Welcome to the Pharmacokinetics UK 2010 Meeting

Wednesday 3rd November – Friday 5th November

The Bristol Hotel Prince Street Bristol BS1 AQF

Programme and Abstract Book

Wednesday 3rd November

12:00 Arrival & lunch

Welcome and session 1: PKPD Bridging Strategies for Emerging Markets 14:00 Welcome: Steve Toon 14:05 Introduction to the First Session: Peter Milligan and Amin Rostami 14:10 Bernadette Hughes, Pfizer Points to consider in the Design, conduct and analysis of Confirmatory Global Trials 14:55 Kimitoshi Ikeda, Novartis Consideration on sample size of Japanese patients in multi-regional trials 15:40 Coffee break 16:00 Ni Khin, US Food and Drug Administration Regulatory and Scientific Issues Regarding the Reliance on Efficacy Data from Foreign Sites to Support New Drug Applications: An FDA Perspective 16:45 Open discussion 17:00 Session close 18:30 Poster session & free bar 20:00 Dinner

Thursday 4th November

Session 2: PKPD Modelling and Simulation; its Impact on the Efficiency of the Drug Development Process 09:00 Introduction to the Second Session: Steve Toon and Terry Shepard 09:05 William Gillespie, Metrum Institute Examples of decision-focused modeling and simulation: Limitations imposed by current methodology and potential remedies 09:40 Valérie Nicolas, Debiopharm PK-PD Modelling and Simulation from a Biotech Company Perspective 10:15 Coffee break 10:45 Richard Lalonde, Pfizer Impact of Modelling and Simulation in Drug Development 11:15 Stephen Senn, University of Glasgow Tumbling Dice - What Simulation Can and Cannot Do For You 11:45 Discussion/debate 12:00 Session close 12:00 Lunch

Session 3: Infectious Disease 13:30 Introduction to the Third Session: Alison Thomson and Leon Aarons 13:35 Kathryn Maitland, Imperial College London Challenges for Treating sepsis in African children 14:10 William Hope, University of Manchester Pharmacokinetics and pharmacodynamics of antifungal agents 14:45 Coffee break 15:15 Alasdair MacGowan, University of Bristol Pharmacodynamics and pharmacokinetics of antibacterials in preventing emergence of resistance 15:50 Laura Dickinson, University of Liverpool Pharmacokinetic Modelling of Antiretrovirals in HIV-Infected Patients 16:25 Gerry Davies, University of Liverpool PK-PD of sterilizing activity in tuberculosis 17:00 Session close

The Peter Coates Lecture, with an introduction by Steve Toon 18:00 Geoff Tucker, University of Sheffield / Simcyp The Virtual Patient: PBPK Applications in Drug Development 20:15 PKUK Banquet

Friday 5th November

Session 4: Open Session 09:25 Introduction to the Open Session: Mike Walker and Geoff Tucker 09:30 Julie Bertrand, INSERM Université Paris Diderot / University College of London Pharmacogenetics of CYP2B6 and population pharmacokinetics of nevirapine in HIV-infected Cambodian patients (ANRS12154 study) 09:55 Mike Smith, Pfizer A Model-Based Framework for Quantitative Decision Making 10:20 Helen Graham, University of Manchester Development of a novel method for updating the predicted partition coefficient values generated by an existing in silico prediction method 10:45 Coffee break 11:10 Camila de Almeida, AstraZeneca Method for the non-compartmental analysis of log normally distributed pharmacokinetic data: estimation and statistics of pharmacokinetic parameters 11:35 Joachim Grevel, BAST Inc The Process of Model-based Drug Development

12:00 Final conclusions, closing remarks and lunch

Presentation Abstracts

Session 1: PKPD Bridging Strategies for Emerging Markets

1. Points to consider in the Design, conduct and analysis of Confirmatory Global Trials

Christy Chuang-Stein¹and Bernadette Hughes²

1Statistical Research and Consulting Centre, Kalamazoo Michigan USA; 2Clinical Sciences, Primary Care Business Unit, Sandwich UK Pfizer Inc

Ideally a global Phase 3 programme should enable the rapid access of valued medicines to patients worldwide, i.e. in both developed and emerging markets in a relatively short time frame. The design and conduct of such a programme is challenging since it has to satisfy the differing demands of regulatory agencies, pricing and reimbursement boards. To meet internal budgetary pressure, a global programme needs to be executed with maximum efficiency and reasonable operating costs.

There are multiple factors to consider in planning a global programme. From the regulatory perspective, these include relevant regulatory guidances and differences across regions with regard to endpoints and choice of active comparator. From the clinical perspective, there may be regional differences in clinical and medical practice for the definition and diagnosis for the condition under study. If there are intrinsic or extrinsic ethnic differences, this needs to be addressed in the clinical programme from both a safety and efficacy perspective. Statistical considerations can be complex if data are to be pooled across regions for interim decisions (e.g. sample size re-estimation or futility in a region) since the consistency of the treatment effect and the placebo response across regions is unknown. This in turn links in with the enrolment strategy across the regions and how this will be handled within each clinical study. For certain indications, it may not be possible to conduct global studies. In the latter case, we need to carefully consider the contribution of each study to support the target indication in multiple regions within the context of a global programme.

The above are just a few of the key aspects that will be critical in underwriting a successful global Phase 3 programme. It is essential that consideration of all these factors begin early in the development lifecycle in order to minimise the risk of a failed global confirmatory study. In this presentation, we will also share some lessons learned from our experience with Phase 3 global clinical programmes.

2. Consideration on sample size of Japanese patients in multi-regional trials

Kimitoshi Ikeda1 and Frank Bretz2,3

1Novartis Pharma KK, Tokyo, Japan, 2Novartis Pharma AG, Basel, Switzerland, 3Hannover Medical School, Hannover, Germany

Background: In Japan, new drugs are often approved several years later than in other regions. The recently published guidance on „Basic Principles on Global Clinical Trials‟ addresses specifically this time lag [1]. The guidance introduces two methods of determining the size of the Japanese population in a global trial. Consequently, multi-regional trials have received increasing attention by pharmaceutical companies in Japan. A global trial has at least two main objectives. First, it is necessary to show a significant benefit in effect of a new drug over placebo in the entire study population. Second, one needs to demonstrate that the results for the Japanese subpopulation are consistent with those from the entire population. There are several possibilities to satisfy these general requirements.

Aim: We discuss critically the methods shown in the guidance document [1] on global trials that was recently issued in Japan to calculate the required sample size for a Japanese subpopulation in multi-regional clinical trials. We provide closed form expressions for the resulting probabilities (see also [2], [3]). Moreover, we propose alternative methods using a hypothesis test in the Japanese population.

Method: We investigate and compare the performance of the current approaches in [1] with the alternative approach. We investigated the operating characteristics for the different methods, such as different probabilities of success and the false-error rate, under different scenarios.

Results: The numerical results indicated that the proposed approach has a better performance compared to the current approaches. That is, it provided higher probabilities to achieve statistical significance in the entire population and consistency between Japanese and entire population, when there is an effect in both Japanese and the entire population. Moreover, the false-positive error rates of the proposed approach were comparable or even lower than those of the current approach when a study drug has no effect in either Japanese of entire population.

Reference 1. Ministry of Health, Labour and Welfare. Basic Principles on Global Clinical Trials, 2007. 2. Quan H, Zhao P-L, Zhang J, Roessner M, Aizawa K. Sample size considerations for Japanese patients in a multi-regional trial based on MHLW Guidance. Pharmaceutical Statistics 2010; 9(2): 100-112. 3. Uesaka H. Sample size allocation to regions in a multiregional trial. Journal of Biopharmaceutical Statistics 2009; 19:580-594.

3. Regulatory and Scientific Issues Regarding the Reliance on Efficacy Data from Foreign Sites to Support New Drug Applications: An FDA Perspective

Ni A. Khin, M.D.

Division of Psychiatry Products, Office of Drug Evaluation 1/OND, CDER, FDA

In recent years, there have been a number of changes in the design and conduct of controlled clinical trials intended to support New Drug Applications (NDAs) submitted to the US Food and Drug Administration (FDA), including a shift in where these trials are conducted. Globalization of clinical trials is rapidly becoming a reality. Although the US remains in the lead regarding the total number of clinical investigators involved in clinical trials in recent years, growth in both the numbers of clinical investigators and clinical trial sites is observed largely outside the US, in particular, Asia, Eastern Europe and Latin America. It is anticipated that, by the year 2012, 65% of FDA regulated trials will be conducted outside the US. While FDA accepts data generated from foreign sites for NDAs as long as they are from adequate and well-controlled trials that are conducted in compliance with the standards of Good Clinical Practice, there are no systematic data on what proportion of data is derived from non-US sites and their impact on marketing approval. Given the observation that there is an apparent reduction of treatment effect in recent multiregional clinical trials (MRCT) of antipsychotic and antidepressant medications, we have been exploring data generated from MRCT schizophrenia and depression trials to see if there are any difference in baseline disease characteristics, placebo and drug responses, and the treatment differences between US versus non-US sites, and also, among various regions. At this PKUK session, a brief regulatory overview on acceptability of foreign data in the US drug approval will be provided. The findings from our exploratory analyses of efficacy data from schizophrenia and depression trials will be presented. Points to consider in conduct of MRCT, including PK/PD modeling to optimize dosing, will be further discussed.

Session 2: PKPD Modelling and Simulation; its Impact on the Efficiency of the Drug Development Process

4. Examples of decision-focused modeling and simulation: Limitations imposed by current methodology and potential remedies

William R. Gillespie, James A. Rogers, Matthew M. Riggs and Marc R. Gastonguay

Metrum Institute, 2 Tunxis Road, Suite 112, Tariffville, CT 06081, USA

Examples of decision-focused modeling and simulation (M&S) are presented that illustrate current practice using both empirical and mechanistic models. This leads to a discussion of the potential for getting more value out of M&S via improvements in methodology and process.

One example involves the use of a largely empirical Bayesian hierarchical model for model- based meta-analysis of ADAS-cog time courses in Alzheimer's disease (AD) patients. The model was developed by simultaneous analysis of both summary data (sample means) and individual data. This approach seeks to leverage all available, relevant data. Simulations based on various generations of the model have been used to explore and optimize the designs of clinical trials. Illustrative examples of such simulations are briefly presented. One explores an ambitious adaptive Phase 2/3 program. Another focuses on optimization of PoC trial design and analysis.

In another example a multi-scale systems biology model for calcium homeostasis and bone remodeling was developed. The model was used to explore the impacts of chronic renal failure, parathyroid disorders and discontinuation of denosumab treatment on biomarkers and bone mineral density. It was also used to investigate potential causal mechanisms for long-term bone remodeling effects of estrogen in menopausal and postmenopausal women. The current model is deterministic. An effort is underway to implement the model in a Bayesian framework that incorporates uncertainty in parameter values and inter-individual variation is selected parameters. This will permit probabilistic inference and model updating via MCMC analysis of new data as it becomes available.

Though both examples contributed value, they also illustrate limitations that suggest directions for method development. The empirical approach used for the AD example permits predictive inferences about the effects of a particular drug only when ADAS-cog data for that drug is available. It does not permit predictions based on biomarker or preclinical information; as a result such predictions are not available during early development.

Mechanistic modeling of calcium homeostasis has much greater potential for making predictive inferences prior to the availability of clinical data. However the model does not account for variability or uncertainty, and the model complexity precludes parameter estimation via conventional regression methods. This also prevents rigorous statistical inference.

These limitations prompt a discussion of existing and proposed methodological development streams to remedy them. Most of those streams coalesce into a call for development of multi-scale mechanistic models for disease progression and drug effects relevant to various therapeutic areas. Effective development of such large scale biostatistical models requires a long-term multidisciplinary effort. Ideally this could be done as an industry/academic collaboration.

5. PK-PD Modelling and Simulation from a Biotech Company Perspective

Valérie Nicolas

Debiopharm, Lausanne, Switzerland

Being a small biotech company, resource constraints force us to prioritise between “Need to have” and “Nice to have” data. As opposed to Big pharmas with teams fully and routinely dedicated to modelling & simulation, we perform PK and PD modelling on a case-by-case basis to support our drug development. I would like to illustrate this approach as it was applied to an antiviral drug providing easy access to PD data and the early exploration of its pharmacokinetics, pharmacodynamics as well as the PK-PD relationship in several phase I and II studies, including population methodology and in silico simulations.

Although the extremely complex and non-linear PK of the drug prevented us from developing a complete and robust PK-PD model in short time, a pragmatic approach allowed us to demonstrate a dose-dependent additive antiviral effect of the drug when combined with standard of care treatment that allowed the clinical development team to chose a safe and effective dosing regimen and move the compound successfully into phase II.

In conclusion, limited but goal-oriented PK-PD modelling and simulation, provided sufficient understanding of the ADME and pharmacological properties of this drug, in order to successfully support decision making during drug development, maintain the momentum and meet the ambitious timelines of the development plan.

6. Impact of Modelling and Simulation in Drug Development

Richard L. Lalonde

Clinical Pharmacology, Pfizer Inc. New London, CT, USA

Most new compounds fail in drug development before ever reaching the market. Kola and Landis (Nat Rev Drug Dis 2004;3:711-5) reported that approximately 90% of compounds that enter clinical trials fail at one stage or another before approval by regulatory agencies. The high cost of new drugs (~$800,000,000/approved drug based on a Tufts University study in 2001) is driven by this high attrition rate. The most common cause of attrition in Phase 3 clinical trials is lack of efficacy relative to placebo and/or active comparators. This high failure rate for efficacy in late stages is surprising because efficacy should be known relatively well prior to Phase 3. This presentation will discuss how quantitative methods, including modelling and simulation, can help address this challenge.

A Nobel Prize was awarded in 2002 for research into why humans don‟t always make the best decisions when there is ambiguity or uncertainty (as there is in the early stages of development of a new drug). A proven solution to address this problem is the quantitative use of prior information. Model-based drug development (Clin Pharmacol Ther 2007; 82:21- 32) is an evolving paradigm that helps to integrate what is known about each new compound and provides a quantitative basis to make more informed decisions. Modelling and simulation have been used successfully in other industries (e.g. aeronautics, automobiles, etc) for decades and the pharmaceutical industry has been a relatively late adopter. However, the high costs of development and high failure rate are providing powerful incentives to implement more innovative methods.

This presentation will provide specific examples of the application of model-based drug development at Pfizer with a focus on the impact on trial design and development strategy. These examples will demonstrate how modelling and simulation have added value and provided insights that were not available by more traditional approaches. The examples will be taken from different therapeutic areas and stages of drug development to illustrate the applicability in many different settings.

7. Tumbling Dice - What Simulation Can and Cannot Do For You

Stephen Senn

School of Mathematics and Statistics, University of Glasgow

Computer simulation is undeniably a flexible and useful tool for modelling but it also has limitations. These limitations must be understood if one is not to be fooled into placing too much importance on the results of the 'thought experiment' that a simulation consists of.

First, it must be understood that simulation is just mathematics by other means. In fact nearly all simulation is (in essence) just a convenient and tractable way of doing high dimension integration. (In fact in some cases the integration is not so high and one would be better off doing the maths!) In consequence it should be understood that simulation cannot substitute for the wet stuff - experiments are needed so that we can truly learn about the world.

Second, it must be understood that reasoning from model to data is not the same as reasoning from data to model. The latter is what statistics is about but the former is what a simulation produces.

I discuss these points, illustrate them with some examples, point to some pitfalls of simulation and make some tentative recommendations.

Session 3: Infectious Disease

8. Challenges for treating sepsis in African children

Professor Kathryn Maitland

Imperial College London Based in KEMRI-Wellcome Trust Programme, Kilifi, Kenya

Admission to hospital with infection and severe illness in children in Africa is common. At our local hospital we have shown that the incidence for community-acquired infection is ~1000/100,000 in children <5 years. Streptococcus pneumoniae, nontyphoidal salmonella species, Haemophilus influenzae, and E. coli account for more than 70 percent of isolates. Of all in-hospital deaths, 26% were due to community-acquired bactaeremia. 71% of invasive bacterial disease deaths occur within 48 hours of admission. One of the major risk factors for poor outcome was severe malnutrition.

Invasive bacterial infection complicates up to 25% of fatalities in children with severe malnutrition with Gram-negative organisms constituting 50% of invasive bacterial pathogens. This high fatality is striking given that in vitro antibiotic susceptibility testing indicates that over 85% of organisms are fully susceptible to the antimicrobial regimen recommended by WHO (ampicillin and gentamicin). These deaths, on seemingly adequate treatment, call for an evaluation of the additional risk factors associated with in-hospital death including an in- depth assessment of the pharmacokinetics of the common antimicrobials in this patient group.

The evidence base for supportive care (resuscitation fluids, glucose correction and transfusion triggers) as well as antimicrobial treatments are weak and unsupported by relevant studies. As a result shock, severe dehydration, hypoglycaemia, anaemia and bacteraemia are important risk factors for fatal outcome. Specific pharmacokinetic studies are warranted in children with severe malnutrition since „one size may not fit all‟ owing to the reduced tissue mass and altered volume of distribution in children with oedema and wide spread dermatosis (hallmarks of kwashiorkor) and number of severe complications that may affect bioavailability and pharmacokinetics of important life saving treatments.

9. Pharmacokinetics and pharmacodynamics of antifungal agents

William Hope

School of Medicine, University of Manchester, Manchester, UK

Invasive fungal infections have a significant impact on human health. The mortality from invasive candidiasis and invasive aspergillosis is approximately 50%. Cryptococcal meningitis kills more people in sub-Saharan Africa than tuberculosis. There are relatively fewer antifungal agents compared with currently available antibacterial and antiviral compounds. There are an increasing number of well validated experimental models of invasive fungal infections that have facilitated a further understanding of PK-PD relationships for antifungal agents. Modern PK-PD approaches can and should be used to further optimise antifungal therapy for these medically important and serious infections.

10. pD and pK of Antibacterials in Prevention of Emergence of Resistance

Alasdair MacGowan

BCARE, University of Bristol and North Bristol NHS Trust, Southmead Hospital, Bristol

Pre clinical infection models have been widely used since the late 1980s to study anti- bacterial exposure relationships to antibacterial effects – most usually bactericidal activity. Such models have now been extensively validated by correlation with human pK-pD studies – most often on developmental antibacterials. pK-pD studies form an integral part of new drug development (i.e. telavancin, daptomycin, tigecycline, doripenem) and are also used to optimise dosing of older agents (i.e. Blactams, aminoglycosides, colistin). Pre clinical pK-pD allows for the development of appropriate doses, dose frequencies, modes of administration, potential target pathogens and also clinical breakpoints to define bacteria as drug susceptible. More recently, pre clinical models have been used to establish the relationship between drug exposures and the risk of emergence of resistance in target pathogens. Such modelling may also help address issues such as duration of therapy which is not studied in conventional pK-pD studies. The risk of emergence of resistance depends on drug exposure, dosing regimen, target pathogens, duration of exposure and bacterial load. At present, good quality clinical correlates are lacking but it is increasingly recognised that most existing dosing regimens carry a risk of emergence of resistance even when optimised for antibacterial effect. In the future, present studies are likely to be extended by front loading simulations, combination regimens, short course therapy and resistance gene transfer studies.

11. Pharmacokinetic Modelling of Antiretrovirals in HIV-Infected Patients

Laura Dickinson1,2

1 NIHR Biomedical Research Centre, Royal Liverpool & Broadgreen University Hospital Trust, Liverpool UK; 2 Department of Pharmacology, University of Liverpool, Liverpool, UK

At the end of 2008 33.4 million adults and children were estimated to be living with human immunodeficiency virus (HIV) worldwide; 2.7 million of which accounted for new infections. Almost 70% of the HIV-infected population resides in Sub-Saharan Africa which also has the highest prevalence of new infections and HIV-related deaths.

Introduction of highly active antiretroviral therapy (HAART) has had a huge impact on those infected resulting in HIV becoming a chronic manageable disease. However, of those thought to require antiretroviral therapy globally, approximately less than half are currently receiving treatment (42%). To date 25 antiretroviral drugs have been approved to treat HIV infection and can be subdivided into 5 specific classes: nucleoside (-tide) reverse transcriptase inhibitors [N(t)RTI], non-nucleoside reverse transcriptase inhibitors (NNRTI), protease inhibitors (PI), entry inhibitors and the novel integrase inhibitors. First-line treatment typically consists of an N(t)RTI backbone and either an NNRTI or a ritonavir-boosted PI, therefore targeting the virus lifecycle at more than one stage. Despite access to combination therapy and new drugs emerging long-term treatment success is hindered by numerous factors related to the virus (e.g. high mutation and replication rate), the drugs (e.g. poor pharmacokinetics and tolerability, development of resistance) and the individual patient (e.g. poor adherence, drug transporters, genetics) leading to therapeutic failure.

The majority of PIs are metabolised by CYP3A4 resulting in poor bioavailability and they were initially associated with high pill burdens and dosing frequencies. Co-administration of PIs with low dose ritonavir (100-200 mg twice daily), a potent CYP3A4 and p-glycoprotein inhibitor and a PI itself, has been successfully implemented as a strategy to overcome inadequate pharmacokinetics, so much so that all licensed PIs (with the exception of one) are recommended to be given in combination with ritonavir. This approach has led to reduced dosing frequencies (three times dailytwice dailyonce daily), decreased pill burdens and improved patient adherence to PI-based therapy. NNRTIs efavirenz and nevirapine are metabolised predominantly by the polymorphic CYP2B6 enzyme and pharmacokinetics have been shown to be influenced by CYP2B6 genotype. The potential for drug-drug interactions between PIs, NNRTIs and concurrent medications is high due to overlapping metabolic pathways and enzyme/transporter inhibition and/or induction properties. Cumulatively, the pharmacokinetics of most antiretrovirals are highly variable. Population pharmacokinetics is an ideal approach to determine sources of variability, whether demographic, genetic or a combination of both and can be applied to aid development of optimal dosing strategies in special populations such as neonates and pregnant women. Furthermore, it allows analysis of sparse data in addition to rich sampling. This is beneficial for investigations in HIV-infected populations in which data are lacking and/or blood sampling is difficult (e.g. elderly, TB or hepatitis co-infection, pregnancy) or in resource-limited settings, where the burden of HIV disease is highest and study costs must be kept to a minimum.

12. PK-PD of sterilizing activity in tuberculosis

G.R. Davies BM MRCP PhD DTM&H

Senior Lecturer

Department of Pharmacology, Institute of Translational Medicine, University of Liverpool, UK

Tuberculosis remains a leading global killer amongst infectious diseases. Current first-line therapy requires six months of a four-drug regimen for stable cure and is critically dependent on rifampicin, with those resistant to this drug requiring at least 24 months of treatment to achieve cure rates of only 60%. Recent activity in drug discovery has renewed the anti-tuberculosis pipeline with a number of agents now entering clinical development. However, both pre-clinical data from animal models and the early clinical stages continue to be plagued by uncertainty and controversy. The pharmacodynamic measures used in Phase II rely heavily on serial quantitative bacteriology. A key question relates to whether these early measures of efficacy accurately reflect the ability of a regimen to prevent treatment failure and relapse. This capacity is informally referred to as “sterilizing activity” but is poorly defined as a pharmacodynamic concept both in vitro and in vivo.

Traditional approaches to early bacteriological data have employed simple measures at fixed time points which do not adequately capture the full pharmacodynamic profile nor facilitate a learning approach to this critical phase of development. Applying a statistical modelling approach has many advantages and appears to greatly improve the efficiency of Phase II trials. Data from multiple studies reveal a biphasic profile of response with a dramatic slow-down in elimination of organisms after the first week. Traditionally, data from the first few days, so-called “early bactericidal activity” has been used to obtain proof-of- concept and intial dose selection but it is known that this approach is insensitive. The later phase of activity may better reflect pharmacodynamics against bacilli in a persistent antibiotic tolerant state and is empirically correlated with ultimate treatment outcome. Studies that have prioritised time-to-event or non-linear regression approaches appear to have greater statistical and predictive power to detect differences in this sterilizing activity.

The development of these approaches opens the way to faster and more comprehensive evaluation of combination regimens in early clinical trials. By utilising optimal design principles and methods based on the population Fisher's information matrix of the pharamcodynamic model it can be shown that such studies could halve the sample size required for current Phase II studies. With creative use of factorial and/or sequential methods, such novel designs should facilitate more reliable dose-finding and selection of optimal companion drugs, thereby ensuring that only the best new regimens enter Phase III trials for which global capacity and resources are currently limited.

Session 4: Open Session

13. Pharmacogenetics of CYP2B6 and population pharmacokinetics of nevirapine in HIV-infected Cambodian patients (ANRS12154 study)

J Bertrand1, M Chou2, C Verstuyft3, O Segeral4, L Borand5, V Ouk6, F Mentré1, Taburet AM7

1 INSERM UMR 738, and Paris Diderot University,2Rodolphe Mérieux Laboratory, Faculty of Pharmacy University of Health Sciences, Phnom Penh, Cambodia, 3Assistance Publique Hôpitaux de Paris, Hôpital Bicêtre, Molecular Genetic, Pharmacogenetic Hormonology department, and EA2706 Univ Paris Sud, France, 4Assistance Publique Hôpitaux de Paris, Hôpital Bicêtre, Internal Medicine Department, Paris, France, , 5Institut Pasteur in Cambodia Phnom Penh, Cambodia, 6Hospital Calmette Phnom Penh, Cambodia, 7Assistance Publique Hôpitaux de Paris, Hôpital Bicêtre, Clinical Pharmacy, France

Background: In resource-limited settings noncompetitive HIV-1 reverse transcriptase inhibitors are the WHO recommended backbone of first-line antiretroviral therapy. In Cambodia at the end of 2008, it is estimated that 69.5% of patients on antiretroviral drug regimens were on a nevirapine backbone regimen and worldwide, most patients living with AIDS and who need antiretroviral treatment are on a nevirapine-based antiretroviral regimen. However, data on factors influencing nevirapine pharmacokinetics and exposure in different populations are lacking.

Aims: To characterize nevirapine pharmacokinetics in a Cambodian population of HIV- infected patients and to identify environmental and genetic factors of variability focusing on the CYP2B6, CYP3A5 and ABCB1 (MDR1) genes.

Methods: One hundred and seventy Cambodian HIV-infected patients were included in an open-label, single-center, and multiple-dose pharmacokinetic study. Nevirapine trough concentrations were measured after 18 and 36 months of starting antiretroviral treatment. In addition, ten patients participated in an extensive pharmacokinetic substudy of five blood samples collected at predose and at 1 h, 2 h, 4 h, 8 h after the morning intake. Demographic, laboratory, medication and genetic polymorphism of CYP2B6, CYP3A5 and ABCB1 were also collected. Population pharmacokinetic modeling was performed using MONOLIX software version 2.4. The genetic component of the variability RGC was computed in order to assess to what extent a model parameter is likely to be under the influence of genetic polymorphisms, and p-values associated with the covariates remaining in the final model were obtained by permutations to correct for the Wald test type I error inflation that has been shown to occur in such design.

Results: Patients carrying homozygous loss of function alleles of CYP3A5 6986A>G, CYP2B6 516G>T, CYP2B6 1459C>T and ABCB1 3435C>T represent 42.4%, 9.2%, 0% and 18% of the population, respectively. The median nevirapine trough concentrations did not differ after 18 and 36 months of treatment (5705 (≤50 ─ 13871) ng/mL and 5709 (≤50 ─ 15422) ng/mL respectively). Interpatient and intrapatient variabilities were estimated only for the nevirapine apparent clearance and were 28% and 17%, respectively with a corresponding RGC of 69.1%. CYP2B6 516G>T and creatinine clearance were found to significantly affect nevirapine apparent clearance. Estimated nevirapine apparent clearance was 2.95 L/h, 2.62 L/h and 1.86 L/h for CYP2B6 516GG, 516GT and 516TT genotype, respectively. Impact of creatinine clearance was small.

14. A Model-Based Framework for Quantitative Decision Making

Mike K Smith1*, Jonathan French1, Kenneth G. Kowalski2, Matthew M. Hutmacher2, Wayne Ewy3

(1: Global Pharmacometrics, Pfizer Inc,2: Ann Arbor Pharmacometrics Group, Michigan, 3: Senior Consultant, Pharmacometrics, Pfizer Inc (Retired)) * Presenter

This presentation proposes a general framework for the formal integration of model-based predictions and their uncertainty in the planning of prospective trials and quantitative decision-making. Standard operating characteristics such as power, which are conditional on a chosen effect size, quantify the performance of the design. Optimizing trials based solely on power does not fully address the needs of drug development teams interested in understanding the performance of the compound as well as the performance of the proposed study design. This is because power does not quantify the likelihood of the compound achieving the assumed treatment effect. Metrics such as probability of a correct decision, probability of a Go decision, and probability of reaching a target value are proposed to evaluate the performance of the compound and trial. A conceptual clinical trial simulation (CTS) approach is outlined for calculating these trial performance metrics and to evaluate the „false positive‟ and „false negative‟ error rates for the proposed metrics. An example is presented to illustrate the CTS procedure and show how different choices of trial design, analytic technique and trial metric influence the probability of making correct decisions.

15. Development of a novel method for updating the predicted partition coefficient values generated by an existing in silico prediction method

Helen Graham1, James Yates2, Aleksandra Galetin1, Leon Aarons1

1School of Pharmacy and Pharmaceutical Sciences, University of Manchester, M13 9PL, 2AstraZeneca, Alderley Edge, SK10 4TG

Background: The use of PBPK modelling is becoming an increasingly important step in the drug development process as it aims to reduce the amount of in vivo and in vitro work needed during the early stages. Tissue-to-plasma partition coefficients (Kps) are a vital input parameter for these models, as they help to describe the distribution of a drug within the body, and can be used to predict volume of distribution. Many in silico methods exist in the literature for the prediction of these Kp values, with varying degrees of accuracy. Six of these methods have been compared in previous work, with the Rodgers & Rowland method [1] found to be the most accurate across all drug classes and in all tissues, and therefore this method has been chosen for further development.

Aim: To update the Kp predictions generated by the Rodgers & Rowland method by using in vivo data and prior knowledge of the prediction error inherent in the method, and use these updated predictions to produce predictions for other pharmacokinetic parameters, such as Vss, in both rat and human.

Methods: A covariance matrix was generated from prior knowledge of the prediction error of the Rodgers & Rowland Kp predictions when compared to experimental values for 32 compounds. A Monte Carlo simulation was performed to produce randomly generated sets of Kp predictions (using the Rodgers & Rowlands predictions as the mean), and these values were then used within a PBPK model to produce a set of predicted concentration- time profiles. Using a conditional log likelihood function, information from experimentally derived plasma conc-time profiles was combined with the Rodgers and Rowland predicted values to produce a set of updated Kp values. This work was all performed using the modelling tool AcslX®.

Results: The updated Kp values were shown to be an improvement upon the Kp predictions generated by the Rodgers & Rowland method when compared to experimental values, and they were shown to produce improved predictions for Vss and predicted iv profiles that more closely mimicked the experimental data.

Conclusion: A novel method has been described that can generate updated Kp values that are an update of the predictions generated by the Rodgers & Rowland method, using information about the prediction error and experimentally-derived iv profiles.

References: [1] Rodgers et al. (2005) J Pharm Sci 94(6):1259-1276; Rodgers and Rowland (2006) J Pharm Sci 95(6):1238- 1257

16. Method for the non-compartmental analysis of log normally distributed pharmacokinetic data: estimation and statistics of pharmacokinetic parameters

Camila de Almeida1, Philip Jarvis2, James Yates1

1 Modelling and Simulations Group, DMPK CIRA, AstraZeneca, Alderley Park, SK10 4TG, 2 Discovery Statistics, DECS, AstraZeneca, Alderley Park, SK10 4TG

Background: Non-compartmental analysis (NCA) is used to characterise drug absorption, distribution, metabolism and excretion in the body. Pharmacokinetic (PK) parameters such as clearance (CL) and volume of distribution at steady state (Vss) depend nonlinearly on the area under the curve (AUC), the area under the first moment curve (AUMC), and are estimated from NCA analysis following the course of drug kinetics in blood. Recently, Wolfsegger & Jaki (2009) [1] proposed two methods, bootstrap and asymptotic, for the estimation of the confidence interval (CI) of such parameters, based on AUC and AUMC from time 0 to infinity obtained from a concentration-time curve. The terminal phase area is extrapolated based on linear regression on the log-scale on the last observed concentrations. Bootstrap results in nominal coverage for a minimum of 3 sample repeats at each time point. The asymptotic method however, underestimates the CI coverage, resulting in nominal coverage only when the number of sample repeats exceeds 10. Aim: Given the practical experimental issues restricting the sample sizes per time point, a novel method is proposed and analysed specifically for log-normally (LN) distributed data. A comparison of the novel method with the ones in the literature was performed, highlighting the positive and negative points of each method. Several factors that have an impact on the coverage of the estimated CI were investigated: the experimental design, the number of sample repeats per time point, the portion of AUC and AUMC extrapolated, the number of points used to estimate the terminal phase. Methods: AUC and AUMC are, according to the trapezoidal rule, a sum of correlated LN variables. These areas can be quantified by approximating this sum to a single LN variable by an extension of Fenton-Wilkinson (FW) approximation [2]. FW provides also an estimation of the variances used to determine the 95% CI for AUC, AUC, CL and Vss. Coverages of estimated CIs were assessed using in silico LN data sets generated with same experimental design as the detailed in Wolfsegger & Jaki (2009) [1] for validation of the bootstrap and asymptotic methods. Results: Point estimates of PK parameters from this novel method were closely related to those obtained by WinNonLin. The novel method exhibited improved CI coverages compared to the asymptotic method, with a slight underestimation of the CI. Changes in the experimental design could improve the coverage of the CI of our method to values close to the expected nominal coverage. Conclusion: The estimation of CI for PK parameters obtained from the novel method showed an improvement in relation to the asymptotic method presented in the literature, for small data sets. The use of bootstrap to small data sets is still questionable, as analysis of real data reveal CI that enclose negative values for PK parameters. Our method also provides a straightforward method to assess parameters involving nonlinear functions of AUC and AUMC values, being important for example a better assessment of bioequivalence of drugs.

References: [1] M. J. Wolfsegger and T. Jaki. Non-compartmental estimation of pharmacokinetic parameters in serial sampling designs. Journal of pharmacokinetics and pharmaco-dynamics, 36(5):479-494, 2009. [2] A. A. Abu-Dayya and N. C. Beaulieu. Outage probabilities in the presence of correlated lognormal interferers. IEEE Transactions on Vehicular Technology, 43(1):164-173, August 1994.

17. The Process of Model-based Drug Development

Joachim Grevel

AstraZeneca R&D Charnwood, Clin. Pharmacology and DMPK, Loughborough, LE11 5RH

Background: In 2004 the U.S. Food and Drug Administration started the Critical Path Initiative1 to boost the success rate and development speed of the pharmaceutical industry. Based on Sheiner‟s Learn versus Confirm Paradigm2, the concept of Model-based Drug Development (MBDD) was defined as the development of pharmaco-statistical models of drug efficacy and safety, based on preclinical and clinical data, to improve drug development knowledge management and decision making. Pfizer Global Research and Development summarized their experience with MBDD in 20073.

Aim: This presentation tries to extract from both the theoretical background and the practical experience an applicable process to base the clinical development of new medications on model-guided decisions.

Methods: Notation: Δ, true treatment effect accessible by simulation with a very large number of subjects; T, measured treatment effect accessible by simulation with an optimal number of subjects; TV, target value; P(go) = P(T ≥ TV), probability of success; PTV = P(Δ ≥ TV), probability of achieving TV. The models used in the simulation of trials should be as mechanistic as necessary. They are continuously updated as new data is available. They link dose rates, functional biomarkers, and outcomes that relate to TV. As such they provide a chain of causal evidence4. They should incorporate models for compliance erosion, dropout rates, and disease progression. The subjects for the simulated trials could be sampled from public databases of the specific disease5. Each simulated trial is analysed (by modelling or standard stats test, dependent upon type of TV) to determine whether T|Δ ≥ TV. A sufficient number of trials (between 100 and 1000) determines P(go) and PTV.

Results: Scenarios are presented how to change the drug development process in order to base investment decisions on predictions of P(go) and PTV. The validity of these predictions depends on: a realistic assessment of all model assumptions, predictive performance characteristics of the models, intelligent choice of TV (absolute or relative to placebo or competitor, static or time-variant, simple or composite or utility or value). MBDD incorporates quantitative stopping rules based on P(go) and PTV and reduces the role of Type I and II error calculations. MBDD can provide a realistic risk management across the portfolio by allowing a different distribution of PTV in the different phases of development.

Conclusion: Model-based drug discovery has been driven by the advances of biotechnology. But new projects entering the clinic were not advance by model-based development. This imbalance led to poor decisions in many large pharmaceutical companies. MBDD is the necessary correction. References: 1. http://www.fda.gov/ScienceResearch/SpecialTopics/CriticalPathInitiative/default.htm 2. Sheiner LB, "Learning VS Confirming in Clinical Drug Development," Clin. Pharmacol. Ther., 1997, 61:275-291. 3. Lalonde RL, Kowalski KG, Hutmacher MM, et al., ”Model-based Drug Development” Clin. Pharmacol. Ther., 2007, 82:21-32. 4. Food and Drug Administration Modernization Act of 1997, Pub. L. No. 105-115, 111 Stat. 2295 (1997): ”...if the secretary determines, based on relevant science, that data from one adequate and well- controlled clinical investigation and confirmatory evidence (obtained prior to or after such investigation) are sufficient to establish effectiveness, the secretary may consider such data and evidence to constitute substantial evidence...” 5. Innovative Medicines Initiative (http://imi.europa.eu/calls-02_en.html) Second Call 2009,Topic 7: Proposal by the Drug Disease Model Resources (DDMoRe) Consortium.

Poster Abstracts

1. WBPBPK models of beta-blockers in rats

Cheung, S. Y. A.1, Rodger, T.2, Rowland, M.3, Aarons, L.3

1 AstraZenceca, Macclesfield, SK10 4TG, UK, 2 ICON Development Solutions, Manchester, M15 6SH, UK, 3 University of Manchester, M13 9PT, UK

WBPBPK models have been widely used in drug development to describe the kinetics of therapeutic agents in animals and humans. The advantages of such modelling techniques are the incorporation of vast amounts of prior physiological information such as blood flow and organ volume to ensure the model can represent the real physical body. Unlike system biology models, which require in depth details of all the pathways, the WBPBPK models sit between system biology models and the traditional 1, 2 and 3-compartmental models to allow unknown quantities that enable inter-species prediction of PK behavior to be estimated with prior physical information.

A series of closed loop WBPBPK models were developed for enantiomers of a series of seven racemic beta-blockers, namely acebutolol, betaxolol, bisoprolol, metoprolol, oxprenolol, pindolol and propranolol, together with S-timolol based on the open loop forcing function model. The previous results indicated that when the compounds are fitted individually, sometimes, the racemic pairs had different distribution models for the testes and gut compartments. It had been thought that the influence of the differences in the structure was due to experimental procedure such that the gut tissues were treated as a single compartment in the experiment which contains parts that are a mixture of perfused and permeable tissues.

To overcome this, in recent work, it was found that when the racemic pairs are estimated simultaneously, a common structure can be obtained. Parameter estimates of the tissue-to- blood partition coefficient from steady state experiments, individually and racemic pair are mostly in agreement. The objective function (OF) obtained from the paired estimation is the sum OF obtained for the pairs estimated individually. This indicates the population behaviors for each of the racemic pair are the same and have the same structure. However, when the data are pooled together to gain common structure, there was no reduction to the value of OF, as both sets of data are required to defined the common structure to filter out any experimental bias to the structure.

Conclusion: Simultaneous fitting of racemate kinetics allows stabilisation of parameter estimates and robust identification of PBPK model structure.

2. Characterisation of Xenograft Response to Taxotere by Nonlinear Mixed Effects Modelling

James Yates1

1AstraZeneca, Alderley Edge, SK10 4TG

Background: The human derived Xenograft tumour model is one of the main pre-clinical disease models used in oncology drug research and development. It is commonly used to investigate combination of novel compounds with standards of care. One such standard of care is Taxotere. In the xenograftted PC3 celline a steep dose response has been observed. This is problematic because if an insufficient dose is administered as monotherapy it means that there is no positive control with which to compare combination therapy. If the dose is too effective, there is insufficient dynamic range in which to observe a combination effect

Aim: To develop a “Population” model of the Pharmacokinetics and Tumour growth inhibition of Taxotere that can be used to simulate the response to untested doses.

Methods: All analysis was performed in NONMEM v6 and R v2.9. Pharmacokinetic data from two mice studies were gathered. Xenograffted tumour growth data was also available from two separate studies where doses of 15 and 20mg/kg were tested. A two compartment model was fitted to the pharmacokinetic data. Mean parameter values and inter-animal variances on the PK parameters resulted. There were no pharmacokinetic data associated with the Tumour growth studies, therefore PK was simulated for these studies using the mean parameter values. A second mixed effects model was developed using the Simeoni model to describe the change in time of tumour volume. The parameter estimates from this model resulted in mean values and inter-animal variances for control growth as well the effect of a given blood concentration of Taxotere. A simulation study was then carried out to simulate a dose response curve for a range of untested doses. One hundred studies with the same design as those used for model building were simulated. Each simulated study contained a number of different dose groups. The simulations were then summarised in terms of mean and standard deviation of tumour growth inhibition for each dose across the studies as well as the proportion of studies for which a particular dose had a significant effect against control. As a form of model validation, data resulting from a previously untested dose 7.5mg/kg) was compared to a monte carlo simulation of the model at that dose. There were some disparities noted. However the response was in the range suggested by the simulations

Results: The simulated dose response suggested that it would be difficult to recommend a dose that would give a moderate response (~50% growth inhibition) whilst being significant against control. However a dose of 5mg/kg gave a mean response of 51.4 percent growth inhibition with an 80% chance of being significant against control.

Conclusion: Using a mixed effects model approach the xenograft response to a chemotherapeutic was characterised. It is clear from the model validation that inter-study variability should be characterised as well.

References: Simeoni et al. 2004. Predictive Pharmacokinetic-Pharmacodynamic Modeling of Tumour Growth Kinetics in Xenograft Models after Administration of Anticancer Agents.

3. A PK/PD approach to predict the effects of Capthepsin K inhibitors in human serum C-terminal Telopeptide of Type I Collagen (s-CTx) levels.

Pablo Morentin Gutierrez, Jonathan Bowyer, Ken Page

AstraZeneca, Alderley Park, Mereside, 3G68a, Macclesfield, SK10 4TG, UK

Abstract

Cathepsin K (Cat-K, a lysosomal collagenase-like enzyme) is believed to play a major role in the pathological breakdown of bone and cartilage in the arthritic joint. A primary activity of Cat-K is to drive bone resorption and therefore serum bone degradation biomarkers such as the type I collagen degradation product, s-CTx, can provide a proof of mechanism marker in volunteers and osteoarthritis patients.

In drug discovery and development, projects depend on PK/PD models capable of predicting the effect of a candidate drugs in man to maximize chances of success. A PK/PD approach, for the development of new Cat-K inhibitors, using measurement of s-CTx in the dog to predict the effect expected in man is presented here. Unlike many PK/PD approaches for new discovery projects, the predictive power (in man) of this approach could be investigated thanks to the published human data available for Balicatib (Cat-K inhibitor).

Following administration of Balicatib to dogs, a clear dose related suppression of the s-CTx levels was observed. An indirect PK/PD model with variable turnover rate was postulated to describe the circadian variation in the baseline levels of s-CTx. The effects of the Cat-K inhibitor were introduced by means of a drug induced inhibitory function of the turnover rate of s-CTx.

Once this PK/PD model for Balicatib was generated for the dog, an initial prediction of the human effects of Balicatib on s-CTx in man was carried out using the dog system parameters (i.e. compound independent) and extrapolated (from dog to man) compound dependant parameters. While the time course of the effect was very well predicted, the accuracy of this initial prediction, including the simulations for placebo treated subjects was relatively poor, clearly indicating that the system parameters of the PK/PD model for human and dog were different.

A modified PK/PD model that included system parameters generated from human placebo data plus compound dependant parameters extrapolated from the dog data was then used and provided a very accurate prediction of the effects of Balicatib in s-CTx in man. Therefore, we now have a generic approach to predict the expected effects in s-CTx in man for new Cat-K inhibitors using the dog model and the PK/PD approach described here.

4. Use of a Public Domain Drug-Disease Modeling Framework and Phase 2 Data to Simulate Phase 3 Studies of Motesanib (125 mg Once Daily) Plus Carboplatin/Paclitaxel (CPM) or Bevacizumab Plus Carboplatin/Paclitaxel (CPB) Versus Carboplatin/Paclitaxel (CP) Alone in First-Line Treatment of Non–Small-Cell Lung Cancer (NSCLC)

Laurent Claret,1 Jian-Feng Lu,2 René Bruno,1 Robert Sikorski,2 Yong-jiang Hei,2 Yu-Nien Sun2

1Pharsight – a Certara™ Company, Marseille, France; 2Amgen Inc., Thousand Oaks, CA, USA

Background: Motesanib is an orally administered small-molecule antagonist of VEGF receptors (1, 2, and 3), PDGF, and c-kit receptors that has shown antitumor activity in a randomized phase 2 study of first-line CPM or CPB in NSCLC (J Thorac Oncol. 4[9 suppl 1]:S354-5, 2009).. Recently, the US Food and Drug Administration (FDA) has developed a drug-disease model based on data from several pivotal studies in NSCLC (Clin Pharmacol Ther. 86:167-174, 2009) that can be used to simulate overall survival for other therapeutic molecules.

Aim: We utilized the drug-disease model developed by the FDA to simulate overall survival for a phase 3 study of CPM or CPB vs CP in NSCLC using data from the motesanib NSCLC phase 2 study.

Methods: In the motesanib NSCLC phase 2 study patients (randomized 1:1:1) received up to six 3-week cycles of CP plus one of the following treatments: motesanib 125 mg orally once daily (QD; Arm A); motesanib 75 mg twice daily 5 days on/2 days off (Arm B); or bevacizumab 15 mg/kg intravenously once every 3 weeks (Arm C). Two models were included in the modeling framework: a longitudinal tumor size model and a survival model that relates change in tumor size at week 8 (first study visit) and patient characteristics (tumor size at baseline and ECOG performance status) to survival time. Using the two models we simulated multiple replicates (n=5000) of studies with ≥700 randomized patients of CPM or CPB vs CP that had a design similar to an ongoing phase 3 study of CPM vs CP. The modeling was applied to patients with characteristics similar to those of patients in the phase 2 study (n=177).

Results: For a phase 3 study with three treatment arms the model predicted a median survival time (95% CI) of 11.0 (9.8–12.3) months, 10.8 (9.7–12.1) months, and 9.3 (7.6– 11.1) months for the CPM, CPB, and CP arms, respectively. Relative to CP hazard ratios for CPM and CPB were similar: 0.87 (95% CI, 0.71–1.1) and 0.89 (95% CI, 0.73–1.1), respectively. The result for the CPB arm was consistent with the hazard ratio of 0.88 (95% CI, 0.78–1.01) which was determined in similar simulations performed by the FDA using data from the E4599 study (N Engl J Med. 355:2542-2550, 2006).

Conclusions: Using an NSCLC disease model, simulations predicted that first-line treatment of NSCLC with CPM or CPB may have similar survival benefits over CP alone. Both treatments may result in an approximately 1.7 month increase in median overall survival. End-of-phase-2 decisions and the design of phase 3 studies may be facilitated using the modeling framework as a tool.

5. Within Subject Variability in Aminoglycoside Pharmacokinetics in Patients with Cystic Fibrosis

Alghanem S1, Paterson I2, Thomson AH1

1Strathclyde Institute of Pharmacy and Biomedical Sciences, University of Strathclyde Glasgow, 2Cystic Fibrosis Unit, NHS Greater Glasgow and Clyde, Pharmacy and Prescribing Support Unit, NHS Greater Glasgow and Clyde

BACKGROUND: Patients with cystic fibrosis are now living longer, with a median predicted survival of 38.8 years in 2008 compared to 35.2 years in 2007. The majority of pulmonary infections in patients with cystic fibrosis involve P. aeruginosa, which is usually treated with a combination of ceftazidime and tobramycin. Due to the chronic nature of this condition, patients often receive multiple courses of aminoglycosides. A clinical pharmacokinetic monitoring service was introduced to the Glasgow Adult Cystic Fibrosis Unit in the early 1990s; the resulting database offers a unique opportunity to investigate inter-occasion variability in aminoglycoside handling over a prolonged period of time.

AIMS: To determine the nature and extent of inter-occasion variability in aminoglycoside pharmacokinetics in patients with cystic fibrosis and to investigate the influence of covariates on aminoglycoside pharmacokinetic parameters.

METHODS: The study involved a retrospective analysis of the aminoglycoside database for patients with cystic fibrosis and covered the period 1993 to 2009. The data were analysed using NONMEM version 6. One and two compartment models were compared and the influence of covariates, including a range of methods for estimating renal function, was examined. Within-subject variability (WSV) was investigated with the assumption that one course of therapy represented one occasion.

RESULTS: A total of 2238 tobramycin concentrations were available from 166 patients aged 14 to 66 years, median 23 years. Concentrations ranged from 0.1 to 18 mg/L, with a median value of 2.8 mg/L. The number of occasions available ranged from 1 to 28 with a median of 5 occasions. The two compartment model provided a better fit of the data (Reduction in the objective function value (∆OFV = -166.1). The inclusion of WSV on clearance (CL) produced a further improvement in fit (∆OFV= -234.2). There was no further improvement when WSV was added to the volume of distribution of the central compartment (V1) model. The final covariate model for CL included creatinine clearance estimated by the Cockcroft and Gault equation(1), with the minimum serum creatinine concentration fixed to 60µmol/L. Between- subject variability (BSV) in CL was 18.5% and WSV was 11%. V1 was best described using height, which reduced BSV from 16% to 12%. BSV could not be estimated for the peripheral volume (V2) or inter-compartment clearance (Q).

CONCLUSION: Since unexplained within subject variability in the handling of aminoglycoside antibiotics in patients with cystic fibrosis is low, patients can be started on a previous individualised dosage regimen if a new course of therapy is required.

REFERENCE 1. Cockcroft D, Gault M. Prediction of Creatinine Clearance from Serum Creatinine. Nephron. 1976;16(1):31-41.

6. The SAEM algorithm in MONOLIX for Non-Linear Mixed Effects Models with Stochastic Differential Equations

Maud Delattre (1), Pierre Del Moral (2), Marc Lavielle (1,3)

(1) Depart. of Mathematics, University Paris-Sud, (2) INRIA Bordeaux, (3) INRIA Saclay

Objectives: The use of stochastic differential equations in non linear mixed effects models enables the decomposition of the intra-patient variability into some residual errors and some dynamical system variability. Several authors already addressed this problem and proposed an approximation method based on the First Order Conditional Estimation (FOCE) method used in non-linear mixed effects models, with the Kalman Filter used for SDEs (see [1], [2], [3]). In [4], the authors propose a stochastic EM algorithm where the diffusion process of the SDE is considered as a component of the non observed data and is simulated at each iteration. This procedure has some appealing theoretical properties but the computational effort is prohibitive for practical applications. The objective of this contribution is to present a new maximum likelihood estimation procedure which is computationally tractable in practical situations and which avoids the linearization of the model.

Methods: We propose a new algorithm based on the Stochastic Approximation EM (SAEM) method with the Kalman Filter for linear SDE systems. We show that a linear SDE system is not relevant when the components of the stochastic system are known to be positive, which is usually the case in a biological perspective. Assuming that the diffusion process randomly perturbs the coefficients of the associated ODE system is more realistic but the SDE system is not linear any more. The extended Kalman filter for non linear SDE systems can be used in such situations. This methodology was implemented in a working version of MONOLIX and tested on several simulated PK examples.

Results: We show with these simulated examples that the proposed method does not reduce to consider some correlated residual errors. Indeed, we show the ability of the method to properly decompose the intra-patient variability into several components.

Conclusions: A new maximum likelihood estimation method for non linear mixed effects models governed by a system of stochastic differential equations is now implemented in a working version of MONOLIX. For nonlinear SDE systems, we aim to develop in a next future a new SAEM based method using a particle filter instead of the extended Kalman filter. This method is expected to exhibit better theoretical and practical properties.

References: [1] S. Mortensen, S. Klim, B. Dammann, N. Kristensen, H. Madsen, R. Overgaard "A Matlab framework for estimation of NLME models using stochastic differential equations", Journal of Pharmacokinetics and Pharmacodynamics vol:34, pages: 623-642, 2007. [2] R. Overgaard, E. Jonsson, C. Tornøe, H. Madsen, "Non-Linear Mixed Effects Models with Stochastic Differential Equations. Implementation of an Estimation Algorithm", PAGE 2004. [3] C. W. Tornøe, H. Agersø, R. V. Overgaard, H. A. Nielsen, H. Madsen, E. N. Jonsson, "Stochastic differential equations in NONMEM", PAGE 2004. [4] Donnet S, Samson A, Parametric inference for mixed models defined by stochastic differential equations, ESAIM P&S, 12:196-218, 2008.

7. Could Plasma Kinetics of L-Theanine in Humans Serve as Basis for Kinetics in the Brain?

P.C. van der Pijl1,1, L. Chen2, T.P.J. Mulder1

1Unilever Research and Development Vlaardingen, Olivier van Noortlaan 120, PO Box 114, 3130 AC, Vlaardingen, The Netherlands; 2Unilever Research Shanghai, 99 Tianzhou Road, Shanghai 200233, China

Tea has been consumed for > 2000 years and has traditionally been associated with mental clarity, and especially the tea ingredients caffeine and L-theanine have been linked to this effect. The pharmacokinetics of caffeine has been described extensively, whereas that of L- theanine is largely unknown. The aim of this study is to describe the pharmacokinetics of L- theanine in humans. This knowledge could support a plausible mechanism of action. To study the PK of L-theanine in humans, volunteers received 25, 50, and 100 mg of L-theanine in tea, in L-theanine-enriched tea, and as L-theanine in an aqueous solution. Plasma L-theanine concentration were determined after which data were fitted with a 1- compartment model with 1st order input and output. For all interventions, disposition was rapid with a t½,a and t½,b of ≈ 15 and 65 min respectively (tlag ≈ 10 min). About 50 min after tea consumption dose-proportional Cmax values between 1.0 - 4.4 mg/L were achieved. For the dose range studied, body weight-corrected Cmax and AUC were linear dose-proportional. Finally, Vhyp/Fabs was 18  3 L. It is assumed that L-theanine is taken up via a common Na+-coupled co-transporter in the intestinal brush-border membrane. L-theanine is metabolised in the kidney to glutamate and ethyl amine by glutaminase. Intact L-theanine in systemic circulation enters the brain via the L-system without metabolic change in the latter compartment. The fate of L- theanine at the site of action remains not fully understood, however. The PKs of L-theanine in systemic circulation are simple (1-compartment model, linear), but the site of action is likely to be the brain. Therefore multi-compartment-based PK simulation techniques should be used to assess the concentration of L-theanine in the brain. This, on its turn, should allow development of hypotheses for the mechanism of action, ultimately supporting design of human trials looking into effects of L-theanine in the brain.

Disclosure; the study was funded by Unilever.

1 Corresponding author: P.C. van der Pijl, Olivier van Noortlaan 120, 3133 AT, The Netherlands. Telephone: + 31104605454, Fax: + 31104605993, E-mail: [email protected] 8. Utilisation of a physiologically-based pharmacokinetic (PBPK) model to explore the effects of a CYP3A4 inhibitor in patients with varying degrees of renal impairment

Fran Stringer1, Nolan Wood1, Cheikh Diack2, Bart Ploeger2

1Takeda Global Research & Development Centre, Ltd., London, UK; 2LAP&P Consultants BV, Leiden, Netherlands.

Background: Clinical studies to evaluate the effects of multiple factors on systemic drug exposure, such as drug-drug interactions (DDIs) in patients with renal impairment (RI), are difficult to perform. However, such interactions may result in clinically relevant changes in systemic exposure necessitating dosage adjustments or cautionary statements in the drug product label. Predictive tools such as physiologically based pharmacokinetic (PBPK) modelling are being increasingly used to estimate the magnitude of these complex interactions. A PBPK model, using in vitro data and in silico simulations, was constructed to predict the systemic exposure of a CYP3A4 substrate in the presence of the potent CYP3A4 inhibitor ketoconazole (KTZ) in patients with mild, moderate and severe renal impairment Aim: To estimate the effects of a CYP3A4 inhibitor on systemic exposure of a CYP3A4 substrate in patients with varying degrees of renal impairment to aid inclusion/exclusion criteria recommendations for phase 3 and supratherapeutic dose selection for proposed future thorough QT studies. Methods: Structural model development and simulations were performed using Berkley Madonna. The applied statistical framework for the PBPK model was based on a previously published model (1). Physiological and anatomical parameters including body weight, tissue blood flows and tissue volumes were derived from published data, with model parameters for the CYP3A4 substrate including volume, clearance (systemic, intrinsic, hepatic and renal), absorption rate, plasma protein binding and fraction absorbed derived using in-house data. Model parameters for ketoconazole were calculated using SimCYP version 8.2. The main assumptions applied during modelling were: rapid equilibration between plasma and all tissue components; KTZ did not influence transporters of the CYP3A4 substrate (e.g. P-gp); renal processes (filtration, secretion and reabsorption) decreased proportionally with renal function change. Plasma concentrations of the CYP3A4 substrate in the absence and presence of KTZ were estimated using the PBPK model and compared to observed values from a clinical DDI study prior to simulating the effects of KTZ in patients with varying degrees of renal impairment. Simulations were conducted assuming no effect of renal impairment on hepatic CYPs; renal impairment modulated hepatic CYPs; KTZ affected CYP activity only and KTZ affected both CYP and renal clearance. Results: Simulated plasma concentration-time profiles of the CYP3A4 substrate in the absence and presence of KTZ described the observed mean data adequately. Assuming no effect of renal impairment on hepatic CYPs, and KTZ affected CYP activity only, the mean model-estimated increase in systemic exposure (AUC) to the CYP3A4 substrate was 2.72- fold for subjects with moderate RI (30-49ml/min) receiving concomitant KTZ. A mean increase of 2.88-fold was estimated in the same patient population when KTZ affected both CYP and renal clearance processes. When RI was assumed to affect hepatic CYPs and KTZ affected CYP activity only, the mean model estimated increase in AUC was 3.72-fold in patients with moderate RI, compared to 3.95-fold when KTZ affected both CYP and renal clearance. Conclusions: The PBPK model adequately described the observed data. A PBPK approach can be used to estimate exposures for complex DDIs in patient populations that are difficult to study.

1. Zhao P, Grillo JA, Young-Moon C et al. Utility of integrating drug disposition data in to a physiologically-based pharmacokinetic (PBPK) model to evaluate a complex drug-drug interaction: CYP/transporter inhibition by ketoconazole in subjects with renal impairment (RI) taking rivaroxaban. ASCPT: March 17-20, 2010 9. Compartmental Modelling of the Pharmacokinetics of an Efflux Transporter

T. R. B. Grandjean 1, J. T. W. Yates 2, M. J. Chappell 1

1 School of Engineering, Warwick University, Coventry, UK, 2 AstraZeneca R&D, Alderley Park, Cheshire, UK

E-mail(s): [email protected] (M. J. Chappell)

A mathematical model has been developed describing the pharmacokinetics of Hoechst 33342 following administration into a culture medium containing a population of transfected cells (HEK293 hBCRP) with a potent inhibitor of the BCRP, Fumitremorgin C (FTC), present. FTC is reported to almost completely annul resistance mediated by BCRP in vitro. The non-linear and multi-compartmental model describes the relationship between the concentration of Hoescht 33342 and FTC initially spiked in the medium and the observed change in fluorescence due to Hoescht 33342 binding to DNA. This model has been extended to consider multi-cell, multi-input responses.

Structural identifiability arises from the inverse problem of inferring from the known properties of a biomedical or biological system a suitable model structure and estimates for the corresponding rate constants and other parameters. Structural identifiability analysis considers the uniqueness of the unknown model parameters from the input-output structure corresponding to proposed experiments to collect data for parameter estimation. This is an important theoretical prerequisite to experiment design, system identification and parameter estimation, since numerical estimates for unidentifiable parameters are effectively meaningless. If parameter estimates are to be used to inform about intervention or inhibition strategies, or other critical decisions, then it is essential that the parameters be uniquely identifiable. Such analysis is highly relevant to large-scale, highly complex systems, typical in chemical kinetics and systems biology.

Structural identifiability analysis has been performed on the Hoechst 33342 pharmacokinetic models developed using a method based on the similarity transformation/exhaustive modelling approach. The analysis demonstrated that all models derived are uniquely identifiable for the experiments/observations available. This permitted subsequent numerical parameter estimation to be performed with greater confidence.

A kinetic modelling software package, FACSIMILE (MPCA Software, UK), was used to obtain numerical solutions for the system equations and for parameter fitting. Model fits gave very good agreement with in-vitro data provided by AstraZeneca across a variety of experimental scenarios. This should ultimately permit predictive analysis to be performed using the model in an attempt to optimise targeting of the compound to cancerous tumours.

Work supported by the UK MRC Capacity Building Intitiative & AstraZeneca.

13. Experimental and Mathematical Analysis of Hepatic Uptake

T. GRANDJEAN3, A. LENCH2, J.W.T. YATES1,M.J.CHAPPELL3, C.J. O‘DONNELL1

1 DMPK Department, Astrazeneca, Alderley Park, Macclesfield, UK 2 Departments of Pharmacy and Pharmacology, University of Bath, Bath, UK 3 School of Engineering, University of Warwick, Coventry, CV4 7AL, UK

E-mail(s): [email protected] (J Yates)

Introduction To investigate the nonlinear kinetics of in vitro hepatic uptake the OATP substrate, Pitivastatin, was used as a probe. Experiments were conducted using freshly isolated rat hepatocytes, utilising the „oil spin‟ methodology described by Hassen et al 1. Briefly, freshly isolated rat hepatocytes were incubated with Pitvastatin (5 – 300µM). At 10 s, 30 s, 50 s and 70 s aliquots were spun through a silicone oil layer to separate the hepatocytes from the media. [Pitivastatin]hepatocyte was detemined using LCMSMS.

Results Uptake to rat hepatocytes was saturable and progressed according to Michaelis-Menten 6 kinetics. The Km and Vmax of Pitvastatin were 2050 pmols/min/10 cells and 33µM respectively, which was in in good agreement with other literature reports 2.

Mathematical Modelling A nonlinear pharmacokinetic model has been derived to characterise the uptake process. A structural identifiablity analysis was performed on the model to establish that all unknown parameters could be identified from the experimental observations available. The model was then subsequently used for parameter estimation and model validation using the data collected. Sensitivity analysis and model robustness analyses were also performed. Once fully validated the model has the potential to perform robust, predictive simulations to ascertain optimal levels of uptake and the effects of the use of appropriate inhibitors.

This work was supported by AZ & TG was supported by AZ and the MRC Capacity Building initiative.

[1] Abdullah M. Hassen, Dennis Lam, Masato Chiba, Eugene Tan, Wanping Geng and K. Sandy Pang. Uptake of sulfate conjugates by isolated rat hepatocytes. Drug Metabolism and Dispositon. 1996; 24: 792- 798. [2] Syunsuke Shimada, Hideki Fujino, Takashi Morikawa, Matsuko Moriyasu and Junji Kojima. Uptake mechanism of pitavastatin, a new inhibitor of HMG-CoA Reductase, in rat hepatocytes. Drug Metab. Pharmacokin. 2003; 18: 245–251.

11. Population Pharmacokinetics of Oral Ciprofloxacin in Paediatric Patients with Severe Malnutrition

Ungphakorn W1, Thuo N2, Muturi N2, Karisa J2, Muchohi S2, Kokwaro G3,4, Thomson AH1,5, Maitland K 2, 6

1SIPBS, University of Strathclyde, Glasgow, 2KEMRI-Wellcome Collaborative Programme, Kilifi, Kenya, 3Dept of Pharmaceutics and Pharmacy Practice, University of Nairobi, Nairobi, Kenya, 4Kenya Medical Research Institute (KEMRI)/Wellcome Trust Research Programme, Nairobi, Kenya, 5Pharmacy Dept, Western Infirmary, Glasgow, 6Imperial College, London.

Background Severe malnutrition in children is a major clinical problem in less developed countries and associated with a high mortality rate in patients who develop sepsis. Parenteral ampicillin and gentamicin are routinely used but practical difficulties with administration and concerns about resistance have prompted the search for alternative options. Ciprofloxacin is a quinolone antibiotic with high oral bioavailability and an antimicrobial activity similar to gentamicin. There is limited information on the pharmacokinetics of oral ciprofloxacin in paediatric patients and none in malnourished children.

Aim To determine the influence of clinical characteristics on the population pharmacokinetics of oral ciprofloxacin in paediatric patients with severe malnutrition.

Methods The study was conducted at Kilifi District Hospital, Kenya in malnourished children >6 months old. Standard WHO guidelines for the management of severe malnutrition were followed and all children received ampicillin and gentamicin. Ciprofloxacin (10 mg/kg 12 hourly for 48 hours) was administered 2 hours before or after (according to the recommended guideline) or with the child‟s nutritional milks (more practical). Up to 4 ciprofloxacin concentrations were measured during the first 24 hours of therapy. The data were analysed using NONMEM1. First order, zero order and transit compartment absorption models were compared. The base model included allometric relationships between oral clearance (CL) and volume of distribution (V) and weight. A range of 24 additional clinical characteristics, including indices of malnutrition, were examined for their influence on ciprofloxacin pharmacokinetics.

Results The data comprised 202 ciprofloxacin concentration measurements from 52 children aged 8 to 102 months. Absorption was generally rapid after a lag of around 45 minutes but highly variable; Cmax ranged from 0.6 to 4.5 mg/L. A one-compartment model with first order absorption and lag adequately described the data. Sodium and potassium concentrations, feeding status, shock and dehydration, were identified as potential covariates on initial screening. A combination of high mortality risk, which reduced CL by 28%, and serum sodium concentration provided the best fit for CL and sodium concentration for V. These factors reduced between subject variability in CL from 50% to 38% and in V from 49% to 43%. Bootstrap, prediction-corrected visual predictive check and normalised prediction distribution error results were satisfactory. Estimates of AUC24 ranged from 8 to 61 mg.h/L, indicating that a target AUC/MIC ratio >125 would only be achieved in all patients if the MIC was <0.06 mg/L

Conclusions The pharmacokinetics of oral ciprofloxacin in malnourished children were influenced by weight, plasma sodium, and an a priori risk of mortality. The wide range of AUC24 estimates suggests that oral ciprofloxacin may have limited application to the management of sepsis in this patient group.

1. Beal SL, Sheiner LB, Boeckmann AJ (Eds), NONMEM Users Guides (1989-2006), Icon Development Solutions, Ellicott City, Maryland, USA. 12. A Time-varying Physiologically Based Pharmacokinetic Model of Caffeine during Pregnancy

L. Gaohua, K. Abduljalil, P. Furness, M. Jamei, A. Rostami-Hodjegan, and H. Soltani

The aim of this study is to develop a time-varying physiologically-based pharmacokinetic (PBPK) model to consider the effects of anatomical and physiological changes during pregnancy on the caffeine disposition in pregnant women. For this purpose, an extensive and structured literature search was carried out to identify the raw data for the pregnancy- induced changes. Then a pregnancy PBPK model was developed, based on a similar structure of Simcyp full-PBPK model. The model consists of 14 compartments representing various tissues, namely, adipose, bone, brain, heart, lung, kidney, gut, muscle, spleen, skin, liver, arterial and venous blood, and a pregnancy tissue compartment. This compartment represents overall fetus, uterus, placenta, amniotic fluid, and mammalian glands. Anatomical and physiological changes during pregnancy were incorporated using gestational time- dependent parameters. The pregnancy PBPK model successfully simulated caffeine concentration-time profile both in pregnant and non-pregnant women. Furthermore, the model is useful to get insight into caffeine disposition in different organs and possible contribution of various physiological parameters in pregnant women.

13. Development of a novel method for updating the predicted partition coefficient values generated by an existing in silico prediction method

Helen Graham1, James Yates2, Aleksandra Galetin1, Leon Aarons1

1School of Pharmacy and Pharmaceutical Sciences, University of Manchester, M13 9PL, 2AstraZeneca, Alderley Edge, SK10 4TG

Please see abstract in Open Session

14. Method for the non-compartmental analysis of log normally distributed pharmacokinetic data: estimation and statistics of pharmacokinetic parameters

Camila de Almeida1, Philip Jarvis2, James Yates1

1 Modelling and Simulations Group, DMPK CIRA, AstraZeneca, Alderley Park, SK10 4TG, 2 Discovery Statistics, DECS, AstraZeneca, Alderley Park, SK10 4TG

Please see abstract in Open Session

15. The Process of Model-based Drug Development

Joachim Grevel

AstraZeneca R&D Charnwood, Clin. Pharmacology and DMPK, Loughborough, LE11 5RH

Please see abstract in Open Session