model-based drug development applied to oncologymodel-based drug development applied to oncology 186...

Review

10.1517/17460441.2.2.185 © 2007 Informa UK Ltd ISSN 1746-0441 185

Model-based drug development applied to oncologyJeffrey S Barrett†, Manish Gupta & John T Mondick†The Children's Hospital of Philadelphia, Abramson Research Center, Rm 916H, 3615 Civic Center Blvd, Philadelphia, PA 19104, USA

Model-based drug development (MBDD) is an approach that is used toorganize the vast and complex data streams that feed the drug developmentpipelines of small molecule and biotechnology sponsors. Such data streamsare ultimately reviewed by the global regulatory community as evidence of adrug’s potential to treat and/or harm patients. Some of this information iscaptured in the scientific literature and prescribing compendiums formingthe basis of how new and existing agents will ultimately be administered andfurther evaluated in the broader patient community. As this data streamevolves, the details of data qualification, the assumptions and/or critical deci-sions based on these data are lost under conventional drug developmentparadigms. MBDD relies on the construction of quantitative relationships toconnect data from discrete experiments conducted along the drug develop-ment pathway. These relationships are then used to ask questions relevant atcritical development stages, hopefully, with the understanding that the vari-ous scenarios explored represent a path to optimal decision making. Oncol-ogy, as a therapeutic area, presents a unique set of challenges and perhaps adifferent development paradigm as opposed to other disease targets. Thepoor attrition of development compounds in the recent past attests to thesedifficulties and provides an incentive for a different approach. In addition,given the reliance on multimodal therapy, oncological disease targets areoften treated with both new and older agents spanning several drug classes.As MBDD becomes more integrated into the pharmaceutical research com-munity, a more rational explanation for decisions regarding the developmentof new oncology agents as well as the proposed treatment regimens thatincorporate both new and existing agents can be expected. Hopefully, theend result is a more focussed clinical development programme, whichultimately provides a means to optimize individual patient care.

Keywords: ADME, clinical pharmacology, in silico, model-based drug development, pharmacodynamics, pharmacokinetics

Expert Opin. Drug Discov. (2007) 2(2):185-209

1. Introduction

Recent insights into the molecular basis of cancer have altered the landscape forcancer drug discovery and development. Advances in molecular targeted cancertherapies offer the promise of personalized medicine for cancer patients. Initialefforts have made it painfully clear that the translation of molecular insights intouseful therapeutic approaches is highly complex and the success of any particularapproach is far from guaranteed. Progress with molecular targeted therapies havealso provided insights into the shortcomings of traditional chemotherapeutic strate-gies and have led to a renewed effort to define optimal dosing strategies, particularlywhen agents of various classes are studied in combination [1]. Likewise, traditionalparadigms of drug development are not optimally suited to these new challenges andmay not fully exploit the potential of new molecular advances.

1. Introduction

2. Oncology targets and drug

development

3. Discovery and preclinical

development

4. Phase I/II

5. Phase III

6. Regulatory communication and

impact

7. Case study: docetaxel

8. Case study: actinomycin D

9. Conclusion

10. Expert opinion

Model-based drug development applied to oncology

186 Expert Opin. Drug Discov. (2007) 2(2)

An overwhelming result of the improvements in discoverystage sciences (e.g., high-throughput screening) and theincrease in oncology targets is the vast amount of data gener-ated in industrial and academic laboratories. The use ofmodeling and simulation to facilitate the development ofnovel anticancer agents and the integration of existingagents into multimodal therapies now has clear precedents.Past examples include the evaluation of docetaxel [2-4], topo-tecan [5-7] and imatinib [1,8,9]. Model-based drug develop-ment (MBDD) involves the union of science-basedintuition (knowledge about a target and the agents investi-gated to hit such targets) and techniques to describe com-plex biological systems (modeling and simulationmethodologies) with the ultimate intention of minimizingrisk to developing an agent and maximizing its clinicalbenefit in conjunction with other compounds.

Many agents, despite promising scientific rationale (includ-ing compelling preclinical data), have shown disappointing effi-cacy in clinical trials. In most cases, it remains unclear whythese agents failed. Did the targeted therapy hit the target? If so,is the target or pathway not as important as postulated? If not,can the agent be used differently, against certain molecularlydefined tumors, combined with other agents, or further modi-fied and optimized for better effect? Were the dose and/orschedule optimal? Was the combination of agents studied opti-mal? Is there a segment of the study population that exhibitedgreater response or toxicity? Although such questions are oftenasked at the conclusion of years of investigation, they are cer-tainly constructed from more specific questions that were orwere not asked at earlier stages of drug evaluation. The vision ofa MBDD paradigm is to construct quantitative expressionsabout target–drug activity → drug-–exposure → expo-sure–(biomarker) response → response–(clinical)outcome rela-tionships, so that these questions promote assumption andscenario testing prior to clinical investigation.

The authors focus this review on defining the landscape foroncology drug development, highlighting factors that differ-entiate oncology from other therapeutic areas. The authorsdefine the MBDD approach and identify modeling and simu-lation (M&S) activities by the development of stages specificto oncology drug targets. These strategies apply to both newand historical agents as both are interwoven into current andprospective treatment modalities. Lastly, the authors illustratethe approach with previously published and recent efforts.

2. Oncology targets and drug development

2.1 Pharmacotherapy as a treatment modalityPharmacotherapy is principally concerned with the safe andeffective management of drug administration. It implies anunderstanding of drug pharmacokinetics (PK) andpharmacodynamics (PD) so that individual dosing guidance,when necessary, can be provided to optimize patient responsewithin their individual therapeutic window. Many oncologydrugs have a low therapeutic index. Pharmacologic and

physiologic changes that occur in oncologic populations andsubpopulations (i.e., elderly and pediatric) can lead to dra-matic consequences such as excessive drug concentrations,unacceptable toxicity or subtherapeutic concentrations andineffective treatment. Unfortunately, in many cases, doses arerarely adapted based on PK or PDs. However, there are meansto elucidate such guidance and several prominent examples ofthe benefit of such approaches.

Dosing in oncological disease states has been guided by sev-eral axioms that are ultimately influenced by PK/PD relation-ships. First and foremost, drugs are more effective incombination and may be synergistic. Second, dosing is likelyto be more effective if drugs do not share common mecha-nisms of resistance. Likewise, it is often beneficial if drugs donot overlap in major toxicities. Drugs should be in adminis-tered near their maximum individual doses and should beadministered as frequently as possible to maximize dose inten-sity (dose per unit time) limiting tumor regrowth. Lastly, andin summary, it is highly desirable to elicit the maximum cellkill within each treatment cycle, using the highest dosepossible, repeating doses as frequently as tolerable.

Given the focus on maximizing cell kill, it is criticallyimportant to maintain therapeutic drug exposures at the siteof action. As indicated previously, the PK attributes of the var-ious drug combinations must be understood so that the dos-ing principles above can be maintained. Although drugclearance is the principal parameter monitored in mostsituations, other PK (protein, tissue and tumor binding, dis-tribution and absorption) and PD (tolerance, receptor orenzyme modulation) processes can be influenced in one wayor another by pharmacological or physiological changeswithin a patient or based on competition for eliminationpathways. In some cases, the potential for such modificationscan be projected based on the properties of the drug sub-stance. However, most are discovered following theexploration of failed therapy.

The uniqueness of oncology is that a therapeutic drug classcomes from both the standpoint of the narrow therapeuticwindow for many of the existing agents as well as thetolerance for toxicity. Dose finding in the past has often beenguided by the maximum tolerated dose (MTD) with littleadditional data to judge effectiveness prior to survival-basedefficacy studies. The emphasis on identifying biomarkers ofdrug activity and toxicity [10,11] has been a valuable additionfor many new agents in development and has certainlyimproved the ability to diagnose prostate cancerprostate-specific antigen (PSA) [12] and prostatic acid phos-phatase (PAP) [13], ovarian and breast cancer (CA 125 [14] andCA 15-3 [15]). However, the generation of biomarkers to iden-tify specific cancer disease progression or help identify thetherapeutic window within which dose selection can be opti-mized has not yet caught up to the expectation (Figure 1). Inaddition, as pharmacotherapy in oncology is multimodal,combination therapy yields temporal, composite therapeuticprogressions (Figure 2), which vary across patients and by

Barrett, Gupta & Mondick

Expert Opin. Drug Discov. (2007) 2(2) 187

disease stage, making treatment dynamic and very muchindividualized to patient response.

2.2 Critical decisions which guide compound advancementThe drug development process is long, arduous and expen-sive. The cost of drug development continues to increase andpharmaceutical sponsors consistently seek to gain efficiency inthe underlying processes. Those companies able to maintainsuccessful portfolios do so by making critical decisions. Thedecisions typically occur at key development milestones, butmay also occur in reaction to an unanticipated outcome.Table 1 lists the critical decision milestones that define a newchemical entity’s (NCE’s) progression through the

development path. Proof-of-mechanism, while desirable, isoften undemonstrated and many drug monographs containwording, which declares that an agent’s mechanism of actionis unknown. During early stages of development thousands ofdrug molecules are screened for activity, safety and ‘druggabil-ity’. Druggability is a relatively new concept and refers to thecharacteristics of the active pharmaceutical ingredient (API)relative to the desired properties of the agent to be developed.It encompasses the physiochemical properties of the agentthat dictate whether it can be compressed into a tablet or evensuitable for an oral formulation, its PK properties that revealdosing requirements, drug interaction potential and thelikelihood of food effect, among other factors. Ultimately, thiscriterion determines the eligibility of a drug product to be

Figure 1. Theoretical therapeutic window for oncology agents.

0

Indices of efficacy Indices of toxicity

Drug exposure

% M

axim

um e

ffect

Figure 2. Potential for temporal effects on the therapeutic window for oncology agents as a function of therapy duration.

0

Delay in clinical response decline

Accumulating toxicity

Plateau in toxicity

Threshold toxicity

Clinical response progression

Duration of therapy

% M

axim

um e

ffect



approved and marketed successfully for an indication thatwould yield financial gain to the sponsor. Hence, the market-place landscape is embedded into this assessment.

Over the past decade or so, anticancer drug discovery hasshifted from an empirical approach, characterized by a ran-dom screening of varied compounds using high-throughputcell-based cytotoxicity assays to a more rational and mechanis-tic, target-based approach. The goal of this new target-basedapproach is to improve the efficacy and selectivity of cancertreatment by developing drugs based on the mechanism ofaction specific to these targets. Validated targets are generallyproteins that play a vital role in malignant transformation,such as the products of the mutated genes that results in again of function (e.g., ras, bcr-abl), or normal receptors andsignaling proteins in pathways that regulate apoptosis [16].Some of the examples of drug development based on tar-get-based approaches include tretinoin (all-trans-retinoicacid), which targets the promyelocytic leukemia (PML) andthe retinoic acid receptor (RAR)α fusion protein in acutepromyelocytic leukemia (APL) and imatinib mesilate, whichtargets the BCR-ABL fusion protein in chronic myelogenousleukemia and the mutated KIT (a protein-tyrosine kinasereceptor specific for stem cell factor receptor) ingastrointestinal stromal tumors (GIST).

2.3 Role of modeling and simulation to facilitate model-based drug developmentM&S approaches applied during preclinical drug develop-ment can provide quantitative predictions of the clinical per-formance of NCEs and help to better plan the early stages ofclinical drug development programmes. The M&S tasks atthis early stage of drug development are primarily designed to‘learn’ about the properties of the molecule that are then ‘con-firmed’ by the data generated during the early clinical drugdevelopment [17]. Table 2 lists and defines the most commonM&S techniques used to construct quantitative relationshipsfor a subsequent query. Many of these are referred to inSections 3, 4 and 5, which describe their application invarious development stages and with the case studies

described. PK/PD modeling approaches are generallywell-appreciated, particularly for their use on guiding trialdesign for oncology agents [18], but M&S activities beginmuch earlier in the development plan under a MBDDapproach and are not limited to drug exposure and activityprediction. Table 3 shows the overlay of M&S activities acrossthe phases of drug development and against some of the com-mon milestones that exist for oncology agents.

From the moment a drug target is defined, assumptionsregarding the optimal characteristics of viable drug candidatesare formed and a target product profile is generated to encap-sulate these features so that the preclinical, clinical and eco-nomic features of minimal, acceptable and superiorcandidates can be ranked based on the scoring of theseattributes. From the moment a drug molecule is synthesized,experimental data is generated to permit assessment of theactual attributes of particular agents. Initially, these may con-stitute physiochemical data, which, when coupled withmolecular structure alone can be used to make predictionsabout drug absorption, distribution, metabolism andexcretion (ADME) properties. ADME properties can then beindexed against activity and/or toxicity data using a techniquesuch as quantitative structure–activity relationships (QSAR).Likewise, as compounds advance through the various stages ofdevelopment, presumably passing predefined criteria foragents of that class as well as regulatory hurdles to ensureaccess to human phase testing, the nature of experimentaldata and the resultant M&S activity evolves as can be seen inTable 3. The salient feature of the MBDD approach is thatthese relationships are connected in such a way that the out-comes or findings from a model-based relationship can betraced to previous models, relationships and data which wereessential for proceeding to the current phase of testing.

The MBDD paradigm can, of course, be applied to anytherapeutic area. For the most part the application ofthe MBDD approach is the same regardless of indication,but there are differences in some disease-drug targets thenecessitate alteration in the types and/or nature of theunderlying models. However, the use of such models to

Table. 1. Critical decision stages for compound progression.

Proof-of-mechanism Establish the correlation between drug mechanism of action and the pathophysiology of the intended disease state or condition

Druggability assessment Determination of whether the formulation can be made consistent with market image and if the PK properties allow achievement of exposure targets

Therapeutic window definition Drug therapy can be administered between an acceptable ‘window’ of drug exposure, which maximizes clinical benefit against any untoward drug action

Proof-of-concept A realization of a method/idea to demonstrate feasibility – determines the effectiveness of a drug and confirms safety in patients that the drug may ultimately treat

Clinical outcome assessment Assessment of change in a patient's ‘status’ over a period of time to determine if the condition is improving, worsening or remains unchanged

Health economics and market valuation

Expected final performance relative to existing medicines used to treat target indications. Analysis of patient costs relative to clinical benefit and health outcomes



Table 2. Modeling and simulation techniques/approaches used in a model-based drug development paradigm.

Techniques/approach Description

Quantitative structure–activity relationship Models that correlate structural or property descriptors of compounds with activities. Physicochemical descriptors would include hydrophobicity, topology, electronic properties and steric effects and include chemical and biological assays

Quantitative structure–pharmacokinetic relationship Models that correlate structural or property descriptors of compounds with PK attributes (e.g., drug clearance)

In vitro–in vivo extrapolation A model-based approach to predict drug clearance, drug–drug interactions, gut absorption and co-variates that determine drug disposition

Response surface model A technique to explore the interaction of several response variables using regression modeling to provide a conclusion for optimal interactions of the response variables of a system

Allometric scaling Models that describe the relationship between PK or physiologic processes (more specifically parameters) and indices of size. Toxicology and risk assessment regularly use allometric models for extrapolation between species and for predicting biological responses in humans based on responses in a laboratory animal

PK modeling Models that explain the relationship between drug dose and exposure (of the parent compound and/or metabolites)

PB/PK Models that define drug kinetics in terms of the physiology, anatomy and biochemistry of the organism and are composed of compartments, which represent body organs and tissues with body compartments are linked together by a flow network. A PB/PK model is defined by a system of deterministic kinetic equations (mass balance equations) of the amount or the concentration of the drug in the compartments as a function of time and initial dose

PK/PD A model that combines the relationship describing concentration versus time (PK) and the relationship describing effect versus drug concentration (PD) to predict the time-course of effect (activity or toxicity) versus time

PPKPopulation PK (and/or PK/PD) modeling

Models that describe the relationships between physiology (both normal and disease altered) and PKs, PDs, the interindividual variability in these relationships and their residual intraindividual variability

CTS A technique for knowledge synthesis and exploration of possible clinical trial results based on a mathematical/stochastic model of the trial execution process, including submodels of the drug action and disease process based on the Monte Carlo simulation

Limited sampling model Models for estimating PK metrics (i.e., AUC and Cmax) based on limited (few) samples collected post drug administration. A regression-based approach in which discrete sampling times (or windows) are evaluated relative to their predictive performance for drug exposure

D-optimal design A regression-based approach to explore the relationship between design information content and sampling location and density

BF BF is a model-based approach incorporating prior data history to improve statistical estimation involved in regulating processes during transition (i.e., when decision making is required), when applied in a clinical pharmacology setting, BF incorporates historical PK or PK/PD data into models which guide dosing

AUC: Area under the curve; BF: Bayesian forecasting; CTS: Clinical trial simulation; PB/PK: Physiologically based pharmacokinetic modeling; PD: Pharmacodynamic; PK: Pharmacokinetic.



drive compound progression, articulate and challengeassumptions and defend decision criteria is still the focusof the approach. Specifically, the availability of biomarkersat early stages, particularly when they can be transitionedinto the patient setting is valuable in the discrimination ofactive doses as well as the definition of the therapeuticwindow. This is the case for many cardiovascular agentsparticularly those that target the vascular space. Incontrast, CNS-mediated targets (for example, depression,

dementia and Parkinson’s disease) rely on psychometricscales, which are often imprecise and have no correlationwith preclinical markers. Similarly in oncology, althoughbiomarker exploration is increasing, most of those definedto date are prognostic for disease status and do notnecessarily track drug actions. This presents challenges todrug developers from the standpoint of biomarkerdevelopment, the assignment of exposure targets based onpreclinical pharmacology data and the development of

Table 3. Modeling and simulation activities used during a model-based drug development approach for oncology agents.

Stage Desired milestone M&S activity

Discovery preclinical

• Mechanism established• Selectivity, specificity targets achieved

• Combination therapies evaluated • IV form stable; SC and oral dosage forms

possible

• CUI favors development

• Single-agent activity models• Synergy/response surface modeling – project drug combinations

• Disease progression: define fundamental assumptionsand baseline time course

• QSARPK and in silico modeling – predict FTIM dose, drug supply needs and desired exposure

• Generate attributes – define CUI and market potential

• Early toxicology acceptable• Input scenarios evaluated (dose, regimen,

sequence and duration)

• Favorable PK based on mechanism; QD dosing

• Mechanistic PK/PD modeling: in vitro – in vivo prediction; tumor binding/exposure models; biomarker performance; xenograft to clinical correlation

• IVIVC; IVIVE; dose-exposure scenarios simulated• Allometric modeling to confirm FTIM dose and drug supply needs

• PB/PK modeling to evaluate peripheral exposure targets and oral absorption

Phase I/II • MTD > therapeutic dose• Optimal dose and schedule determined

• Low interaction potential• Formulation/regimen can deliver desired

exposure profile

• Initial PK/PD models to define dose-exposure and exposure response (considerations for PET and toxicity/AE models)

• Project Phase II trial designs (dose/regimem/ combination selection for Phase II/III)

• Phase I window portable to Phase II

• Biomarker response confirms preclinical mechanistic studies

• PoC confirmed•‘Targeted’ populations identified• Dose selection supported

• Confirmatory PK/PD models (relevance to target population)

• Project Phase III trial designs (based on submission strategy) via CTS; LSS (D-optimal design)

• Identify potential design conflicts (non-response, adherence, ITT, etc.)

• Assessment of need and performance of individualized dosing strategies/designs

Phase III • Phase III clinical outcomes match predictions

• Biomarker response confirms preclinical mechanistic studies

• PoC confirmed• Favorable R&D investment recovery

projected

• Forecasting special or suspected varied populations (e.g., Japanese)

• Potential validation strategy (CTS) for bio-/surrogate- marker

• Cost analysis/pharmacoeconomics

Phase IV • Favorable and informative labeling attained• Favorable R&D investment recovery

confirmed

• Assess potential populations requiring dose modifications (pediatrics, pregnancy, etc.)

• New indications defined• Product line extensions

• PK/PD and CTS to evaluate new indications• PK/PD and CTS to explore therapeutic advantages to product line extensions

CTS: Clinical trial simulation; CUI: Clinical utility index; FTIM: First-time-in man; ITT: Intention-to-treat; IVIVC: In vitro–in vivo correlation; IVIVE: In vitro–in vivo extrapolation; LSS: Limited sampling strategy; MTD: Maximum tolerated dose; PoC: Proof-of-concept; QSARPK: Quantitative structure–activity pharmacokinetic relationships.



models that satisfy mechanistic versus empiricalplausibility.

Table 4 lists examples of M&S applications in the develop-ment/investigation of oncology agents. Whereas this list isneither comprehensive nor prototypical, it does illustrate thediversity of recent applications across many classes ofoncology agents consistent with a MBDD paradigm.Rombout et al. provide a thorough review of the applicationof M&S strategies used during the development of anticanceragents [19]. One of the many benefits of the approach is theability to examine clinical study designs prior to conducting atrial (clinical trial simulation). Figure 3 shows a typical trialsimulation schematic for an oncology agent. Many of themodel compartments are actually quantitative relationshipsthat are defined based on preclinical and/or pooled Phase I/IIdata, which approximate the time dependencies on drugactions. Likewise, knowledge regarding the intended patient

population can be obtained from previous trials with the sameor related agents and the impact of filtering criteria (placebo,drop-out, inclusion–exclusion criteria) evaluated.

3. Discovery and preclinical development

During the preclinical phase of the drug development, thePK/PD properties of candidate molecules are generally com-pared. In addition, potential relationships among dose,concentration, efficacy, and/or toxicity are modeled, whicheventually helps in the selection of the doses chosen for earlyclinical testing. In oncology specifically, the evaluation of doseinput, sequence with other agents and duration of input (e.g.,short, intermittent or continuous infusions) are essential forlater phase testing. Thus, one of the deliverables of M&S atthis stage, is to project the Phase I dose based on thepreclinical, biopharmaceutical and PK/PD properties of the

Table 4. Representative modeling and simulation efforts across various classes of oncology agents.

Class Drug M&S activity Outcome

DNA alkylating agents

Cyclophosphamide BF [121] Individual dose adjustment to reach target exposure

Ifosfamide PPK [56] Described effects of autoinduction on ifosfamides disposition

Busulfan PPK [70] Recommendation of adjusted weight-based doses in children< 12 kg

Antimetabolites Methotrexate PPK [122] Serum creatinine and age explained variability in clearance

Pemetrexed PPK [72,73] Renal function affected drug clearance

PK/PD [74,75] Supported folic acid and vitamin B12 administration to manage neutropenia

Trimetrexate RSM [123] Combination of trimetrexate, LY-309887 and tomudex demonstrateded synergistic inhibition of cancer cell growth

Topoisomerase inhibitors

Irinotecan PK/PD [64] Irinotecan hematologic toxicity profiles predicted using a general PK/PD model developed using other oncology agents

Topotecan PPK [6] Development of limited-sampling strategy for clinical monitoring

Antimitotics Paclitaxel PPK [77] Individual predicted AUC from sparse sampling scheme was a predictor of granulocytopenia and survival

Docetaxel CTS [2] Dose intensification would not benefit patients with high baseline α1-acid glycoprotein

Podophyllotoxins Etoposide LSM [124-127] Development of limited PK sampling model to predict etoposide AUC

Antibiotics Epirubicin PPK [128] Performed simulations to support fixed doses based on AST levels

Target-specific agents

Capecitabine PK/PD [76] Systemic exposure was poorly predictive of safety and efficacy

PB/PK [37] Activation by dThdPase, elimination by DPD and blood flow rate determine 5-FU production in tumor tissue

Oblimersen RSM [129,130] Using response surface modeling, the authors identified the concentration region where maximum synergy or antagonism was achieved at varying concentrations of cytotoxics used in combination with oblimersen

BF: Bayesian forecasting; CTS: Clinical trial simulation; LSM: Limited sampling models; M&S: Modeling and simulation; PB/PK: Physiologically based PK models; PK/PD: Pharmacokinetic–pharmacodynamic models; PPK: Population pharmacokinetics; RSM: Response surface modeling.



NCE and to establish the concentration–effect relationship ofthe NCE in animals. This is generally achieved using acombination of M&S approaches, including populationanalysis of preclinical data, allometric scaling to predicthuman PK and empirical efficacy scaling can be used toproject the anticipated human dose and/or dosing regimen forPhase I studies.

As with other therapeutic areas, the discovery/preclinicalstage is focussed on identifying target activity metrics andscreening compounds against established, often mecha-nism-specific assays. In oncology, the activities during preclin-ical stage include the molecular characterisation of modelsalong with an appreciation of the PD and PK properties ofcompounds defining both toxicity and antitumor efficacy [20].Different PD models can be used to examine the influence ofexposure-time and drug concentration on drug effect in vitro,which can be subsequently used for optimizing regimens con-sidered during in vivo testing. Such models might also beuseful in providing insights into mechanisms of drug action.

The preclinical program in oncology usually involvesin vitro evaluation of compounds from chemical libraries forantitumor effect against various cell lines. Clinical potencycan be estimated from preclinical models, which can be usedto further screen candidate drugs. The preclinical system caneither be an in vitro or an in vivo system; for example, myeloidcolony-forming unit progenitor cell systems (CFU-GM) maybe used for estimating relative potency of different drugs [21].These models are used during the preclinical drugdevelopment to rank compounds and optimize the

experimental designs, allowing consistent savings in terms ofboth resources and time.

Due to the financial and logistic limitations to clinicallyevaluate all possible drug combinations, different drugcombinations of older (cytotoxic) drugs and new anticancerdrugs are evaluated using in vitro modeling approaches [22]

such as response surface modeling during the preclinical stageof oncology drug development [23]. The authors have identi-fied synergistic interactions using MTT assays (in vitro)between G3139 (active agent in the oligonucliotide Bcl-2inhibitor oblimersen sodium) and cytotoxic agents, for exam-ple, etoposide based on response surface modeling (Figure 4)during preclinical investigations of combination therapy withG3139 and cytotoxic agents (projecting antitumor effect).The zero-interaction response surface for the drug combina-tions was generated using a variation of the Loewe additivityequation. Using this approach, the authors were able to iden-tify the region where maximum synergy/antagonism wasachieved at varying concentrations of drugs used in combina-tion. These model-based approaches are aimed at identifyingmore safe and effective drug combinations and elucidating therationale for such combinations. There are several publishedexamples demonstrating the effects of combination therapyusing a model-based approach. For example, trastuzumab andtamoxifen were evaluated in both in vitro and in vivo modelsof breast cancer, which indicated that this combination repre-sented a promising treatment strategy [24]. Some of the ques-tions that need to be answered during this stage of the drugdevelopment process are: i) what are the parameters that

Figure 3. Schematic for a clinical trial simulation model used to project study outcomes based on drug-, patient- anddesign-specific factors.

Side effects

Prognostic factors, gene expression, protein profile

Dose

Tumor size dynamics

Biomarkers

Clinical response

Progression

Survival

Pharmacokinetic Mechanism of actionresistance

Exposure



define synergy or antagonism and the best model for assessingthese in vitro?; ii) do combination data generated in vitrocorrespond to the response of in vivo models?; iii) what addi-tional information can help improve the predictability ofthese models (sequence of drug administration, duration ofexperimental conditions and number of models required todefine a credible interaction)?; iv) how important is it to eval-uate combinations based on previously determined pharma-cology of the drugs? and v) what classes of drugs should becombined for potential therapeutic advantage? [25]. This com-bination data generated in vitro or in vivo is later tested forevaluation of clinical response during the clinicaldevelopment of drugs [26].

A fundamental step of the preclinical development ofoncology drugs is the in vivo evaluation of the antitumoreffect, mostly in mice bearing xenograft models [27]. Neverthe-less, data obtained from these studies often fails to truly pre-dict clinical activity, particularly to assess which cancerindication or histology would be most suited for a drug candi-date. Lately, potentially more clinically relevant in vivo tumormodels such as orthotopic, metastatic and genetically engi-neered mouse models are gaining more popularity [20]. To aidin the selection of a proper dosing schedule for clinical trials,preclinical studies in vivo are generally performed to comparethe antitumor effect and toxicity after different drug concen-trations and drug exposure times. A PK/PD model, linkingthe administration regimen of a candidate to tumor growthdynamics, would greatly improve the preclinical developmentof oncology drugs. The first step in this direction is to developa mathematical model describing the progression of disease. A

number of tumor growth models describing the progressionof a tumor are described in the literature [28,29]. Some of thesemodels are more empirical (e.g., sigmoid functions, logistic,Verhulst, Gompertz and von Bertalanffy, etc.), whereas othersare more mechanistic, which makes certain assumptionsabout tumor growth, involving cell-cycle kinetics and bio-chemical processes, such as those related to angiogenesisand/or immunological responses. Predictive PK/PD models oftumor growth kinetics in xenograft rodent models have beenstudied for several anticancer agents [27,30,31].

During the preclinical stage, it is also important to comparethe PK/PD properties of the candidate molecules and estab-lish the concentration–effect relationship of the NCE. Theprediction of human PK from preclinical data is essential toreduce the evaluation of compounds with poor PKs in thehuman phase of testing. The common practice is to estimatePK parameters from observed plasma data by empiricalapproaches, such as non-compartmental analysis (NCA) or asimple compartmental model of the data. There are severalexamples in the literature where such approaches have beenapplied, although limited examples exist where a popula-tion-based approach has been applied during the preclinicalstage of drug development. Rouits et al. used non-linearmixed effects modeling approaches to evaluate the linearity ofCPT-11 PK and the influence of the schedule of administra-tion in mice [32]. Although these models may have some useand pragmatism, PK/PD modeling is progressing from anempirical, descriptive discipline into a more mechanistic sci-ence that can be applied to all stages of drug development [33].

M&S methodologies can be used to capture uncertaintyand use the expected variability in biopharmaceutical andPK/PD data generated in preclinical species for projectionof the plausible range of clinical dose based on expo-sure–response (antitumor and toxicity) relationships;clinical trial simulation can be used to forecast the proba-bility of achieving a target response in patients based onthe information obtained in the early phases of develop-ment [17,34]. Preclinical studies are based on certainassumptions, including: i) the relative efficacy and potencyobserved in the in vitro systems and the studied animalspecies, given the projected uncertainty, is predictive of therelative efficacy and potency later observed in humans andii) allometric scaling most often provides a reasonable esti-mate of the clearance of the compound in humans. Inter-species scaling, based on allometric relationships, havebeen used to bridge preclinical and clinical PK informationby prediction of drug clearance and the volume of distribu-tion. For the purposes of estimating starting doses inhumans, the human clearance and volume of the NCE isprojected with uncertainty based on allometric scaling ofthe clearance values obtained in different animal species.The prediction from animal species (e.g., rat) to humancan be further improved by scaling by the relative drugpotency between rat and human myeloid cells in vitro, inaddition to accounting for species differences in PKs, for

Figure 4. Response surface modeling illustrating thesynergy achieved with combination therapy relative to thesingle agent profiles. Dark spheres correspond to experimentalcombination effects and light (smaller) spheres to thezero-interaction surface.

Effe

ct (=

1-Su

rviv

al)

1

0.5

510

1520

0

0

50

100

150

200

G-3139 (µM) Etoposide (nM)



example, in protein binding and the formation ofmetabolites [21]. Physiologically based approaches might bemore useful in cases where allometric scaling fails, mostlyfor active processes, such as metabolism and transport.

Physiologically based PK models (PB/PK) offer a more sci-entific approach to help rationalise the ‘first into human’ deci-sion-making process [33]. PB/PK models are increasingly beingused in preclinical drug development to enhance the under-standing of the complex interactions of biological and physio-logical parameters that affect drug behaviour at the molecularlevel in multiple physiological compartments [35]. These mod-els provide several benefits, including: i) the prediction ofplasma (blood) and tissue PK of drug candidates prior toin vivo experiments; ii) supporting a better mechanisticunderstanding of PK properties, as well as helping the devel-opment of more rational PK/PD relationships from tissuekinetic data predicted and, hence, facilitating a more rationaldesign during clinical candidate selection and iii) extrapola-tion across species, routes of administration, duration andtiming of drug input and dose levels [36]. These models arealso used to screen potential clinical candidates from the poolof compounds in drug discovery [34]. For cytotoxic drugs,PB/PKs model can be used to identify factors that govern thedose-limiting toxicity in gastrointestinal tissue and antitumoractivity. The developed PB/PK model can also be applied tohuman cancer xenografts in rodents, which could be used fur-ther to plan clinical development strategies from preclinicaldata. Mechanism-based population PK/PD models can beapplied to several important milestones during oncologic drugdevelopment such as candidate selection, first-in-man studies,prodrug and formulation development, dose finding andschedule optimization, assessing influence of modifyingagents, drug combination studies, subgroup identificationand feedback individualisation [18,21]. There is an increasinglygreater number of examples in oncology, where the PB/PKmodel has been successfully used during the preclinical stage.Tsukamato et al. has shown that the PB/PK model integratingbiochemical parameters in vitro (enzyme activity and tissuebinding) was suitable for describing the PKs of capecitabineand its metabolites in human xenograft models. They alsodemonstrated that the 5-FU total exposure in tumor tissue atthe effective dose for human cancer xenograft models wascomparable with that at the clinical dose in humans [37]. Inanother example, Baxter et al. described the PB/PK model formonoclonal antibodies in human tumor xenografts in nudemice, which could potentially be used to scale antibody PKsfrom mouse to man [38].

During the preclinical stage of drug development,biomarkers are used to assess target activity and/or safetythresholds and, ultimately, as a means of bridging the transi-tion from preclinical to clinical studies [39,40]. Preclinically,biomarkers can facilitate the selection of animal models andof lead compounds tested in those models. They can also aidin the demonstration of the mechanism of action from bothin vitro and in vivo models. Some biomarkers evaluated in this

manner, such as those that measure apoptosis or signalingpathways, can be used to quantify the effects of anticancerdrug combinations and subsequently to predict optimumclinical treatment regimens. Biomarkers such as PSA are beingused routinely to monitor the treatment response of prostatecancer in animal models [41]. One of the recent advances thatpromises to shorten the preclinical drug development processfor oncology agents is the incorporation of positron emissiontomography (PET) into early drug studies to identifynon-invasive disease specific biomarkers as PK/PD endpoints, so that the best candidate can be selected for clinicaldevelopment. Non-invasive molecular imaging probes havebeen successfully used to detect tumor hypoxia for Hsp90molecular chaperone inhibitor 17AAG, as well as the develop-ment of SR-4554 [42]. Likewise, such techniques yield richdata from which quantitative models can be derived and usedto define activity targets and screen development candidates.

4. Phase I/II

In the oncology setting, there is seldom a clear distinctionbetween the design and conduct of Phase I and Phase II stud-ies. Unlike most therapeutic areas, Phase I oncology trials areusually performed in the target population. The principalgoals of Phase I/II studies are to describe safety and toxicity,pharmacologic behavior and to define an optimal and safedose and schedule for subsequent efficacy studies. The designand analysis of Phase I/II oncology studies are particularlywell suited to a population-based approach, as these studiesare generally conducted in patients with metastatic cancer,who are being treated with a combination of therapeuticagents. Typically, PK and PD data are pooled across studiesfor analysis via non-linear mixed-effects modeling. This mod-eling effort is vital as the first step to learn about the PKcharacteristics and quantify the variability of the NCE inhuman subjects.

The transition from preclinical to clinical drug develop-ment most often resides in setting an optimal starting dose inhumans. There is no gold standard yet for estimating the ini-tial dose for oncology phase-I dose escalation trials. Generally,the starting dose of a cytotoxic agent is calculated as one-thirdof the toxic low dose (TDL) expressed as mg/m2 in a largeanimal species (either dog or monkey). Collins et al. used apharmacologically guided dose escalation (PGDE) approachdefined as one-tenth of a dose (expressed in mg/m2) that islethal to 10% of non-tumor bearing mice (LD10) [43].Recently, there has been a shift to transform the starting-dosecalculation from an empirical approach to a more mechanisticpharmacokinetically and pharmacodynamically driven one.Gallo et al. used physiologically based hybrid PK modelsderived from preclinical data and then scaled it to humans topredict human tumor carboplatin, topotecan andtemozolomide concentrations [44]. These efforts have focussedon tumor-based PK/PD relationships, which provide arationale means to select the most efficacious drug dosing



regimens [30,44]. It has also been suggested that PK/PD modelsdescribing the hematologic toxicity time profile may be usefulfor dose escalation decisions [21]. Simulations to explore vari-ous dose escalation scenarios could be predictive of likelyhematologic toxicity outcomes in response to the proposedescalation schemes.

Historically, much of the efforts put forth in modelinganalyses for oncology agents have been focussed on charac-terising and explaining the variability in PKs. Numerousexamples exist to illustrate this point: melphalan clearance issomewhat higher in males than in females and decreaseswith decreasing creatinine clearance [45], variability in epiru-bicin clearance could be attributed to sex and also to age inwomen [46], carboplatin clearance in children is related toweight and serum creatinine [47], mild-to-moderate renalinsufficiency has no effect on gemcitabine PKs [48].

More recent modeling studies have focussed on investigat-ing patient specific genotypes in order to identify the effectsof genetic polymorphisms on the disposition of chemothera-peutic agents. The association of cytochrome P450(CYP)2D6 polymorphism and ipifarnib PKs was investigatedusing a mixed effects modeling approach [49]. The effects ofCYP2C8, CYP3A4, CYP3A5 and ABCB1 polymorphism onpaclitaxel PKs have also been investigated using the sameapproach [50]. Although results for both of these studies dem-onstrated that no influence on drug disposition was apparentfor the enzyme polymorphisms studied, several examples out-side the oncology setting have shown genetic factors to be sig-nificant predictors of drug disposition. A recent study ofcyclophosphamide pharmacogenetics used a mixed-effectsmodeling approach to demonstrate that patients with theCYP2B6 G516T variant had a higher cyclophosphamideclearance than wild-type patients [51]. Such studies for oncol-ogy agents are warranted to explain the substantial interindi-vidual variability in PK and PD that may be attributed togenotype for these agents.

In addition, there are several reported studies in the litera-ture where an integrated population PK and/or PD model hasbeen developed during Phase I/II studies to elucidate the rela-tive contributions of parent and metabolite to the pharmaco-logic and toxic effects of anticancer therapy. The populationPK model describing the disposition of irinotecan and itsmetabolites has been reported by several investigators [52-55].Population PKs of ifosamide and its 2- and 3-dechloroethyl-ated and 4-hydroxylated metabolites in resistant small-celllung cancer patients has been reported in the literature [56]. Inthis example, the PKs of ifosamide was described by concen-tration-dependent autoinduction and clearly showed thatfractionation of the dose of ifosamide resulted in an increasedexposure to 2-dechloroethylifosamide. These models are use-ful in predicting the time course and inter-individualvariability of characterised substances.

PK/PD models developed at this stage are also important incharacterising the effect of drug–drug and drug–diseaseinteractions that might be expected to perturb the PK/PD

relationships of the drug. Population PK model suggested amutual drug–drug interaction between cyclophosphamideand thiotepa, most likely due to induction of thiotepametabolism with time [57]. In another study, a popula-tion-based approach was used to evaluate the PKs of paclitaxelwhen it was concomitantly administered with the P-glycopro-tein (P-gp) inhibitor, zosuquidar trihydrochloride. The modelclearly showed an increase in paclitaxel exposure, which canmost likely be attributed to P-gp inhibition [58]. The sameauthors have also shown an increase in doxorubicin and doxo-rubicinol exposure in the presence of zosuquidar hydrochlo-ride [59]. This approach has also been used to show that thereis no significant PK interaction between 5-fluorouracil andoxaliplatin [60].

Efforts to estimate the relationship between the exposureand magnitude of myelosuppression have used a variety ofempirical models (e.g., linear, log-linear, logistic regressionand E-max). Using mechanistic PK/PD models, it is possibleto appropriately assess both the impact on PK interaction aswell as the PD (potential efficacy or toxicity) of the combina-tion. A key minimum objective of these early PK/PD studiesis to verify the in vivo biological activity of the drug againstthe in vivo intended target and to ensure that the action onthe target is of suitable intensity and duration with respect tothe dosing interval. However, parameters from these mecha-nistic models are generally not available when analyzing clini-cal data and, thus, more simplistic semi-mechanistic modelsof myelosuppression to optimize anticancer therapies arebecoming increasingly popular among the scientific commu-nity. These models use the whole concentration-time profileand not a summary variable of exposure (empirical models) asinput to the model.

Myelosuppression is the most frequent toxicity in responseto chemotherapy. Much of the effort in developing PK/PDmodels has been focussed on describing myelosuppression inresponse to chemotherapy [61]. Models of chemotherapeu-tic-induced neutropenia have been reported for indisulam [62]

and topotecan [63]. Friberg et al. have reported a semi-mecha-nistic approach of modeling chemotherapy-induced myelo-suppression that partitions the model into drug specificparameters and system parameters that are generalizable acrosschemotherapeutic agents [64]. This model was developed frompatient data resulting from docetaxel, paclitaxel and etoposideadministration. The model was then applied to myelosuppres-sion data resulting from irinotecan, vinflunine and DMDCadministration. Application of a generalisable model across avariety of chemotherapeutic agents would be beneficial for theinvestigation of myelosuppression in clinical development.Likewise, a simulation approach for study design, dose escala-tion studies in Phase I could be useful to limit the probabilityof patients experiencing severe myelosuppression. Discreteevent simulation (DES) is a mechanism used to study trialperformance and efficiency. The approach decomposes studyevent into a set of logically separate processes thatautonomously progress through time. Each event occurs on a



specific process and is assigned a logical time (a timestamp).Recently, DES has been used to investigate doseescalation/de-escalation and stopping rules based on thefrequency of dose-limiting toxicity (DLT) occurrence in pedi-atric oncology trials [65]. The classic Phase I oncology trialdesign was decomposed into a series of discrete time events(accrual/enrolment, evaluation and/or time to DLT or ineval-uability) with outcome probabilities (DLT or inevaluability)assigned to each subject based on historical data from Phase Ipediatric oncology trials. The probability of DLT occurrencewas correlated with the trial cohort and varied distributionswere used to simulate/assign the discrete time elements. Met-rics for study efficiency (time to reach MTD, time to com-plete various trial designs and the number of patientsnecessary to complete the trial) were defined and comparedfor the classic 3+3 design [65].

Models constructed during Phase II drug development aretypically used to guide the decision-making process foradvancement to Phase III studies. The primary use of thesemodels has been to support dose selection for large-scalePhase III efficacy studies. Likewise, knowledge of drug inputand timing (dose, duration and sequence of various treatmentregimens) is incorporated into both Phase II trial designs.Subsequently, the design and evaluation of proposed Phase IIIstudies can be based on models developed in Phase II [66]. Asthere is no straightforward method to conduct sample size cal-culations to appropriately power a Phase III study, PK/PDsimulation lends itself as a valuable tool for the assessment ofthe impact trial design factors and study conduct variances onstudy outcomes. The number and distribution of plasma col-lection time points, the number of doses and dose range, thenumber of patients included (total and subpopulation), theeffect of compliance and/or dropouts on statistical power andthe clinical efficacy/outcomes are all aspects of the trial thatmay be addressed by simulation. The actinomycin case studyreported in this paper describes an example of a modeling andsimulation approach to trial design in oncology.

Predicting Phase III outcomes based on models developedin Phase I/II has also recently been advocated [67,68]. Thismethodology allows for the investigation of probable trial out-comes for a particular design before incurring the time andexpense of a large-scale Phase III trial. For example, a simula-tion study has been described that compared bodyweight-adjusted and fixed doses of darbepoetin and the effectson hemoglobin levels [69]. Simulations showed that forpatients weighing 45 – 95 kg, the impact of the fixed dose wasinsignificant, with a minor weight effect present outside thisrange. Simulation studies to examine trial outcomes have alsobeen described for docetaxel [2]. A higher docetaxel dose(125 mg/m2 versus 100 mg/m2) demonstrated no advantagein treating patients with high baseline α1-acid glycoproteinlevels who exhibit a shorter time-to-progression and death.Based on the result of these simulations, the decision wasmade to forego the Phase III trial.

Despite the advantages of using a rigorous modeling andsimulation effort during the Phase I/II drug developmentstage, the full potential of this approach has not been fullyrealised [18,21,70]. Traditional, well-defined markers of clinicalsuccess in the oncology arena include overall survival,response rate, time to disease progression and disease-free sur-vival. These markers are often examined in a modeling con-text using statistical models such as survival and time-to-eventanalysis. However, recent efforts in Phase I/II trials have con-centrated on the evaluation of novel biomarkers that would beindicative of pharmacological activity of a chemotherapeuticagent. Molecular targeted agents afford the opportunity to usebiological markers as measures of efficacy. For these agents,the primary end-point used to measure the dose–effect rela-tionship is likely to be biological rather than toxicity, depend-ing on whether a biological end-point is achieved at anon-toxic dose [71]. Thus, for molecular targeted drugs, themaximum therapeutic effect may be achieved at doses that arewell below the MTD. For example, imatinib mesilate’s thera-peutic effects result from the inhibition of BCR-ABL tyrosinekinase activity. However, it also inhibits other receptor tyro-sine kinases. If inhibition of other tyrosine kinases is responsi-ble for one or more of its toxicities, then the dose–responserelationship for toxicity may not parallel the dose–responserelationship for BCR-ABL inhibition because of the differingaffinities of the receptor for the drug [16]. Although drugdevelopment by way of a biological marker approach wouldnot be likely to replace traditional methods, it will be animportant adjunct. Modeling and simulation approachesincorporating well-defined and appropriate biomarkers willfacilitate the development of oncological agents as thisapproach would circumvent some of the problems with tradi-tional end-points, such as long follow-up periods and the con-fusing of multiple treatments. Not only would thismodel-based drug development paradigm support the deci-sion process while developing a potential oncology candidate,but also assists in identifying candidates, which may prove tobe unsuccessful before incurring the time and expense ofconducting a Phase III trial.

Pemetrexed is an example of an oncology agent whosePhase II clinical development has significantly benefited froma model-based drug development approach. Pemetrexed is anantimetabolite that inhibits thymidylate synthase (TS), dih-drofolate reductase (DHFR) and glycinamide ribonucleotideformyl transferase (GARFT). The initial pemetrexed popula-tion PK model was reported by Ouellet et al. [72]. PK datafrom four Phase II studies was pooled for the analysis with theobjectives of characterising pemetrexed PKs and to estimatethe effects of demographic factors that may be relevant predic-tors of variability in pemetrexed PKs. Pemetrexed PKs wasdescribed by a two-compartment model. Creatinine clear-ance, body weight, alanine transaminase and folate deficiencywere demonstrated to have a significant impact onpemetrexed clearance.



A second pemetrexed population PK analysis was reportedby Latz et al. that pooled data from 10 Phase II studies [73].The goals of this analysis were similar to the previous study,with an additional objective of providing individual Bayesianestimates of PK parameters for the development of a PK/PDmodel to describe pemetrexed induced neutropenia. Renalfunction was again identified as an influential covariate forclearance. This study suggested that pemetrexed dose adjust-ments might be warranted based on an individual patient’srenal function.

A population PK/PD model to describe neutropenia inpatients has also been reported [74]. The objectives of thisanalysis were to describe the time course of the neutro-penic response to pemetrexed and to identify any demo-graphic factors that explain variability in the neutropenicresponse. The data set used for the modeling consisted ofeight Phase II trials. The response was described by asemi-mechanistic physiological population PK/PD modelthat estimated baseline neutrophil count, mean transittime and a feedback parameter. The analysis supportedfolic acid and vitamin B12 supplementation to managepemetrexed-induced neutropenia. The clinical relevance ofthe population PK/PD model was further investigatedthrough simulation studies [75]. Time profiles for absoluteneutrophil counts (ANC) were generated using the modelto examine the incidence and severity of myelosuppressionin response to two pemetrexed doses administered accord-ing to both body surface area and renal function inpatients with normal and elevated vitamin deficiencymarkers. It was concluded from this study that vitaminsupplementation was necessary to manage the risk of mye-losuppressionin response topemetrexed administration.Results from the Phase II modeling and simulation effortshave been included in the pemetrexed product label.

5. Phase III

The primary goal of Phase III clinical studies is to showefficacy and safety in a larger patient population represent-ative of the target population. Phase III studies are typi-cally not as well controlled as Phase I and Phase II trials,with a widely varying patient population that may includepatients of various disease states, additional medicalconditions not associated with oncology, patients adminis-tered numerous concomitant medications, differing radio-therapy treatments and broadly different demographicfactors. These trials are performed after data collected inPhase I and Phase II suggest effectiveness and safety.Another necessity of the Phase III stage is to gather theappropriate efficacy and safety information to providedosing guidance and product label specifications.

The Phase III process provides further opportunity toevaluate the relevance of a given biomarker to achemotherapeutic agent’s pharmacologic action. Given thelarge-scale nature of the Phase III study, this point in the

development process usually provides the necessary data toeither confirm the ability of a proposed marker to functionas a surrogate for efficacy/safety or to demonstrate little orno clinical use of the biomarker.

As a wide patient diversity is studied in Phase III trials, thisaffords the opportunity to retrospectively study any specialpopulations and patient demographics that may influencedrug disposition. For example, within the hundreds or thou-sands of patients dosed in a Phase III trial, a small portionmay show characteristics of renal impairment ranging frommild to severe. For a drug that is eliminated via the renalroute, the disposition of the chemotherapeutic agent in thesepatients may be compared with those patients with normalrenal function. Based on the model developed from the PhaseIII data, inferences can be made regarding the magnitude ofthe renal effect and what dose adjustments may be necessaryin this population. If an additional trial is deemed necessaryto study the special population, simulation efforts may beused to not only design the trial, but also to predict the trialoutcomes and determine the probability of a successful trial.

Phase III studies also afford an opportunity to examine theclinical response to a chemotherapeutic agent in a large popu-lation of patients. Population PK/PD modeling is the pre-ferred method of analyzing PK/PD data arising from thesestudies due to the sparse sampling designs used in Phase IIIstudies. For example, capecitabine exposure has beendetermined using a population PK model developed fromdata collected in two Phase III studies in colorectal cancerpatients [76]. Exposure metrics were then explored as predic-tors of efficacy and safety for logistic regression andtime-to-event analysis. Results from the study showed thatexposure to capecitabine and its metabolites were generallynot predictive of outcomes. A population PK analysis ofpaclitaxel in endometrial cancer patients has also been per-formed to predict individual patient exposure from sparse PKsamples [77]. It was found that survival and granulcytopeniawere predicted by paclitaxel AUC.

Another goal of the Phase III trial may be to learn the nec-essary PK and PD properties to individualize chemotherapyto the particular patient. The narrow therapeutic index, com-bined with the lack of appropriate surrogate markers of toxic-ity or response, adds to the empiricism in the administrationof cancer chemotherapy. In addition, the PKs of cancer chem-otherapy agents is highly variable between patients. However,recent studies have established relationships between systemicexposure to cancer chemotherapy and both response and tox-icity. These relationships have been used to individualizechemotherapy dose administration a priori and a posteriori.Drugs which can be individualized based on their PKs includemethotrexate [78-80], busulfan [81-83] and carboplatin [84-86].Other examples of anti-neoplastic agents, which may eventu-ally be individualized based on their PKs are mercaptopurine[87], fluorouracil [88-90], etoposide [91,92] and topotecan [93,94].Population PK/PD modeling has played a large role indefining individualized dosing metrics for all of these drugs.



The FDA’s accelerated development and review process wasintroduced in 1992 as a means to expedite the drug develop-ment process for those indications that are of a serious orlife-threatening nature for which no therapy exists. Includedin this mechanism would also be drug candidates that wouldprovide significant benefits relative to existing therapies. Asurrogate end point may be used in place of a traditional,well-defined end point to demonstrate efficacy. This surrogatemust be further evaluated in post marketing studies in orderto verify the clinical benefit of the drug. Modeling and simu-lation approaches may be instrumental in the decision-mak-ing process for drug development in the case of acceleratedapproval. Capecitabine, imatinib and topotecan are examplesof drugs that were approved under the accelerated approvalprogramme which contain information in the product labelpertaining to population PK modeling. These analyses havebeen limited to characterising PK and the contribution ofdemographic factors to variability. Little data is available atpresent on the use of a surrogate marker in a modeling andsimulation approach to accelerated approval, but this willlikely change based on ongoing activities at many pharmaceu-tical sponsors and the ongoing and collaborative work of theFDA and the National Cancer Institute (NCI).

6. Regulatory communication and impact

As pharmaceutical sponsors and clinical investigators exploretreatment modalities which may offer advantages to existingoncology therapies, an important partner in this endeavour is,of course, the regulatory community. In the US the FDA hastaken an active role in this process and have formed collabora-tions, developed tools and established communication path-ways with the drug industry in the hope of providing bettercancer drugs to patients sooner. Specifically, in 2003, underan agreement between the FDA and the NCI, the two agen-cies initiated a plan to share knowledge and resources [201].

This agreement is intended to broaden existing programmesand to create additional joint programmes between these twoDepartment of Health and Human Services (HHS) agencies.NCI believes that the timely and considered application ofscience across the continuum of discovery, development anddelivery, will optimize and accelerate the delivery of safer,more efficacious, drugs to patients. In addition, the FDA hascreated a ‘tools’ website [202], which contains a variety of infor-mation related to cancer and approved cancer drug therapies.Although still at the pilot stage, the intention is clearly to pro-mote communication and education regarding novel therapiesand oncology disease biology. Most importantly, the FDA isconsidering the creation of a library of drug/disease modelsthat could eventually be used by industry to simulate anddesign clinical trials [203]. A disease model for oncologic indi-cations will certainly have to address the hundreds of differentdiseases that can arise from virtually any tissue or organ in thebody and despite sharing similar properties of local invasionand distant spread may exhibit different causative factors,

natural history of disease, method of diagnosis and treatmentmodality. A drug/disease model that accommodates suchdiversity would likewise by the funnel of any MBDDplatform [95].

Two case studies are described in the subsequent section.Both illustrate the MBDD paradigm from different startingpoints with respect to their development and continued use asanticancer agents vital to their respective indications. Thedocetaxel experience is more well-known regarding the his-toric perspective and broad base of clinical application. Actin-omycin D is a work in progress for a drug which has beeninvestigated for > 50 years without the benefit of a moderndrug development plan.

7. Case study: docetaxel

The history of the research and development of the prototypetaxanes paclitaxel and docetaxel is well known particularlygiven the natural origin of their discovery. Docetaxel was firstsynthesised in 1986 following the identification of the com-pound in the needles of the European Yew tree in 1981. By1990 Phase I studies had been initiated followed closely byPhase II and III studies in 1992 and 1994, respectively [96].The first regulatory approval from the US FDA came in 1996for locally advanced or metastatic breast cancer. Additionalapprovals ensued following the demonstration of survival ben-efit in various target populations. Table 5 lists the major mile-stones in the discovery and development of docetaxel. It alsolists many of the concurrent M&S activities conducted insupport of docetaxel’s development. Many of the M&Sanalyses were in fact used in the decision making and subse-quently submitted as part of the regulatory transactions withFDA in support of label claims and in response to targetedquestions regarding activity and especially safety. An interest-ing aside is that several of these efforts were championed byscientists outside the sponsor company (initially Rhône-Pou-lenc Rorer) and the composite data including in vitro activity,IVIVC and clinical outcome data are continually used byother companies seeking to make the next generation taxanes[97], evaluate their product’s performance when added to a reg-imen including docetaxel or comparing safety and/or efficacyresults with docetaxel.

Much of the early phase testing was not published, butincluded PK and PK/PD model-based analyses to support doseselection for human phase testing (personal communication).These studies and family of models then became the startingpoint for Phase I study designs. Six human trials were con-ducted to support Phase I [96]. These studies [98-103] character-ised the safety and PK of docetaxel administered intravenouslyat different infusion lengths (1, 2, 6 and 24 h) up to 5 days andwere consistent in defining the MTD between 80 and115 mg/m2. The primary dose-limiting adverse effect was neu-tropenia, whereas mucositis was the dose-limiting toxicity whendocetaxel was administered over a longer duration [96]. Based onthe Phase I results, the recommended dose for Phase II testing



was 100 mg/m2. Whereas population-based PK analyses wereinitiated to partition sources of variation from the pooledPhase I data [104], the Phase II data were used to explore expo-sure–response relationships with the intention of understandinghow to manage patient toxicity to docetaxel [105]. Several popu-lation-based PK and PK/PD analyses were published [5,104,105]

along with one of the earliest, detailed examples of populationPK model building and validation [106]. With respect to thePKs alone, it was learned that interpatient variability is relatedprimarily to body surface area and hepatic function. Clinically,this information was extended to recognise that patients with

impaired liver function are at increased risk of serious adverseeffects (febrile neutropenia, severe infections, severe stomatitisand toxic death) during docetaxel therapy at 100 mg/m2, whichcould be reduced by adjusting the dose to 75 mg/m2.

The understanding of exposure–response relationshipswas then used examine dose optimization with docetaxel.A series of clinical trial simulations [2] were initiated to testwhether a specific subset of adult patients withnon-small-cell lung cancer might benefit from doseintensification. PK and PD models for time to progression,death and drop-out were developed and validated with the

Table 5. Development milestones for docetaxel relative to modeling and simulation activities that facilitated development and regulatory acceptance.

Date Milestone M&S activity

1960 NCI creates program to screen plant species for new anticancer drugs

1962 Samples collected from the Pacific yew tree

1963 Anticancer activity in samples of the Pacific yew tree bark documented

1971 Active ingredient identified leading to isolation of paclitaxel, the first taxane

Many QSAR studies/analyses conducted in support of back-up program

1981 Researchers in France examine European yew tree

1986 Discovery of docetaxel, a taxane drug derived from needles of the European yew tree

Preclinical ADME studies conducted; initial PK models developed

1989 Human studies of docetaxel begin Human PK models developed

1993 Laboratory research shows that docetaxel is 2 – 4 times as potent as paclitaxel due to an increased ability of docetaxel to bind to tubulin

Patient PK/PD models developed; exposure–response with target toxicities explored.

1996 Docetaxel approved by FDA (locally advanced or metastatic breast cancer)

Population PK models developed and submitted with the NDA [104,106]; clinical use and pharmacoeconmic models developed [131]

1999 Docetaxel approved by FDA as second-line treatment for locally advanced or metastatic non-small-cell lung cancer

Pooled Phase I PPK [5]

2002 Docetaxel approved by FDA as first-line treatment for non-small-cell lung cancer in combination with cisplatin

2003 Clinical trial in patients with metastatic breast cancer failing prior chemotherapy shows median survival advantage of docetaxel over paclitaxel (15.4 months versus 12.7 months)

2004 Docetaxel approved by FDA as treatment for patients with androgen-independent (hormone-refractory) metastatic prostate cancer, in combination with prednisone

2004 Docetaxel, in combination with doxorubicin and cyclophosphamide, is FDA-approved for the adjuvant (supplemental) treatment of operable node-positive breast cancer

Present Docetaxel continues to be tested in ongoing clinical trials for treatment of other cancers

QSAR models developed for next generation taxanes [97]

M&S: Modeling and simulation; NCI: National cancer institute; PPK: Population pharmacokinetic modeling; QSAR: Quantitative structure–activity relationship.



use of Phase II data from 151 patients with non-small-celllung cancer. The simulation process was evaluated by com-parison of the original Phase II data with the predictedresults. Simulations were undertaken for the evaluation ofwhether a Phase III trial of two different dose of docetaxelin these patients would result in improved survival. In thesimulated Phase III trial, although median survival wasslightly longer in the 125 mg/m2 docetaxel group than inthe 100 mg/m2 group, the difference was significant inonly 6 of 100 trials. Hence, given the small likelihood thata meaningful difference in clinical outcomes wouldactually exist, the simulation was the basis for not conduct-ing such a trial. Much of the evolved modeling effort, par-ticularly the relationships which describe docetaxelexposure versus toxicity are described in the drugmonograph (package insert,) which confirms both thevalue the sponsor company (now sanofi-aventis) and theFDA place on this information.

As mentioned previously, the progression of models thatcharacterise various facets of docetaxel as a medicinecontinue to benefit drug development. In an effort toinform the next generation of taxanes, biochemicalanalyses for a series of 20 butitaxel analogs, paclitaxel anddocetaxel were used to build two- and three-dimensionalquantitative QSAR models to investigate the propertiesassociated with microtubule assembly and stabilisation [97].

Comparative molecular field analysis (CoMFA) modelsbuilt using steric and electrostatic fields and hologramquantitative structure–activity relationship (HQSAR)modeling of this same data were used to examine theability to predict activity outcomes based on structuralfeatures. Results are promising and may be indexed withmodels predicting safety indices to project the therapeuticwindow of development compounds prior to clinicalevaluation. Hence, although not part of a modern MBDDparadigm, docetaxel remains an excellent example ofthe MBDD concept applied to an oncology agent, whichcontinues to be used to treat patients and has beenexplored as an adjunctive or comparative agent.

8. Case study: actinomycin D

Actinomycin D is a member of the antibiotic class ofantineoplastic agents (actinomycins) produced by theStreptomyces species of fungus. Actinomycins were firstdiscovered in 1940 [107] and actinomycin D was first isolatedand introduced into the clinical oncology setting in 1954.During its initial clinical evaluation it was studied inpatients with gestational choriocarcinoma and in primaryWilms’ tumor in combination with surgery andradiotherapy. Actinomycin D is one of the olderchemotherapy drugs, having gained approval from the FDAin 1982. Actinomycin D is presently used in the treatmentof several childhood sarcomas including Ewing’s sarcoma,rhabdomyosarcoma, soft tissue sarcomas and Wilms tumor.

Based on the origin and era of actinomycin D’s discoveryand clinical evaluation, now-standard physiochemical andpreclinical evaluation were not performed in a comprehensivemanner. Despite its ongoing use in the above-mentionedchildhood sarcomas, much information is still lacking.Unfortunately, the pediatric patient has been the recipient ofthis ignorance as patient management on actinomycin D is acontinual dilemma. At the root of this problem is theunknown relationship between dose and exposure. At present,dosing guidance provided by the drug sponsor cites an adultdosage of 500 µg daily i.v.for a maximum of 5 days. The dos-age for adults or children is recommended not to exceed15 µg/kg or 400 – 600 µ/m2 meter of body surface daily basedon the 5-day intravenous regimen. The only guidance for dos-ing special populations recommends that the calculation ofthe dosage for obese or edematous patients should be on thebasis of surface area in an effort to relate dosage to lean bodymass. Unfortunately, none of the dose information/guidanceprovided in based on the PKs of actinomycin D.

As a result, close clinical evaluation and monitoring ofpatient response to actinomycin D is critical to maintain ade-quate therapy and only partially successful in managingpatient care. Toxic reactions following actinomycin D admin-istration are frequent and may be severe, limiting in manyinstances the amount that may be administered. Deaths havebeen reported as well. The severity of toxicity varies markedlyand is only partly dependent on the dose used. One hypothe-ses yet untested would be that the time course and/or severityof such reactions more closely correlates with individualpatient exposure. Again, formal guidance provided in thepackage insert states that the dosage of actinomycin D variesdepending on the tolerance of the patient, the size and loca-tion of the neoplasm, and the use of other forms of therapy. Itfurther advises that it may be necessary to decrease the usualdosages suggested when additional chemotherapy or radiationtherapy is used concomitantly or has been used previously. Anadditional concern is that a wide variety of single agent andcombination chemotherapy regimens with actinomycin D areused. Because chemotherapeutic regimens are constantlychanging, dosing and administration mandates direct supervi-sion of physicians familiar with current oncologic practicesand new advances in therapy often without the benefit ofanticipated drug interaction potential.

In 2002 the Children’s Oncology Group (COG) sus-pended three active protocols for the treatment of childrenwith rhabdomyosarcoma after four actinomycin-associateddeaths from hepatotoxicity. Despite this event, actinomycinD is an integral component of rhabdomyosarcoma andWilms’ tumor therapy and pediatric oncologists continue toadminister the drug despite the gap in knowledge. Followingthe award to the COG through the NCI in September of2004, U10 CA098543-0251, ‘actinomycin-D/vincristineexposure–response characterisation in children with cancer,’a comprehensive strategy to identify critical factors thatdefine the therapeutic window for these agents in children



has been undertaken. The original proposal contains fourprojects, one of which has the objective of simulating expo-sure–response and ultimately clinical outcomes resultantfrom the assembly of models that define dose–exposure,exposure–response and response–outcome relationships forvarious trial designs. Infants under the age of 12 months areknown to exhibit higher toxicity rates than older childrenadministered equivalent body-weight adjusted doses ofactinomycin D. Likewise, it is not known at present if thishigher toxicity rate is due to dispositional differences indeveloping infants. Elucidating these potential differences inactinomycin PKs would be a key aspect in providing dosingguidance in this population.

Given the limited knowledge of actinomycin D PKs, amodel-based approach to designing such a study would befeasible. The objectives of the modeling exercise conductedby the authors’ laboratory were to: i) construct a popula-tion PK model to describe actinomycin D disposition inchildren and young adults; ii) perform clinical trial simula-tions incorporating parameter uncertainty for the designand evaluation of a prospective large-scale actinomycintrial in pediatric cancer patients and subsequent sensitivityanalysis; and iii) perform simulations to calculate thenumber of subjects under the age of 1 year old required toreliably estimate a clinically meaningful change in clear-ance. Hence, the application of the MBDD approach inthis setting is based on the needs assessment regarding the‘unknowns’ which have confused the ability to managepatient pharmacotherapy. The approach is focussed on pri-oritisation of the available (albeit limited) prior informa-tion to construct models exploring the relationships ofinterest (e.g., exposure-toxicity) and guide targetedexperimentation when critical to model assumptions.

Distribution studies of actinomycin D in rat, monkeyand dog had shown that equivalent doses based on bodysurface area (BSA) of 0.6 mg/m2 result in nearly equal tis-sue exposures in the three species [108]. This work wasexpanded further to characterise the distribution andkinetics of actinomycin D in the beagle dog using aflow-limited PB/PK model [109]. Early published humanactinomycin D PK studies consisted of actinomycin D inthree adult cancer patients at a dose of 15 µg/kg [110] andtwo doses in children of 0.75 mg/m2 (n = 1) and1.5 mg/m2 (n = 1) [111]. More recently, Veal [112] has pub-lished preliminary PK results in children receiving actino-mycin D. This limited data represented the priorinformation on dose–exposure relationships withactinomycin D.

Initially, PB/PK models were constructed from physiologicdata in children with the actinomycin D PK scaled from thebeagle. Specifically, mean parameter values for organ bloodflows (Q) and organ volumes (V) and actinomycin D specificPK data from the respective animal models were used to con-struct pediatric PB/PK models [113,114]. Intersubject variabilityaround physiologic parameters (e.g., partition coefficients,

etc.) was assumed to incorporate a measure of precisionaround the predicted exposures. This model agreed well withthe pediatric data obtained in pilot studies ongoing at TheChildren’s Hospital of Philadelphia [115] and will be revisitedon historical toxicity data from previous clinical trials that canbe mined and indexed to dose and demographic subgroups.

Additional population PK models were developed fromactual pediatric PK data recently made available. PK data(33 patients from the United Kingdom Children’s CancerStudy Group, UKCCSG) were provided by Veal et al. [112].This data (165 plasma concentration observations) were col-lected from children aged 1.58 – 20.3 years receiving actino-mycin D as part of their standard chemotherapy (doses of0.7 – 1.5 mg/m2). A three-compartment model withfirst-order elimination was chosen as the structural modelfor the data. The population PK data was analysed usingnon-linear mixed-effects modeling. Actinomycin D andvincristine PK parameters were allometrically scaled by totalbody weight, normalised to a weight of 70 kg [116].

The final population PK model was used to simulate newstudies. Simulations incorporated uncertainty in the parame-ter estimates by probability density functions for all modelparameters. The uncertainty in parameters was implementedas intertrial variability. From these distributions, 500 sets ofpopulation PK parameters were simulated [116]. The 500parameter sets and the final population PK model were usedto simulate 500 replicate data sets for a preliminary clinicaltrial construct. The original PK model was then fitted to eachreplicate data set and the bias and precision in parameter esti-mation was evaluated for a given trial design. Deficiencies inthe trial designed were then identified and iteratively refineduntil a construct was identified that produced minimal biasfor all PK parameters, particularly total drug clearance and thevolume of distribution in the central compartment (Figure 5).

The simulation study yielded a final design consisting of200 patients, which will be randomised to one of two sam-pling schemes. Once the study design was defined, ahypothetical age effect was incorporated using a power modelto describe the possible maturational changes in clearance forchildren < 1 year old [116]. The results from a simulation studythat included 20 subjects below the age of 1 were used to con-struct sample size curves over the range of possible clearancevalues. From the sample size analysis, it was concluded that 50subjects under the age of 1 would be required to detect a 30%change in clearance in children under the age of 1 (Figure 6).Simulations performed to confirm this sample size showedthat CL could be accurately estimated in children under1 year old.

A feasible and informative trial design was identified for anactinomycin and vincristine clinical trial in pediatric cancerpatients. The results of this effort have been incorporated intoa prospective trial protocol to be conducted through theCOG’s Phase I Consortium. PK, safety and clinical responsedata will be evaluated with the existing mixed-effects modeland the final population PK analysis derived from this study



will be used to revise actinomycin D labeling. Whereas theeffort has been uniquely focussed on deriving dosing guidancefor pediatric patients, the MBDD approach applied in thissetting has also facilitated experimentation which hasenhanced the overall information content on this importantagent. Sensitive and specific analytical methods weredeveloped for use in pediatric populations [115,117,118],information on metabolism and protein binding [115] andmethodologies to enhance patient enrolment [115] were devel-oped as outcomes of the MBDD approach in that theyremoved uncertainty in the respective models that they areused in (PK and trial design models, respectively). Thisexample should also highlight the notion that this approach isconfined to a big Pharma budget as all of this research isfunded by the NIH [119].

9. Conclusion

The term ‘model-based drug development’ has been used torefer to the approach of constructing quantitativerelationships to define and explore key decision criteria thatconstitute development milestones or key hurdles for adevelopment compound. The approach has received theendorsement of the FDA [120] and the global regulatory com-munity [70] and is various stages of implementation at manylarge Pharma companies. Oncologic drug targets offer andexcellent application to this approach. The application ofmodels in oncology has been a necessity for some time giventhe toxicity of many of the drug classes, the severity of theindications and the status of the patients. Given: i) howpoorly characterized some of the older agents are; ii) the

Figure 5. Bias and precision in parameter estimation from final trial design based on evaluation of 500 replicate data setsgenerated from 500 sets of PK parameters simulated from a population PK/PD model for actinomycin D. Individual parametersfrom the population model are shown on the x-axis.CL: Systemic clearance; OMCL: Intersubject variation in CL; OMV1: Intersubject variation in V1; Q2: Intercompartment 2 clearance; Q3: Intercompartment 3 clearance;V1: Central compartment volume of distribution; V2: Peripheral compartment 2 distribution volume; V3: Peripheral compartment 3 distribution volume.

V1 V2 V3 CL Q2 Q3 OM V1 OM CL

-50

-30

-10

10

30

50

% B

ias



reality of multimodal strategies to treat the variety of cancertypes; and iii) the heterogeneity of the various patient popula-tions, a MBDD paradigm must, of course, accommodateexisting agents for which there is often sparse data or limitedbridging of translational information. This presents an addi-tional challenge to pharmaceutical sponsors as they mustoften ‘back-fill’ information gaps for historical agents devel-oped when development programs, particularly for oncologydrugs, were less comprehensive.

10. Expert opinion

Aspects of MBDD, particularly those pertaining to M&Sapproaches and techniques, have been a part of drugdevelopment for some time [19]. The FDA’s Critical Path doc-ument [204] and NIH Roadmap [70] were important for thepublic recognition of what is clear to many – the nextgeneration of medicines will require more rigorous preclinicaland clinical investigation. Consumer tolerance is low withrespect to adverse effects and adverse drug reactions

appreciated for the first time after the drug is marketed (espe-cially when there was safety signals identified early in develop-ment). Hence, scientific and ethical accountability are notonly called into question post approval, but examined criti-cally along the drug development pathway as never before.The potential for post marketing litigation has also raised cor-porate awareness regarding record-keeping and likewisedecision making.

Therefore, MBDD in addition to its obvious use as a drugdevelopment paradigm, provides a mechanism to examinethe traceability of drug development decisions. One wouldassume that good science needs no conscious. However, thevast amount of data generated in support of NCEs, the dataand knowledge gaps regarding historic agents and thecomplexity of modern disease targets provides a challenge tothe brightest scientists. The MBDD approach necessitatesmultidisciplinary project teams and, more importantly,demands new skill sets to be nurtured. The Critical Pathdocument clearly identifies clinical pharmacology as anessential discipline for the future given that dose finding has

Figure 6. Power sample size analysis required to detect a 30% change in clearance in children under the age of 1 based ontrial simulation model. Individual curves for 70, 80 and 90% power shown. The x-axis refers to the actual difference in actinomycinclearance between children < and ≥ 1 year of age.

1 2 3 4 5 6

∆ CL (L/h)

0

100

200

300

400

70% Power

80% Power

90% Power

n p

er g

rou

p



been poorly performed in many regulatory submissions andis often the culprit when drugs are removed from themarket. Most important is that the clinical pharmacologydiscipline evolves to appreciate and exploit the quantitativetechniques that MBDD relies on. The additional challengeto drug developers seeking to ‘sell’ this approach is tomaintain the focus on improving the information content ofthe NDA. There are obvious economic gains to be made indevelopment efficiency both from the standpoint ofomitting less informative as well as redundant trials wheremodel-based approaches can be relied on. However, thesesame models can and have been used to skip Phase IIA/Btrials based on presumed acceptable risk: benefit and thedesire to accelerate Phase III trials. This is contrary to thesentiments of the Critical Path document and can only beviewed as a void in the ability to defend a proposed dosageand/or regimen.

The molecularly targeted mechanisms that define manypresent oncology pipelines necessitate a much greaterdependence on informatics approaches. Despite this feature,there remains a reluctance to connect gene- and cell-basedresults to in vivo data. Some of this reluctance is due to poor

correlation between xenograft models and clinical data. How-ever, quite often, the reason is also due to experimental condi-tions that are neither accurately defined nor standardized, orpoor experimental designs that prohibit data pooling requiredfor generalizations across molecular entities. The root cause inmost cases is that investigators are comfortable with makingdecisions regarding discrete experiments and unwilling toinvest the time to understand how experimental data can belinked especially if it involves a challenge to establishedexperimental methods. A major hurdle in the extensionof MBDD to early development phases will be theappreciation of these factors and the ‘leap of faith’ required tochange preclinical decision making. This is essential to deliv-ering on the promise of translational medicine as well giventhat many among the academic medical research communityare participating in oncology-based research in closecollaboration with the pharmaceutical industry.

Acknowledgements

This work was supported in part by the NIH/NCI,U10CA098543 – 0251.

Bibliography1. CASALI PG, BERTULLI R,

FUMAGALLI E et al.: Some lessons learned from imatinib mesylate clinical development in gastrointestinal stromal tumors. J. Chemother. (2004) 16 Suppl. 4:55-58.

2. VEYRAT-FOLLET C, BRUNO R, OLIVARES R, RHODES GR, CHAIKIN P: Clinical trial simulation of docetaxel in patients with cancer as a tool for dosage optimization. Clin. Pharmacol. Ther. (2000) 68(6):677-687.

3. BISSERY MC, NOHYNEK G, SANDERINK GJ, LAVELLE F: Docetaxel (Taxotere): a review of preclinical and clinical experience. Part I: Preclinical experience. Anticancer Drugs (1995) 6(3):339-55, 363-368.

4. BRUNO R, HILLE D,RIVA A et al.: Population pharmacokinetics/pharmacodynamics of docetaxel in Phase II studies in patients with cancer. J. Clin. Oncol. (1998) 16(1):187-196.

5. GALLO JM, LAUB PB, ROWINSKY EK, GROCHOW LB, BAKER SD: Population pharmacokinetic model for topotecan derived from Phase I clinical trials. J. Clin. Oncol. (2000) 18(12):2459-2467.

6. LEGER F, LOOS WJ, FOURCADE J et al.: Factors affecting pharmacokinetic variability of oral topotecan: a population analysis. Br. J. Cancer (2004) 90(2):343-347.

7. MOULD DR, HOLFORD NH, SCHELLENS JH et al.: Population pharmacokinetic and adverse event analysis of topotecan in patients with solid tumors. Clin. Pharmacol. Ther. (2002) 71(5):334-348.

8. FABBRO D, RUETZ S, BUCHDUNGER E et al.: Protein kinases as targets for anticancer agents: from inhibitors to useful drugs. Pharmacol. Ther. (2002) 93(2-3):79-98.

9. JUDSON I, MA P, PENG B et al.: Imatinib pharmacokinetics in patients with gastrointestinal stromal tumour: a retrospective population pharmacokinetic study over time. EORTC Soft Tissue and Bone Sarcoma Group. Cancer Chemother. Pharmacol. (2005) 55(4):379-386.

10. CHATTERJEE SK, ZETTER BR: Cancer biomarkers: knowing the present and predicting the future. Future Oncol. (2005) 1(1):37-50.

11. KELLOFF GJ, LIPPMAN SM, DANNENBERG AJ et al.: Progress in chemoprevention drug development: the promise of molecular biomarkers for

prevention of intraepithelial neoplasia and cancer – a plan to move forward. Clin. Cancer Res. (2006) 12(12):3661-3697.

12. EFSTATHIOU JA, CHEN MH, CATALONA WJ et al.: Prostate-specific antigen-based serial screening may decrease prostate cancer-specific mortality. Urology (2006) 68(2):342-347.

13. WANG Y, HARADA M, YANO M et al.: Prostatic acid phosphatase as a target molecule in specific immunotherapy for patients with nonprostate adenocarcinoma. J. Immunother. (2005) 28(6):535-541.

14. LEONARD GD, LOW JA, BERMAN AW, SWAIN SM: CA 125 elevation in breast cancer: a case report and review of the literature. Breast J. (2004) 10(2):146-149.

15. TAMPELLINI M, BERRUTI A, BITOSSI R et al.: Prognostic significance of changes in CA 15-3 serum levels during chemotherapy in metastatic breast cancer patients. Breast Cancer Res. Treat. (2006) 98(3):241-248.

16. FOX, E, GA CURT, BALIS FM: Clinical trial design for target-based therapy. Oncologist (2002) 7(5):401-409.

17. CHIEN JY, FRIEDRICH S, HEATHMAN MA, DE ALWIS DP, V SINHA: Pharmacokinetics/pharmacodynamics and the stages of drug development: role of



modeling and simulation. Aaps. J. (2005) 7(3):E544-559.

18. VAN KESTEREN C, MATHOT RA, BEIJNEN JH, SCHELLENS JH: Pharmacokinetic-pharmacodynamic guided trial design in oncology. Invest. New Drugs (2003) 21(2):225-241.

19. ROMBOUT F, AARONS L, KARLSSON M et al.: Modeling and simulation in the development and use of anti-cancer agents: an underused tool? J. Pharmacokinet. Pharmacodyn. (2004) 31(6):419-440.

20. SUGGITT M, BIBBY MC: 50 years of preclinical anticancer drug screening: empirical to target-driven approaches.Clin. Cancer Res. (2005) 11(3):971-981.

21. KARLSSON MO, ANEHALL T, FRIBERG LE et al.: Pharmacokinetic/pharmacodynamic modeling in oncological drug development. Basic Clin. Pharmacol. Toxicol. (2005) 96(3):206-211.

22. STRAETEMANS R, O'BRIEN T, WOUTERS L et al.: Design and analysis of drug combination experiments. Biom. J. (2005) 47(3):299-308.

23. BEIJNEN JH, SCHELLENS JH: Drug interactions in oncology. Lancet Oncol. (2004) 5(8):489-496.

24. WANG CX, KOAY DC, EDWARDS A et al.: In vitro and in vivo effects of combination of Trastuzumab (herceptin) and Tamoxifen in breast cancer. Breast Cancer Res. Treat. (2005) 92(3):251-263.

25. GITLER MS, MONKS A, SAUSVILLE EA: Preclinical models for defining efficacy of drug combinations: mapping the road to the clinic. Mol. Cancer Ther. (2003) 2(9):929-932.

26. MCLEOD HL: Clinically relevant drug-drug interactions in oncology. Br. J. Clin. Pharmacol. (1998) 45(6):539-544.

27. ROCCHETTI M, POGGESI I, GERMANI M et al.:A pharmacokinetic–pharmacodynamic model for predicting tumour growth inhibition in mice: a useful tool in oncology drug development. Basic Clin. Pharmacol. Toxicol. (2005) 96(3):265-268.

28. QUARANTA V, WEAVER AM, CUMMINGS PT, ANDERSON AR: Mathematical modeling of cancer: the future of prognosis and treatment. Clin. Chim. Acta. (2005) 357(2):173-179.

29. BORKENSTEIN K, LEVEGRUN S, PESCHKE P: Modeling and computer simulations of tumor growth and tumor response to radiotherapy. Radiat. Res. (2004) 162(1):71-83.

30. LIANG H, SHA N: Modeling antitumor activity by using a non-linear mixed-effects model. Math. Biosci. (2004) 189(1):61-73.

31. SIMEONI, M, MAGNI P, CAMMIA Cet al.: Predictive pharmacokinetic–pharmacodynamic modeling of tumor growth kinetics in xenograft models after administration of anticancer agents. Cancer. Res. (2004) 64(3):1094-1101.

32. ROUITS, E, GUICHARD S, CANAL P, CHATELUT E: Non-linear pharmacokinetics of irinotecan in mice. Anticancer Drugs (2002) 13(6):631-635.

33. ROWLAND M, BALANT L, PECK L: Physiologically based pharmacokinetics in drug development and regulatory science: a workshop report (Georgetown University, Washington, DC, May 29-30, 2002) AAPS PharmSci (2004) 6(1):E6.

34. PARROTT N, JONES H, PAQUEREAU N, LAVE T: Application of full physiological models for pharmaceutical drug candidate selection and extrapolation of pharmacokinetics to man. Basic. Clin. Pharmacol. Toxicol. (2005) 96(3):193-199.

35. BLESCH KS, R GIESCHKE, TSUKAMOTO Y et al.: Clinical pharmacokinetic/pharmacodynamic and physiologically based pharmacokinetic modeling in new drug development: the capecitabine experience. Invest. New Drugs (2003) 21(2):195-223.

36. THEIL FP, GUENTERT TW, HADDAD S, POULIN P: Utility of physiologically based pharmacokinetic models to drug development and rational drug discovery candidate selection. Toxicol. Lett (2003) 138(1-2):29-49.

37. TSUKAMOTO Y, KATO Y, URA M et al.: A physiologically based pharmacokinetic analysis of capecitabine, a triple prodrug of 5-FU, in humans: the mechanism for tumor-selective accumulation of 5-FU. Pharm. Res. (2001) 18(8):1190-1202.

38. BAXTER LT, ZHU H, MACKENSEN DG, JAIN RK: Physiologically based pharmacokinetic model for specific and nonspecific monoclonal antibodies and fragments in normal tissues and human tumor xenografts

in nude mice. Cancer Res. (1994) 54(6):1517-1528.

39. KELLOFF GJ,KROHN KA, LARSON SM et al.: The progress and promise of molecular imaging probes in oncologic drug development. Clin. Cancer Res. (2005) 11(22):7967-7985.

40. KELLOFF GJ, SIGMAN CC: New science-based endpoints to accelerate oncology drug development. Eur. J. Cancer (2005) 41(4):491-501.

41. FLOYD E, MCSHANE TM: Development and use of biomarkers in oncology drug development. Toxicol. Pathol. (2004) 32 Suppl. 1:106-115.

42. SEDDON BM, WORKMAN M: The role of functional and molecular imaging in cancer drug discovery and development. Br. J. Radiol. (2003) 76 Spec No 2:S128-138.

43. COLLINS, JM, GRIESHABER CK, CHABNER BA: Pharmacologically guided phase I clinical trials based upon preclinical drug development. J. Natl. Cancer Inst. (1990) 82(16):1321-1326.

44. GALLO, JM, VICINI P, ORLANSKY P: Pharmacokinetic model-predicted anticancer drug concentrations in human tumors. Clin. Cancer Res. (2004) 10(23):8048-8058.

45. MOUGENOT P, PINGUET P, FABBRO M et al.: Population pharmacokinetics of melphalan, infused over a 24-hour period, in patients with advanced malignancies. Cancer Chemother. Pharmacol. (2004) 53(6):503-512.

46. WADE, JR, KELMAN AW, KERR DJ, ROBERT J, WHITING B: Variability in the pharmacokinetics of epirubicin: a population analysis. Cancer Chemother. Pharmacol. (1992) 29(5):391-395.

47. CHATELUT E, BODDY AV, PENG Bet al.: Population pharmacokinetics of carboplatin in children. Clin. Pharmacol. Ther. (1996) 59(4):436-443.

48. DELALOGE S, LLOMBART A, PALMA M DI et al.: Gemcitabine in patients with solid tumors and renal impairment: a pharmacokinetic Phase I study. Am. J. Clin. Oncol . (2004) 27(3):289-293.

49. PEREZ-RUIXO JJ, ZANNIKOS P, OZDEMIR V et al.: Effect of CYP2D6 genetic polymorphism on the population



pharmacokinetics of tipifarnib. Cancer Chemother. Pharmacol. (2006) 58(5):681-691.

50. HENNINGSSON A, MARSH S, LOOS WJ et al.: Association of CYP2C8, CYP3A4, CYP3A5, and ABCB1 polymorphisms with the pharmacokinetics of paclitaxel. Clin. Cancer Res. (2005) 11(22):8097-8104.

51. XIE, H, GRISKEVICIUS L, STAHLE L et al.: Pharmacogenetics of cyclophosphamide in patients with hematological malignancies. Eur. J. Pharm. Sci. (2006) 27(1):54-61.

52. XIE R, MATHIJSSEN RH, SPARREBOOM A, VERWEIJ J, KARLSSON MO: Clinical pharmacokinetics of irinotecan and its metabolites: a population analysis.J. Clin. Oncol. (2002) 20(15):3293-3301.

53. CHABOT GG, ABIGERGES D, CATIMEL G et al.: Population pharmacokinetics and pharmacodynamics of irinotecan (CPT-11) and active metabolite SN-38 during Phase I trials. Ann. Oncol. (1995) 6(2):141-151.

54. KLEIN CE, GUPTA E, REID JM et al.: Population pharmacokinetic model for irinotecan and two of its metabolites, SN-38 and SN-38 glucuronide. Clin. Pharmacol. Ther. (2002) 72(6):638-647.

55. MA MK, ZAMBONI WC, RADOMSKI KM et al.: Pharmacokinetics of irinotecan and its metabolites SN-38 and APC in children with recurrent solid tumors after protracted low-dose irinotecan. Clin. Cancer Res. (2000) 6(3):813-819.

56. KERBUSCH, T, VANPUTTEN JW, GROEN HJ et al.: Population pharmacokinetics of ifosfamide and its 2- and 3-dechloroethylated and 4-hydroxylated metabolites in resistant small-cell lung cancer patients. Cancer Chemother. Pharmacol. (2001) 48(1):53-61.

57. DE JONGE ME, HUITEM AD, RODENHUIS S, BEIJNEN JH: Integrated population pharmacokinetic model of both cyclophosphamide and thiotepa suggesting a mutual drug–-drug interaction. J. Pharmacokinet. Pharmacodyn. (2004) 31(2):135-156.

58. CALLIES S, DE ALWIS DP, HARRIS A et al.: A population pharmacokinetic model for paclitaxel in the presence of a novel P-gp modulator, Zosuquidar Trihydrochloride

(LY335979) Br. J. Clin. Pharmacol. (2003) 56(1):46-56.

59. CALLIES S, DE ALWIS DP, WRIGHT JG et al.: A population pharmacokinetic model for doxorubicin and doxorubicinol in the presence of a novel MDR modulator, zosuquidar trihydrochloride (LY335979) Cancer. Chemother. Pharmacol. (2003) 51(2):107-118.

60. JOEL SP, PAPAMICHAEL D, RICHARDS F et al.: Lack of pharmacokinetic interaction between 5-fluorouracil and oxaliplatin. Clin. Pharmacol. Ther. (2004) 76(1):45-54.

61. FRIBERG LE, KARLSSON MO: Mechanistic models for myelosuppression. Invest. New Drugs (2003) 21(2):183-194.

62. VAN KESTEREN C, ZANDVLIET AS, KARLSSON MO et al.: Semi-physiological model describing the hematological toxicity of the anti-cancer agent indisulam. Invest. New Drugs (2005) 23(3):225-234.

63. LEGER F, LOOS WJ, BUGAT R et al.: Mechanism-based models for topotecan-induced neutropenia. Clin. Pharmacol. Ther. (2004) 76(6):567-578.

64. FRIBERG LE, HENNINGSSON H, MAAS H, NGUYEN L, KARLSSON MO: Model of chemotherapy-induced myelosuppression with parameter consistency across drugs. J. Clin. Oncol. (2002) 20(24):4713-4721.

65. BARRETT JS, PATEL D, SKOLNI S, ADAMSON PC: Discrete event simulation (DES) as a technique to study decision rule efficiency in event-driven clinical designs.J. Clin. Pharmacol. (2006) 46(9):1092.

66. KOWALSKI KG, HUTMACHER MM: Design evaluation for a population pharmacokinetic study using clinical trial simulations: a case study. Stat. Med. (2001) 20(1):75-91.

67. DE RIDDER F: Predicting the outcome of Phase III trials using Phase II data: a case study of clinical trial simulation in late stage drug development. Basic. Clin. Pharmacol. Toxicol. (2005) 96(3):235-241.

68. GIRARD P: Clinical trial simulation: a tool for understanding study failures and preventing them. Basic Clin. Pharmacol. Toxicol. (2005) 96(3):228-234.

69. JUMBE N, YAO B, ROVETTI R, ROSSI G, HEATHERINGTON HC: Clinical trial simulation of a 200-microg fixed dose of darbepoetin alfa in chemotherapy-induced anemia. Oncology

(Williston Park) (2002) 16(10 Suppl. 11):37-44.

70. BHATTARAM, VA, BOOTH BP, RAMCHANDANI RP et al.: Impact of pharmacometrics on drug approval and labeling decisions: a survey of 42 new drug applications. Aaps. J. (2005) 7(3):E503-512.

71. HOEKSTRA, R, VERWEIJ J, FESKENS FA: Clinical trial design for target specific anticancer agents. Invest. New Drugs (2003) 21(2):243-250.

72. OUELLET D, PERICLOU AP, JOHNSON RD, WOODWORTH JR, LALONDE RL: Population pharmacokinetics of pemetrexed disodium (ALIMTA) in patients with cancer. Cancer Chemother. Pharmacol. (2000) 46(3):227-234.

73. LATZ JE, CHAUDHARY A, GHOSH A, JOHNSON RD: Population pharmacokinetic analysis of ten Phase II clinical trials of pemetrexed in cancer patients. Cancer Chemother. Pharmacol. (2006) 57(4):401-411.

74. LATZ JE, KARLSSON MO, RUSTHOVEN JJ, GHOSH A, JOHNSON RD: A semimechanistic-physiologic population pharmacokinetic/pharmacodynamic model for neutropenia following pemetrexed therapy. Cancer Chemother. Pharmacol. (2006) 57(4):412-426.

75. LATZ, JE, RUSTHOVEN JJ, KARLSSON MO, GHOSH A, JOHNSON RD: Clinical application of a semimechanistic-physiologic population PK/PD model for neutropenia following pemetrexed therapy. Cancer Chemother. Pharmacol. (2006) 57(4):427-435.

76. GIESCHKE, R, BURGER HU, REIGNER B, BLESCH KS, STEIMER JL: Population pharmacokinetics and concentration-effect relationships of capecitabine metabolites in colorectal cancer patients. Br. J. Clin. Pharmacol. (2003) 55(3):252-263.

77. MOULD DR, FLEMING GF, DARCY KM, SPRIGGS D: Population analysis of a 24-h paclitaxel infusion in advanced endometrial cancer: a gynaecological oncology group study. Br. J. Clin. Pharmacol. (2006) 62(1):56-70.

78. BORCHERS, AT, KEEN CL, CHEEMA GS, GERSHWIN ME: The use of methotrexate in rheumatoid arthritis. Semin. Arthritis Rheum. (2004) 34(1):465-483.



79. PAULEY JL, PANETTA JC, SCHMIDT J et al.: Late-onset delayed excretion of methotrexate. Cancer Chemother. Pharmacol. (2004) 54(2):146-152.

80. PINCUS T: Guidelines for monitoring of methotrexate therapy: ‘evidence-based medicine’ outside of clinical trials. Arthritis Rheum. (2003) 48(10):2706-2709.

81. LINDLEY C, SHEA T, MCCUNE J et al.: Intraindividual variability in busulfan pharmacokinetics in patients undergoing a bone marrow transplant: assessment of a test dose and first dose strategy. Anticancer Drugs (2004) 15(5):453-459.

82. SANDSTROM M, KARLSSON MO, LJUNGMAN P et al.: Population pharmacokinetic analysis resulting in a tool for dose individualization of busulphan in bone marrow transplantation recipients. Bone Marrow Transplant (2001) 28(7):657-664.

83. BLEYZAC N, SOUILLET G, MAGRON P et al.: Improved clinical outcome of pediatric bone marrow recipients using a test dose and Bayesian pharmacokinetic individualization of busulfan dosage regimens. Bone Marrow Transplant (2001) 28(8):743-751.

84. ENGLISH MW, LOWIS SP, PENG B et al.: Pharmacokinetically guided dosing of carboplatin and etoposide during peritoneal dialysis and haemodialysis. Br. J. Cancer (1996) 73(6): 776-780.

85. PRIEST ER, VOGELZANG NJ: Optimal drug therapy in the treatment of testicular cancer. Drugs (1991) 42(1):52-64.

86. ROUSSEAU, A MARQUET P: Application of pharmacokinetic modeling to the routine therapeutic drug monitoring of anticancer drugs. Fundam. Clin. Pharmacol. (2002) 16(4):253-262.

87. GISBERT, JP, GOMOLLON F, MATE J, PAJARE JMS: [Individualized therapy with azathioprine or 6-mercaptopurine by monitoring thiopurine methyl-transferase (TPMT) activity]. Rev. Clin. Esp. (2002) 202(10):555-562.

88. SADEE W, WONG CG: Pharmacokinetics of 5-fluorouracil: inter-relationship with biochemical kinetics in monitoring therapy. Clin. Pharmacokinet. (1977) 2(6):437-450.

89. SANTINI, J, MILANO G, THYSS A et al.: 5-FU therapeutic monitoring with dose adjustment leads to an improved

therapeutic index in head and neck cancer. Br. J. Cancer (1989) 59(2):287-290.

90. SCHLEMMER, HP, BACHERT P, SEMMLER W et al.: Drug monitoring of 5-fluorouracil: in vivo 19F NMR study during 5-FU chemotherapy in patients with metastases of colorectal adenocarcinoma. Magn. Reson. Imaging (1994) 12(3):497-511.

91. MILLER AA, TOLLEY EA, NIELL HB: Therapeutic drug monitoring of 21-day oral etoposide in patients with advanced non-small cell lung cancer. Clin. Cancer Res. (1998) 4(7):1705-1710.

92. PFLUGER KH, SCHMIDT L, MERKEL M, JUNGCLAS H, HAVEMANN K: Drug monitoring of etoposide (VP16-213) Correlation of pharmacokinetic parameters to clinical and biochemical data from patients receiving etoposide. Cancer Chemother. Pharmacol. (1987) 20(1):59-66.

93. MONTAZERI A, CULINE S, LAGUERRE B et al.: Individual adaptive dosing of topotecan in ovarian cancer. Clin. Cancer Res. (2002) 8(2):394-399.

94. BOUCAUD M, PINGUET F, CULINE S et al.: Modeling plasma and saliva topotecan concentration time course using a population approach. Oncol. Res. (2003) 13(4):211-219.

95. RUSSEL J: FDA mulls drug/disease model library, Bio-IT World. (2005).

96. BRUNO R, RIVA A, HILLE D, LEBECQ A , THOMAS L: Pharmacokinetic and pharmacodynamic properties of docetaxel: results of phAse I and Phase II trials. Am. J. Health Syst. Pharm. (1997) 54(24 Suppl 2):S16-9.

97. CUNNINGHAM SL,CUNNINGHAM AR, DAY BW: CoMFA, HQSAR and molecular docking studies of butitaxel analogues with beta-tubulin. J. Mol. Model (2005) 11(1):48-54.

98. BISSETT D, SETANOIANS A, CASSIDY J et al.: Phase I and pharmacokinetic study of taxotere (RP 56976) administered as a 24-hour infusion. Cancer Res. (1993) 53(3):523-527.

99. EXTRA JM, ROUSSEAU F, BRUNO R et al.: Phase I and pharmacokinetic study of Taxotere (RP 56976; NSC 628503) given as a short intravenous infusion. Cancer Res. (1993) 53(5):1037-1042.

100. BURRIS H, R IRVIN, J KUHN et al.: Phase I clinical trial of taxotere administered

as either a 2-hour or 6-hour intravenous infusion. J Clin. Oncol. (1993) 11(5):950-958.

101. PAZDUR R, NEWMAN RA, NEWMAN BM et al.: Phase I trial of Taxotere: five-day schedule. J. Natl. Cancer Inst. (1992) 84(23):1781-1788.

102. TAGUCHI T, FURUE H, NIITANI H et al.: [Phase I clinical trial of RP 56976 (docetaxel) a new anticancer drug]. Gan To Kagaku Ryoho (1994) 21(12):1997-2005.

103. TOMIAK E, PICCART MJ, KERGER J et al.: Phase I study of docetaxel administered as a 1-hour intravenous infusion on a weekly basis. J. Clin. Oncol. (1994) 12(7):1458-1467.

104. LAUNAY-ILIADIS MC, BRUNO R, COSSON V et al.: Population pharmacokinetics of docetaxel during Phase I studies using nonlinear mixed-effect modeling and nonparametric maximum-likelihood estimation. Cancer Chemother. Pharmacol. (1995) 37(1-2):47-54.

105. BRUNO R, VIVIER N, VEYRAT-FOLLET C, MONTAY G, RHODES GR: Population pharmacokinetics and pharmacokinetic-pharmacodynamic relationships for docetaxel. Invest. New Drugs (2001) 19(2):163-169.

106. BRUNO R, VIVLER N, JVERGNIOL JC et al.: A population pharmacokinetic model for docetaxel (Taxotere): model building and validation. J. Pharmacokinet. Biopharm. (1996) 24(2):153-172.

107. WAKSMAN SA, WOODRUFF HB: The soil as a source of microorganisms antagonistic to disease-producing bacteria. J. Bacteriol .(1940) 40(4):581-600.

108. GALBRAITH WM, MELLETT LB: Tissue disposition of 3H-actinomycin D (NSC-3053) in the rat, monkey and dog. Cancer Chemother. Rep. (1975) 59(6):1601-1609.

109. LUTZ RJ, GALBRAITH WM, DEDRICK RM, SHRAGER R, MELLETT LB: A model for the kinetics of distribution of actinomycin-D in the beagle dog. J. Pharmacol. Exp. Ther. (1977) 200(3):469-478.

110. TATTERSALL MH, BROWN B, FREI E 3RD: The reversal of methotrexate toxicity by thymidine with maintenance of antitumour effects. Nature (1975) 253(5488):198-200.



111. VEAL, GJ, ERRINGTON J, SLUDDEN J et al.: Determination of anti-cancer drug actinomycin D in human plasma by liquid chromatography-mass spectrometry. J. Chromatogr. B. Analyt. Technol. Biomed. Life Sci. (2003) 795(2):237-243.

112. VEAL GJ, COLE M, ERRINGTON J et al.: Pharmacokinetics of dactinomycin in a pediatric patient population: a United Kingdom Children's Cancer Study Group Study. Clin. Cancer Res. (2005) 11(16):5893-5899.

113. MONDICK J, BARRETT J: A physiologically-based pharmacokinetic model to predict tissue distribution of actinomycin-D in humans. J. Clin. Pharmacol. (2004) 44:1213.

114. BARRETT, J, SKOLNIK J, GASTONGUAY M, ADAMSON P: The value of priors and prior uncertainty in clinical trial simulation: case study with actinomycin-D in children with cancer. in PAGE 13. 2004. Uppsala, Sweden.

115. SKOLNIK J, PACCALY D, BARRETT J, ADAMSON P: Mechanisms to enhance enrollment in pediatric trials: an in vitro procedure to clear residual chemotherapeutics from indwelling catheters. J. Clin. Pharmacol. (2005) 45:1085.

116. MONDICK J, GASTONGUAY M, GIBIANSKY L, BARRETT J: Acknowledging parameter uncertainty in the simulation-based design of an actinomycin-d pharmacokinetic study in pediatric patients with wilms' tumor or rhabdomyosarcoma. population approach group europe, 15th Annual Meeting. Bruuge, Belgium. (2006).

117. SKOLNIK J, BARRETT J, SHI H, ADAMSON P: Pre-clinical studies in support of the clinical pharmacologic evaluation of actinomycin-D in children with cancer. J. Clin. Pharmacol. (2004) 44:1192.

118. SKOLNIK JM, BARRETT JS, SHI H, ADAMSON PC: A liquid chromatography-tandem mass spectrometry method for the simultaneous quantification

of actinomycin-D and vincristine in children with cancer. Cancer Chemother. Pharmacol. (2006) 57(4):458-464.

119. BARRETT, J, MONDICK J, SKOLNIK J, ADAMSON P: Clinical trial simulation in an academic research setting: engaging the clinical team and generating the simulation plan. J. Clin. Pharmacol. (2005) 45:1087.

120. MILLER, R,EWY W, CORRIGAN BW et al.: How modeling and simulation have enhanced decision making in new drug development. J. Pharmacokinet. Pharmacodyn. (2005) 32(2):185-197.

121. SALINGER DH, MCCUNE JS, REN AG et al.: Real-time dose adjustment of cyclophosphamide in a preparative regimen for hematopoietic cell transplant: a Bayesian pharmacokinetic approach. Clin. Cancer. Res. (2006) 12(16):4888-4898.

122. FALTAOS DW, HULOT JS, URIEN S et al.: Population pharmacokinetic study of methotrexate in patients with lymphoid malignancy. Cancer Chemother. Pharmacol. (2006) 58(5):626-633.

123. WHITE DB, SLOCUM HK,BRUN Y, WRZOSEK C, GRECO WR: A new nonlinear mixture response surface paradigm for the study of synergism: a three drug example. Curr. Drug Metab. (2003) 4(5):399-409.

124. STROMGREN AS, SORENSEN BT, JAKOBSEN P, JAKOBSEN A: A limited sampling method for estimation of the etoposide area under the curve. Cancer Chemother. Pharmacol. (1993) 32(3):226-230.

125. TRANCHAND B, AMSELLEM C, CHATELUT E et al.: A limited-sampling strategy for estimation of etoposide pharmacokinetics in cancer patients. Cancer Chemother. Pharmacol. (1999) 43(4):F316-322.

126. HOLZ JB, KOPPLER H, SCHMIDT L et al.: Limited sampling models for reliable estimation of etoposide area under the curve. Eur. J. Cancer (1995) 31A(11):1794-1798.

127. LUM BL, LANE KJ, SYNOLD TW et al.: Validation of a limited sampling model to determine etoposide area under the curve. Pharmacotherapy (1997) 17(5):887-890.

128. RALPH LD, THOMSON AH, DOBBS NA, TWELVES C: A population model of epirubicin pharmacokinetics and application to dosage guidelines. Cancer Chemother. Pharmacol. (2003) 52(1):34-40.

129. ROBINSON BW, BEHLING KC, BARRETT JS et al.: Pre-clinical in vitro evaluation of BCL-2 antisense compound Genasense TM (G3139; Genta, Inc.) as a targeted pro-apoptotic agent for leukemias with t(4;11)(q21;q23) translocations. ASH Annual Meeting Abstracts (2005) 106(11):3360.

130. GUPTA M, ROBINSON B, FELIX C, BARRETT J: Strategies for assessment of drug combination therapy: response surface modeling of MTT assays to target agent synergy via cell survival. J. Clin. Pharmacol. (2005) 45:1093.

131. LAUNOIS R, REBOUL-MARTY J, HENRY B, BONNETERRE J: A cost-utility analysis of second-line chemotherapy in metastatic breast cancer. Docetaxel versus paclitaxel versus vinorelbine. Pharmacoeconomics (1996) 10(5):504-521.

Websites201. http://www.cancer.gov/newscenter/pressreleas

es/NciFdaCollabNational Cancer website (2003).

202. http://www.fda.gov/cder/cancer/index.htmFDA oncology tools website (2006).

203. http://www.bio-itworld.com/newsitems/2005/dec2005/12-19-05-news-fda-modelsBio-IT world website (2005).

204. http://www.fda.gov/oc/initiatives/criticalpath/whitepaper.htmlCDER: innovation and stagnation: challenge and opportunity on the critical path to new medicinal products. US Food and Drug Administration (2004).



AffiliationJeffrey S Barrett†1,2 PhD, FCP, Manish Gupta1,3,4 PhD & John T Mondick1 PhD†Author for correspondence1Laboratory for Applied PK/PD, Clinical Pharmacology & Therapeutics Division, The Children’s Hospital of Philadelphia, USATel: +1 215 590 7544;Fax: +1 67 426 5479;E-mail: [email protected] Department, University of Pennsylvania Medical School, USA3Pfizer Fellow, Pharmacometrics4Pfizer Pharmaceuticals, Ann Arbour, Michigan, USA

model-based drug development applied to oncologymodel-based drug development applied to oncology 186...

Documents