Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Pharmaceutical Sciences

Research Advisor

Bernd Meibohm, Ph.D.


Marc Gastonguay, Ph.D. Duane D. Miller, Ph.D. John C. Panetta, Ph.D. Charles R. Yates, Pharm.D., Ph.D.


Bayesian Markov Chain Monte Carlo, maximum likelihood, monoclonal antibody, nonlinear mixed effects modeling, population pharmacokinetics, therapeutic protein


Over the past two decades, there has been an increase in the research and development and therapeutic application of monoclonal antibodies (mAbs). An application of pharmacokinetic (PK) and pharmacodynamic concepts that has likely contributed to the success of the pre-clinical and clinical drug development of therapeutic mAbs is population PK, which attempts to quantify the typical disposition characteristics and sources of PK variability (such as between-subject, within-subject, and inter-occasion variability) within study populations. Population PK also attempts to identify and quantify the impact of covariates on systemic drug exposure and assess their potential implications for clinical dosing. The general theme of my dissertation research was population PK modeling of therapeutic mAbs, which focused on the population PK modeling of cetuximab, and the evaluation of different estimation methods for population PK modeling of therapeutic mAbs with nonlinear PK characteristics.

Cetuximab is a therapeutic mAb directed against the epidermal growth factor receptor and is indicated in the treatment of squamous cell carcinoma of the head and neck (SCCHN). I performed a population PK analysis of cetuximab using nonlinear mixed effects modeling and the software NONMEM. A total of 912 cetuximab concentrations were available from 143 patients with recurrent and/or metastatic SCCHN enrolled in two phase I/II studies. The PK of cetuximab was best described by a two-compartment model with Michaelis-Menten type saturable elimination. Mean population estimates (between-subject variability, %CV) of the PK parameters were: Vmax 4.38 mg/hr (15.4%), Km 74 μg/ml, V1 2.83 L (18.6%), V2 2.43 L (56.4%), and Q 0.103 L/hr (97.2%). Ideal body weight and white blood cell count were identified as predictors of Vmax, and total body weight as a predictor of V1. My findings suggested that clinical dose adjustments beyond the approved body surface area-based dosing of cetuximab may be warranted in patients with extreme deviations of their actual body weight from ideal body weight. Agreement between simulated and measured concentrations for up to 43 weeks of therapy indicated that the final population PK model was able to adequately describe the nonlinear PK of cetuximab in patients with SCCHN at the currently approved dosage regimen, and that the cetuximab PK parameters remained constant during prolonged therapy.

Nonlinear eliminaton is a common characteristic of the PK of therapeutic mAbs. Accordingly, PK models with nonlinear elimination are commonly used in population PK analyses of mAbs, but difficulties detecting and characterizing this nonlinear PK have been reported in a number of studies. The challenge with detecting and characterizing the nonlinear elimination of therapeutic mAbs may not only be dependent on the clinical study design, but also on the estimation method used for the population PK analysis. However, little work has been done so far evaluating population estimation methods using PK models that are representative of the typical disposition characteristics of therapeutic mAbs. In order to address this question, I conducted a simulation study to compare the parameter estimation performance of the first-order (FO), first-order conditional estimation with interaction (FOCE-I), and Laplacian estimation with interaction (LAP-I) methods in NONMEM and a Bayesian Markov Chain Monte Carlo (MCMC) method in WinBUGS when applied to population PK modeling of therapeutic mAbs with nonlinear PK. The Bayesian MCMC method was evaluated with both vague and informative priors. Published findings of population PK analyses of therapeutic mAbs were used to define the informative priors. Simulations were performed with uncertainty included simultaneously on all parameters in the population PK model in order to evaluate the sensitivity of estimation performance to uncertainty in the simulation model parameters. The impact of study design on estimation performance was explored by evaluating the methods under a dose-ranging design (‘informative design’) and four different single dose level designs at different dose levels (‘uninformative designs’).

Under all study designs, the FO method generally produced larger bias and lower precision for all model parameters compared to the other estimation methods. Comparison between the methods in NONMEM and WinBUGS was limited to the informative and uninformative 600 mg dose level study designs due to prolonged run times with WinBUGS. Under the informative study design, bias and precision for all model parameters was less than ±25% and 52%, respectively, for FOCE-I, LAP-I, and Bayesian MCMC with both sets of priors. Under the uninformative 600 mg dose level design, the estimation performance of FOCE-I and LAP-I decreased as bias and precision for many of the model parameters, in particular those related to nonlinear elimination, significantly increased to ±40-173% and 53-173%, respectively, while Bayesian MCMC with informative priors provided a clear performance advantage producing results that were comparable to those under the informative design. Under both informative and uninformative 600 mg dose level designs, the estimation performance of FOCE-I, LAP-I, and Bayesian MCMC showed sensitivity to uncertainty in the simulation parameter values for one or more parameters. This was especially evident when informative Bayesian priors were used under the uninformative 600 mg dose level design, which was expected given the relative uninformativeness of the data under this particular design.

The findings from this work should be of value to pharmacometricians involved in population PK modeling of therapeutic mAbs. When sufficient concentration-time data are available to characterize the nonlinear elimination of the mAb, then FOCE-I, LAP-I, or Bayesian MCMC would likely be suitable for the populatin PK analysis. In situations where insufficient data are available to characterize the nonlinear elimination of the mAb, and relevant prior information is readily available, the use of a Bayesian MCMC method with informative priors should be considered.