COMMENTARY
`
`Statistical and Ethical Issues in the Design
`and Conduct of Phase I and II Clinical
`
`Trials of New Anticancer Agents
`
`Mark J. Ratain, Rosemarie Mick, Richard L. Schilsky,
`Mark Siegler*
`
`The development of new anticancer agents is a complex,
`stepwise process proceeding from discovery to demonstra-
`tion of
`antitumor
`activity
`in
`preclinical models
`and
`evaluation of nonnal
`tissue toxicity prior to initiation of
`clinical trials. The purpose of the initial clinical trials (phase
`I)
`is
`to define the toxic effects of
`the agent and the
`recommended dosage for subsequent
`testing (1). The vast
`majority of drugs that reach phase I studies go on to phase II
`testing, which is aimed at finding evidence of efficacy in
`human cancer
`(1). Rarely, drugs
`are withdrawn at
`the
`completion of phase I
`testing, usually due to unpredictable
`and/or severe nonhematologic toxic effects that may be
`irreversible (2-4).
`Although there is general agreement regarding the overall
`research goals of phase I and phase II
`studies of new
`anticancer agents,
`two problems emerge:
`I) The research
`goals may differ from the patient’s goals, and 2) there is no
`consensus on how to achieve the researcher’s goals in the
`most
`efficient
`and ethically appropriate way.
`In
`this
`commentary, we discuss both statistical and ethical issues of
`early anticancer drug development and suggest an alternative
`framework that may improve the protocol design and address
`the ethical
`issues of phase I and phase II clinical
`trials.
`
`Phase I Trials: Statistical Issues
`
`Background
`
`is to
`trial
`The major scientific objective of the phase I
`determine the recommended phase II dose of the drug being
`studied. Thus, phase I trials address an estimation problem
`rather than the testing of a hypothesis. The theoretically
`optimal dose for any individual patient is the dose resulting
`in the maximally achievable antitumor response with an ac-
`ceptable level of toxicity. Since one cannot predict efficacy
`prior to treatment,
`the theoretically optimal
`individual dose
`is the maximal dose that does not exceed an acceptable level
`
`are designed to
`trials
`I
`toxicity. However, phase
`of
`determine the recommended phase II dose for a population,
`not an individual. Thus, since there is usually substantial
`interpatient variability in toxic effects,
`the recommended
`phase II dose will always be an imprecise estimate of the
`optimal dose for the individual patient (5). As a result, many
`phase II trials utilize intrapatient dose modifications based
`on observed toxicity in an attempt to treat each patient at the
`optimal dose.
`A second problem with the design of phase I trials is that
`the optimal dose is usually administered to only a minority
`of the patients treated in the phase I trial. The reason for this
`occurrence is complicated. For detemiination of the recom-
`mended phase II dose, it is standard research practice to treat
`cohorts of patients beginning at a dose that is believed to be
`nontoxic
`and then escalating the dose for patients
`in
`successive cohorts until defined grades of toxic effects are
`observed. Table l
`lists the variables defined by the in-
`vestigator who uses the traditional phase I design and some
`of the commonly considered alternatives.
`Statistical issues in phase I trials were initially outlined by
`Schneiderman (6). Other
`investigators have subsequently
`considered such issues with regard to the traditional design,
`such as the optimal starting dose (7-10) and whether real-
`time pharmacokinetics can more quickly lead to dose escala-
`tions
`(11-13). However,
`investigators have not carefully
`focused on criteria for tennination of phase I studies, which
`should only occur when the recommended phase II dose has
`been adequately defined.
`If all
`three patients of a hypothetical cohort experience
`dose-limiting toxicity, most
`investigators would agree that
`the recommended phase II dose has been exceeded. How-
`ever, even in that setting, the investigator must acknowledge
`that the 95% exact confidence interval for the incidence of
`
`‘See “Notes" section following “Rcfercnccs."
`
`Journal of the National Cancer Institute, Vol. 85. No. 20, October 20, 1993
`
`COMMENTARY 1637
`
`Amerigen Exhibit 1102
`Amerigen Exhibit 1102
`Amerigen v. Janssen IPR2016-00286
`Amerigen V. Janssen IPR20l 6-00286
`
`

`
`
`
`Table 1. Variables to be defined for traditionally designed phase I trials‘
`
`
`Variable
`
`Staning dose
`Patients per nontoxic dose level
`Definition of DLT
`Subsequent dose levels
`Patients per toxic dose level
`Definition of RPTD
`Patients at RPTD
`
`Example
`
`‘/io LD,0
`3
`Any grade 3 toxicity
`“Modified Fibonacci"
`6
`<2 of 6 patients with DLT
`6
`
`Suggested alternatives
`
`Higher
`Fewer
`Grade 3-41’
`Pharmacokinetically guided dosing
`Greater
`<3 of 6 patients with DLT
`Much greater
`
`‘DLT = dose-limiting toxicity; RPTD = recommended phase II dose.
`TGrade 3 organ toxicity is dose-limiting, but grade 3 myelosuppression or nausea/Vomiting is not.
`
`rate of
`If the target
`dose-limiting toxicity is 37%—l00%.
`dose-limiting toxicity is 50%, then the recommended phase
`H dose may be underestimated. Many investigators will not
`accept a 33% rate of dose-limiting toxicity (two of six
`patients) at the recommended phase II dose. Yet,
`the 95%
`exact confidence interval for this dose-limiting toxicity rate
`is quite broad (6%-73%).
`
`Alternatives to “Standard” Phase I Trial Design
`
`Because of the relatively small number of patients who
`actually receive the recommended phase II dose, several
`authors (5,14-22) have suggested substantial modifications to
`the traditional phase I trial design. These modifications have
`been aimed at efficiently determining the recommended
`phase II dose, while minimizing the risk of both undertreat-
`ment and excessive toxicity. Since most responses in phase I
`studies occur at dose levels that are 80%-120% of the
`
`recommended phase II dose (23), such modified trial designs
`would result in a greater response rate during phase I. For
`example, several
`investigators (14.16) have suggested “up
`and down” designs,
`in which the dose for each patient
`is
`assigned on the basis of the experience in the patient(s) most
`recently treated and may be adjusted either higher or lower.
`These designs are relatively simple modifications to the
`traditional design, but they base major protocol decisions on
`relatively small numbers of patients.
`In contrast
`to the traditional design and these modifica-
`tions of it, O’Quigley et al.
`(17,18) proposed a Bayesian
`approach,
`the continual
`reassessment method. A Bayesian
`method begins with assumptions about the main end points
`of the study, known as “priors,” which are derived from the
`investigators’ prior observations and/or beliefs based on their
`own experience and that of others.
`In a phase I trial
`that
`uses
`a Bayesian approach,
`infonnation 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 H dose. Patients are then treated
`at
`the current estimated recommended phase II dose, and
`these
`estimates
`are
`continually updated to reflect
`the
`accumulating dose—toxicity data. When the sample size
`(determined in advance) has been reached, the final estimate
`of the recommended phase II dose is made from all available
`data. Although it
`is desirable to meet
`the ethical goal of
`treating each patient at the hypothesized recommended phase
`
`II dose, there is a danger that patients would be exposed to
`highly toxic doses
`if major errors were made
`in
`the
`assumptions about
`the end points of the study.
`is a bridge
`We recently proposed a design (22)
`that
`between the traditional design and the Bayesian design
`suggested by O’Quigley et al. (17,18). The scheme utilizes a
`cohort-based escalation approach similar
`to that used in
`traditional phase
`I
`studies. As dose—toxicity data
`are
`accumulated, a pharmacodynamic model
`is fit
`to the data.
`Model-guided dosing commences only after a prespecified
`number of patient cohorts have been treated and evaluated
`for toxic effects and after a dose—toxicity relationship is
`statistically defined. Thus, model-guided dosing is
`less
`dependent on assumptions and more dependent on observed
`toxic effects. Unlike the design proposed by O‘Quigley et al.
`(17,18),
`each patient
`is not
`treated at
`the
`estimated
`11
`recommended phase
`close, but
`instead at
`the dose
`calculated to yield a target nadir white blood cell count for
`the individual patient. This design is only useful when
`myelosuppression is dose limiting, although modifications to
`include graded nonhematologic toxicity would be feasible.
`Based on computer simulation studies, this method performs
`better than the traditional design because it requires entry of
`fewer patients
`in the phase I
`study and yields more
`reproducible estimates of the recommended phase II dose for
`an average patient. As pointed out by Mathe and Brienza
`(24),
`there may be substantial
`interstudy variability in the
`recommended phase II dose.
`
`Phase I Trials: Ethical Issues
`
`Background
`
`Investigators conducting phase I trials must adhere to the
`ethical norms of clinical
`research (25-30) and therefore
`encounter a number of potential ethical
`issues:
`1) Minimizing patients treated at
`ineffective doses;
`2) Minimizing patients treated at
`toxic doses;
`3) Historically low probability of response in phase I
`trials;
`
`4) Unknown toxicity and benefit of new agent; and
`5) Difficulty
`in obtaining true
`informed consent
`vulnerable patient populations.
`Some have argued that since phase I nials are designed to
`define the qualitative and quantitative aspects of toxicity,
`
`in
`
`1638 COMMENTARY
`
`Journal of the National Cancer Institute, Vol. 85, No. 20, October 20,
`
`l993
`
`

`
`patients should only enter such trials “for the benefit of
`future cancer victims” (26). While this situation may apply
`to normal volunteers without disease who enroll
`in phase I
`trials of some drugs, it surely does not apply to phase I trials
`of anticancer agents, since both the control and treatment
`arms of the trial only involve patients with cancer (31).
`Thus, cancer treatments in phase I
`trials are always ad-
`ministered with therapeutic intent, even though the major
`scientific objective of a phase I study is to define toxicity.
`Most importantly, the ethical concerns related to therapy for
`patients with advanced cancer do not depend on whether or
`not the treatment is administered in the context of a clinical
`
`trial or whether such a trial is phase I, phase II, or phase HI.
`
`Risk—Benefit Ratio
`
`Patients offered treatment in the context of a phase I trial
`are informed that
`the purpose of the study is to find the
`optimal drug dose and that some patients will be treated at a
`dose that
`is too low or too high (i.e.,
`too toxic). They are
`also informed that
`there is a small possibility that
`the
`treatment will be beneficial. The probability of benefit,
`however, does not depend on the type of trial, but rather on
`the disease being treated, the drug being tested, and the dose
`of the drug that
`is administered. For example,
`there is
`probably a greater response rate for lymphomas in phase I
`trials than for pancreatic cancer in phase III trials, since the
`latter is rarely responsive to any therapy.
`Phase I studies often raise ethical concerns because of the
`
`is exposed to drug toxicity
`the patient
`perception that
`without being offered a reasonable expectation of benefit
`(30). Another common concern is the potential for under-
`treatment of those patients enrolled at dose levels that are
`eventually determined to be subtherapeutic. These concerns
`have led some to suggest
`the use of
`intrapatient dose
`escalation as a solution to this problem (24). With this
`approach, patients who experience little or no toxicity at
`their entry dose receive increased doses on subsequent
`cycles. However, intrapatient dose escalation only allows for
`the possibility that patients who do not rapidly succumb to
`their disease might eventually receive a “therapeutic” dose
`of the new agent. This approach does not recover the time
`lost while the patient received ineffective therapy, and it
`does not clearly increase the chance of tumor response, since
`the tumor is likely to be resistant even to doses that result in
`toxic
`effects
`unless
`such
`dose
`escalation
`is marked
`
`(>100%). In addition, it increases the period of time that a
`patient
`remains on a particular
`study, possibly limiting
`opportunities for treatment in subsequent trials. Traditionally
`designed phase I studies do have a unique property: At least
`one patient must experience dose-limiting toxicity for the
`trial
`to be completed. A statement
`to this effect
`is now
`included in the consent fonn of phase I cancer trials at our
`institution.
`
`substantial and rewarding response, as was observed during
`the phase I
`trials of cisplatin (33) and paclitaxel
`(Taxol)
`(34).
`In fact, all 17 drugs that were eventually marketed
`after beginning National Cancer Institute-sponsored phase I
`testing from 1970 through 1983 manifested activity in at
`least one phase I
`trial
`(22). Thus,
`the potential range of
`outcomes, both beneficial and harmful,
`is extremely broad.
`At study initiation, the type and degree of toxicity likely
`to occur are unknown, although toxicology studies are often
`predictive for drug toxicity to organs. Patients must
`take
`risks
`that cannot be
`easily measured. Since both the
`probability of benefit and toxicity are unknown, estimating
`the therapeutic index is impossible. At low dose levels, there
`is a low probability of both toxicity and benefit. At dose
`levels near the recommended phase H dose, there is a high
`probability of severe toxicity and a maximal probability of
`benefit
`(23). The magnitude of the probability of benefit
`cannot be ascertained until phase II
`trials have been
`completed.
`
`Informed Consent
`
`An issue of great concern to both investigators and
`institutional
`review boards
`is
`the
`consent process
`(27,29,30,35,36), with the suggestion by some of procedural
`safeguards
`such as
`third-party consultation with nonin-
`vestigators (i.e., primary care physicians) as patient advo-
`cates (30). The major obstacle to true “informed consent” is
`that
`the investigator often has little information. Consent
`forms usually contain a litany of possible side effects, when
`in fact the clinical experience has been almost exclusively
`without toxicity. Furthermore, centers with a major interest
`in phase I trials may have 10 or more trials available to an
`individual patient. Should patients be offered all such trials
`as alternatives? Since there may be substantial deficiencies
`in participants’ perceptions of the consent process in even
`less complex trials (36), subjects may be confused regarding
`what
`is involved in any specific trial.
`Freedman (35) has suggested a cohort—specific approach to
`phase I cancer
`studies.
`In this approach,
`the informed
`consent process is different for the first patient entered in the
`study compared with that used to enroll patients after
`significant toxicity has been noted. This strategy requires a
`dynamic consent form, which varies according to the patient
`cohort. One expects
`that
`investigators would ordinarily
`communicate this information to the patient verbally, but
`inclusion of this information as an appendix to the standard
`consent form would ensure that
`the patient is informed of
`the actual experience to date.
`
`Phase II Trials: Statistical Issues
`
`Background
`
`the low
`is
`trials
`Another major concern in phase I
`probability of response. There is no question that, on the
`basis of bidimensional
`tumor measurements, most patients
`entered in phase I trials do not achieve a partial or complete
`response (32). However,
`there is the small possibility of a
`
`Phase II trials are generally studies with no control group
`that are aimed at estimating the antitumor efficacy of a new
`agent in a particular disease. The design of such trials has
`been discussed extensively in the literature;
`the discussion
`has recently focused on issues of sample size and hypothesis
`
`Joumal of the National Cancer Institute, Vol. 85, No. 20, October 20, I993
`
`COMMENTARY 1639
`
`

`
`testing as a basis for guidelines relating to early discontinua-
`tion of therapy if there is adequate evidence of inefficacy
`(15,37—51).
`a
`test
`trials
`trials, phase II
`to phase I
`In contrast
`hypothesis: “New drug X is active against disease Y.” The
`investigator must still define “active” as well as select
`which subset of patients with disease Y will be studied. For
`example,
`the investigator may wish to determine whether
`drug X has a 20% response rate in women with metastatic
`breast cancer
`refractory to doxorubicin.
`In the simplest
`design, based on binomial probability, 14 patients would be
`evaluated initially. If no patient responds,
`the investigator
`may conclude with 95% certainty that drug X has a response
`rate less than 20% in this patient population. If at least one
`patient
`responds,
`a second cohort of patients would be
`treated to better estimate the response rate. The greater the
`number of patients in this second cohort, the more precisely
`one can estimate the response rate.
`
`Alternative Designs
`
`to this design have been
`A number of variations
`suggested. Fleming (43) proposed testing two hypotheses
`simultaneously, one for a minimally acceptable response rate
`and one for a maximally unacceptable response rate (i.e., the
`highest response rate an investigator would be willing to
`miss). Simon et al.
`(45) suggested randomized phase II
`trials. This design is similar to the Fleming design in that it
`tests two (or more) hypotheses simultaneously: a) drug A is
`active against disease Y, and b) drug B is active against
`disease Y.
`
`End Points
`
`the
`Virtually all phase II trials use a common end point,
`objective response rate, which is defined as the percentage
`of assessable subjects who demonstrate a partial or complete
`response. Is screening new agents for a l5%-20% objective
`response rate the best approach? One could argue that we
`are using needlessly strict criteria and that we should simply
`be looking for improvement in quality of life or decrease in
`the rate of tumor growth, possibly by utilizing sequential
`bidimensional measurements (52). Clearly,
`the end point
`should be based on the patient population to be studied.
`
`Phase II Trials: Ethical Issues
`
`Probability of Response
`
`lung cancer. In a separate analysis (54),
`for non—small—cell
`the overall response rate for new agents in phase II trials in
`non—small—cell
`lung cancer was 4%, which is less than the
`overall objective response rate in phase I
`trials of 6%
`reported by Von Hoff and Turner
`(23). Thus, one can
`conclude that
`the vast majority of patients with advanced
`solid tumors that are either refractory to standard therapy or
`for which no standard therapy exists will not respond to any
`treatment, whether
`it be noninvestigational, phase II, or
`phase I.
`
`Informed Consent
`
`trials is the
`issue associated with phase II
`One ethical
`construct used for hypothesis testing,
`i.e., attempting to
`disprove inactivity. The null hypothesis being tested in phase
`II
`trials is
`that “Drug X has less than a Z% objective
`response rate in disease Y.” Thus, responses disprove the
`investigator’s null hypothesis, and, usually, an expansion of
`the trial
`is then required to confirm the response rate. But
`how do investigators view their evolving data from phase II
`trials? What do we tell the 14th patient if all of the prior 13
`have failed to respond? Do we say, “If you don’t respond,
`we will be 95% certain that drug X has less than a 20%
`objective response rate in disease Y”?
`Sordillo and Schaffner (55) have previously addressed this
`issue and have recommended that a statement be added to
`
`the consent form regarding the lack of activity to date. But
`do we really need the 14th patient? The investigator has
`already concluded, albeit with only 94% certainty, that drug
`X is inactive (<20% objective response rate) in disease Y.
`But
`if we accept 13 patients as the final sample size,
`then
`what do we tell
`the 13th patient
`if none of the first 12
`responded? This recursive reasoning can be iterated back to
`the beginning of the study!
`
`Vulnerability of Subjects
`
`Patients entering phase II trials are as vulnerable as those
`entering phase I trials. Whereas patients may often select a
`phase I
`trial
`from multiple studies available at a major
`center, investigators have been cautioned against conducting
`more than one phase II trial for the same patient population,
`because of the risk of bias in patient selection in individual
`studies. Thus, patients offered a phase II trial at a particular
`center may feel
`that
`they have
`no good altematives,
`regardless of the data accrued to date.
`
`Patient Selection
`
`trials have received little
`in phase II
`issues
`Ethical
`attention in the past. Whereas it
`is commonly known that
`very few patients achieve partial or complete responses
`during phase I studies (32), little concern has been expressed
`among investigators regarding the generally poor results of
`phase II trials. Marsoni et al. (53) tabulated the results of all
`National Cancer Institute-sponsored phase II trials from 1970
`to 1985 and found a single active drug from among 42 phase
`II trials for colon cancer and from among 33 phase II trials
`
`recognize the problems with current
`investigators
`As
`phase II designs, new ethical
`issues must be addressed.
`Currently, an important
`issue under debate is whether to
`include in phase II trials previously untreated patients with
`tumors that are sensitive to chemotherapy but incurable, such
`as metastatic breast cancer or extensive small-cell
`lung
`cancer (56-61). This issue was recently considered in detail
`by Moore and Korn (62).
`
`1640 COMMENTARY
`
`Jouma] of the National Cancer Institute, Vol. 85, No. 20, October 20, 1993
`
`
`
`

`
`Comparison of Ethical Issues for Phase I and
`Phase II Trials
`
`trials place the investigator/
`Both phase I and phase II
`physician under “ethical stress” (26); in both types of trials,
`the statistical and clinical objectives are in conflict. The
`conflict
`in phase I between the scientific goals of defining
`toxicity,
`recommended dosage, and clinical pharmacology
`and the therapeutic goal of antitumor effect
`is obvious to
`many and is generally recognized by both investigators and
`Institutional Review Board chairs (30). The conflict in phase
`II is less obvious, depending on the extent of accrual to the
`study and the results in the assessable patients to date (55).
`Patients may enroll
`themselves
`in
`clinical
`trials
`for
`altruistic reasons, but most do so primarily in the hope of
`therapeutic benefit (3036). Patients seek the benefit but may
`in turn benefit
`from the hope itself. Those patients who
`enroll
`in phase I
`trials are usually aware of the lack of
`knowledge of efficacy regarding drug X at dose Z in disease
`Y. However, one must also be concerned about the consent
`
`process in phase II trials. It is not clear that patients are told
`that “drug X has been shown to be safe, but there are no
`data yet to suggest that it is effective in disease Y.” Even if
`there are no clinical data, the patient might feel that drug X
`is likely to be effective. It is even less clear that the patient
`will be informed of negative data as the trial evolves.
`The decision to enroll in a clinical trial should include an
`
`assessment of the probability of both incremental benefit and
`harm. A major difference between phase I and phase H trials
`is
`in the assessment of toxicity.
`It may be paradoxical
`to
`some that most patients in phase I trials experience less drug
`toxicity than those in phase II
`trials. This paradox occurs
`because most patients in phase I studies are undertreated
`(23.30), whereas patients in phase II
`trials are generally
`treated at a dose that results in moderate to severe toxicity.
`Thus, on the average, toxicity in phase I studies is less than
`in phase II, but there is a much greater degree of variation.
`Most patients in phase I studies have a low probability of
`both benefit and toxicity. Their major costs are the in-
`convenience of being in the study and “opportunity risk” if
`there are other possible treatments. In contrast, most patients
`in phase II
`trials have a significant risk of experiencing
`toxicity and an unknown likelihood of benefit, at
`least at
`study initiation. From the standpoint of the patient and
`treating physician, the late stages of a phase I trial, in which
`the dose being used is near the recommended phase II dose,
`are virtually equivalent to the early stages of a phase II trial,
`because the dose and toxicity have been identified and the
`benefit
`is unknown.
`
`Considerations for the Future
`
`Phase I Trials
`
`The traditional cohort design is clearly imperfect, and it is
`important to consider implementation of new study designs,
`such as the design proposed by O’Quigley et al. (17,I8) or
`the model-based dosing design that we have suggested (22).
`Phase
`I
`studies
`should attempt
`to define more
`than
`
`emphasizing instead the
`recommended phase H dose,
`development of a preliminary pharmacodynamic model to be
`tested prospectively in phase II
`trials (63), particularly in
`conjunction with limited sampling strategies based on prior
`pharmacokinetic studies (64-66).
`How much patient autonomy should there be in phase I
`studies? The opportunity for a patient
`to choose among
`multiple phase I
`trials is one example. A more difficult
`autonomy issue is whether patients should have a say in the
`dose of the drug they wish to receive.
`If the investigator
`does not know the optimal dose, might it be reasonable to
`treat patients willing to take greater risks with higher doses?
`Suppose 10 dose levels have been defined for drug X,
`beginning at the standard starting dose (one-tenth of the dose
`lethal
`to 10% of treated animals). Patients might be given
`the choice of receiving the current dose level N, which has
`been incompletely evaluated, or dose level N + 1, which has
`not been evaluated. If a patient selected level N + 1 and
`experienced no toxicity,
`level N could be dropped. Then,
`patients would choose between level N + 1 and level N + 2.
`This approach could be further expanded to allow patients
`even greater autonomy, such as being allowed to choose
`level N + 2 or even N + 8. Thus, more aggressive patients
`could expedite completion of a study, assuming that
`the
`consent process can be completed in a fully infonned,
`ethical
`setting. An important concern, however,
`is
`the
`patient’s
`ability to make
`such difficult
`and
`complex
`decisions. Clearly,
`this would require a dynamic consent
`fonn (35) and probably third—party consultation with a
`noninvestigator as patient advocate.
`
`Phase II Trials
`
`the ethical
`Investigators must become more aware of
`issues in phase II trials. A dynamic consent process could be
`implemented to inform patients of results available to date,
`particularly when there is a high probability that the drug is
`ineffective (55). This may make single-institution phase II
`trials more difficult to complete unless patients are accrued
`rapidly. It would also require better communication among
`those institutions participating in multi—institution studies.
`Phase II trials involving multiple institutions and multiple
`drugs (tested as single agents) would be a possible solution
`to this
`issue.
`In order to maximize patient autonomy, a
`randomized
`consent
`design could
`be
`used
`(67),
`or,
`alternatively, the patient and treating physician could jointly
`select an agent on the basis of data accrued to date. Since
`the purpose of phase II is to decide whether further testing is
`warranted, this self—selection by patients would not necessar-
`ily alter conclusions
`(55). Patients would be
`likely to
`“follow the winner” once activity is
`identified.
`If this
`activity is not confirmed, subsequent patients would be less
`likely to select
`the agent.
`In addition, patients would be
`guided by the type and degree of toxicity to be expected.
`The major drawback to such innovative designs, which
`require a high level of patient participation,
`is the concern
`that the complexity of the consent process could lead to the
`exclusion of
`those patients
`intellectually or emotionally
`unable to participate.
`
`Joumal of the National Cancer Institute, Vol. 85, No. 20, October 20, 1993
`
`COMMENTARY 1641
`
`

`
`
`
`We must also reconsider the value of demonstrable data
`
`than a partial
`less
`is
`that
`regarding therapeutic benefit
`should be considered
`response. Palliation of
`symptoms
`highly important, and new agents with little toxicity and
`documented symptomatic benefit might warrant approval
`with a relatively low objective response rate. For example,
`the potential usefulness of a drug with 15 minor responses
`associated with symptomatic benefit
`in 18 patients is very
`different from that of a drug with no responses at all, yet
`both might have an objective response rate of 0%. The
`determination of a minimum target objective response rate
`depends on the clinical setting; a 20% objective response
`rate in primary pancreatic cancer is much more exciting than
`a 20% objective response rate in untreated metastatic breast
`cancer. Since evidence of activity in phase I has
`some
`predictive value (23). this infonriation could also be utilized
`in the implementation of new phase 11 designs (48).
`A more innovative approach to phase I and phase II
`clinical
`trials
`is
`important. We must expand beyond a
`hypothesis-testing framework driven by a defined P value
`(68,69). Biostatisticians and clinicians should both consider
`
`phase I evaluation as an ongoing process by which a dose-
`toxicity relationship is refined and understood. Implementa-
`tion of
`frequent data evaluation with proper Bayesian
`analysis
`(19,22,48,68,70,7I) will allow us
`to efficiently
`evaluate new agents and preserve our patients’ dignity and
`autonomy in the process.
`
`References
`
`IJ. et al: EORTC New Drug
`(1) Schwartsmann G, Wanders J. Koier
`Development Office coordinating and monitoring programme for
`phase I and 11
`trials with new anticancer agents. Eur
`J Cancer
`272ll62—1l68, 1991
`trial of
`(2) Brown TD, Ettinger DS, Donehower RC: A phase I
`spirohydantoin mustard (NSC 172112)
`in patients with advanced
`cancer. J Clin Oncol 4:1270—1276. 1986
`(3) Margolin K. Doroshow J, Leong L, et al: 3-Deazaguanine: report of a
`phase I
`trial and drug-related cardiac toxicity.
`Invest New Drugs
`81369-376. 1990
`(4) Schilsky RL, Ratain MJ, Janisch L, et al: Phase I clinical and
`pharmacologic study of 502U83 administered as a 24-hour continuous
`intravenous infusion. Cancer Chemother Pharmacol 31:283-288. 1993
`(5) Edler L: Statistical requirements of phase I studies. Onkologie 13:90-
`95, 1990
`(6) Schneidennan MA: Mouse to man: statistical problems in bringing a
`drug to clinical trial. In Proceedings of the 5th Berkeley Symposium
`on Mathematical Statistics and Probability, vol. 4. Berkeley: Univ
`Calif Press, 1967. pp 855-866
`(7) Freireich EJ, Gehan EA. Rall DP. et al: Quantitative comparison of
`toxicity of anticancer agents in mouse. rat. hamster. dog, monkey. and
`man. Cancer Chemother Rep 50:219-244. 1966
`(8) Penta JS, Rozencweig M, Guarino AM, et al: Mouse and large-animal
`toxicology studies of twelve antitumor agents: relevance to starting
`dose for Phase I clinical
`trials. Cancer Chemother Pharmacol 3:97-
`101. 1979
`toxicology to
`(9) Grieshaber CK, Marsoni S: Relation of preclinical
`findings in early clinical trials. Cancer Treat Rep 70:65-72. 1986
`Penta JS. Rosner GL, Trump DL: Choice of starting dose and
`escalation for phase 1 studies of antitumor agents. Cancer Chemother
`Pharmacol 312247-250, 1992
`for
`roles
`(ll) Collins JM, Laharko DS, Dedriclr RL, et al: Potential
`preclinical pharmacology in phase 1 clinical
`trials. Cancer Treat Rep
`70:73-80, 1986
`trials.
`(12) Pharmacokinetically guided dose escalation in phase I clinical
`Commentary and proposed guidelines. EORTC Pharmacokinetics and
`Metabolism Group. Eur J Cancer Clin Oncol 211083-1087. 1987
`
`(10)
`
`(13) Collins JM, Grieshaber CK, Chabner BA: Pharmacologically guided
`phase I clinical trials based upon preclinical drug development. J Natl
`Cancer Inst 8221321-1326. 1990
`(14) Bolognese JA: A Monte Carlo comparison of three up-and-down
`designs for dose ranging. Controlled Clin Trials 42187-196, 1983
`(15) Geller NL: Design of phase I and II clinical
`trials in cancer: a
`statistician's view. Cancer Invest 2:483—491, 1984
`(I6) Storer BE: Design and analysis of phase I clinical trials. Biometrics
`45:925-937. 1989
`(17) O'Quigley J, Pepe M, Fisher L: Continual reassessment method: a
`practical design for phase I clinical trials in cancer. Biometrics 46:33-
`48. 1990
`(I8) O'Quigley J, Chevret S: Methods for dose finding studies in cancer
`clinical trials: a review and results of a Monte Carlo study. Stat Med
`10:l647—1664. 1991
`(19) Gatsonis C, Greenhouse JB: Bayesian methods for phase I clinical
`trials. Stat Med ll:l377—l389. 1992
`(20) Gordon NH. Willson JK: Using toxicity grades in the de