Robert J. Bauer (1), Andrew C. Hooker (2), France Mentré (3)
(1) ICON Clinical Research LLC, Gaithersburg, MD, (2) Uppsala University, Sweden, (3) Inserm and Université de Paris, France
Introduction: An evaluation and optimization tool for clinical trial design has been incorporated into NONMEM 7.5, to be released in August 2020. The purpose of this feature is to allow the evaluation and optimization of clinical trial designs, such as assessing the impact of selected or optimized time points, doses, or covariates, on the expected relative standard error of model parameters for a given number of subjects. A modeller may now create a model using familiar NMTRAN control stream code, and use it to perform design evaluations and optimizations. The same code may be subsequently used for clinical trial simulations as well as population analysis of real clinical data in NONMEM, saving time in code writing. The design characteristics and optimization options are specified using the record $DESIGN, which accepts various options that are also available in design evaluation tools such as POPED and PFIM. One may select the optimality criterion (D-, A-, DS-, R-, individual Bayesian), Fisher Information matrix (FIM) type (such as Full/block diagonal and FO/FOCE/Laplace approximation), or specify interesting/uninteresting parameters.
Objectives: Validate this trial design algorithm in NONMEM by comparing its results with those of other design software previously published in [1].
Methods: The two example models described in [1] were tested. The first model was that of a one compartment pharmacokinetic (PK) model of warfarin with first order absorption from a depot compartment, with 8 sampling times per subject. A proportional residual error model for data and exponential inter-subject random effects model for the PK parameters were assumed. The second model consisted of a multiple response PK/PD model with repeated dosing with the same sample times across responses. The model describes the effect of pegylated interferon given once weekly for 4 weeks on hepatitis C virus (HCV) kinetics, with 12 sampling times per subject for serum PK measurement of interferon and serum PD measurement of viral load. An additive residual error model for interferon concentration and log viral load, with exponential inter-subject random effects model for the PK and PD parameters, were assumed. To assess the design, the population Fisher Information Matrix (FIM) was computed based on first order (FO) linearization, with covariances between population parameters (thetas), and random effects scale parameters (omegas, sigmas) set to zero. The standard errors of the parameters were obtained from the diagonal elements of the inverse of this block diagonal FIM. The block diagonal FIM had been shown in [1] to be the most accurate in predicting the standard errors to a full clinical trial simulation (CTS), with population analysis on the full set of simulated data. We also performed design optimisation of the sampling times using Nelder Mead optimization in NONMEM $DESIGN, using D-optimality criteria.
Results: The warfarin design evaluation took 0.02 seconds to complete on a Latitude E6440 computer, with -7-4610M 3.0 Ghz processors, and the HCV design evaluation took 0.12 seconds to complete. The relative standard errors and FIM obtained from NONMEM’s $DESIGN record evaluation were nearly identical to those reported in [1] for both models. Additionally, the time samples were subjected to Nelder Mead optimization using NONMEM $DESIGN, using the –log(block diagonal FIM) as the objective function to be minimized. The warfarin design optimization took 0.7 seconds to complete, yielding similar optimized sample times to those obtained in [2]. The HCV design optimization took 88 seconds to complete. For both models, the number of distinct optimal time points were equal in number to the thetas to be estimated in the model which is expected when optimizing the block diagonal FIM [2].
Conclusion: The $DESIGN record in NONMEM provides a useful tool in rapid evaluation and optimization of clinical trial design that provide results identical to other optimal design software, providing seamless integration into subsequent data analysis using NONMEM’s many population parameter estimation methods. Also, design evaluations and optimization are much faster to compute than full CTS and estimation analyses, each of which may take several minutes to hours depending on the model type and amount of data, and many scenarios may be tested in a short period of time using the optimal design and evaluation tool.
References:
[1] Nyberg J, Bazzoli C, Ogungbenro K, Aliev A, Leonov S, Duffull S, Hooker AC, Mentre F. Methods and Software Tools for Design Evaluation in Population Pharmacokinetics-Pharmacodynamics Studies. British J. Clinical Pharmacology. 79:6-17, 2014.
[2] Stromberg EA. Nyberg J, Hooker AC. The effect of Fisher information matrix approximation methods in population optimal design calculations. J. Pharmacokinetics and Pharmacodynamics. 43(6):609-819, 2016.
Reference: PAGE () Abstr 9497 [www.page-meeting.org/?abstract=9497]
Poster: Oral: Methodology - New Tools