Application of Optimal Design to Reduce the Sample Costs of a Dose-finding Study
Angelica L. Quartino (1), Joakim Nyberg (1), Marie Cullberg (2), Lena E. Friberg (1)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden; (2) Clinical Pharmacology & DMPK, AstraZeneca R&D, Mölndal, Sweden.
Objectives: A dose-finding study was planned with 3 treatment arms and 120 patients in each arm. To obtain an early readout of a pharmacodynamic effect, each patient was to receive one test dose as an immediate release tablet. For maintenance dosing an extended release tablet was to be given twice daily for 8 weeks. The planned PK sampling schedule consisted of 18 samples per patient (reference design) to assess the predictability in PK between the test dose and the repeated dosing.
The aim of the project was to optimize the PK sampling times and particularly to reduce the number of samples, without losing precision of the parameter estimates.
Methods: Based on data from Phase I-IIa studies a 3-compartment model had been developed that included transit compartments describing the absorption process. Non-linearities were identified and characterized for absorption and distribution. Day-to-day variability (IOV) was included in clearance.
Several scenarios were optimized based on the relatively complex PK model. The number of samples was fixed to 18, 16, 14, 12 or 10 samples per patient. For each sample size, the sampling times were optimized both with and without restrictions to the practical aspects of study conduct. The D-optimization was performed in PopED v. 2.10 (http://poped.sourceforge.net/).
Furthermore, simulations (n=40) and re-estimations were performed using NONMEM 6.2. The precision (RSE%) and mean absolute error (MAE) were calculated in order to evaluate the performance of the designs on parameter estimations.
Results: The optimized design for the 18 samples per patient scenario increased the efficiency with 70%, which may be translated into 150 fewer patients needed, compared to with the reference design. The proposed optimal design incorporating clinical restrictions had a similar efficiency as the reference design, but included only 14 samples per patient. Thereby the study cost could be reduced by ~100 000 Euro.
The simulations showed comparable RSE% on average per parameter as was predicted by the optimal design for both the reference and proposed design. The MAE was <20% on average per parameter and was similar between the two designs.
Conclusions: Optimal design theory allowed identification of a design for a complex population PK model that is more informative than the original design, despite fewer samples. Thereby, the study cost could be significantly reduced.
