V O L U M E 2 6 䡠 N U M B E R 8 䡠 M A R C H 1 0 2 0 0 8
`
`JOURNAL OF CLINICAL ONCOLOGY
`
`R E V I E W A R T I C L E
`
`Review of Phase II Trial Designs Used in Studies of
`Molecular Targeted Agents: Outcomes and Predictors of
`Success in Phase III
`Robert H. El-Maraghi and Elizabeth A. Eisenhauer
`
`A
`
`B
`
`S
`
`T
`
`R
`
`A
`
`C
`
`T
`
`Purpose
`Because the appropriate design and end points for phase II evaluation of targeted anticancer
`agents are unclear, we undertook a review of recent reports of phase II trials of targeted agents
`to determine the types of designs used, the planned end points, the outcomes, and the
`relationship between trial outcomes and regulatory approval.
`Methods
`We retrieved reports of single-agent phase II trials in six solid tumors for 19 targeted drugs. For
`each, we abstracted data regarding planned design and actual results. Response rates were
`examined for any relationship to eventual success of the agents, as determined by US Food and
`Drug Administration approval for at least one indication.
`Results
`Eighty-nine trials were identified. Objective response was the primary or coprimary end point in
`the majority of trials (61 of 89 trials). Fourteen reports were of randomized studies generally
`evaluating different doses of agents, not as controlled experiments. Enrichment for target expression
`was uncommon. Objective responses were seen in 38 trials; in 19 trials, response rates were more
`than 10%, and in eight, they were more than 20%. Agents with high response rates tended to have
`high nonprogression rates; renal cell carcinoma was the exception to this. Higher overall response
`rates were predictive of regulatory approval in the tumor types reviewed (P ⫽ .005).
`Conclusion
`In practice, phase II design for targeted agents is similar to that for cytotoxics. Objective response
`seems to be a useful end point for screening new targeted agents because, in our review, its
`observation predicted for eventual success. Improvements in design are recommended, as is
`more frequent inclusion of biological questions as part of phase II trials.
`
`J Clin Oncol 26:1346-1354. © 2008 by American Society of Clinical Oncology
`
`INTRODUCTION
`
`Recently, there has been a veritable explosion of
`knowledge with respect to the molecular biology of
`malignancy. This has led to the identification of po-
`tential new targets for cancer therapy and, subse-
`quently, to the rational design of agents created to
`affect those targets in a clinically meaningful way.
`However, as a growing number of agents targeting
`molecular pathways are tested in the clinic, there has
`been increasing pressure to rethink the standard
`drug development paradigm, specifically early trial
`design, to ensure that these promising new drugs are
`appropriately evaluated.
`The ultimate goal of drug development in on-
`cology is to identify new agents that provide a mean-
`ingful clinical benefit for patients, with the gold
`standard being the prolongation of patient survival.
`
`This is both a time- and resource-intensive process,
`with an estimated monetary cost to bring new drugs
`to market of $800 million (or more)1,2 with equally
`important, although perhaps less definable, costs to
`participating clinical subjects. With this in mind, it is
`understandable that there is pressure to optimize
`early clinical trial design so as to minimize the re-
`sources expended on drugs that are likely to fail in
`later development.
`The primary objective of phase II trials, regard-
`less of the nature of the compound, is to screen for
`preliminary evidence of efficacy in a given tumor
`type. For cytotoxic agents, the standard approach
`has been to enroll small numbers of patients in a
`nonrandomized design, often of two or more ac-
`crual stages, and use objective tumor regression as-
`sessed by standard criteria3,4 as the end point to
`identify drugs with potential efficacy. Retrospective
`
`From the Royal Victoria Hospital, Barrie;
`and National Cancer Institute of Canada
`Clinical Trials Group, Queen’s Univer-
`sity, Kingston, Ontario, Canada.
`
`Submitted July 18, 2007; accepted
`November 1, 2007; published online
`ahead of print at www.jco.org on
`February 19, 2008.
`
`The National Cancer Institute of Canada
`(NCIC) Clinical Trials Group (CTG) is
`supported by a core grant from the
`National Cancer Institute of Canada
`with funds received from the Canadian
`Cancer Society. R.H.E.-M. was a Drug
`Development Fellow at NCIC CTG from
`2005 to 2006 whose salary was
`supported by a Transdisciplinary Train-
`ing Program Grant from the Canadian
`Institutes of Health Research as well as
`a Drug Development Fellowship grant
`from AstraZeneca.
`
`Presented in part at the 42nd Annual
`Meeting of the American Society of
`Clinical Oncology, June 2-6, 2006,
`Atlanta, GA; and the National Cancer
`Research Institute Cancer Conference,
`October 8-11, 2006, Birmingham,
`United Kingdom.
`
`Authors’ disclosures of potential con-
`flicts of interest and author contribu-
`tions are found at the end of this
`article.
`
`Corresponding author: Elizabeth Eisen-
`hauer, MD, FRCPC, National Cancer
`Institute of Canada Clinical Trials Group,
`Queen’s University, 10 Stuart St, King-
`ston, Ontario, Canada K7L 3N6; e-mail:
`eeisenhauer@ctg.queensu.ca.
`
`© 2008 by American Society of Clinical
`Oncology
`
`0732-183X/08/2608-1346/$20.00
`
`DOI: 10.1200/JCO.2007.13.5913
`
`1346
`
`MYLAN - EXHIBIT 1023
`
`

`
`Phase II Designs Used for Targeted Therapy
`
`data support the use of the response end point for studies of cytotoxic
`agents5-7 and the efficiency of the nonrandomized design. The inher-
`ent differences in the mechanism of action between traditional cyto-
`toxic drugs and molecularly targeted agents, coupled with interest in
`increasing the reliability of phase II results in identifying truly active
`agents, have led to considerable discussion about the so-called tradi-
`tional approach to screening trials, particularly in regards to patient
`selection criteria, end points, and study design.
`With respect to patient selection, it is reasonable to expect that
`not all patients with a given tumor type will have similar levels of target
`protein activity or expression, and thus, efficacy of the targeted agent
`may vary according to which subpopulation is evaluated. Enriching
`the population to maximize possible activity could be achieved
`through restriction of study entry to those with a specific tumor
`histology or those whose tumors (over)express a molecular target.
`However, at the time of early clinical studies, such predictors of activity
`are by definition hypothetical, unless the agent affects a target that has
`had validated predictors identified through earlier clinical trials of
`other drugs affecting the same target.
`The appropriate end point for phase II trials of targeted agents
`has also been debated.8-12 Because many of these agents may affect
`tumor cells by reducing proliferation, rather than by causing cell
`death, the impact on tumor growth may be stabilization of disease or
`minor tumor shrinkage. Thus, it is argued that focusing only on
`objective response could result in overlooking some agents that could
`improve survival by causing disease stabilization. Indeed, in lung can-
`cer, attainment of stable disease, in addition to responses, has been
`shown to contribute to improved survival.13
`The design of phase II trials and the choice of which end point is
`to be used are closely linked; the use of a nonrandomized design in the
`traditional phase II trial is reasonable because objective responses are
`unlikely to be due to natural disease processes, rendering a control
`group unnecessary. However, even if response is believed to be a
`reasonable end point for trials of particular agents, the sample size of a
`nonrandomized trial may need adjustment. This is necessary if the
`hypothesized response rate of interest for the targeted agent is lower
`than what might be considered standard for cytotoxic agents, as would
`be the case if the trial population is unselected for predictive markers.
`When objective response alone is not foreseen to be a useful end
`point, novel designs have been proposed for phase II screening trials.
`Included is a multinomial design in which decisions about early stop-
`ping and conclusions on activity are based not only on the number of
`responses seen, but also on the proportion of patients demonstrating
`early disease progression.14,15 Another approach of interest is the
`randomized discontinuation trial where patients are treated with a
`new agent for a specified period of time, after which those with stable
`disease are randomly assigned to continue or discontinue therapy.16,17
`The end point for this type of trial is either time to event (eg, progres-
`sion) or the proportion of patients progression free at a specific time
`point after random assignment. This design also provides information
`on the activity of the drug in terms of rates of response and progression
`at the end of the run-in phase. The randomized discontinuation de-
`sign has been promoted to be of particular help in screening cytostatic
`drugs, such as molecular targeted agents, by permitting early assess-
`ment of whether delays in progression are related to treatment or
`disease and whether they are of sufficient magnitude to suggest that
`the drug may be effective.
`
`The issues we have just highlighted illustrate the fact that numer-
`ous proposals regarding changes to traditional phase II trial design and
`end points have been made in recent years. However, there is little
`information on the impact of these discussions on the actual end
`points and designs used for the evaluation of novel targeted agents in
`phase II trials. In this article, we report the results of a review of phase
`II designs, end points, and outcomes of a series for targeted agents that
`have been studied in clinical trials in the last few years. In particular, we
`were interested in determining the planned design and end point(s) in
`these trials, what the observed results were, and, when possible,
`whether the results of phase II trials of individual agents were useful in
`predicting which agents achieve regulatory approval. Our interest in
`this last point speaks to the ultimate goal of phase II screening trials
`regardless of design: Is the output of the trials useful in identifying
`those agents likely to succeed in phase III trials?
`
`METHODS
`
`Agents
`For this review, we focused on the 31 targeted agents that were the basis of
`a review of phase I trial design published in 2004.18 These agents have distinct
`intracellular or extracellular targets in pathways and can be grouped into major
`classes on the basis of the chemistry of the agent (small molecule, antisense
`oligonucleotide, or antibody) as well as the expected molecular target (Table
`1).19-83 All of these agents completed phase I investigation long enough in the
`past that phase II trials, if undertaken, should now be reported.
`
`Tumor Types and Search Strategy
`We defined six common solid tumor types (breast, colorectal, lung,
`ovarian, prostate, and renal cell carcinomas) on which to focus. We performed
`a MEDLINE search for published articles of completed phase II single-agent
`studies for all of the 31 drugs limited to the six tumor types noted. Studies using
`the agents in combination and studies in progress were excluded from the
`review. The search terms included the name of the agent (including trade
`name, if applicable) and the molecular target, with the limit of clinical trials,
`phase II. A cutoff date of April 30, 2006 was used for retrieving publications.
`In addition to full publications, we also attempted to identify final reports
`of phase II studies published as abstracts that were not yet reported in full
`article form. Abstracts were identified using an electronic search of the pro-
`ceedings of the American Society of Clinical Oncology, the American Associ-
`ation for Cancer Research, and the San Antonio Breast Cancer Symposia
`meetings until the end of 2005. Identified abstracts were excluded if they
`indicated continuing patient accrual or incomplete efficacy results. Abstract
`data were verified against data presented at the actual scientific meeting,
`when possible.
`
`Definition of Trial Versus Report
`Each publication, whether abstract or full article, was considered a re-
`port. A small number of reports included more than one trial. Examples were
`reports in which multiple tumor types were included (each analyzed sepa-
`rately), multiple targeted agents were evaluated (in a noncomparative ran-
`domized design), or multiple doses of the same targeted agent were tested. To
`describe the outcomes of each active arm of targeted therapy appropriately, we
`considered a study with n prospective active treatment groups in which there
`was no intent to conduct a formal statistical comparison to be n trials within a
`single report. For those few reports where random assignment was to standard
`therapy or placebo, the study arms with no active targeted therapy or placebo
`were not considered as separate trials, and the reported outcome was that
`which was planned for the overall study.
`
`Data Abstraction
`All data were abstracted independently for each trial by both authors,
`with any discrepancies resolved by consensus. A summary of abstract data is
`found in Appendix Table A1 (online only). All planned primary and secondary
`
`www.jco.org
`
`1347
`
`

`
`El-Maraghi and Eisenhauer
`
`Table 1. Targeted Agents From 2004 Phase I Review Subjected to Search for Phase II Results
`
`Phase I Review Agents
`
`This Phase II Review
`
`Target Category
`
`Target
`
`Agent
`
`EGFR/HER-2 signaling pathways
`
`Farnesyltransferase
`
`MEK
`mTOR
`Raf kinase
`
`Cell surface receptors
`
`EGFR
`
`Angiogenesis
`
`HER-2
`c-kit
`VEGF
`VEGFR (plus other targets)
`
`Extracellular matrix
`
`Other
`Matrix metalloproteinase
`
`Other
`
`Total No.
`
`BCL-2
`PKC␣
`DNA methyltransferase
`
`BMS-214662
`R115777
`L778, 123
`SCH 66336
`CI-1040
`CCI-779 (temsirolimus)
`ISIS 5132
`BAY 43-9006 (sorafenib)
`ZD1839 (gefitinib)
`OSI-774 (erlotinib)
`C225 (cetuximab)
`MAb225
`EMD 72000
`EKB 569
`RG83852
`Trastuzumab
`STI-571 (imatinib)
`Bevacizumab
`ZD6474
`PTK787
`SU6668
`SU5416 (semaxanib)
`SU11248 (sunitinib)
`Endostatin
`BB-2516 (marimastat)
`BAY 12-9566 (tanomastat)
`COL-3
`BMS-275291
`G3139
`ISIS 3521 (aprinocarsen)
`MG98
`31†
`
`Drug
`Type
`
`SM
`SM
`SM
`SM
`SM
`SM
`AS
`SM
`SM
`SM
`AB
`AB
`AB
`SM
`AB
`AB
`SM
`AB
`SM
`SM
`SM
`SM
`SM
`Other
`SM
`SM
`SM
`SM
`AS
`AS
`AS
`
`No. of Single-Agent
`Phase II Reports
`Identified
`
`No. of Single-Agent
`Phase II Trials in
`Selected Tumors*
`Reviewed
`
`Reference
`No.
`
`5
`
`1
`1
`3
`4
`1
`14
`5
`3
`
`8
`7
`2
`1
`
`2
`2
`
`1
`
`1
`
`5
`1
`65‡
`
`6
`
`1
`3
`7
`5
`2
`17
`5
`3
`
`9
`9
`5
`2
`
`2
`2
`
`3
`
`2
`
`5
`1
`89
`
`19-23
`
`24
`25
`26-28
`29-32
`33
`34-47
`48-52
`53-55
`
`56-63
`64-70
`71,72
`73
`
`74,75
`76,77
`
`78
`
`79
`
`29,32,80-82
`83
`
`Abbreviations: EGFR, epidermal growth factor receptor; HER-2, human epidermal growth factor receptor 2; SM, small molecule; MEK, mitogen-activated
`protein-Erk kinase; mTOR, mammalian target of rapamycin; AS, antisense oligonucleotide; AB, antibody; VEGF, vascular endothelial growth factor; VEGFR, vascular
`endothelial growth factor receptor; PKC␣, protein kinase C alpha.
`*Selected tumors were breast, lung (small-cell and non–small-cell), prostate, colorectal, renal, and ovary.
`†Nineteen agents identified for this review.
`‡Actual total is 67, but two reports have been counted twice because they each included two agents and thus are found twice in this table (ISIS 5312 and ISIS 3521).
`
`end points (eg, response, change in target expression, progression-free sur-
`vival, either individually or in a multinomial combination) were abstracted, as
`were the actual outcome measures (eg, number of patients enrolled/eligible,
`number of patients with complete response, partial response, stable disease,
`progressive disease [PD], median time to progression, and overall survival).
`Data were not collected with respect to sex, age, or nature of the prior systemic
`treatment of patients.
`Calculations
`Once data abstraction was completed, several computations were under-
`taken. To determine the total response rate for a given trial, the total number of
`patients achieving either complete or partial response was divided by all eligi-
`ble patients. Trials without response reported or collected were classified as not
`reported for this outcome. Although we initially planned to report stable
`disease rate, because substantially differing duration requirements were used
`to define stable disease across the trials reviewed, we elected to calculate the
`nonprogression rate as a means of better standardizing the output of our
`review. To determine the percentage of patients with nonprogression, patients
`with PD as best response were subtracted from the number of eligible patients
`
`entered to give the total number of nonprogressors (non-PD). This figure was
`then divided by the total number of eligible patients. Although we recognized
`that assignment of PD as a best response was somewhat dependent on the
`timing of follow-up (usually between 6 and 12 weeks), this was less variable
`than the duration used to define stable disease (which ranged from a few weeks
`to ⬎ 6 months).
`Once the response rate and PD rate for each trial was calculated, addi-
`tional summary information was generated for presentation in tabular form.
`This included the calculation of overall rates of response (and nonprogres-
`sion), which were determined for each drug by grouping all patients in all trials
`of a given agent with either complete or partial response (or nonprogression)
`and then dividing by the total number of eligible patients.
`Response rates for trials and for drugs were categorized in the following
`range groupings: 0%, more than 0% to ⱕ 10%, more than 10% to ⱕ 20%, and
`more than 20%. Nonprogression rates were categorized as follows: ⱕ 30%,
`more than 30% to ⱕ 50%, more than 50% to ⱕ 65%, and more than 65%. The
`numbers of trials with response rates or non-PD rates in the ranges noted were
`displayed in table format according to a variety of groupings of the trial data.
`
`1348
`
`JOURNAL OF CLINICAL ONCOLOGY
`
`

`
`Phase II Designs Used for Targeted Therapy
`
`This included tables representing the number of trials in the various response
`categories by disease type, by individual drugs, by drugs grouped by target, and
`by population enrichment.
`Drug Approval
`Information on whether the agents under evaluation had received accel-
`erated or full US Food and Drug Administration (FDA) approval for use in any
`of the six tumor types as of June 2007 was also identified. A table was then
`created that listed the number of drugs with overall response rates in the ranges
`described earlier versus the number of drugs in each grouping that had
`achieved regulatory approval. The relationship between the four response
`categories and the probability of FDA approval was assessed by the exact
`Cochran-Armitage linear trend test.
`Agents With No Phase II Trials
`There were 12 agents from the original list of 31 for which no phase II
`trials in any of the six tumor types were found. Every reasonable effort was
`made to determine the reasons for this (eg, drug stopped development,
`skipped phase II altogether, was evaluated in other diseases than those we
`focused on, or went into combination phase II immediately after phase I).
`However, not all agents could be traced.
`
`RESULTS
`
`Agents and Trials
`Of the 31 agents surveyed, reports on single-agent phase II eval-
`uation were retrieved for 19 in at least one of the prespecified tumor
`types (Table 1).19-83 Altogether, 65 reports were identified (53 articles
`and 12 abstracts). Several reports contained results of evaluation of
`more than one tumor type or involved several different agents or dose
`levels; thus, the final tally of trials was 89. These were spread across all
`six tumor types, with the largest numbers in breast (21 trials) and lung
`cancers (13 trials in non–small-cell lung cancer and nine trials in
`small-cell lung cancer), followed by renal (15 trials), prostate (14
`trials), and ovarian cancers (six trials; Appendix Table A2, on-
`line only).
`
`Trial Design
`In the majority of
`Randomized versus nonrandomized designs.
`reports (51 of 65 reports; 78%), the investigational agent was evaluated
`using a nonrandomized single-arm design (Table 2). Randomization
`
`Table 2. Trial Design and End Points
`
`Design and End Point
`
`Nonrandomized
`Randomized
`Comparator arms:
`Placebo/standard
`Other investigational drug
`Other dose of same agent
`Primary end point
`Objective response
`Multinomial (response and
`progressive disease)
`Proportion progression free
`Progression-free survival
`Other
`
`No. of Reports
`(N ⫽ 65)
`
`51
`14
`
`Trials
`(N ⫽ 89)
`
`No.
`
`62
`27
`
`3
`4
`20
`
`51
`10
`
`8
`8
`12
`
`%
`
`70
`30
`
`57
`11
`
`9
`9
`13
`
`was used in 14 (22%) of 65 reports. However, in only two reports was
`there random assignment to a placebo (this included the randomized
`discontinuation phase of the BAY 43-9006 renal cell carcinoma trial33
`and a randomized phase II study of gefitinib v placebo in prostate
`cancer45). One study in prostate cancer randomly assigned patients
`between an experimental agent and an active corticosteroid control
`arm.74 In two reports,29,32 patients were randomly assigned between
`two different investigational drugs; in both, the random assignment
`was between ISIS 3521 (aprinocarsen) and ISIS 5132 in a noncom-
`parative phase II design. Finally, in nine reports, the random assign-
`ment was between various dose levels of the same investigational drug.
`End points. The primary end point on which the trial design was
`based was most commonly objective response (51 trials; 57%; Table
`2). In addition, 10 studies used a multinomial end point incorporating
`both response and nonprogression. Thus, objective response was the
`primary or coprimary end point in 61 trials. Only 16 trials were
`designed with end points of progression-free survival or the propor-
`tion of patients progression free at a prespecified time point. Some of
`the remaining studies were in prostate cancer and used measures of
`prostate-specific antigen (PSA) change (eg, PSA response, change in
`slope of PSA increase) as the primary end point. In addition, in two
`reports, toxicity was the primary end point.
`Population enrichment. Efforts to enrich the population under
`evaluation by restricting entry to patients of a particular molecular or
`histologic tumor subtype were undertaken in 18 (20%) of 89 trials. As
`can be seen in Table 3, in 14 studies, enrichment was on the basis of a
`molecular marker assessed in tumor.
`Hypotheses used in design. For those trials in which objective
`response was the primary end point, we attempted to identify the
`hypotheses used to derive the sample size from the methods sections of
`articles because we postulated that response rates of interest might be
`lower for targeted agents than those that have been traditionally used
`for phase II trials of cytotoxic drugs. Unfortunately, of the 51 trials in
`which response was indicated to be the primary end point, only 27
`described the hypotheses that had led to design and the planned
`sample size. Of these, 20 based the sample size and, thus, stopping rules
`on response rates of interest of 20% or higher. In seven trials, response
`rates in the 10% to 15% range were targeted.
`Sample size. Not surprisingly, the mean planned sample size for
`the trials reviewed depended on the design. For trials in which objec-
`tive response was the end point, the mean maximum sample size
`planned was 56 patients (based on the data reported for 35 trials).
`When progression-free survival or non-PD was the end point used,
`the mean maximum sample size planned was 115, and for the multi-
`nomial design, the mean maximum sample size was 41.
`
`Trial Results: Response and Nonprogression
`Response rates. Seventy-six of 89 trials reported objective re-
`sponse outcomes. In total, 38 trials had overall response rates of 0%. In
`the other 38 trials, objective responses were seen; in 19 trials, response
`rates were more than 10%, and in eight trials, response rates were
`more than 20%.
`Appendix Table A3 (online only) shows the reported response
`rates categorized in the ranges shown for all trials sorted by agent.
`Appendix Table A4 (online only) displays the same data but sorted by
`tumor type. Trial response results for all agents affecting the same
`target are shown in Appendix Table A5 (for example, all epidermal
`growth factor receptor inhibitors trials are displayed in one line in the
`
`www.jco.org
`
`1349
`
`

`
`El-Maraghi and Eisenhauer
`
`Table 3. Population Enrichment
`
`Basis for Enrichment
`
`EGFR expression
`HER-2 expression
`Histology
`BAC
`Clear cell
`c-kit
`Other (SD at 12 weeks)
`Total
`
`No. of Trials
`(n ⫽ 18)
`
`3
`7
`
`1
`2
`4
`1
`18
`
`Agent*
`Erlotinib (n ⫽ 2); cetuximab (n ⫽ 1)
`Trastuzumab
`
`Tumor Type*
`Lung (n ⫽ 1); ovary (n ⫽ 1); CRC (n ⫽ 1)
`Breast (n ⫽ 5); lung (n ⫽ 1); ovary (n ⫽ 1)
`
`Erlotinib
`Bevacizumab
`Imatinib
`Sorafenib
`Trastuzumab (n ⫽ 7); imatinib (n ⫽ 4);
`erlotinib (n ⫽ 3); bevacizumab (n ⫽ 2);
`sorafenib (n ⫽ 1); cetuximab
`
`Lung
`Renal cell carcinoma
`Lung (n ⫽ 3); ovary (n ⫽ 1)
`Renal cell carcinoma
`
`Reference
`No.
`
`50,51,53
`56-59,61,62
`
`49
`72
`65,67,68
`33
`
`Abbreviations: EGFR, epidermal growth factor receptor; CRC, colorectal cancer; HER-2, human epidermal growth factor receptor 2; BAC, bronchioloalveolar
`carcinoma; SD, stable disease.
`*Numbers in parentheses represent No. of trials.
`
`table; table online only). Finally, Table 4 provides the overall response
`rate by agent, pooling results for all trials (across all tumor types).
`Nonprogression rates. Results for nonprogression rates were also
`tabulated, but not all are shown. Table 4 lists the non-PD rates by agent
`(pooling across all trials for each particular agent). As can be seen,
`non-PD rates were variable, but several agents (sorafenib, cetuximab,
`temsirolimus, trastuzumab, and gefitinib) had non-PD rates of 50%
`or more overall.
`Although in most tumor types the ranking of agents by response
`rates or non-PD rates was similar (data not shown), renal cell carci-
`noma trials seemed to display a different pattern. Appendix Table A6
`(online only) shows overall response rate and non-PD rates in renal
`cell carcinoma studies by agent. High non-PD rates were seen with
`four agents (sunitinib, sorafenib, temsirolimus, and imatinib), but
`
`only one of these, sunitinib, had a response rate that was more than
`20%; the remainder had observed response rates less than 10%.
`
`Response and Non-PD Rates by Disease Type and
`Regulatory Approval
`To identify whether there were tumor-related patterns in the
`response or non-PD results and regulatory approval (as of June
`2007), we examined overall rates of non-PD and response by agent
`in each tumor type in the trials reviewed, as shown in Appendix
`Tables A7 and A8 (online only). Numbers within each tumor type
`are too small to apply statistics, but it was observed that no agent
`with a 0% response rate in a given tumor type received approval in
`that tumor type. Similarly, no agent with non-PD rates less than
`30% in a given tumor type received approval in that tumor type.
`
`Table 4. Overall Response and Non-PD Rates by Agent
`
`No. of Trials
`
`Overall Response
`Rate (%)
`
`Total No. of Patients
`in Response Rate
`Denominator*
`
`Overall Non-PD
`Rate (%)
`
`Total No. of Patients
`in Non-PD Rate
`Denominator*
`
`2
`3
`5
`2
`3
`7
`3
`5
`5
`1
`5
`6
`1
`9
`2
`2
`9
`17
`2
`89
`
`4
`NR
`5.3
`0
`3
`8
`0
`0
`0
`0
`12
`5
`0
`0
`28
`5
`18
`10
`0
`
`202
`NA
`97
`80
`169
`190
`52
`87
`71
`15
`200
`208
`21
`112
`127
`45
`562
`698
`46
`
`75
`NR
`NR
`36
`60
`66
`12
`13
`27
`40
`39
`23
`14
`21
`0
`0
`53
`50
`NR
`
`202
`NA
`NA
`80
`58
`190
`52
`60
`71
`15
`150
`208
`21
`112
`63
`29
`420
`627
`NA
`
`Agent
`
`Sorafenib
`Marimastat
`Bevacizumab
`BMS-275291
`Cetuximab
`Temsirolimus
`CI-1040
`Aprinocarsen
`ISIS 5132
`MG98
`Erlotinib
`R1155777
`SCH 66336
`Imatinib
`Sunitinib
`Semaxanib
`Trastuzumab
`Gefitinib
`ZD6474
`Total
`
`Abbreviations: PD, progressive disease; NR, not reported in any trial; NA, not applicable.
`*Response and non-PD rate denominators may not match if some trials had one or the other not reported. Only trials with data reported were used to calculate rates.
`
`1350
`
`JOURNAL OF CLINICAL ONCOLOGY
`
`

`
`Phase II Designs Used for Targeted Therapy
`
`What seems of interest is that, despite the suggestion that targeted
`drugs could not be screened by assessment of tumor shrinkage, the
`evidence suggests otherwise; of the 89 trials, 38 documented objective
`tumor responses, and 19 had response rates of more than 10% (and
`eight had response rates of ⬎ 20%). When examined by drug, of the 19
`drugs reviewed, 10 produced overall response rates (across all tumor
`types) ranging from 3% to 28%. Of these, seven have received full FDA
`approval for one or more indications. Furthermore, the data suggest a
`relationship between the level of the overall phase II response rate seen
`in the tumor types reviewed and the likelihood of achieving regula-
`tory approval.
`These results, although based on a relatively small number of
`targeted agents, are similar to a recent review of cytotoxic agents by
`Goffin et al.7 In that review, phase II outcomes of 46 cytotoxic drugs
`were compiled. The authors found a relationship between eventual
`regulatory approval and phase II response rate; drugs with overall
`response rates greater than 20% had a higher likelihood of approval
`than those with response rates of 10% to 20%, which, in turn, were
`more likely to receive approval than those with response rates of 0% or
`0% to 10%. The relationship seemed to hold true for breast, non–
`small-cell lung, ovary, and colorectal cancers but not for renal cell
`carcinoma or melanoma, although the numbers of randomized trials
`actually conducted in the latter two tumor types were fewer. In an
`accompanying editorial, Ratain11 argued that looking at the data as if
`the phase II response rates were diagnostic test results might be more
`appropriate. When he reanalyzed the data in this fashion, he identified
`that the negative predictive value of phase II results was quite high, but
`the positive predictive value was lower, being 0% in melanoma and
`renal cell carcinoma and from 33% to 75% for breast, ovarian, colo-
`rectal, and non–small-cell lung cancers when overall response rates
`were 20% or more. The problem with the approach of viewing the
`response rate outcome as if it were a diagnostic test is that it is guaran-
`teed to achieve high negative predictive values if no randomized trials
`are ever performed to assess survival impact when response rates are
`zero or low.
`Our review has limitations. First, not all targeted agents studied in
`the last decade were included. We identified phase II reports from the
`list of 31 targeted agents reviewed recently with respect to phase I
`design as a convenient approach to following up on the clinical devel-
`opment of those drugs. Second, we did not attempt to identify whether
`trials were conducted but never reported. Furthermore, we arbitrarily
`focused on six common solid tumor types. Thus, the overall single-
`agent response rates cited for each agent could be subject to some
`variation if data from unpublished studies or if trials in all tumor types
`had been included. In addition, our list of the original 31 agents was
`reduced to 19 because 12 drugs had no reported phase II single-agent
`studies in the tumor types that were of interest. Several of those agents
`were dropped from development after phase I, and others either
`remain in early phase II development or skipped phase II altogether to
`proceed into randomized combination studies.
`Nonetheless, there are interesting messages here. First, even rela-
`tively low rates of objective response may signal that an agent has
`potential for achieving regulatory approval on the basis of subsequent
`randomized data. It can be inferred from this observation that agents
`affecting targets that are meaningful in one or more cancer types
`should reasonably be expected to cause tumor shrinkage in at least
`some patients. Failing to see any evidence of response at all suggests
`that the drug is likely to fail in subsequent development. There were no
`
`Noteworthy was the observation that three of four agents in renal
`cell carcinoma with non-PD rates of more than 65% have been
`approved for renal cell carcinoma.
`
`Overall Response Rate and Regulatory Approval
`A number of the agents reviewed in this article had received FDA
`approval by June 2007. Figure 1 graphically displays the relationship
`between overall response rate for each agent (including all trials and all
`tumor types) and regulatory approval. Table 5 provides the same
`information clustering agents into the four response categories used
`earlier. Overall, seven of the 19 agents reviewed have received full FDA
`approval; one (gefitinib) received accelerated approval but failed to
`achieve full approval on the basis of randomized data. The P value of
`the exact Cochran-Armitage trend test for the relationship between
`response category into which agents fell and regulatory approval was
`P ⫽ .005 if the accelerated approval for gefitinib was not counted as
`FDA approved and was P ⬍ .0001 if the accelerated approval for
`gefi