232
`
`Part III I Clinical Testing
`
`5. PHARMACOKINETICS VS PHARMACODYNAMICS
`It is imperative to consider not simply the pharmacokinetics of a new agent, but its
`overall pharmacodynamics. Relatively few clinical studies have examined the relation(cid:173)
`ship between therapeutic, or toxic, response drug dose and exposure time (12).
`Rather, a great deal of emphasis has been placed on defining the interdependent rela(cid:173)
`tionship of concentration and time on antitumor activity (13). Much conflicting data
`addressing the complex relationship of antitumor activity to extracellular drug con(cid:173)
`centrations (q vs exposure time (f) exist. Several investigators have determined that
`the cytotoxicity profile of many agents, such as methotrexate, 5FU, or adriamycin,
`was not a simple function of C x T (14-16), whereas others convincingly show a
`strong dependence of activity on C x T (17-20). Skipper and Adams have suggested
`that new drugs could be developed based on a differential concentration coefficient
`for tumor cell kill vs host toxicity (12,21). This hypothesis is a direct extension of the
`established principle governing bacterial disinfectant action: survival as a function of
`Cn x T (12). Specifically, log transformation of concentration-time curves yields a
`line whose slope is defined as the concentration coefficient, n, and whose intercept is
`the exposure constant, k, for that level of cell kill. Although n may approach unity
`for many agents and thereby demonstrate direct dependence on C x T, many others
`may deviate significantly from unity, providing a very different relationship between
`activity and C x T. This particular pharmacodynamic principle would take into
`account other complex processes that introduce heterogeneity not taken into account
`by the simple C x T model. When this principle was applied to agents that did not
`conform to the simple C x T equation, it was shown that the data conformed to the
`en X T model. Pinedo and Chabner concluded that methotrexate depletion of
`nucleated murine bone marrow cells in vivo was not a simple function of C x T (16).
`By reanalzying their data employing this principle, Adams determined an n of 0.33
`(R 2 = .992) which indeed conformed to the Cn x T parameter (12).
`This pharmacodynamic principle also provides a potential parameter for comparing
`antitumor agents. This parameter represents the minimum exposure conditions re(cid:173)
`quired for a specific level of antitumor activity and accounts for differences in concen(cid:173)
`tration coefficients among agents. Adams has demonstrated this principle elegantly in
`his examination of Crisnatol, a new DNA intercalator in comparison to other agents,
`including doxorubicin, 5-FU cisplatin, etoposide, and tamoxifen. Analysis of these
`agents in the MCF-7 model shows that the minimum en x T parameter gives a relative
`cytotoxicity profile distinct from that found with the standard IC,. end point (12).
`
`6. THE STARTING DOSE
`The initial starting dose is a critical parameter in the design of any Phase I trial.
`Anticancer drugs are inherently toxic agents operating in a very narrow therapeutic
`window. Thus, an aggressive starting dose can result in irreversible toxicity and mor(cid:173)
`tality, leading to a premature tabling of a potentially useful agent. On the other hand,
`a particularly conservative initial dose will lead to poor utilization of resources and ex(cid:173)
`cessive prolongation of a trial. Ideally, a Phase I study should not require more than
`six escalation steps (22). Furthermore, given that patients in the study are treated with
`therapeutic intent, many will be given homeopathic or subtherapeutic doses.
`
`;
`i"!I
`~,
`
`\l
`!
`:~
`11-•
`' _ti"~ Ii
`·~
`I~ • \
`
`Chapter 12 /Phase I Trial Design
`
`233
`
`The starting dose on a Phase I agent is generally extrapolated from preclinical tox(cid:173)
`icology data obtained in rodents, dogs, and less frequently, primates. Typically,
`single- and multiple-dose treatment schedules are employed. Despite the inherent
`problems and assumptions generalizing from animals to humans, a safe and accurate
`starting dose can usually be determined. Interspecies comparisons of dose are most
`• This basic tenet of Phase I trial design
`reliably based on body surface area or mg/m2
`stems from elegant work performed by Pinkel and Freireich, who showed that the
`toxic dose of an agent is similar among species when the dose is measured on the basis
`of body surface area or mg/m2 (23,24). This parameter, unlike body weight, appears
`to have a tighter correlation with physiological functions and inter- and intraspecies
`(adults vs children) toxicity thresholds (24).
`The use of preclinical toxicology studies in multiple species for all anticancer agents
`dates back to the inception of the Drug Development Program at the NCI in the
`mid-1950s. In 1973, the Laboratory of Toxicology, Division of Cancer Treatment,
`NCI, developed guidelines for preclinical toxicology testing of cancer chemothera(cid:173)
`peutic agents (25). The requisite studies were extensive, and involved single-dose
`schedules to assess acute toxicity and lethality in dogs, monkeys, and rodents. Sub(cid:173)
`acute toxicity was determned on dogs using multiple-dose schedules. Following a pre(cid:173)
`determined observation period, the animals underwent a full histological evaluation
`to determine the drug's toxicity profile. The starting doses for a clinical trial of a
`Phase I agent were generally derived by calculating one-third of the lowest toxic dose
`for the most sensitive species (dog or monkey).
`The preclinical toxicology studies were streamlined in the late 1970s, when inter(cid:173)
`species toxicity comparisons could be made because of the extent of toxicology data
`compiled in animals and humans. In October of 1979, the Food and Drug
`Administration Oncologic Drugs Advisory Committee agreed that preclinical studies
`in the mouse could predict quantitative drug toxicity in humans. Thus, the starting
`dose for clinical trials was established as s 1/10 the LD,. (the lethal dose to lOO!o of
`nontumor-bearing animals), determined from statistically reliable murine lethality
`studies. For validation of the human starting dose and to compile further qualitative
`toxicology data, the beagle dog was the species chosen for additional preclinical
`testing (25).
`In 1980, toxicology protocols were developed by the Division of Cancer Treatment
`incorporating these recommendations. Typically, lethality studies are performed on
`mice in single (daily x 1) and multiple dose (daily x 5) schedules, and establish the
`dose causing lethality in 10, 50, and 90% (LD,., LD,., and LD,.) of animals tested
`(25). The LD,. dose is converted to mg/m2 and 1/10 the equivalent dose or 1110
`MELD, 0 is given to the beagle dog. If no toxicity is encountered, the dose is escalated
`until minimal reversible toxicity, i.e., the toxic dose low (TDL) is noted in the dog.
`The TDL has been defined as the lowest dose that produces drug-induced pathologic
`alterations in hematologic, chemical, clinical, or morphologic parameters; doubling
`the TDL produces no lethality (5,26). The human equivalent of 1/3 the TDL in dogs
`in mg/m2 would then be the starting dose in Phase I clinical trials. Conversely, if the
`1110 MELD,. is not tolerated in the dog, the dose is reduced until the dog TDL is
`established. This approach yields a statistically more reliable toxicology profile
`because a larger number of mice in comparison to dogs can be studied.
`Grieshaber and Marsoni subsequently performed a retrospective analysis on seven
`antineoplastic agents that had completed clinical Phase I trials to determine the cor-
`
`
`
`19 of 36
`
`Celltrion, Inc. 1053
`Celltrion v. Genentech
`IPR2017-01122
`
`

`

`!t ;; \(
`'i .t
`
`'.f
`• ,l.
`l
`
`1
`
`' ~
`
`!
`'I
`
`!'
`;
`¥
`1t
`1k
`
`I ~ ~·
`
`~
`, I~
`'I·
`'t· ,,.
`
`t
`I ~-
`;'
`
`234
`
`Part III I Clinical Testing
`
`relation of preclinical toxicology findings to the DLT in humans. They showed that
`1/10 MELD,. reliably predicted a safe human starting dose in 6/7 drugs tested,
`whereas the beagle dog toxicology data had little predictive value in establishing the
`starting dose. However, the preclinical toxicology studies in dogs revealed the human
`DLT IOO"lo of the time with 7SOfo accuracy in providing TDLs. The preclinical mouse
`studies revealed that DLT 7SOfo of the time with SOOfo accuracy in predicting TDLs in
`the human (25). Thus, the beagle dog appears to be a much better model for predict(cid:173)
`ing toxic effects in the human, whereas the mouse may be a better model for
`establishing an initial starting dose.
`Similarly, Goldsmith et al. analyzed mouse, dog, and monkey data on 30 agents
`for which there was correlative Phase I qualitative and quantitative human toxicity
`data (27). However, using 1/3 the TDL in dogs would have underpredicted the
`qualitative toxicity of the drug in 6/30 drugs analyzed. Similarly, Homan's analysis
`indicates a S .90fo probability of exceeding the safe human dose if the 1/3 TDL rule is
`applied to every drug (28). Additional valuable information was drawn from rodent
`toxicity data that would have underpredicted the toxicity of the agent in only 2/29
`cases analyzed. In their retrospective analysis of 12 Phase I agents, Penta et al. have
`also confirmed the potential utility of murine toxicology data predicting an accurate
`and safe starting dose for Phase I chemotherapy trials (29). These data suggest that
`the starting doses in clinical studies should not be based exclusively on the 1/3 TDL in
`the most sensitive large animal and speak to the importance of collecting comparative
`interspecies pharmacokinetic data. There can be significant interspecies differences in
`absorption, distribution metabolism, and excretion of drugs resulting in both quan(cid:173)
`titative and qualitative differences in pharmacodynamics (27). The analysis of such
`comparative data would be significant in determining the human starting dose.
`The rat is also being used in preclinical toxicology studies, since laboratory data
`can be collected with greater ease in comparison to its smaller murine cousin. In fact,
`the European Organization for Research and Treatment of Cancer (EORTC) and the
`United Kingdom Cancer Research Campaign Phase I/II Trials Committee (UKCRC)
`has designated the rat rather than the dog as the second species to confirm the murine
`LD,. (30).
`Despite the utility of preclinical toxicology data in establishing qualitative effects of
`new agents, there are limitations. Although preclinical studies have good predictive
`value for bone marrow, gastrointestinal, hepatic, and renal toxicity, they predict
`poorly for dermatological and cardiac side effects. Moreover, owing to the difficulty
`in performing adequate neurological evaluations in animals, preclinical studies pre(cid:173)
`dict unreliably for central and peripheral nervous system toxicity (23,31).
`
`7. DOSE ESCALATION
`The issue of dose escalation in a Phase I trial is complex, stemming from the
`relative lack of knowledge regarding dose and probability of toxicity for novel anti(cid:173)
`cancer agents. Typically, anticancer agents operate in a very narrow therapeutic win(cid:173)
`dow following nonlinear kinetics, and thus, the potential for life-threatening toxicity
`at effective dosage levels is significant. For this reason, ethics would dictate that the
`agent cannot be tested in a normal healthy volunteer, but must be examined in a
`cancer patient with refractory disseminated disease for whom no other therapeutic
`
`Chapter 12 I Phase I Trial Design
`
`235
`
`Table 3
`Modified Fibonacci Dose
`Escalation Scheme for Phase I Trials
`Percentage increase
`above preceding
`dose level
`
`Drug dose
`n
`2.0n
`S.On
`7.0n
`9.0n
`12.0n
`16.0 n
`
`100
`67
`so
`40
`30-35
`30-35
`
`option exists. The examination of Phase I agents in this study population pose many
`ethical dilemmas that impact directly on the choice of a dose escalation scheme. Since
`therapeutic intent underscores all Phase I trials, it would follow that using a conserva(cid:173)
`tive schema will result in patients receiving subtherapeutic or ineffective doses. Con(cid:173)
`versely, employing an aggressive dose escalation model will lead to the greater
`likelihood that patients will experience toxicity at or beyond the allowable level.
`
`7.1. The Modified Fibona.cci
`Initially proposed by Schneiderman (32), the modified Fibonacci method is an em(cid:173)
`pirical approach to dose escalation that has become the standard methodology in
`Phase I trials. It is based on the Fibonacci sequence, which begins with the integers 1,
`1, 2, 3 and so on, with each number representing the sum of its two predecessors. Let
`n mg/m2 be the dose received by the initial group of patients. In the Fibonacci scheme
`of dose escalation, successive groups of patients receive 2n, 3.3n, Sn, 1n, 9n, 12n,
`16n, and then doses increase by a third of the previous dose (Table 3) (22). In
`other words, this method uses successive percentage increases above the preceding
`dose level of 100, 67, SO, and 40 with subsequent increments of 30-3SOfo (5,33). The
`modified Fibonacci scheme thus slows the Fibonacci sequence, so that a greater
`percentage of increase in the dose escalation occurs earlier in the trial. Obviously, if
`preclinical toxicity studies demonstrate a steep dose/toxicity curve, dose escalation
`would be more conservative than that dictated by the Fibonacci schema.
`Although not always ideal in efficiently predicting an accurate MTD, i.e., the true
`dose that would yield a specified levd of dose-limiting toxicity, the modified Fibonacci
`generally yields a fair approximation of the MTD while operating in a range of safety
`for the majority of patients enrolled (2,22,27,34). Goldsmith et al. have examined the
`efficiency of the modified Fibonacci in the context of the number of dose escalations
`required for 28 different Phase I studies to reach the MTD on a particular schedule.
`Using 1/3 the TDL in dogs as the human starting dose (after excluding those for
`which the starting dose was too high (vide supra), the number of dose escalations for
`20/28 schedules tested ranged from 1-S, whereas only 8 schedules required more than
`6 dose escalations (range 7-13) (27).
`The typical Phase I design dictates the enrollment of cohorts of 3-6 patients/ dose
`level until a predetermined level (generally 330fo) (35) of toxicity is experienced. This
`
`
`
`20 of 36
`
`Celltrion, Inc. 1053
`Celltrion v. Genentech
`IPR2017-01122
`
`

`

`236
`
`Part III I Clinical Testing
`
`stopping dose level, or more frequently the previous dose level, will then be the MTD
`and the starting dose for most Phase II trials of the drug.
`lntrapatient dose escalation (dose escalation within individual patients) at nontoxic
`dose levels is generally not performed. The rationale for this conservative approach is
`the potential for compromising adequate interpretation of toxicity data. Specifically,
`one could not reliably attribute an observed toxic effect to either cumulative toxicity
`or immediate toxicity from the escalated dose (22,36).
`Underlying the modified Fibonacci approach of dose escalation is a layer of con(cid:173)
`servatism that may result in an inefficient use of patients and resources. This, in tum,
`has significant impact on ethical and practical issues in Phase I trial design and drug
`development. Again, inherent in a Phase I trial design is therapeutic intent, which dic(cid:173)
`tates that patients receive therapeutic or near-therapeutic doses of the Phase I agent.
`Still, this goal must be balanced by the threat of unanticipated and unacceptable tox(cid:173)
`icity from aggressive dose escalation. There have, therefore, been several other pro(cid:173)
`posed methodologies to address the aforementioned limitations of the modified
`Fibonacci strategy.
`
`7.2. Pharmacologically Guided Dose Escalation
`The pharmacologically (or phannacokinetically) guided dose escalation (PODE)
`schema was originally proposed by Collins and coworkers in the Blood Level Work(cid:173)
`ing Groups of the Division of Cancer Treatment at the NCI (37). The foundation of
`PODE rests on the basic concept of toxicity and efficacy of a drug being derived from
`its concentration vs time curve. This concept originated from the work of German
`pharmacologists during World War I (13,21). They observed that mustard agents
`were equally toxic whether inhaled for a short time at a high concentration or for a
`long time at a lower concentration. A narrower application of this principle to anti(cid:173)
`cancer agents with regard to mechanism of action and schedule was performed by
`Skipper et al. Their work demonstrated that the effect of antimetabolites on macro(cid:173)
`molecular systems is more dependent on time of exposure above a certain threshold
`concentration than absolute concentration or C x T (38). ('11te importance of con(cid:173)
`sistency of schedule across species on reliable interpretation of PK data and applica(cid:173)
`bility to the design of Phase I trials was stressed. It is interesting to note that in his
`retrospective review of preclinical toxicity data on 30 drugs [vide supra] in com(cid:173)
`parison with the actual clinical doses and schedules used in humans, only four drugs
`had identical schedules in humans, monkey, dog, and mouse (27). With these prin(cid:173)
`ciples as a basis, Collins proposed that C x T (the area under the curve for plasma
`concentration vs time) would be a more realistic index of toxicity than dose, citing
`three potential explanations for variation in toxicity between mouse and humans
`using dose-based comparisons:
`
`j:;
`
`t-f;i~
`I ·~
`·1:
`I ~ ~~
`~~:·
`i'l
`:Ii~
`~
`·~ I.,
`\~;
`
`l ;~ ti
`l
`.i
`I~
`b!;
`.c./ u; ,i:
`l.
`~ l~i
`. 1.1
`,}p.
`1;~
`·~~
`j:"
`1.lf
`Is:
`Le
`i';.
`!
`l•I
`I·~
`'l \
`
`'4
`'11
`
`1
`
`I. Species differences in drug metabolism, elimination, and binding;
`2. Schedule dependency owing to exposure time differences; and
`3. Species differences in target cell sensitivity.
`
`'•
`
`Collins constructed a pharmacodynamic hypothesis that similar biological effects
`(e.g., toxicity) would occur at similar plasma levels in mice and humans (39). In other
`words, the ratios of C x Tin human plasma at the MTD to C x Tat the LD, 0 of an
`
`Chapter 12 I Phase I Trial Design
`
`237
`
`agent should approximate unity if this hypothesis is valid. Retrospectively, Collins
`analyzed 13 anticancer agents for which there were comparative pharmacokinetic and
`pharmacodynamic data. In 10/13 drugs examined, the hypothesis was correct with
`C x T ratios ranging from 0.6-1.3, demonstrating near-equivalent toxicity. The
`other three agents did not conform to the model presumably because of interspecies
`differences in target organ sensitivities (39) (vide infra).
`In addition to using interspecies PK data for efficient dose escalation, Collins
`stressed the importance of examining preclinical quantitative data on schedule
`dependency with regard to its impact on accelerated entry dose in Phase I trials. In
`addition to total drug exposure, the rate of drug delivery can have a significant effect
`on the toxicity profile of a drug. Collins attributes this relationship to a "threshold"
`or peak-level phenomenon (39). The toxicity incurred by a rapid delivery or bolus ad(cid:173)
`ministration of a drug may be abrogated by a longer or continuous infusion. Such a
`schedule would permit greater drug delivery without the toxicity from high peak
`concentrations. The PODE approach can be optimally applied to drugs given on a
`continuous infusion schedule by replacing C x T values or AUC by measured steady(cid:173)
`state concentration of the drug. Steady-state concentration targets are more easily
`grasped and avoid the difficulties of exposure-time differences caused by pharmaco(cid:173)
`kinetic factors (13,40).
`Collins prospectively applied the pharmacodynamic and pharmacokinetic prin(cid:173)
`ciples to eight agents consecutively entered into Phase I trials under NCI sponsor(cid:173)
`ships. Three of the eight agents were too far along in trial to benefit from this pharma(cid:173)
`cologically guided method. Nevertheless, the remaining five trials were significantly
`shortened either because of an increased entry dose or accelerated escalation schema
`(39). For example, merbarone demonstrated an LD,. in mice based on bolus adminis(cid:173)
`• This translated into a starting dose of 12 mg/m2. However,
`tration of 123 mg/m2
`because of acute toxicity noted, continuous infusion schedules in dogs were explored,
`resulting in a revised starting dose of 96 mg/m2/d. Since the human MTD was 1500
`mg/m2/d, the eightfold-higher entry dose resulted in 48 fewer total patients being
`accrued and the study being completed 16 mo earlier than initially anticipated.
`Similarly, for flavone acetic acid, the entry dose was based on the standard 1/10 the
`LD,., but dose escalation proceeded more rapidly based on plasma levels, thereby
`saving four unnecessary Fibonacci steps (39). Assuming the C x T ratio was an effec(cid:173)
`tive predictor for toxicity, Collins concluded that the PODE method could result in a
`20-500/o reduction of dose escalation steps.
`Intrigued by Collins' observations and proposals, the Phannacokinetics and Metab(cid:173)
`olism (PAM) group of the EORTC (now the EORTC Pharmacology and Molecular
`Mechanisms Group) further evaluated the PODE method by retrospectively cor(cid:173)
`relating their extensive human and murine phannacokinetic data with known toxicity
`data in the Phase I setting (37,41). A careful review of their data identified problems
`with this methodology, but provided encouragement for the further evaluation of
`PODE in Phase I study designs. Cited technical problems included PK data obtained
`in animals using disparate routes of administration compared with the human data,
`the lack of an accurate LD,. in mice, and the use of doses other than the LD,. for
`pharmacokinetic studies. Cited pharmacokinetic factors that would affect the basic
`tenet of PODE approach, i.e., toxicity being solely a function of total parent drug
`plasma levels, included:
`
`
`
`21 of 36
`
`Celltrion, Inc. 1053
`Celltrion v. Genentech
`IPR2017-01122
`
`

`

`Chapter 12 /Phase I Trial Design
`
`239
`
`data alone. Instead, its value should be considered in the wider context of deploying
`pharmacokinetic and pharmacodynamic information to optimize and shape early
`clinical trials.
`
`7.3. Other Statistical Strategies
`Additional methodologies to address the shortcomings of the standard Phase I
`design included several statistical approaches, including variations on escalation/
`de-escalation schemes ("up and down") (42-45), stochastic approximation methods
`(44,46), and the application of Bayesian principles to estimate more accurately the
`probability of toxicity at the Mm (35,47,48). Bayesian methods are based on assump(cid:173)
`tions about the main end point of the study, known as "priors," which are derived
`from the investigators' prior observations and/or beliefs stemming from their own
`experience and that of others (49). An example of a classic Bayesian approach that
`proposes model-guided dosing includes the Continual Reassessment Method (35,
`48,50). Using such an approach, information from preclinical studies and/or clinical
`studies of similar drugs is used to make an educated guess regarding the dose-toxicity
`curve and the recommended Phase II dose. Patients are treated at the current esti(cid:173)
`mated recommended Phase II dose that is continually updated to reflect the accumu(cid:173)
`lating dose-toxicity data. When the predetermined sample size has been reached, the
`final estimate of the recommended phase II dose is made from all available data (49).
`A potential drawback of this approach is overestimating the starting Phase I dose,
`thereby placing the initial patients enrolled at great risk for drug toxicity. The Quanti(cid:173)
`tative Assessment Phase I Escalation Study Design, a related approach, uses an itera(cid:173)
`tively updated pharmacodynamic model for dose escalation. Unlike the Continual
`Reassessment Method, dosing is more dependent on true observations of toxicity
`rather than assumptions or "priors." The scheme utilii.es a cohort-based escalation
`approach similar to that used in a traditional Phase I design. As dose-toxicity data
`are accumulated, a pharmacodynamic model is fit to the data. Model-guided dose
`escalation begins only after a specific number of cohorts of patients have been observed
`for toxicity and a dose-toxicity relationship has been statistically defined (49,51). This
`strategy, although currently applicable only to drugs with myelosuppression as the
`DL T, shows significant promise in more accurately and efficiently defining the Mm
`in Phase I studies of myelosuppressive agents. With modification for nonhematologic
`toxicity, such a scheme could have more widespread application.
`
`7.4. Patient Choice Dose Cohort
`A more innovative approach to Phase I design incorporating issues influencing
`patient participation has been proposed by Daugherty et al. This design incorporates
`such less-well-defined aspects as patient motivation and comprehension and the influ(cid:173)
`ence of patients' aversion to risk. In this particular scheme, the patient would play an
`integral role in determining his or her entry dose level.
`Using the Patient Choice Dose Cohort (PCDC) Phase I trial design, patients choose
`from one of three possible doses:
`
`I. The standard Phase I trial starting dose of 1/10 LD .. in mice;
`2. Some multiple (M) of 1110 LD .. ; or
`3. Twice the (M) dose.
`
`i
`~
` f
`t'
`)
`\;.
`
`Ii
`
`l
`~
`j
`\.
`
`~ I '
`
`n
`
`If
`
`~ i
`~
`~
`ii Ii
`M u
`ii 1 I~
`11 Ii
`I;
`
`II
`
`•i
`
`238
`
`Part III I Clinical Testing
`
`1.
`
`Interspecies differences in plasma protein binding resulting in disparate levels of free
`drug and hence different toxicity profiles;
`2. Differences in metabolism of the drug that may affect levels of active drug in the case of
`a prodrug that requires metabolism to the active form or may impact on levels of toxic
`metabolites;
`3. Nonlinear pharmacokinetics in which a small increase in dose would have a significant
`effect on AUC;
`4. Patient heterogeneity resulting in significant variance of measured AUC; and
`S. Lack of adequate PK data on tissue drug levels that may more accurately reflect the
`pharmacodynamic effect of a drug.
`
`Interspecies differences in target organ or cell sensitivity were also stressed relative
`to observed disparate pharmacodynamic effects. Collins cited antimetabolites as an
`example of a class of agents that are not amenable to the PGDE approach because of
`the differences in target cell sensitivity. Antimetabolites are highly dependent on the
`target cell for activating or inactivating enzymes and extracellular nucleic acid precur(cid:173)
`sors. As such, antimetabolite toxicity cannot reliably be predicted by AUC or dose.
`Despite the shortcomings of the PGDE method, the EORTC concluded that it was
`valuable especially when applied to certain classes of drugs. In particular, the anthra(cid:173)
`cyclines demonstrated AUC to be a better predictor of toxicity than dose. Similarly,
`platinum agents showed a reasonable correlation between AUC and toxicity. As
`anticipated, several N-methyl drugs that require metabolism to the active species
`demonstrated poor correlation between AUC and toxicity.
`Graham and Workman critiqued the PGDE approach applied prospectively to
`several new anticancer agents, including amphethinile, brequinar sodium, iodo-doxo(cid:173)
`rubicin, the anthrapyrazoles, rhizoxin, and aphidicolin glycinate (37). Although their
`review will not be detailed here, their analysis supported the global value of this ap(cid:173)
`proach in providing critical information for the rational design of Phase I trials. As
`demonstrated, the PGDE was not rigidly applicable to many agents, but did provide
`information that often gave valuable insight into a new drug's pharmacologic effect
`and clinical potential. For example, in a prospective Phase I study of iodo-doxorubi(cid:173)
`cin, an anthracycline derivative, a PGDE strategy was attempted, incorporating
`Collins' proposals and the guidelines drafted by the EORTC. Near the outset of the
`study, it became clear that marked species differences exist in the pharmacokinetics
`and metabolism of the drug. Principally this was attributed to the rapid metabolism
`of 1-Dox to 1-Doxol by an aldo-keto reductase absent in mice. Thus, although the
`correlative PK studies were modified to measure the metabolite as well, a conven(cid:173)
`tional modified Fibonacci scheme was used. Additional elegant PK investigations
`demonstrated that the toxicity of the parent drug and metabolite had comparable
`LD,. values in mice. As a result of these findings, a PGDE scheme was reintroduced
`and subsequent doses escalated by the summation of the 1-Dox and 1-Doxol AUCs,
`utilizing the original target 1-Dox AUC to guide the dose increments. In addition to
`providing valuable pharmacological data on 1-Dox, by using the PGDE strategy with
`pre-existing knowledge of the comparative pharmacological information on 1-Dox
`and 1-Doxol at the outset of the study, the Mm could have been reached in only five
`steps (37).
`Graham and Workman concluded that the value of this approach should not be
`judged on its ability to predict accurately the MTD based on correlative animal AUC
`
`
`
`22 of 36
`
`Celltrion, Inc. 1053
`Celltrion v. Genentech
`IPR2017-01122
`
`

`

`240
`
`Term
`Subtoxic dose
`
`Minimal toxic dose
`
`Recommended dose for therapeutic
`(Phase II trial)
`MTD
`
`Part III I Clinical Testing
`
`Table 4
`Definltloo of Dose Toxicity lo Phase I Studies
`Definition
`A dose that causes consistent changes of
`hematologic or biochemical parameters
`and might thus herald toxicity at the
`next higher dose level or with prolonged
`drug administration
`The smallest dose at which one or more
`of three patients show consistent, readily
`reversible drug toxicity
`The dose that causes moderate, reversible
`toxicity in most patients
`The highest safely tolerable dose
`
`Once a PCDC had been filled and the patient treated without toxicity, that selected
`PCDC would become the lowest PCDC (X). Subsequent patients would then choose
`from X, MX or 2MX for their dose selection. Of note, Daugherty et al. also proposed
`that cohorts of patients be limited to one, unless a grade III or IV toxicity was
`experienced (52). This method has the advantage of more efficiently utilizing patient
`resources, but a meaningful interpretation of the confidence interval for a potential
`toxicity would be sacrificed.
`In summary, it is evident that additional methodologies to address the inadequa(cid:173)
`cies and shortcomings of the current Phase I trials design must be explored. The two
`factors most frequently modified in the new designs are improving patient safety or
`maximizing the potential for response. However, the very nature of cytotoxics demon(cid:173)
`strating a narrow therapeutic range requires that a change in one factor be accom(cid:173)
`panied by a reciprocal change of the other. Hence, patients who have the greatest
`likelihood of responding to a new anticance agent are also the patients with the great(cid:173)
`est risk for toxicity. As Hawkins has so succinctly stated in his critique of the newer
`methods, "the feasibility of any new trial design must ultimately be demonstrated by
`the successful conduct of the clinical study" (53). Inherent in the definition of success
`is maintaining an ethically reasonable safety zone for patients enrolled in the study.
`
`8. TERMINATING THE PHASE I CLINICAL TRIAL
`
`8.1. De(i.ning The MID
`The definition of the MTD is somewhat arbitrary and varies from investigator to
`investigator. Some define it to be the maximal dose administered during the trial with(cid:173)
`out eliciting any toxicity. Others define the MTD as the minimal administered dose
`associated with a toxic response. These definitions, however, arc strongly dependent
`on sample size. Storer takes the MTD to mean some percentile of the dose toxicity
`relationship, about the 33rd (43). More generally, estimating some arbitrary percen(cid:173)
`tile chosen by the experimenter seems desirable (44).
`
`8.2. Toxicity Grading System
`Reasonable definitions of toxic levels in humans have been established by Carter
`et al. and shown in Table 4 (2,33). However, the TDL and the MTD in human trials
`
`,,~
`
`4
`ff
`p
`I
`~
`~ ~
`u ~ 4
`
`IT
`~
`
`~ e
`
`I &
`'
`~ w i
`9 " ~
`
`I ' i
`
`~ j
`)i
`~ u