Introduction A SAS/IML program for simulating pharmacokinetic data

Computer Methods and Programs in Biomedicine 2005 78, 39—60 A SASIML program for simulating pharmacokinetic data Estelle Russek-Cohen a , 1 ,Marilyn N. Martinez b , ∗ ,Anna B. Nevius b a Department of Animal and Avian Sciences, University of Maryland, College Park, MD 20742, USA b Center for Veterinary Medicine, US Food and Drug Administration, 7500 Standish Place, Rockville, MD 20855, USA Received 21 July 2004; received in revised form 6 October 2004; accepted 19 October 2004 KEYWORDS Monte Carlo simulation; SAS; Bioequivalence; Population kinetics; Veterinary pharmacokinetics Summary Data simulation can be an invaluable tool for optimizing the design of bioequivalence trials. It can be particularly useful when exploring alternative ap- proaches for assessing product comparability especially in the context of encounter- ing various complex experimental situations that can occur in veterinary medicine. With this in mind, we designed a novel SASIML program to generate pharmacoki- netic datasets that reflect the various kinetic, population, and study design charac- teristics that complicate the bioequivalence evaluation of animal health products. Developing this simulation program within SAS provides an opportunity to utilize the statistical capabilities of this software platform. Published by Elsevier Ireland Ltd.

1. Introduction

Traditionally, product bioequivalence determina- tions are rooted in the use of blood level metrics that reflect the rate and extent of drug absorp- tion. As stated in the Food and Drug Administration Center for Veterinary Medicine FDA CVM Guidance 35 [1] , these attributes are most frequently de- scribed by the observed peak drug concentration C max and the area under the concentration ver- sus time curve AUC. More recently, exposure con- Corresponding author. Tel.: +1 301 827 7577. E-mail addresses: erussekumd.edu E. Russek-Cohen, mmartin1cvm.fda.gov M.N. Martinez, aneviuscvm.fda.gov A.B. Nevius. 1 Tel.: +1 301 405 1403; fax: +1 301 405 7980. cepts have gained acceptance as the perspective from which to consider product comparability [2] . In this regard, the same bioavailability parameters AUC and C max are used but are considered mea- sures of the extent of systemic exposure and peak exposure, respectively. Whether dealing with human or veterinary phar- maceuticals, several challenging issues can com- plicate the evaluation of product bioequivalence. While product bioequivalence trials in human and veterinary medicine share many of the same sta- tistical and study design attributes, there are also many unique challenges confronted within the framework of veterinary medicine. For exam- ple, difficulties arise when assessing the relative bioavailability of products delivered in feed and water. In these situations, drug intake is essen- 0169-2607 — see front matter. Published by Elsevier Ireland Ltd. doi:10.1016j.cmpb.2004.10.007 40 E. Russek-Cohen et al. tially random, being dependent upon the feed- ing behavior of the animal. Gavage dosing, while useful to demonstrate the impact of the dosage form on product bioavailability, does not provide information on product palatability and other con- sumption variables. Investigators can control the time when the drug is introduced or removed, but cannot control when the drug is consumed. In these situations, the development of an opti- mal blood sampling strategy [3] can be enormously challenging. The presence of multiple absorption maxima can also complicate bioavailability comparisons. These complex profiles may be attributable to factors such as enterohepatic recirculation [4] , gastric drug retention and gastric motility cycles [5,6] and ion trapping [7] . Apparently random peaks and troughs are also associated with certain long-acting im- plants, such as those containing the growth pro- motants zeranol [8] and trenbolone [9] . These prod- ucts can release drug for periods exceeding 90 days, during which time huge fluctuations in serum drug concentrations can occur. Assessing the bioequiva- lence of these products is problematic because no singular absorption rate constant or peak exposure can be defined. Large variability in treatment effects is often observed when a parallel rather than a crossover study design is employed. In these cases, the population distribution characteristics of pharma- cokinetic parameter values can strongly influence the bioequivalence assessment. Within veterinary medicine, parallel study designs are needed when treatments are administered to growing animals where substantial physiological changes can oc- cur, thereby biasing period 1 versus 2 treatment comparisons, when the products are associated with very long residence times e.g., implants, and when an animal’s blood volume is limited e.g., fish and poultry. Furthermore, confounding the evalu- ation parallel design bioequivalence studies are oc- casions when subpopulations exhibit markedly dif- ferent drug pharmacokinetic characteristics. These subpopulations may be defined by differences in drug clearance or volume of distribution, both be- ing formulation-independent effects. For example, within veterinary species, differences in drug phar- macokinetics have been observed between breeds of dogs [10,11] , chickens [12,13] sheep [14] , and cattle [15] . Pharmacokinetics can also be affected by animal age, sex, and diet [15—17] . Such vari- ations are not a concern if a crossover study de- sign is employed and if there are no subject-by- formulation interactions. With regard to subject-by-formulation interac- tions, while this possibility and its therapeutic sig- nificance has been widely discussed [18] , there are few examples where they have actually been observed. One known example is the subject-by- treatment interaction seen when the bioavailabil- ity of certain earlier formulations of diazepam were compared to Valium [19] . In this case, human sub- jects with low gastric acidity had difficulty absorb- ing generic diazepam formulations, but such prob- lems were not observed in subjects with normal gastric acidity. Therefore, the generic and inno- vator diazepam formulations were bioequivalent in normal healthy volunteers but inequivalent in achlorhydric subjects. To explore the influence of the aforementioned kinds of concerns on bioequivalence determina- tions, Monte Carlo simulation methods are fre- quently employed e.g., [20—24] . The use of in silico techniques provides an inexpensive venue for evaluating the relationship between study at- tributes, the pharmacokinetics of a compound, and the sensitivity of a bioequivalence determination to changes in either study design or kinetic assump- tions. As with any in silico examination, conclu- sions are a function of the basic model assumptions such as the pharmacokinetic model, the shape of the distribution of the pharmacokinetic param- eters, the magnitude of the variability about each of these parameters, and the covariance structure of the model parameters. Computer-based simulation procedures can be particularly useful if employed as a tool for in vivo study protocol development. Within the framework of the simulation model, alternative study designs can be evaluated from the perspective of the abil- ity of these designs to accurately describe product bioavailability and the sensitivity of the study con- clusions to variations in model assumptions. To en- courage the use of these tools, the program should be both easy to use and based upon widely available software. In addition, several important attributes influence the utility of the program. For example, it needs to have the capability to rapidly gener- ate a large number of replications, study design attributes should be easily modified during subse- quent iterations, and the simulation output should summarize the probability of success and failure in an easy-to-follow manner. The program should also provide a summary of the distribution characteris- tics of the input and output variables so that the investigator can confirm that these frequency dis- tributions are consistent with the underlying model assumptions. Finally, the program should be com- patible with statistical software platforms used by both sponsors and regulatory agencies, and the out- put results should be easy to export to other soft- ware programs. A SASIML program for simulating pharmacokinetic data 41 Fig. 1 Basic pharmacokinetic models and corresponding parameters used in developing the SAS pharmacokinetics simulation program. In contrast with other simulation programs, we provide open source code so that users may modify the program to meet their particular need. Thus, a user may wish to compare existing and novel ap- proaches to the evaluation of bioequivalence while another user may be more focused on traditional comparisons of drug formulations. Based upon our evaluation of bioequivalence study datasets that have been submitted to FDACVM, we have identified a number of condi- tions that can complicate the design and analysis of these trials. Using SAS programming language, we have developed a tool for exploring the conse- quences of these various challenges. This project reflects a first stage in our efforts to explore the performance of alternative metrics for evaluating product bioequivalence.

2. Computational methods and theory