Part 2 Resident and Fellows Lecture Series A pril 12 2016 Elizabeth GarrettMayer PhD Hollings Cancer Center Current State of Phase II Lots of changes in the past several years Phase I ID: 756322
Download Presentation The PPT/PDF document "phase ii STUDY designs:" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
phase ii STUDY designs: Part 2Resident and Fellows Lecture SeriesApril 12, 2016Elizabeth Garrett-Mayer, PhDHollings Cancer Center
Slide2
Current State of Phase IILots of changes in the past several yearsPhase I Phase II Phase III FDA approval ??Ceritinib: approved based on “phase I” trial data.Nivolumab phase I trial: 94 melanoma patients (5 doses)76 NSCLC patients (3 doses)
Pembrolizumab
phase I trial: >1200 patients
Phase I objectives have changed…what happens to phase II objectives?Slide3
Goals of Phase II StudiesProvide initial assessment of efficacy or ‘clinical activity’Screen out ineffective drugsIdentify promising new drugs for further evaluationFurther define safety and toxicity TypeFrequencyNewer goals:Define patient subgroups that respond to therapyIdentify need for biomarker-driven phase III designIdentify correct dose for further study based on efficacySlide4
Single Arm Phase I StudiesComparison is “fixed” constantBinary endpoint (clinical response vs. no response)Often one-sided testSimple set-up:Slide5
Phase II studies: Testing a hypothesis Null hypothesis: H0Usually the status quo (e.g. current response rate under the standard of care)If this is the response rate, then the treatment is not worth pursuingAlternative hypothesis: H1 or HaThe expected response rate with this new treatment.Or, lowest response rate we would consider sufficient for further study of the treatment.At the end of the study, we hope to reject the null hypothesis.Slide6
Review of alpha and power
Type I error
Alpha =
α
= probability
of Type I error (level of significance)
Beta =
β
= probability
of Type II error
Power =
1 -
β
Accept H
o
Reject H
o
H
o
is True
H
o
is NOT True
Type II errorSlide7
Single Arm Phase I StudiesComparison is “fixed” constantBinary endpoint (clinical response vs. no response)Often one-sided testSimple set-up:
Based on design parameters:
N=39
Conclude effective if 12 or more responses (i.e., observed response rate of ≥0.31)Slide8
Single Arm Phase I StudiesWhat if by the 15th patient you’ve seen no responses?Is it worth proceeding?Maybe you should have considered a design with an early stopping ruleTwo-stage designs:
Stage 1:
enroll N
1
patients
X
1
or more respond
Stage 2: Enroll an
additional N
2
patients
Stop trial
Fewer than X
1
respondSlide9
Revised Design: Simon Two-StageStage 1: enroll 19 patientsIf 4 or more respond, proceed to stage 2If 3 or fewer respond, stopStage 2: enroll 20 more patients (total N=39)If 12 or more of total respond, conclude effectiveIf 11 or fewer of total respond, conclude ineffectiveDesign properties? What about power? Slide10
Question 1:The power of this two stage design is:higher than in the single stage studylower than in the single stage studythe same as in the single stage studyI can’t remember what power is…Slide11
Two-stage designsSimon two-stage (1989)Used in exampleMANY designs fit the criteria“Optimal”Minimum expected sample size under H0Minimum maximum sample sizePreserves alpha and power, and permits early lookOther two stage designsGehan two-stage: first stage always stops if 0 responses
Balanced design: sample size in first and second stages are the same (or close). Ye and Shyr, 2007
Fleming two-stage: allows for early stopping for efficacy (not just futility).Slide12
Multi-stage designsPredictive probability design (Liu and Lee, 2008)Bayesian DesignLooks multiple times during the study to stop for futilityStopping is based on the probability of a successful trial at the end of the trial based on the accumulated data so far in the trial.Sequential designs (frequentist)Allow multiple looks at the dataControl the timing of looks and ‘spends’ some of the type I or II error at each look.O’Brien-Fleming spending functions.Slide13
Early StoppingFUTILITY stoppingMost of the designs discussed so far ONLY allow stopping if there is strong evidence that the treatment is not efficaciousCan also have early stopping for efficacyGenerally not popular in single arm studiesImportant to accumulate evidence to support claim of efficacyBut, not stopping prolongs time to launch phase IIISlide14
Safety stoppingDon’t forget to consider early stopping plans for excessive toxicityOften there is little information about safety of the phase II dose.What if the dose is too toxic?Data safety monitoring planCan be rule based and included in the protocolOften times there is an independent data safety committee that reviews safety dataSlide15
Frequentist vs. BayesianFrequentists: α and β errorsBayesians:
Quantify designs with other properties
General philosophy
Start with prior information (“prior distribution”)
Observe data (“likelihood function”)
Combine prior and data to get “posterior” distribution
Make inferences based on posteriorSlide16
Bayesian inferenceNo p-values and confidence intervalsFrom the posterior distribution: Posterior probabilitiesPredictive probabilities (e.g., Liu and Lee, 2008)Prediction intervalsCredible intervalsBayesian designsCan look at data as often as you like (!)Use information as it accumulates
Make “what if?” calculations
Helps decide to stop now or notSlide17
Bayesian DesignsRequires ‘prior’Reflects uncertainty about the response rateCan be ‘vague’, ‘uninformative’Can be controversial: inference may change
Prior Distribution
Response RateSlide18
Question 2:Which prior makes the most sense?
1
2
3
4Slide19
Bayesian design exampleSlide20
Posterior ProbabilitiesSlide21
Other priorsWhat if we had used a different prior?Assume informative “orange” priorSlide22
What about PFS as an endpoint?Progression-free survival as the primary endpoint in a single arm phase II trial has been controversialWhy?Slow growing cancer cannot be distinguished from truly stable or modestly shrinking tumor burden. Recall RECIST definition of stable disease. Interval censoring (next two slides).With biomarker defined patient populations, there may not be a good comparator study as a historical control. Without a comparator arm, the inferences due to the above issues are tricky. Slide23
Randomized Phase II Trial of Ridaforolimus in AdvancedEndometrial Carcinoma (Oza et al, JCO June 2015)
http://jco.ascopubs.org/content/early/2015/06/15/JCO.2014.58.8871Slide24
Interval-censoringProgression is generally observed at 2 to 3 month intervals at clinic visits.“8 week” clinic visit may occur at 7.5 weeks, 8.1 weeks, 9 weeks, etc.Slide25
Alternative to PFS in single arm study?If lack of progression is of interest, consider a “clinical benefit” (i.e. CR or PR or SD) at a fixed point in time.The “time” should be defined by a clinic visit. Example:The primary endpoint is clinical benefit, defined as having complete or partial response, or stable disease at the 4 month visit after initiation of therapyThe 4 month visit may occur with +/- one week of 4 months on study.Slide26
Randomized Phase II StudiesWhy randomized?Want some “control” or “prioritization”Primarily two different kinds of randomized phase II studiesPhase II selection design (prioritization)With recent immunotherapies, this approach is becoming a more attractive optionMultiple doses from phase I need to be exploredThese are often done in conjunction with a Phase IPhase II designs with reference control arm (control)Slide27
Phase II selection design (prioritization)Two parallel one arm studies (classic case)Do not directly compare arms to each other.Compare each to “null rate”Why? To compare to each other, you’ll need a study at least two times as large. Nivolumab, Pembrolizumab,
etc
:
The expansion cohorts ended up as randomized selection designs
Doses of the same agent were evaluated for optimal efficacy but not statistically compared.
Appropriate to use when:
Selecting among multiple new agents
Selecting among different schedules or dosesSlide28
“Pick the Winner” designs for selection (Simon et al., 1985; Sargent & Goldberg, 2001 )NOT appropriate when trying to directly compare treatment efficacies (not powered)Simon et al: Uses 2+ Simon two-stage designsAm arm must satisfy efficacy criteria of Simon design
Move the “winner” to phase III
Only have to pick winner if more than one arm shows
efficacy
Sargent & Goldberg:
Winner has to win by a margin, d, for the primary outcome.
Otherwise, additional information (e.g. safety, other clinical outcomes) is used to choose the winner.
Can be used when the goal is prioritizing which (if any) experimental regimen
or dose should
move to
the next phase when
no
a priori
information to favor one. Slide29
Randomized Phase II designs with reference arm (control)Includes reference arm to ensure that historical rate is “on target”Reference arm is not statistically compared to experimental arm(s) due to small NFrequent goal is to get information for planning a phase III trial.
Be careful not to confound the phase III endpoint.
Example: Randomized phase II with a cross-over option.
At progression, patients can switch from standard of care arm to novel treatment.
What happens to assessment of overall survival?Slide30
Randomized Phase II: Stopping RulesSimilar considerations for ‘early stopping’ should be included in randomized phase II studiesStopping is often specific to terminating one or more arms.If one arm is standard of care, stopping rules should:Consider if the novel treatment arm outcomes are significantly worse than standard of care armEnsure that patients in novel treatment arm are not experiencing unacceptably high toxicity due to treatment.Slide31
Question 3: Adaptive TrialsAdaptive trials:Always use Bayesian designsAllow the study team to alter the planned sample size based on low accrual even if they did not plan to change the sample size in advanceAllow prospectively planned modifications of aspects of the study design.Have smaller sample sizes than non-adaptive trials.Ensure patients will have a better chance of getting a more effective therapy.
3, 4, and 5
All of the aboveSlide32
FDA’s draft guidance on adaptive designs in drug development (Feb 2010) …, an adaptive design clinical study is defined as a study that includes aprospectively planned and specified modification or potential for modificationof one or more specified aspects of the study design and hypotheses, basedon analysis of data from subjects in the study. Analyses of the accumulatingstudy data are often performed at prospectively planned points within thestudy, may be performed in a fully blinded manner or in a non-blindedmanner, …
The term “prospective” here means that the adaptation was planned before
data were examined in an
unblinded
form by any personnel involved in
planning for the revision.
Revisions
made or proposed after an
unblindedinterim analysis raise major concerns about study integrity, possibledesign changes need to be
prospectively defined
and carefully implemented
to avoid risking irresolvable uncertainty in the interpretation of study results.
Robert T. O’Neill, PhD / OTS, CDER, FDASlide33
FDA’s draft guidance on adaptive designs in drug development (Feb 2010)The range of possible study design modifications that can be planned in the prospectively written protocol is broad. Examples include changes in the following: • study eligibility criteria (either for subsequent study enrollment or for a subset selection of an analytic population) • randomization procedure
•
treatment regimens of the different study groups (e.g., dose level,
schedule)
•
total sample size of the study (including early termination)
•
concomitant treatments used
•
planned schedule of patient evaluations for data collection
•
primary endpoint (e.g., which of several types of outcome assessments, which
timepoint
of assessment)
•
selection and/or order of secondary endpoints
• analytic methods to evaluate the endpoints (e.g., covariates of final analysis, statistical
methodology, Type I error control) Slide34
Adaptive RandomizationRandomization is “adapted” based on accumulated informationAdaptive on Outcome Assign treatments according to accumulated information about best treatment. Assign with higher probabilities to better therapiesSlide35
Example: Randomized phase II trial with two treatments: A and BSample size is 100First 30 patients are evenly randomized (15 per arm) After 30 patients, patients are adaptively randomized:Arm with the higher response rate has a higher likelihood of assignment of next patientHow much higher depends on: The difference in response ratesA ‘tuning parameter’ that is determined as part of the trial planning to ensure that imbalance is not too large. Can depend on the current sample size.Slide36Slide37
Standard Design
Adaptive Randomization
Begin
by randomizing
with
equal chance per
arm.
Then
, adjust probability
of
assignment to reflect
the
knowledge of the
best
treatment.
Stopping rules: drop an arm when
there
is “strong” evidence that
It has low efficacy OR
It has lower efficacy than
competing treatments
RANDOMIZE
Idarubicin
Ara-C
TroxIdarubicin
Trox
Ara-C
N=25
N=25
N=25
N=?
N=?
N=?Troxacitabine
in AML (Giles et al. 2003)Slide38
Adaptive Design: Troxacitabine in AMLSummary of trial results:TI dropped after 24th patientTrial stopped after 34 patients (TA dropped)
IA
10/18 = 56%
TA
3/11 = 27%
TI
0/5 = 0%
Complete responses by 50 daysSlide39
Buyer bewareAdaptive randomization does not always perform well and can lead to biased estimatesDesigns need to be fully vettedSee Thall et al. 2015
Pathologic example: N=15 in Arm A
N=85 for Arm BSlide40
Baskets and UmbrellasRecent years have seen an emphasis on molecular groupings of tumorsUmbrella trials: Basket trials:
Cancer type (e.g. NSCLC)
Molecular signature A
Molecular signature D
Molecular signature B
Molecular signature C
Lung
Colon
BreastSlide41
Umbrella Trial: Personalizing Therapy for Non-Small Cell Lung Cancer
BATTLE:
B
iomarker-integrated
A
pproaches of
T
argeted
T
herapy of
L
ung
Cancer
E
limination
Kim ES et al, Cancer Discovery 2011; 1(1): 44-53Slide42
BATTLE TrialPatients of same cancer type (here NSCLC) assigned different treatments based on molecular profilePrimary endpoint = Disease control at eight weeks Four parallel phase II studiesEqual randomization followed by adaptive randomization Based on disease control rate for molecular subgroupPatients are assigned at higher probability to treatment with higher disease control rate for their molecular profile.
Slide courtesy of Betsy HillSlide43
Slide from Jack Lee:
http
://www.winsymposium.org/wp-content/uploads/2013/09/L-3.02-J.-Jack-Lee.pdfSlide44
Kim ES et al, Cancer Discovery 2011; 1(1): 44-53Slide45
Other Umbrella TrialsBATTLE-2I-SPYI-SPY-2http://www.ispy2trial.org/Slide46
What is a basket trial?New and evolving form of trial design based on hypothesis that presence of a molecular marker predicts response to targeted therapy independent of tumor histology.Mutations identified, patients assigned to specific treatment arm (or randomization to subset of treatments) based on mutation statusSeveral independent parallel phase II trials in one study~10-15 patients per disease typeTreatment assignment based on molecular profile
Redig
and
Janne
, JCO, Feb 9
2015.Slide47
Basket TrialsBasket design may start with the use of a targeted therapy in an unselected population followed by Next Gen Seq (NGS) in patients who respond to identify genetic biomarkers for subsequent prospective screening.Basket trials have generated an enormous amount of interest: they implement a hypothesis-driven strategy incorporating precision medicine into clinical trials even for mutations that are rare or difficult to study solely within a disease-specific context.
Redig
and
Janne
, JCO, Feb 9
2015.Slide48
Example trial: CUSTOMMolecular Profiling and Targeted Therapies in Advanced Thoracic Malignancies Trial Seeking to identify molecular biomarkers in (1) advanced NSCLC, (2) small cell lung cancer, and (3) thymic malignancies5 targeted agents:Erlotinib against EGFR mutationsSelumetinib (MEK inhibitor) against KRAS, HRAS, NRAS and BRAF mutationsMK2206 (AKT inhibitor) against PIK3CA, AKT1 and PTEN mutationsLapatinib against HER2 mutationsSunitinib
against KIT and PDGFRA
mtuations
.
FIFTEEN (3 x 5) study arms.
Lopez-Chavez et al., JCO, Feb 9, 2015 (
epub
ahead of print)Slide49
CUSTOM statsEach arm is an independent phase II trial using an optimal Simon two-stage design.14 arms: null response rate = 10%; alternative = 40%EGFR mutant NSCLC arm: null response rate = 30%; alternative = 60%.Lopez-Chavez et al., JCO, Feb 9, 2015 (epub ahead of print)Slide50
CUSTOMLopez-Chavez et al., JCO, Feb 9, 2015 (epub ahead of print)Slide51
Lopez-Chavez et al., JCO, Feb 9, 2015 (epub ahead of print)Slide52
StrengthsAbility to identify favorable response to targeted therapy with small N and validate targetOnly 15 pts with NSCLC with EGFR mutation enrolled, yet significant results (60% ORR).(However, Arm closed early: overwhelming evidence of efficacy of erlotinib elsewhere)Proof-of-principle validation of putative targetSlide53
ChallengesAccrualPrevalence of Mutations
Lopez-Chavez et al., JCO, Feb 9, 2015
Figure 2: Frequency of
genetic abnormalities in
(A) NSCLC and (B) small
cell lung cancer.Slide54
Other basket trialsNCI-MATCH:Plans to screen 3000 pts with enrollment of at least 1000 pts for targeted drug combinations. Independent of histology. Launching early 2015.Expected to have 20 different treatments, 20 pharma companies, and as many as 2400 sites.NCI-IMPACT:Randomly assigns pts with a mutation in specific genetic pathway to either targeted therapy for pathway or treatment not known to be pathway specific (NCT0182784)Slide55
Basket Trial SuccessSuccess depends on strength of data linking target and targeted therapyTwo key conditions:Tumor must depend on pathwayTargeted therapy must be able to reliably inhibit the targetSlide56
SummaryPhase II is changing.Lots of talk about meshing Phase I/II/III so keep your eyes and ears open.Consider the following when designing your study:Early stopping for futility and toxicityAccrual….harder than you think!Adaptive design? Performance characteristics are important.Talk to your statistician as early as possible in the planning stages.Slide57
References Ceritinib phase I trial: Shaw et al., NEJM 2014; 370:1189-1197.Nivolumab trials: Topalian et al., NEJM 2012; 366 (26):2443-2454 (see also trial protocol as supplementary material)Topalian et al., JCO 2014; 32(10): 1020-1031Phase II design guidance: Seymour et al., Clinical Cancer Research 2010; 16(6):1764-1769.Two-stage designs Simon R. Control Clin Trials. 1989 Mar;10(1):1-10.Gehan
EA. J Chronic Dis. 1961 Apr;13:346-53
Ye
and Shyr. Clinical Trials 2007; 4(5): 514-24.
FDA draft guidance on adaptive designs:
http://www.fda.gov/downloads/Drugs/Guidances/ucm201790.pdf
Predictive probability design: Liu and Lee, Clinical Trials. 2008; 5(2):93-106.
PFS example: Oza
et al., JCO 2015;
Epub
ahead of print.Slide58
References Randomized prioritization designs: Simon R, Wittes RE, Ellenberg SE. Cancer Treatment Rep 1985;69:1375–1381. Sargent D and Golberg R. Statistics in Medicine 2001; 20: 1051-1060.Adaptive Randomization example: Giles FJ et al. J Clin Oncol. 2003 May 1;21(9):1722-7.Adaptive randomization:
Thall
PF and
Wathen
JK European Journal of Cancer 2006;
Dangers of AR:
Thall
PF et al., Annals of Oncology 2015, Epub ahead of print. doi:10.1093/annonc/mdv238BATTLE trial:
Kim
ES et al, Cancer Discovery 2011; 1(1): 44-53.
Zhao, et al. Clinical Trials 2008; 5:181-193
Basket Trials :
Redig
and
Janne
, JCO, Feb 9 2015.Lopez-Chavez et al., JCO, Feb 9, 2015 (epub ahead of print)