Design Concept for a Confirmatory Basket Trial

Transcription

1 Design Concept for a Confirmatory Basket Trial Robert A. Beckman, MD 1 and Cong Chen, PhD 2 1 Professor of Oncology & of BiostaDsDcs, BioinformaDcs, and BiomathemaDcs Lombardi Comprehensive Cancer Center and InnovaDon Center for Biomedical InformaDcs Georgetown University Medical Center Founder and Chief ScienDfic Officer, Oncomind, LLC 2 Director, BiostaDsDcs and Research Decision Sciences, Merck & Co, Inc.

2 Acknowledgements Co- authors on the present work: Cong Chen led group; co- led concept development; led all sta6s6cal and simula6on work Zoran Antonijevic, Amgen Rasika Kalamegham, Genentech Pathway design subgroup, addidonal members: ChrisDne Gausse, Merck SebasDan Jobjorrnsson, Chalmers Lingyun Liu, Cytel Sammy Yuan, Merck Yi (Joey) Zhou, Ultragenyx Advisor: Sue- Jane Wang, FDA Pathway design subgroup is one of 5 working subgroups of the DIA Small Popula6ons Workstream, a group of 50 stadsdcians and clinicians from industry, academia, and nadonal health authorides (FDA and EMEA) Small populadons workstream is part of DIA Adap6ve Design Scien6fic Working Group (ADSWG), a group of 180 stadsdcians and clinicians from industry, academia, and nadonal health authorides (FDA and EMEA)

3 Small PopulaDons Within A Common Disease The increasing discovery of molecular subtypes of cancer leads to small subgroups that actually correspond to orphan or niche indicadons, even within larger tumor types Enrolling enough padents for confirmatory trials in these indicadons may be challenging. The shia to a molecular view of cancer requires a corresponding paradigm shia in drug development approaches Exclusive use of one indicadon at a Dme approaches will not be sustainable

4 Approaches to development based on predicdve biomarkers OpDmized co- development of a single drug and its companion diagnosdc Gives a clear hypothesis and answer and sdll has a role in selected instances Will be challenging to do in niche indicadons Umbrella trials One tumor type with muldple drugs and predicdve biomarkers PaDents are matched to drugs based on predicdve biomarkers CooperaDon among muldple sponsors Examples: BATTLE, I- SPY, Lung- MAP Basket or bucket trials MulDple tumor types with one drug and predicdve biomarker Approval based on pooled analysis Premise is that molecular subtype is more fundamental than histology Single sponsor

5 Agenda IntroducDon General Design Concept for a Basket Trial Challenges of Basket Trials and RecommendaDons for Overcoming Them Detailed Design ConsideraDons Conclusions

6 The Original Basket: ImaDnib B pa6ents with 40 different malignancies with known genomic mechanisms of ac6va6on of ima6nib target kinases KIT, PDGFRA or PDGFRB Imatinib mg BID primary Endpoint ORR Synovial Sarcoma Aggressive Fibromatosis Dermato- fibrosarcoma Protuberans Aggressive Systemic mastocytosis Hyper- eosinophilic syndrome Myelo- prolifera6ve disorder 1/16 (6%) 2/20 (10%) 10/12 (83%) 1/5 (20%) 6/14 (43%) 4/7 (58%) 13 centers in consortium: North America, Europe, Australia Lead to supplemental indications for these 4 subsets after pooling with other trials and case reports October Blumenthal. 25, 2017 Innovative trial designs to accelerate the availability of highly effective anti-cancer therapies: an FDA perspective, AACR 2014

7 Basket Trials to Date A similar design to ImaDnib B2225 was endorsed at a Brookings/Friends Conference in 2011 Common features: Exploratory and opportunisdc in nature Single- arm trials with ORR as primary endpoint Intend to use pooled populadon for primary analysis to gain broader indicadon across tumor types (individual tumor type is not adequately powered) Involve possibly transformadve medicines in padents with great unmet need and seemingly excepdonally strong sciendfic radonale

8 Issues Clinical data to support pooling my be limited, and treatment effect may differ between tumor types Vemurafenib works in melanoma with BRAF V600E mutadon but not colorectal cancer with same mutadon Not all drugs hoped to be transformadonal live up to this promise Response rate may not predict overall survival Single arm trials are subject to padent selecdon bias PredicDve effect of a biomarker is confounded with the prognosdc value which is oaen unknown Health authorides can be non- commijal upfront

9 DIA Small PopulaDon Pathway Subteam Can we develop a generalizable confirmatory basket design concept with stadsdcal rigor? Applicable not only to excepdonal cases, but to all effecdve medicines in any line of therapy Follow exisdng accelerated and standard approval pathways to increase drug approvability This would have muldple benefits Increase and accelerate access to effecdve medicines for padents in niche indicadons Provide sponsors with cost- effecdve opdons for development in niche indicadons Provide health authorides with more robust packages for evaluadon of benefit and risk

10 GENERAL DESIGN CONCEPT

11 SELECTION PRUNING (External Data) PRUNING (Interim endpoints) Accelerated Approval Op6on Consistent trend in definidve endpoint FULL APPROVAL (Pooled analysis of defini6ve endpoint)

12 Features of the Design (I) Tumor histologies are grouped together, each with their own control group (shared control group if common SOC) Randomized control is preferred Single arm cohorts with registry controls may be permijed in excepdonal circumstances as illustrated by imadnib B225 and others In an example of pardcular interest, each indicadon cohort is sized for accelerated approval based on a surrogate endpoint such as progression free survival (PFS) This may typically be 25-30% of the size of a Phase 3 study IniDal indicadons are carefully selected as one bad indicadon can spoil the endre pooled result

13 Features of the Design (II) IndicaDons are further pruned if unlikely to succeed, based on: External data (maturing definidve endpoint from Phase 2; other data from class) Internal data on surrogate endpoint Sample size of remaining indicadons may be adjusted based on pruning Type I error threshold will be adjusted to control type I error (false posidve rate) in the face of pruning Pruning based on external data does not incur a stadsdcal penalty Discussed in more detail later in talk Study is posidve if pooled analysis of remaining indicadons is posidve for the primary definidve endpoint Remaining indicadons are eligible for full approval in the event of a posidve study Some of the remaining indicadons may not be approved if they do not show a trend for posidve risk benefit as judged by definidve endpoint

14 CHALLENGES OF BASKET DESIGNS AND RECOMMENDATIONS FOR OVERCOMING THEM

15 Challenge 1: Risks of Pooling One of more bad indicadons can lead to a failed study for all indicadons in a basket Histology can affect the validity of a molecular predicdve hypothesis, in ways which cannot always be predicted in advance Vemurafenib is effecdve for BRAF 600E mutant melanoma, but not for analogous colorectal cancer (CRC) tumors This was not predicted in advance but subsequently feedback loops leading to resistance were characterized

16 Addressing challenge 1 Basket trials are recommended primarily aaer there has been a lead indicadon approved (by opdmized convendonal methods) which has validated the drug, the predicdve biomarker hypothesis, and the companion diagnosdc Example, melanoma was lead indicadon preceding Brookings trial proposal in V600E mutant tumors IndicaDons should be carefully selected IndicaDons should be pruned in several steps before pooling

17 Challenge 2: AdjusDng for Pruning Pruning indicadons that are doing poorly on surrogate endpoints may be seen as cherry picking This can inflate the false posidve rate, an effect termed random high bias Addressing the challenge: Emphasize use of external data, especially maturing Phase 2 studies, for pruning Pruning with external data does not incur a penalty for random high bias Apply stadsdcal penalty for control of type I error when applying pruning using internal data Methods for calculadng the penalty are described in stat methods papers (see key references) Rules for applying penalty must be prospecdve Penalty is not large enough to offset advantages of design

18 Challenge 3: Will the companion diagnosdc assay generalize across indicadons? AnalyDcal properdes of assay may depend on Dssue type Cutoff between biomarker posidve and negadve may vary between Dssue types for a condnuous biomarker Addressing the challenge: AnalyDcal validadon of the assay for all relevant indicadons prior to study start Prior to study start, recommend biomarker stradfied randomized phase 2 studies to set provisional cutoffs for condnuous biomarkers in each indicadon to the extent feasible

19 Challenge 4: Availability of Dssue Tissue sampling and processing are variables that can greatly affect the outcome of a study based on a predicdve biomarker Basket studies will require cooperadon and uniformity across departments organized by histology Addressing the challenge: The sponsor must have extensive contact with the pathology department and relevant clinical departments at all invesdgadve sites and provide standard methods for Dssue sampling, handling, and processing The sponsor should engage an expert pathologist who is dedicated to training prior to trial start, and troubleshoodng during the trial

20 Challenge 5: Clinical validity of the predicdve biomarker hypothesis The clinical validity of the predicdve biomarker can only be verified by inclusion of biomarker negadve padents in the confirmatory study Addressing the challenge Recommend a smaller pooled, stradfied cohort for biomarker negadve padents, powered on surrogate endpoint Would need to expand the biomarker negadve cohort (to evaluate definidve endpoint) if surrogate endpoint shows possible benefit Prior evidence should permit this if: An approved lead indicadon has already provided clinical evidence for the predicdve biomarker hypothesis Prior phase 2 studies support the predicdve biomarker hypothesis in other indicadons

21 Challenge 6: High Screen Failure Rate Pro: padents will have access to tailored therapy Con: padent has a high risk of being a screen failure if biomarker posidve subgroup is low prevalence Addressing the challenge: Study should provide a broad- based test like NGS which will give the padent some guidance on alternadve therapies if they are screen failures for basket study

22 Challenge 7: Interim endpoints may not predict definidve endpoints Addressing the challenge: Prefilter indicadons based on maturing definitve endpoint data from phase 2 See Figure 2 Require consistent trend in definidve endpoint for final full approval

23 Phase 2 Influencing Phase 3 AdaptaDon: The Phase 2+ Method Beckman, R.A., Clark, J. & Chen, C. IntegraDng predicdve biomarkers and classifiers into oncology clinical development programmes. Nature Reviews Drug Discovery 10, (2011)

24 Another Possible Source of External Data Real World Data (RWD) from Off- Label Use Impact of RWD on basket trial performance is currently under study in a project led by postdoctoral fellow Daphne Guinn

25 DETAILED DESIGN CONSIDERATIONS

26 Designs to Be Compared Sample size changes aaer pruning D0: No pruning and no change (benchmark) D1: No increase to sample size aaer pruning D2: Sample size in pooled analysis aaer pruning remains same as planned for the trial (SS) D3: Sample size for trial remains same aaer pruning as planned for the trial (SS) Designs Overall Trial Pooled Popula6on D0 SS SS D1 <SS <SS D2 >SS SS D3 SS <SS

27 Type I error control k tumor indicadons each with sample size of N and all with 1:1 randomizadon An interim analysis is conducted at informadon fracdon t for each tumor indicadon and a tumor will not be included in the pooled analysis if p- value>α t The pooled analysis will be conducted at α* so that the overall Type I error is controlled at α when there is no treatment effect for any tumor (H0) What is α*?

28 Solving for adjusted alpha (α*) Let Y i1 be the test stadsdcs based on informadon fracdon t, and Y i2 be the test stadsdcs based on the final analysis of data in the i- th cohort (i=1, 2,,k) Suppose that m cohorts are included in the final analysis (m 1), and let V m be the corresponding test stadsdcs. The probability of a posidve outcome in pooled analysis is or Q 0 (α* α t, m)= Q 0 (α* α t, m)= Pr H ( {Y 0 i1 > Z α t 1 for i=1,,m}, {Y j1 < Z 1 α for j=m+1, k}, V t m > Z 1- α* ) α* is solved from below where c(k, m) = k!/((k- m)!m!) Pr H ( {Y 0 i1 > Z α t k m=1 1 for i=1,,m}, V m > Z 1- α* )(1- α t ) (k-m) c(k, m) Q 0 (α* α t, m) = α

29 α* under different design opdons α* decreases with increasing k as expected, but its relationship with α t is complicated with the interplay between cherry-picking and futility stopping.

30 Strong control of Type I error This problem is sdll open Challenge: one or more strongly posidve indicadons can drive an overall pooled posidve result and negadve indicadons are carried along ConsideraDons: Should a basket trial for 5 indicadons have lower type I error than those 5 indicadons studied individually (i.e two- sided X 5)? Is this different from a convendonal phase 3 trial in a broad populadon? A posidve result in such a trial is oaen driven by a subgroup, known or unknown SimulaDon of type I error under specific cases is planned in my group

31 Comparison of operadng characterisdcs k=6 tumor indicadons with total planned event size (kn) ranging from The true treatment effect is log(0.6), or hazard rado of 0.6 in a Dme- to- event trial Pruning occurs at when half of the events have occurred Number of acdve indicadons (g) with target effect size ranges from 3 to 6, with remaining ones inacdve

32 Study power and sample sizes under different pruning and pooling strategies Planned events Number of active tumors Power (%) for a positive study Exp. number of events for pooled population Exp. number of events for overall study D0 D1 D2 D3 D0/D2 D1 D3 D0/D3 D1 D

33 An ApplicaDon of Special Interest A randomized controlled basket trial with 1:1 randomizadon in 6 tumor indicadons, each targedng a hazard rado of 0.5 in PFS with 90% power at 2.5% alpha 88 PFS events and 110 padents planned for each indicadon PFS analysis is conducted when all are enrolled D2 is applied to keep total sample size at 660 in pooled populadon targedng 430 death events The study has ~90% power to detect a hazard rado of 0.7 in OS at 0.8% alpha (aaer taking the penalty) assuming ρ=0.5 Observed hazard rado ~0.79 or lower for a posidve trial in pooled populadon (vs ~0.84 under D0) PotenDal to gain approvals in 6 indicadons based on comparable sample size to a convendonal Phase 3 trial

34 Conclusions It is feasible to create a general design concept for a basket study that is suitable for many agents MulDple challenges can be addressed with careful planning Benefits include: Increased and earlier padent access to targeted therapies for small subgroups Cost- effecdve methods for sponsors to develop targeted agents in small subgroups More robust datasets for health authorides to assess benefit- risk in these small padent groups

35 Key References Li, Wen, Chen, Cong, Li, Xiaojun, and Beckman, Robert A. EsDmaDon of treatment effect in two- stage confirmatory oncology trials of personalized medicines. StaDsDcs in Medicine, in press (2017). Beckman, Robert A., Antonijevic, Zoran, Kalamegham, Rasika, and Cong Chen. AdapDve Design for a Confirmatory Basket Trial in MulDple Tumor Types Based on a PutaDve PredicDve Biomarker. Clinical Pharmacology and TherapeuDcs, 100: (2016). Yuan, Shuai S, Chen, Aiying, He, Li, Chen, Cong, Gause, ChrisDne K, and Robert A. Beckman. On Group SequenDal Enrichment Design for Basket Trial. StaDsDcs in BiopharmaceuDcal Research, 8: (2016). Chen, Cong, Li, Nicole, Yuan, Shuai, Antonijevic, Zoran, Kalamegham, Rasika, and Robert A. Beckman. StaDsDcal Design and ConsideraDons of a Phase 3 Basket Trial for Simultaneous InvesDgaDon of MulDple Tumor Types in One Study. StaDsDcs in BiopharmaceuDcal Research, 8: (2016). Magnusson BP, Turnbull BW. Group sequendal enrichment design incorporadng subgroup selecdon. Stat Med. 2013;32(16): Heinrich MC, Joensuu H, Demetri GD, Corless CL, Apperley J, Fletcher JA, et al. Phase II, open- label study evaluadng the acdvity of imadnib in treadng life- threatening malignancies known to be associated with imadnib- sensidve tyrosine kinases. Clin Cancer Res. 2008;14(9): Demetri G, Becker R, Woodcock J, Doroshow J, Nisen P, Sommer J. AlternaDve trial designs based on tumor genedcs/pathway characterisdcs instead of histology. Issue Brief: Conference on Clinical Cancer Research 2011; October hjp:// clinical- cancer- research , 2017