`
`A. David Rodrigues
`Bristol Myers Squibb Co.
`P.O.Box 4000
`Plinceton
`NJ 08543
`USA
`david.rodrigues @bms.com
`
`Editors
`K. Sandy Pang
`Leslie Dan Faculty of Pharmacy
`University of Toronto
`144 College Street
`Toronto ON M5S 3M2
`Canada
`ks.pang@utoronto.ca
`
`Rairnund M. Peter
`AstraZeneca UK
`Alderley Park
`Macclesfield, Cheshire
`United Kingdom SKlO 4TF
`raimund. peter@astrazeneca.com
`
`ISBN 978-1-4419-0839-1
`DOl 10.1 007/978-1-4419-0840-7
`Springer New York Dordrecht Heidelberg London
`
`e-ISBN 978-1-4419-0840-7
`
`Library of Congress Control Number: 2009938736
`
`© American Association of Pharmaceutical Scientists 20 I 0
`All rights reserved. This work may not be translated or copied in whole or in part without the written
`permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York,
`NY 10013, USA), except for blief excerpts in cormection with reviews or scholarly analysis. Use in
`connection with any foml of information storage and retrieval, electronic adaptation, computer software,
`or by similar or dissimilar methodology now known or hereafter developed is forbidden.
`The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are
`not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject
`to proprietary rights.
`While the advice and information in this book are believed to be true and accurate at the date of going
`to press, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
`elTors or omissions that may be made. The publisher makes no wananty, express or implied, with respect
`to the material contained herein.
`
`Printed on acid-free paper
`
`Springer is part of Springer Science+Business Media (www.springer.com)
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 2
`
`
`
`
`
`626
`
`D.1. Greenblatt and L.L. von Moltke
`
`1999; Greenblatt et aI., 2008). At the same time, in vitro techniques for studying
`human drug metabolism became increasingly developed and refined, including pre(cid:173)
`dictive models for in vitro-in vivo scaling, and the availability of heterologously
`expressed individual human CYPs. At a clinical level, polypharmacy was becoming
`increasingly prevalent, as the population aged, the number of patients with multi(cid:173)
`ple illnesses increased, and our capacity to provide pharmacologic treatments for
`serious disorders became more and more effective. Some newly introduced classes
`of medications - such as the azole antifungal agents and the selective serotonin
`reuptake inhibitor (SSRI) antidepressants - offered unique therapeutic options, but
`also had the secondary property of inhibiting certain human CYPs, thereby ele(cid:173)
`vating the risk of DDls (Greenblatt et aI., 1999; Hemeryck and Belpaire, 2002;
`Venkatakrishnan et aI., 2000). A dramatic and widely publicized event was the inter(cid:173)
`action of the nonsedating antihistamine terfenadine with potent CYP3A inhibitors
`such as ketoconazole and erythromycin (Honig et aI., 1993b; Honig et aI., 1992;
`Honig et aI., 1994; Honig et aI., 1993a). Under usual circumstances, terfenadine
`itself served only as a prodrug, being essentially completely transformed via hepatic
`and enteric CYP3A into fexofenadine, which was the entity having antihistaminic
`properties. Although terfenadine had effects on the cardiac QTc interval (Rampe et
`aI., 1993; Crumb et aI., 1995), this was of minimal concern since intact terfenadine
`does not ordinarily reach the systemic circulation. However, during co-treatment
`with CYP3A inhibitors, conversion of terfenadine to fexofenadine is blocked, and
`potentially hazardous levels of the parent drug reach the circulation (Honig et aI.,
`
`60
`
`..
`
`......
`
`50
`
`en z
`o
`~ o 40
`::i
`III
`::::>
`ll.
`is 30
`C
`LL o
`a: 20
`w
`III
`:2
`::::>
`z
`
`10
`
`1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
`YEAR
`
`Fig.24.1 Number of articles indexed as DDI studies published per year in the Journal of Clinical
`Pharmacology, 1990-2008
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 4
`
`
`
`24 Clinical Studies of Drug-Drug Interactions: Design and Interpretation
`
`627
`
`1994; von Moltke et ai., 1994b). A few cases of serious and even fatal cardiac
`arrhythmias were reported as a consequence (Monahan et aI., 1990; Woosley et al.,
`1993). The "terfenadine affair" led to an acutely increased awareness of the poten(cid:173)
`tial importance of DDIs. Terfenadine was withdrawn from clinical practice, and a
`number of regulatory reforms increased the requirements for DDI assessments as
`a component of drug development. The overall shift in focus of the scientific and
`drug development community is clearly evident from the prevalence of DDI studies
`among scientific publications (Fig. 24.1).
`
`24.2 Epidemiology of Drug-Drug Interactions
`
`Given the prevalence of polypharmacy in contemporary clinical practice, the num(cid:173)
`ber of possible DDIs can become very large. If an individual patient is taking 11
`drugs concurrently, the number of pairwise combinations of these two drugs can be
`calculated as follows:
`
`n1
`(n - 2)!2!
`
`(24.1)
`
`The larger the value of 17, the greater the number of different drug combi(cid:173)
`nation pairs, and potential pairwise DDIs (Table 24.1). A patient with diabetes,
`hypertension, ischemic heart disease, and depression might well be taking 10
`drugs concurrently, in which case the number of possible drug interactions is 45.
`Considering this large "denominator" of possibilities, the number of clinically
`important DDIs encountered in contemporary therapeutics actually is relatively
`small.
`
`Table 24.1 Relation of
`number of drugs concunently
`administered to the number of
`possible pairwise drug-drug
`interactions
`
`Number of drugs
`
`Possible pairwise drug interactions
`
`2
`3
`4
`5
`6
`7
`8
`9
`10
`11
`12
`
`1
`3
`6
`10
`15
`21
`28
`36
`45
`5S
`66
`
`The outcome options following concurrent administration of two drugs can be
`constructed based on a probability hierarchy (Fig. 24.2). The most probable outcome
`is that the two drugs act independently, with no evidence of any interaction. Less
`probable is a DDI which can be demonstrated in a controlled laboratory setting,
`but is not detectable in clinical practice either because the magnitude of the change
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 5
`
`
`
`
`
`24 Clinical Studies of Drug-Drug Interactions: Design and Interpretation
`
`629
`
`with CYP2D6 substrate drugs such as desipramine (Hemeryck and Belpaire, 2002;
`von Moltke et al., 1994a; von Moltke et al., 1995; Preskorn et al., 1994; Alderman
`et a1., 1997). Yet clinically important drug interactions are rarely reported in actual
`practice (Davies et al., 2004; deVane, 2006; Molden et al., 2005). One possible
`explanation is that the therapeutic index of the victim drug or drugs is large enough
`that even a substantial change in plasma levels is not clinically relevant. Another
`explanation is that clinicians recognize the potential DDI, and make a pre-emptive
`downward adjustment in the dose of the victim to prevent the DDI.
`
`24.3 Drug Interaction Mechanisms and Terminology
`
`We have used the term "perpetrator" to indicate the drug that is causing the DDI,
`while "victim" or "substrate" is the drug that is being interacted with (Greenblatt
`and von Moltke, 2008). In a pure phmmacodynamic DDI, the perpetrator does
`not alter the plasma concentrations or systemic pharmacokinetics of the victim.
`Instead, the two drugs produce either additive or antagonistic pharmacodynamic
`effects. The interaction may occur via additive or opposite actions on the same
`receptor systems that yield additive or opposite clinical actions. Ethyl alcohol and
`benzodiazepines produce additive sedation through their actions on the gamma(cid:173)
`aminobutyric acid (GAB A) receptor system (Chan, 1984; Greenblatt and von
`Moltke, 2008); the pharmacokinetic interaction between alcohol and benzodi(cid:173)
`azepines, if any, is small, and does not account for the additive sedative effects
`(Greenblatt et al., 1978; Greenblatt and von Moltke, 2008; Ochs et a1., 1984;).
`Benzodiazepine agonists and caffeine have antagonistic pharmacodynamic actions.
`Benzodiazepines produce sedation via the GABA-benzodiazepine receptor sys(cid:173)
`tem, whereas caffeine produces alertness due to its action as an adenosine receptor
`antagonist (Biaggioni et al., 1991; Kaplan et al., 1992a, b; Kaplan et al., 1993).
`When caffeine is given together with a benzodiazepine agonist such as zolpidem,
`the sedative effects of zolpidem are partially reversed (Cysneiros et aI., 2007).
`However, there is minimal, if any, pharmacokinetic interaction between these two
`agents.
`A pure phmmacokinetic interaction involves only the effect of the perpetrator
`on the systemic clearance of the victim drug, causing plasma levels of the victim
`to increase or decrease. The clinical actions of the victim may be correspondingly
`increased or decreased, but only because of the indirect effects of the perpetrator
`on systemic clearance, rather than a direct effect of the perpetrator on the target
`receptor mediating clinical action.
`Pharmacokinetic DDIs involving drug-metabolizing enzyme systems (such as
`the CYPs) are generally classified as inhibition or induction. With metabolic inhibi(cid:173)
`tion, the perpetrator impairs the clearance of the victim drug, systemic exposure
`increases, and the clinical concern is toxicity. With induction, clearance of the
`victim increases, systemic exposure decreases, and the clinical concern is lack of
`efficacy (Table 24.2). However, inhibition and induction are not simply the same
`process in opposite directions - they involve fundamentally different mechanisms.
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 7
`
`
`
`630
`
`D.J. Greenblatt and L.L. von Moltke
`
`Inhibition of victim drug clearance happens rapidly upon exposure to the perpe(cid:173)
`trator, and represents a direct effect of the perpetrator on the drug-metabolizing
`enzyme. Metabolic inhibition can be studied in vitro using cell homogenates from
`human liver, or cells expressing human metabolic enzymes (Venkatakrishnan et aI.,
`2001; Venkatakrishnan et aI., 2003). From these in vitro systems itis straightforward
`to derive metrics of inhibitory potency such as the inhibition constant (Kj) or the
`50% inhibitory concentration (ICso). In contrast, induction is an indirect process -
`the perpetrator (inducer) initiates a signal for the cell to produce more metabolic
`protein. This is slower than inhibition, and requires cultures of intact cells to study
`in vitro. The metric of induction potency is not so straightforward. Generally, the
`inductive effect of a candidate inducer is expressed as the fractional degree of induc(cid:173)
`tion relative to the hypothetical "maximum" induction by an index inducer such as
`rifampin.
`
`Table 24.2 Comparison of metabolic inhibition and induction
`
`Inhibition
`
`Induction
`
`Effect on victim drug
`Clearance
`Plasma levels
`Principal clinical concern
`Onset (after exposure to
`perpetrator)
`Offset (after perpetrator is
`discontinued)
`Mechanism
`
`In vitro system
`Metric of potency in vitro
`
`Decreased
`Increased
`Toxicity
`Rapid
`
`Rapid
`
`Increased
`Decreased
`Loss of efficacy
`Slow
`
`Slow
`
`Direct chemical
`effect
`Cell homogenates
`Ki or ICso
`
`Indirect signal to increase
`protein synthesis
`Cell culture
`Induction relative to maximum
`
`24.4 The Design of Clinical Drug Interaction Studies
`
`The general objective of DDI studies is to answer the following scientific questions:
`
`1. Given candidate "victim" and "perpetrator" drugs, is there a pharmacokinetic
`interaction between these two drugs that is not a chance event?
`2. What is the magnitude of the pharmacokinetic interaction?
`3. Is the interaction likely to be of clinical importance?
`
`Answers to the first two questions are largely objective and numerical, with little
`need for subjective interpretation or supplemental information. The third question is
`different - unless the DDI study incorporates measures of pharmacodynamic effect
`that are applicable to the target patient population, some supplemental information
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 8
`
`
`
`24 Clinical Studies of Drug-Drug Interactions: Design and Interpretation
`
`631
`
`on the exposure-response relationship for the victim substrate drug is needed before
`a judgment can be made.
`
`24.4.1 Study Rationale
`
`The majority of clinical DDI studies involve healthy volunteers who do not have
`a medical need for the drugs under study. As such, study participation for these
`individuals is of no clinical benefit, but does entail some risk (though presumably
`low, and acceptable to an Institutional Review Board). There is also a dollar cost
`involved in the conduct of DDI studies. The cost is borne by the pharmaceutical
`sponsor in the case of an investigational dmg, by the general public in the case of an
`NIH-supported study, or by some other entity. The core assumption is that the risk
`and cost of the DDI study are justified based on the potential public health benefit
`of the information to be acquired.
`Clinical observations raising the possibility of a DDI may form the basis for
`initiating a formal study to either confirm or rule out a DDI. In the course of drug
`development, in vitro data is commonly used to identify dmg pairs for which a DDI
`needs to be evaluated in a clinical study. "Drug X" may be identified as an inhibitor
`of a certain human CYP isofoml in vitro, with a quantitative potency metric of Ki or
`ICso. If [1] is a typical plasma concentration of Drug X encountered during treatment
`with the highest approved dosage, then the ratio [1]/Ki or [1]/ICso is used to judge
`whether a clinical DDI is unlikely, possible, or probable, based on FDA guidelines.
`A DDI is termed "possible" if
`
`[I]/Ki > 0.1
`
`(24.2)
`
`This boundary is arguably too conservative on scientific grounds and triggers a
`large number of clinical DDI studies which tum out to be negative. Nonetheless that
`boundary reflects the current regulatory outlook, and sponsors often will initiate a
`DDI study on that basis.
`A second category of rationale for DDI studies is not directly scientific, but rather
`epidemiologic, based on a high probability of concurrent dmg therapy. Dmg X may
`be under development for a medical condition (such as diabetes, hypertension, or
`hyperlipidemia) that has high co-morbidity with ischemic heart disease. The sponsor
`may choose to initiate DDI studies of Drug X with digoxin or with warfarin because
`the probability of concurrent therapy is high, and because digoxin and warfarin (as
`potential victim dmgs) have a narrow therapeutic index. Even if there is no direct
`scientific rationale raising the possibility of a DDI, it could be argued that clinical
`data excluding DDIs with digoxin or walfarin is needed to assure safe co-treatment
`of Drug X with these potentially hazardous medications.
`Finally, an inevitable consequence of the industry-based system of drug devel(cid:173)
`opment is that research may be initiated solely for business reasons. Within a given
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 9
`
`
`
`632
`
`D.1. Greenblatt and L.L. von Moltke
`
`drug class, a number of therapeutic options may be available, for which differences
`in therapeutic efficacy or toxicity may be subtle at most (Table 24.3). Competitive
`advantage then may turn on pharmacokinetic properties, such as mechanism of
`clearance, elimination half-life, and the risk of DDls. A clinical study may be initi(cid:173)
`ated to show that the sponsor's Drug X is not an inhibitor of a specific CYP isoforn1,
`whereas a competitor drug within the same class is in fact a significant inhibitor of
`the same CYP. These properties can be included in a product label, and used by
`pharmaceutical representatives or advertising matelials for competitive advantage.
`An example is the interaction of macrolide antimicrobials with human CYP3A.
`Erythromycin, clarithromycin, and telithromycin are significant CYP3A inhibitors,
`whereas azithromycin is not (Greenblatt et al., 1998a).
`
`Table 24.3 Examples of
`drug classes for which
`individual drugs can be
`distinguished based on
`phatmacokinetic properties or
`drug interaction potential
`
`Newer antidepressants
`Fluoxetine
`Sertraline
`Paroxetine
`Fluvoxamine
`Citalopram
`Venlafaxine
`Drugs to treat erectile dysfunction
`Sildenafil
`Tadalafil
`Vardenafil
`Macrolide antimicrobials
`Erythromycin
`Clarithromycin
`Azithromycin
`Telithromycin
`Hypnotics
`Triazolam
`Zolpidem
`Eszopiclone
`Temazepam
`
`24.4.2 Protocol Construction
`
`The customary design is a typical DDI protocol that involves a randomized, two(cid:173)
`way crossover study in a series of healthy volunteers. On one occasion, the victim
`substrate is administered in the control or baseline condition, without coadministra(cid:173)
`tion of the perpetrator. Total area under the plasma concentration curve from zero to
`infinity is calculated CAUCo). On a separate occasion, area under the curve is deter(cid:173)
`mined in the same subjects during coadministration of inhibitor (AUCr). The AUC
`ratio (RAUc) is calculated as
`
`AUCr
`RAUC = AUCo
`
`(24.3)
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 10
`
`
`
`24 Clinical Studies of Drug-Drug Interactions: Design and Interpretation
`
`633
`
`This represents the fractional increase in substrate AUe attributable to coad(cid:173)
`ministration of the perpetrator. The reciprocal of RAUC is the fractional change in
`clearance of the substrate.
`A key requirement is that the exposure of the volunteer subject to the perpetrator
`has to span the duration of blood sampling for plasma concentrations of the sub(cid:173)
`strate. The shorter the half-life of the substrate, the shorter the duration of sampling,
`and the lower the cost and risk of the DDI study. If the perpetrator drug is a metabolic
`inhibitor, this will prolong the necessary exposure duration and sampling time, but
`short half-life victim drugs, nonetheless, are easier to deal with in DDI studies. If
`the perpetrator is an inducer, this if anything decreases the necessary sampling dura(cid:173)
`tion, but this advantage may be offset by the need for a period of pretreatment with
`the inducer due to the time required to attain maximum induction.
`An alternative design is to study the kinetics of the victim drug at steady state.
`With this design, Equation (24.3) is modified to represent the ratio of substrate AUe
`values over a dosage interval segment at steady state. If the intrinsic kinetics of the
`victim drug are nonlinear, this may constitute support for the steady-state DDI study
`design. Beyond that, the steady-state design only has a "showcase" advantage in
`that it more closely mimics the usual therapeutic situation in which the substrate
`is given on an extended basis. However, if the kinetics of the substrate victim are
`linear (dose independent), single-dose kinetics are predictive of what will happen at
`steady state, and the single-dose design provides DDI data of equivalent quality. An
`obvious drawback of the steady-state design is that duration, cost, and risk of the
`study are substantially increased, since the substrate drug must be dosed to steady
`state both in the control condition and during coadministration of the perpetrator.
`
`24.4.3 Studies of Specific Drug Pairs
`
`The initiator of a DDI study may have a clinical or research question that applies
`only to a specific drug pair, without the objective of information that is more general(cid:173)
`izable. With the limited research objective, the study design involves administration
`of the substrate victim on two occasions, with and without perpetrator, as described
`above. The forthcoming research outcome applies to that drug pair, but not neces(cid:173)
`sarily applicable to any other pair. An example is the pharmacokinetic interaction of
`diazepam and fluvoxamine (Perucca et ai., 1994), applicable to that particular com(cid:173)
`bination of substrate and perpetrator, but with no obvious connection to other drug
`combinations.
`
`as Victim
`
`"Drug X" may be identified as a potential DDI victim either through in vitro data,
`clinical observations, or both. The in vitro model may have identified the one or
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 11
`
`
`
`634
`
`DJ. Greenblatt and L.L. von Moltke
`
`more CYP isoforms responsible for clearance of Drug X. The commercial entity
`developing Drug X, or a group of academic investigators, then pose the question:
`what happens to the in vivo clearance of Drug X if one or more of the CYP iso(cid:173)
`forms responsible for clearance is either induced or inhibited by a perpetrator. This
`question may represent a critical point in drug development. The outcome could
`influence the drug's clinical safety profile, and the degree of restrictiveness of the
`product label if the drug is eventually approved. The DDI study outcome could even
`lead to discontinuation of the drug as a development candidate.
`The choice of perpetrator in the DDI study usually will be whatever produces
`the "worst case" - that is, the interaction of largest possible magnitude. The scien(cid:173)
`tific community and the FDA largely agree on what those specific perpetrators are,
`sometimes termed "index inhibitors" or "index inducers" (Table 24.4). Whatever the
`degree of inhibition or induction produced by the index compound, no other perpe(cid:173)
`trator will be any worse. Ketoconazole and ritonavir are typical choices of index
`inhibitor for studies of substrate victims metabolized by CYP3A (Lee et aI., 2002;
`Tsunoda et aI., 1999; Knox et aI., 2008; Greenblatt et aI., 2000). The sponsor or
`investigator may also wish to concurrently study a less potent perpetrator, in which
`case the DDI trial design would be modified to become a three-way crossover. For
`example, a candidate drug that is a CYP3A substrate may be studied with ketocona(cid:173)
`zole and erythromycin as perpetrators, representing strong and moderate CYP3A
`inhibitors, respectively.
`The impact of a DDI on the clearance of a victim drug is greatest when that
`drug is extensively metabolized, and a single CYP isoforrn mediates clearance.
`Candidate victim drugs metabolized mainly by CYP3A isoforrns are a target of
`concern, since inhibition of CYP3A by a strong inhibitor such as ketoconazole or
`ritonavir may produce large values of RAve (Equation (24.3» (Lee et aI., 2002;
`Tsunoda et aI., 1999; Knox et aI., 2008; Greenblatt et aI., 2000). Concern is aug(cid:173)
`mented when the substrate victim has high clearance, and undergoes significant
`presystemic extraction after oral dosage (Fig. 24.3).
`An important feature of study design is the optimal duration of pre-exposure to
`the perpetrator drug prior to administration of the substrate victim. To minimize
`study cost and risk, exposure duration should be the minimum necessary to pro(cid:173)
`duce maximum inhibition or induction. In the case of CYP3A inhibition studies,
`there is strong data to indicate that 24 h of pre-exposure to ketoconazole or ritonavir
`is sufficient to produce maximal inhibition (Fig. 24.4) (Stoch et aI., 2009). For a
`time-dependent (mechanism-based) CYP3A inhibitor such as erythromycin or clar(cid:173)
`ithromycin, 48 h of pre-exposure is sufficient (Okudaira et aI., 2007). On the other
`hand, if the perpetrator is an inducer (rifampin), a pre-treatment period of 5-7 days
`is needed for induction to become maximal (Ohnhaus et aI., 1989; Lin, 2006).
`
`24.4.5 Candidate Drug as Pelpetrator
`
`If the candidate drug is being evaluated as a possible perpetrator of DDIs, the
`study design requires selection of an index substrate - that is, a substrate victim
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 12
`
`
`
`
`
`
`
`24 Clinical Studies of Drug-Drug Interactions: Design and Interpretation
`
`637
`
`Table 24.4 Representative index substrates, inhibitors, and inducers applicable to the design of
`drug interaction studies*
`
`CYP isoform
`
`Index substrates
`
`Index inhibitors
`
`Index inducers
`
`CYPIA2
`
`Caffeine
`
`Fluvoxamine
`
`CYP2B6
`CYP2C9
`CYP2C19
`CYP2D6
`
`CYP3A
`
`Bupropion, efavirenz
`Flurbiprofen
`Omeprazole
`Desipramine,
`dextromethorphan
`Midazolam, triazolam,
`buspirone
`
`Clopidogrel
`Fluconazole
`Fluvoxamine
`Quinidine, paroxetine
`
`[Cigarette
`smoking]
`Rifampin
`Rifampin
`Rifampin
`[None known]
`
`Ritonavir, ketoconazole Rifampin
`
`*Table entries are intended to be representative, not inclusive.
`
`et ai., 2003; Zhu et aI., 2001; Chainuvati et aI., 2003; Blakey et aI., 2004; Gurley
`et aI., 2002; Chow et aI., 2006; Zhou et al., 2004; Christensen et aL, 2003). Instead
`of separate studies, volunteer subjects receive a mix of substrates concurrently, or in
`close temporal proximity, in a single study. Many possible substrate combinations
`have been proposed and utilized in cocktail DDI studies. A key piece of prelimi(cid:173)
`nary infOlmation is an unequivocal demonstration that each pairwise combination
`of substrates in the cocktail does not itself create DDIs with each other.
`
`6 Approach to Analysis of Data
`
`If the clearance of a substrate drug is not dependent on a polymorphically regu(cid:173)
`lated process, the population distribution of AUC values following a fixed dose of
`that substrate will not be consistent with a normal distribution, but rather will have
`a positive skew. Generally the skewed pattern is consistent with a log-normal dis(cid:173)
`tribution (Fig. 24.5) (Greenblatt, 2008; Friedman et al., 1986; Lacey et al., 1997;
`Greenblatt et al., 1989). In any given DDI study, the number of AUC values is
`usually not sufficient for a stable characterization of the underlying statistical dis(cid:173)
`tribution. Nonetheless a log-normal distribution is generally assumed, based on
`experience with larger population studies.
`Going "by the book," a calculation of arithmetic mean and standard deviation
`(SD) of AUC is theoretically precluded if the underlying distribution is non-norma1.
`In practice, however, the values of mean and SD calculated based on the assump(cid:173)
`tion of underlying normal or log-normal distributions are nearly identicaL Statistical
`theory reportedly justifies calculation of geometric mean AUC, along with a 90%
`confidence interval. However, the geometric mean value will underestimate the
`"real" mean value calculated using the assumption of a normal or log-normal
`distribution (Fig. 24.5).
`Statistical analysis of the significance of the DDI - that is, whether the aggre(cid:173)
`gate value of AUCo differs from AVCI -
`is most straightforwardly done via a
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 15
`
`
`
`
`
`24 Clinical Studies of Drug-Drug Interactions: Design and Interpretation
`
`639
`
`Compare mean AUCo and AUCr using nonparametric test
`Compare arithmetic mean RAUC VS. l.00 using Student's t-test
`Determine whether 1.00 falls within 90% CI around geometric mean of RAUC
`In our experience, the different statistical options yield conclusions that are essen(cid:173)
`tially identical.
`
`24.5 Is a Drug Interaction of Clinical Importance?
`
`A key and critical point is that none of these options for data aggregation and
`statistical analysis described above provides insight on 'whether a DDI is clini(cid:173)
`cally important. That judgment must be based on supplemental knowledge of the
`exposure-response relationship for the victim drug. A quantitatively large inter(cid:173)
`action is more likely to be clinically significant, but this is not necessarily so
`(Culm-Merdek et al., 2005). Also, statistical significance is not equivalent to clinical
`significance.
`If a perpetrator drug, acting as a CYP inhibitor, causes a 50% increase in
`exposure to the substrate victim, this would be numelically evident as
`
`meanAUCr-meanAUCo - 0 5
`mean AUCo
`- .
`
`or
`mean RAUC = 1.5
`
`To judge clinical importance of the interaction, one would need another data
`source to determine whether a 50% increase in mean plasma levels of the sub(cid:173)
`strate drug was sufficient to cause a meaningful change in efficacy or the occunence
`of toxicity. The most convincing data on this question comes from clinical trials
`evaluating efficacy and adverse events in relation to dosage among patients tak(cid:173)
`ing the substrate drug for clinical purposes. If the daily dosage is increased by
`a factor of l.5 - on average, the equivalent consequence of the DDI - is there
`greater efficacy ancl!or an increased frequency of adverse events? In some DDI
`studies, pharmacodynamic endpoints are included as part of the study design, in
`which case the DDI study itself may provide data on the clinical consequences of
`the interaction (Cysneiros et al., 2007; Greenblatt et at, 1998a; Greenblatt et al.,
`2000; Culm-Merdek et al., 2005; Greenblatt et al., 1984; Greenblatt et al., 1998b;
`Greenblatt et a1., 2003; Culm-Merdek et al., 2006; von Moltke et al., 1996). The
`limitation of kinetic-dynamic studies
`healthy volunteers is that extrapolation to
`patient populations taking the substrate drug on an extended basis is not necessarily
`straightforward.
`Investigators with a pre-existing understanding of dose/concentrationiresponse
`relationships for the substrate victim may have the option of incorporating this infor(cid:173)
`mation into the statistical inference plan for the DDI study. Suppose it is established
`that RAUC can range from 0.7 to 1.4 with no evident clinical consequence. The DDI
`protocol could then adopt this range as pre-determined "no-effect boundaries." If
`
`Teva Pharmaceuticals USA, Inc. v. Corcept Therapeutics, Inc.
`PGR2019-00048
`Corcept Ex. 2049, Page 17
`
`
`
`640
`
`D.J. Greenblatt and L.L. von Moltke
`
`the 90% CI around the mean RAUC falls entirely within the 0.7-1.4 range, then the
`DDI - even if statistically significant - can be deemed as unlikely to be clinically
`important.
`
`24.6 Are Clinical Drug Interactions Predictable from In Vitro
`Models?
`
`In vitro data now is commonly used in the course of drug development to identify
`potential DDIs that may require clinical studies to rule out or confirm, and to quan(cid:173)
`titate the magnitude of the DDI if there is one. Current FDA guidelines deem that a
`DDI is possible if [1]/Ki is greater than 0.1, as discussed above (Huang et ai., 2008;
`Huang et aI., 2007; Tucker et ai., 2001). These guidelines do allow more informed
`targeting of the extensive resources needed to conduct clinical studies, but the
`guidelines nonetheless are very conservative and minimally quantitative. For more
`than a decade, there has been substantial investment of research energy into the
`question of whether in vitro models can provide more specific quantitative infor(cid:173)
`mation that can forecast not only whether or not a clinical DDI will happen, but
`also how small or big the interaction will be (Venkatakrishnan et aI., 2001, 2003;
`Greenblatt et aI., 2008; von Moltke et aI., 1994a; Brown et aI., 2006; Galetin et aI.,
`2005; Lin, 2006; Obach et aI., 2005, 2006; Ragueneau-Majlessi et aI., 2007; von
`Moltke et aI., 1998; Youdim et aI., 2008; Zhou, 2008; Williams et aI., 2004; Ito et
`a1., 2003; Rostami-Hodjegan and Tucker, 2007; Galetin and Houston, 2006; Galetin
`et aI., 2006, 2007; Ito et aI., 2004; Ohno et aI., 2007; Kanamitsu et aI., 2000; Brown
`et aI., 2005; Bachmann and Lewis, 2005; Bachmann, 2006; Lin, 2000; Volak et al.,
`2007; Farkas et aI., 2008).
`The basis for all predictive in vitro-in vivo scaling models is a link between what
`is observed in a clinical DDI study (as in Equation (24.3», what is available from
`the in vitro model (such as Ki or IC50), and some measured or assumed in vivo
`exposure to the inhibitor ([1]). The most straightforward linkage relationship is
`
`AUCr
`[1]
`RAUC = - - = 1 + -
`AUCo
`Ki
`
`(24.4)
`
`RAUC is objectively determined in a clinically DDI study, and Ki is determined in
`vitro based on generally accepted procedures. A principal uncertainty is the inhibitor
`exposure. Theoretical model validity requires that [1] be the inhibitor c