3 Introduction ICH - General Purpose E5 Guideline: New GCP Guideline Unification of necessary documentation and its formats for NDA submissionE5 Guideline:Extrapolation of foreign clinical dataAvoidance of unnecessary clinical trialsNew GCP GuidelineQuality assurance of clinical trial dataSimultaneous Global Drug DevelopmentBetter drugs in a timely fashion
4 Regulatory Environment Review timeA number of approved drugs by application of E5 guideline
13 Hollowing out of Clinical Trials Domestic companies conduct their clinical trialsoutside of JapanHigh cost toconduct clinical trialsSlow speedof clinical trialsHollowing out ofClinical Trials
14 Recent R&D Trend From bridging to global studies Importance of basic science
15 Concept: Avoidance of Unnecessary Clinical Trials Bridging studiesForeigndataNewRegionsSimultaneous global studiesUSEUASIA
16 Intra variability >> Inter variability Issues to be shownIntrinsic factorsExtrinsic factorsIntra variability >> Inter variabilityConduct of a proposed clinical trial among regionsDifference in Medical Practice- Different study design- Different adverse event reporting system
17 Intrinsic factors (Influence of Genotype) Fukuda et. al.(2000) investigated whether the disposition of venlafaxine was affected by the CYP2D6 genotype.# subject=36 blue(*10/*10) = 6 red(*1/*10,*2/*10)=13 orange(*1/*1,*1/*2,*2/*2)=16 green(others)=1may affect efficacy and safety – adjustment of dosage
18 Mixture of Target Disease Population DNA micro array: NEJM,2002- Target Population: diffuse large-B-cell lymphoma - Efficacy：anthracycline chemotherapy -35% - 40% -mixture of target disease populationInstead of application of those sensitive statistical adjustment, NEJM reported the application of DNA microarray information to investigate the existence of mixture of targeted disease population in clinical trials.The success rate of anthracycline chemotherapy on diffuse large B-cell lymphoma is approx %. The researchers employed information of microarray to define a clear targeted disease population by grouping targeted population.Gene expression:- grouped target population- clearly defined target disease population
19 Mixture of Target Disease Population DNA micro array: NEJM,2002Cox regressionThen the researcher applied gene-expression signatures and found 4 distinct gene-expression signatures. The graph shows clear drug effect between the groups.Gene-expression signatures: 4 distinct gene-expression signaturesscore by the combination of the 4 signatures
20 Extrinsic factors Different medical practice Ex: Depression Trials US and EU: Placebo Controlled TrialJapan: Non-inferiority Trial orPlacebo Controlled Relapse Trial
21 3 Major Studies Drug Source Indication Type of Study Tolterodine Presentation by Dr.Kong Gans at the 3rd K-H Sympo.Overactive BladderAsian Study(Japan and Korea)IrresaReview report by PMDECNon-small Lung CancerGlobal phase II study(Japanese vs. Non-Japanese)LosartanNEJMRenal DiseaseGlobal study
22 Lessons Intrinsic factors: design (phase I and II) Importance of basic scienceClear definition of a target population- P450 information: investigate individual variationw.r.t. efficacy and safety- pharmacogenomics: possibly identified individualcharacteristics- surrogate markers: quick detection of efficacydifferent angles of profile- PPK analysis: investigation of possible factors
23 Lessons Extrinsic factors Regulatory aspects: Realization of conductivity of a planned trialRegulatory aspects:New GCP implementationregulatory science practice – depends on structure of a review systemDesign aspects:study design: different medical practiceindependent data monitoring committeeSimulation studies probably play an important role for future prediction
24 Statistical Issues and Potential Approaches How can statistics play a role in extrapolation of foreign clinical data?
25 Statistical Issues Intrinsic factors Statistical Issues: Clearly defined target populationintra-variability >> inter-variabilityRandomization SchemeStatistical Issues:Definition of similarityStatistical test vs point estimationVariability within a regionRequired sample size?
26 Practical Issues Extrinsic factors Statistical Issues: Conductivity of a proposed clinical trialRegulatory agenciesDifferent medical practiceStatistical Issues:What should be shown?Similarity: dose response, efficacyRegulatory sciencePlacebo response: how to estimate
27 Kitasato-Harvard-Pfizer-Hitachi project Under various settings, using real data sets and simulation techniques, we are trying to figure out how to deal with the important issues concerning design and analysis of global clinical trials.Project team member[Kitasato] M. Takeuchi, X. M. Fang, F. Takahashi, H. Uno[Harvard] LJ Wei[Pfizer] C. Balagtas, Y. Ii, M. Beltangady, I. Marschner[Hitachi] J. MeheganThe 6th Kitasato-Harvard Symposium, Oct 24-25, 2005, Tokyo, Japan
28 Global/Multi-national Trials Global trials involve many regions/countries.Global trials provide us information about investigational drug worldwide simultaneously.As to getting new drug approval, there is the fact that each region/country has its own regulatory policy.A lot of statistical issues for DESIGN, ANALYSIS and MONITORING of global trials still remain.we are trying to figure out how to deal with these issues, using real data sets.Today’s talk is concerning with the analysis issues regarding local inference.
29 Questions → One of the challenging statistical issues Although a single summary of the treatment difference across countries is important, but local inference is also desirable.What can we say about the treatment difference in one country, for example, in Japan (with ONLY 14 subjects)?Can we think of the treatment difference derived from “pooled analysis” as that in Japan?Should we believe the results derived from “by-country analysis” ?Can we borrow the information from other countries? How to borrow information?→ One of the challenging statistical issues
30 Analysis model for local inference One extremePooled Analysis(borrowing directly)another extremeBy-country Analysis(borrowing NO info)Compromisedapproaches in between(borrowing information)Suppose Cox-modelFit the stratified Cox model (strata=country)An empirical Bayes approachFit Cox model to each countryNormal approximation of MLE for the treatment differenceFit a Normal-Normal hierarchical model (next page)Get the posterior distribution of and Confidence Set.Fit the Cox model to each countryGet CI forGet CI for: treatment difference: covariate1=treatment group0=control group: baseline hazardfunction for k-thcountry: treatment differencefor k-th country
31 A normal-normal hierarchical model Distribution of random parameter of interestTrue treatmentDifferencein each countryIndividualSampling Density
32 A normal-normal hierarchical model Distribution of random parameter of interestTrue treatmentdifferenceIn each countryNormal Approx.of MLEIndividualSampling Density
33 A normal-normal hierarchical model Empirical Bayes:Estimating UNKOWNhyper parameter using observed dataDistribution of random parameter of interestTrue treatmentdifferenceIn each countryNormal Approx.of MLEIndividualSampling Density
34 A reason why we picked a N-N model on EB There is a well-known issue on EBCI:“Naive” EBCI fails to attain their nominal coverage probability.“Naive” EBCI is constructed from the posterior distribution ofwith plugging-in the estimates to unknownHowever, since are random, the posterior variance should beThe term under the square root is just an approximation of the first term of RHS in above equation.There are a lot of literature concerning EB for a N-N model. Some theories are available to correct “Naive” EBCI especially for a N-N model. (Morris (1983), Laird & Louis (1987), Carlin & Gelfand (1990), Datta et al (2002), etc.) We applied the Morris’ correction in the following analysis.
35 Approximated likelihood / Posterior distribution Pooled AnalysisEmpirical BayesBy-Country Analysis
36 Simulation studiesA small simulation study was conducted to evaluate the performance of this approach under the Cox model.The number of countries and the sample size in each country were fixed,evaluated the coverage probability and average length of confidence interval were evaluated based on 10,000 iterations.Simulation scheme:Parameter of interest (treatment difference):Survival time of group A:Survival time of group B:Censoring time of both groups:Thus, generated data for group A:generated data for group B:, the coverage probability of 95% CI is calculated
37 ConclusionThis empirical Bayes approach (Normal-Normal hierarchical model coupled with normal approximation of the estimator of the treatment difference) can be used in a wide variety of situations.From a simulation study, the performance of this approach was not bad in terms of both coverage probability and length of CIs.As to RALES data, this analysis provides shorter CIs and suggests that the treatment differences among each country are toward the same direction.In global clinical trials, performing this kind of intermediate analysis can be encouraged as a planned sensitivity analysis in addition to the pooled analysis and by-country analysis for better understanding of the treatment difference in a specific country.
38 ReferencesBerger, J. O. (1985). Statistical Decision Theory and Bayesian Analysis, 2nd ed. New York: Springer-Verlag.Carlin, B. & Gelfand, A. (1990). Approaches for empirical Bayes confidence intervals. JASA 85,Carlin, B. & Louis, T. (2000). Bayes and Empirical Bayes Methods for Data Analysis, 2nd ed. London: Chapman & Hall/CRC.Datta, G et al (2002). On an asymptotic theory of conditional and unconditional coverage probabilities on empirical Bayes confidence intervals. Scand. J. Statist 29,Laird, N. & Louis, T. (1987). Empirical Bayes confidence intervals based on bootstrap samples. JASA 82, 739—750.Morris, C. (1983a). Parametric empirical Bayes inference: theory and applications. JASA 78,Morris, C. (1983b). Parametric empirical Bayes confidence intervals. In Scientific inference, data analysis, and robustness, 25—50, New York: Academic Press.Pitt, B et al. (1999) The effect of spironolactone on morbidity and mortality in patients with severe heart failure. NEJM 341, 709—717.
39 Safety Issues Intrinsic/Extrinsic factors How can we ensure the safety of the drug if a drug is approved based on a small clinical data in a region?Need a type of a phase IV study after a approval, i.e., electronic data capturing system, and how can we analyze the data and what is a appropriate interpretation.
40 Safety Issues Network system among Hospitals Research Grant from MHLW Network system among hospitals by EDC to monitor patientsDetection of unexpected AEsBuild data base regarding pats` background for signal detection, pharmacoepidemiology
41 Overall Picture Step 1 Step 2 Medical Facility 1 Medical Facility 2 Medical Facility NStep 2Data CenterMedical Facility 3Medical Facility 5Medical Facility 4
42 Step 1: Within a MF Connect Necessary Medical Records per Patient Unification of Medical Recordsper Patient regarding-Patient`s background- Dosage and durationEfficacySafety
43 Step 2: Among MFs (i) Unification of Data base from different MFs and Medical Facility 1Medical Facility 2Medical Facility NStep 2Data Center(i) Unification of Data base from different MFs andEstablishment of Patients` data base at Data Center(ii) Detect unexpected AEs and analyze safety profileaccording to actual dosage and duration
44 Conclusion Asian and Global Studies are a future direction Design and Statistical Issues must cope with basic sciencePhase IV studies based on EDC are necessary for assurance of safety