U.S. Food and Drug Administration

Transcription

1 U.S. Food and Drug Administration Notice: Archived Document The content in this document is provided on the FDA s website for reference purposes only. It was current when produced, but is no longer maintained and may be outdated.

2 Clinical Trial Designs Sue Jane Wang, Ph.D. Associate Director, Adaptive Design & Pharmacogenomics Office of Biostatistics, Office of Translational Center for Drug Evaluation and Research, US Sciences FDA Presented at CDER Small Business Assistance Clinical FDA White Oak Campus, Building#31, April 21, 2011 Trials Forum

3 Disclaimer The views expressed in this presentation are not necessarily of the US FDA 2

4 Outline Overviewof drug development Key Elementsof StudyDesigns Exploratory studies Adequate andwell Controlled Studies SomeRegulatoryGuidance 3

5 Traditional paradigm Drug Research &Development 4

6 Emerging Trend for Drug R&D Search for Search for Search for promising promising promising Statistical Analysis COMPOUND DRUG PLAN Stage Stage 5

7 (1) Key Elements of StudyDesign Whatquestion(s)does (do) the studywant to address? Whatis the experimentalunit? Whatmeasures clinicalbenefit and/or clinical risk? Whatdose regimens to study? 6

8 Key Elements of StudyDesign (2) Doesthe study have a comparator? Ifso, are there available (approved) therapies? Isthe study design ethically sound? Forpatients in the study Forfuture patients to be studied Arethere important prognostic factors of disease? 7

9 (A&WC) Key Elements of StudyDesign (3) Howlarge is the number of experimental units needed? Will the experimental treatment be evaluated sequentially or concurrently? Will the design elements be fixed or may change? Whatconstitutes a successful trial? Trialsin earlyphase or early stage Trialsthat are adequate and well controlled 8

10 Trials Multiple Objectives in Exploratory Clinical 9

11 Aims in Early Phase or Early Stage Understand Is Is there drug there dose tolerability of activity response?? Research mode Plenty of Little interest to commit large generally small sample size Not unusual to a compound flexibility resources see PI initiated phase I yet trial 10

12 Types of Exploratory Studies Dose Ranging Exposure response Dose escalation Single vs. multiple doses Placebo controlled? Dose Response Target dose(s) estimation 11

13 Dose Regimen of Interest Phase I primary Maximum Phase II Minimum primary Maximum interest is Tolerated Dose Tolerability (MTD) interests: POC, Dose Effective Dose Safe Dose Maximum Utility Dose Minimally (MSD) (MED) (MUD) based on Selection current info acceptable dose (MAD): the lowest dose that has a utility of at least, say, 70% 12

14 Garrett-Mayer, Clinical Trials

15 Continual Reassessment Method DefineDLT, e.g., any grade 3or higher toxicity occurred in 1 st 4 wks on study TITE CRM DefineDLT, e.g., any grade 3or higher toxicity occurred in three monthsor longer of thepatients beingon study Allow staggered entryî shorten studyduration MTD, e.g.,dose level achieved adlt closest toλ% 14

16 Point est. vs. Interval est. DLTs per patient treated at a dose level Toxicity Point Estimate Exact 95% Confidence Interval 0 of 3 0% 0 71% 0 of 6 0% 0 46% 1 of % % 2 of % % ide intervalestimatesbased on3 and6patients.when2of6patientshavedlt, he MTDmayhavebeen exceeded,but there is a very good chance that it hasnot. ith the3+3design, the summarydatamaybe too sparse tobe reliable. 15

17 Continual Reassessment Method Often PI-Initiated in oncology 22 patients studied at 4dose levels (30 45 mg/m2), JCO patients studied JCO2004 at 5dose levels (4 mg/m2 to 20 mg/m2/wk) 37 patients (from2 6400ng/kg) EJC 2006 centers) studied at14 dose levels (10 to 16

18 CRM and its Modifications Objective: Estimate MTD Sequential Design Bayesian or Maximum Likelihood Assume DoseResponse Curve Data Source for Dose ResponseCurve Assumptions for Dose Response Model Start Dose Target Dose Estimation based on % DLT 17

19 Phase 1 Trial CRM Method Patient population: Androgen Independent Prostate Cancer Proposed Dose Levels 30, 35, 40 DLT: any toxicity resulted in a delay of 1wk in Docetaxel or Gleevec or resulted in a dose reduction during Estimated MTD Dosethat achieved a DLT closet to 30%,45 combo JCO 22: , 3329,

20 To determine MTD In principle, theprobability of DLT shoulddependon the contextof thedisease and the treatment, the toxicity rates that are seen in alternative therapies available to thepatient (if any) and thebalancingof the relationship between potential toxicities and potentialbenefitof the treatment. In some trials, a 20% DLT ratemay be toohigh,while inothers it may be too low 19

21 Early Assessment of Drug Activity Dose ranging, dose response studies What is an appropriate dose range? Will there be dose response? Plenty of learning and exploration Might add higher dose(s) or lower dose(s) Preliminary assessment of drug activity Internal decision making for further development Fixed, Adaptive vs Model based Design 20

22 Study Designs Balanced Adaptive Design Frequentist Optimal Design Adaptive Bayesian 21

23 POC and Dose-Response POC: Is there evidence ofdose response? Any evidence of treatment effect? Howwell dose response curve is estimated? Dose Finding:Which dose tobring tonext next phaseof thedrugdevelopment? stageor ICH E 4: the purposeof dose response information is to find the smallest dosewith adiscernible useful effect 22

24 23

25 Utility Function an example 24

26 Miller et al

27 When is adaptive design useful compared to optimal design under Bayesian framework? If differences between possible dose effect large (in relation to variability of data there is gain from adaptive dosing scenarios are in interim analysis), If scenarios similar enough or variability large, decisions based on interim data could lead into wrong direction especially if assumed dose range is the relatively uninformative part, given true DR unknown Miller et al,

28 The purpose of dose-response information is to find the smallest dose with a discernible useful effect (ICH E-4) 27

29 Early Phase or Stage of Drug Development Uncertainty aboutdrug activity,moa Shouldbe an exploratory or modeof investigation learning Trade off between false negative vs. false positive vs. estimationproblem Prerequisitesbefore launching confirmatory trial a 28

30 Early stage in Drug Development Dose ranging; Dose response; promising doseprofile Exposure response; have plenty of flexibility early uncertainties for learning & for quantifying many AIM: hope to maximize probability of for future planning correct selection and 29

31 Dealing with Learning (i) Combine Learn that Formalizes Learn in Exploratory Trials Hypothesis generation Learn Change H 0 Selection Global H 0 Wide Flexibility Final Analysis Patient pop n? Corr (Early,late endpoints)? Dose range? Dose regimen(s)? Effect size? etc. Probability of correct selection Estimation & Quantify Uncertainty 30

32 Dealing with Learning (ii) (ii) Formalize learning to plan confirmatory trial Use of point estimate of effect size from ph II to plan ph III can be valuable in prediction of useful doses. But, may be too optimistic with usual α-level and 1-β level Æpossibly regression toward mean *Wang et al. (2006, Pharmaceutical Statistics) 31

33 Learn Trial versus Confirm Trial Clinical questions to be addressed in early stage trials for decision making are naturally different from late stage trials for rigorous inference Caveats: if pursued as one trial for inference, learning data is a part of inference data that are subject to multiplicity if pursued as learn vs. confirm, adaptive elements can be built in within learn trial versus within confirm trial, but, separately 32

34 Learn and Confirm Within Trial of Most Enthusiastic Interest Statistical theoryhas shown that learn and confirmwithin the same study yield liberal type I error rate and overestimate performance characteristics ifmultiplicity adjustment is not formally accounted for, e.g., commonly used internal cross validationof model built using the same or apartof thedata e.g., in gene expression orwhole genome screening or gene association studies, rigorous confirmationof theprediction accuracy should beperformed in an independent dataset Multiplicity issue and regression to themean issue 33

35 Adaptive Design Prospectivelyplanned opportunity for modification of one or more specified aspects of the study design and hypotheses based on analysis of data (usually interim data) from subjects in the study Analysisof the accumulating study data are performed at prospectively planned time points within the study Analysescan be performed in a fully blinded manner or in an unblinded manner, and can occur with or without formal statistical hypothesis testing 34

36 Non inferiority and Objectives Clinical Trial Designs Randomized Crossover Randomized Enrichment Group Superiority trials Controlled withdrawal strategy sequential trials Trials trials 35

37 Concepts & Terminology Design: Conventional vs.adaptive Plan: ProspectivePlan vsreactiveunplannedchanges Adaptations: Unblinded vs Blinded non comparative InterimAnalysis:beyond ICH E9 Bias: Statistical vsoperational Study: Exploratory vsa&wc (can have expl element) PhI, II, III, confirmatory, seamlessph 2/3 notused GroupSequential Trial &Beyond: FirewallsAdaptive MonitoringProcess/Procedure/Documentation 36

38 Some DesignConsiderations When large amount of data in early phase studies collection is not feasible When large practiced in amount of data collection late phase studies is routinely Accurate modeling relies on large amount of data Desired to pursue modeling of little data in early phases because 37

39 Designwith Little vsmore Philosophy 1 Mine the data and what do the data tell? Many slices of the data, then, give the clinical question and simultaneously the answer. What the new compound/drug behaves in the trial, not what the design should be to answer if the new compound/drug is useful Philosophy 2 What is the clinical question or objective? Choice of study design to address the Qs. 38

40 Learn/preliminary confirm for Decision Correct go/no go decision is critical To improve the probability of correct selection based on early phase exploration relies upon Being able to make no go decision Being able to also make go decision Patient population starts narrow cannot anticipate degree of heterogeneity, effect size(s) Dose groups start a few (or more) in exploratory trials If still in a narrow patient population, even if picked promising dose, uncertainty in phase III with broad patient population M&Sfor planning A&WC 39

41 &Well Controlled (A&WC)21CFR Adequate Not exploratory adaptive design clinical trial In addition to experimentwise type I error rate control Should possess the following characteristics clear statement of the objectives, proposed and actual methods of analysis in protocol, SAP, and reports design that permits a valid comparative evidence of T-effect methods of adequate assurance of patient selection patient assignments that minimize bias, group comparability minimize bias on all parties: pts, investigator, data analyst endpoints well-defined that address clinical primary hypo. analysis results interpretability of the effects of drug 40

42 Regions in SchizophreniaMRCT* patients from 37 countries

43 in Mortality SOC Regional Variability A Scenario of an Patientdistribution RDSmortality Europe (50%) 3 countries LatinAmerica (50%) 7 countries 14d 11.7% 7.2% All causemortality 14d 20.5% 12.1% Non RDS relatedmortality8.8% 5.7% All causemortality 28d 26.2% 14.6% Q: Are region specificmortality intrinsic? 42

44 Scientific Principles For a trial that is exploratory in nature statistical validity may not necessarily be controlling the statistical error of making a wrong statement of at least one possible clinical conjecture Not intended as primary basis for efficacy A stage to better quantify uncertainty and parameter estimates as such the study is well designed to learn, explore or address plausible effect sizes, dose regimens, patient populations, primary efficacy endpoints, etc. If active controlled, explore useful study objectives evaluation 43

45 Scientific Principles For a trial that is adequate and well controlled (A&WC) or otherwise known as confirmatory trial First principle: statistically valid (ICH E 9) Design induced bias vs operationally induced biasdue to trial conduct Study results are interpretable 44

46 45