.org/?abstract=1371]
[2] PopED (http://poped.sf.net)
