Averaged Model Based Decision Making for Dose Selection Studies
Yasunori Aoki (1), Bengt HamrÚn (2), Daniel R÷shammar (2), and Andrew C. Hooker (1)
(1) Uppsala University Pharmacometrics Research Group, Uppsala, Sweden, (2) AstraZeneca R&D, Quantitative Clinical Pharmacology, M÷lndal, Sweden
Objective: Approximately one third of the cost of a drug development program incurs in Phase III clinical studies, hence the correctness of the dose selection decisions at the end of Phase IIb studies largely influence the overall cost of drug development.
In this work, we introduce an averaged model based methodology for selecting the minimum effective dose from Phase IIb study data. We demonstrate this methodology through simulation studies of an asthma drug, with FEV1 as a biomarker, and confirm that it accurately quantifies the risk of choosing the wrong dose hence can guide more accurate dose selection decisions than the traditional pairwise dose selection criterion.
Methodology: The key idea of this methodology is to incorporate both the model parameter estimation uncertainty and the model structure uncertainty in dose selection. We average the parameter estimation uncertainty over the multiple candidate models weighted by the Akaike Criterion to compute the likelihood of each candidate dose being the correct minimum effective dose.
We have generated 5 000 sets of simulated datasets by adding artificial dose response relationships to the FEV1 placebo data. For each simulated dataset, the model structure of the artificial dose response relationship was chosen randomly from different models and model parameters were also chosen randomly.
For each data set, we have conducted the dose selection by following the traditional study protocol, and also by using the proposed methodology. After selecting the doses for all the simulated datasets using both criterions, we have counted the number of times each criterion has chosen the correct minimum effective dose.
In addition, to increase the numerical stability of the parameter uncertainty quantification calculation, we have implemented a preconditioning technique.
Results: We have confirmed through the simulation study that the proposed averaged model based dose selection criterion chooses the correct dose consistently more often than the traditional pairwise approach in various scenarios. On average, the probability of choosing the right dose was 40% using the traditional approach while it has increased to 60% by using the proposed methodology.
Conclusion: We have constructed the model averaged based dose selection methodology and demonstrated through simulation studies that the proposed methodology has a potential to increase the accuracy of the dose selections at the end of the Phase IIb clinical studies.
The link to the quicktime movie of the presentation can be found here.