"Statistics Speaker Agenda 5"
Statistical Issues for Medical Devices and Diagnostics Bethesda Marriott Hotel Bethesda, MD April 16 - 17, 2008 April 16, Wednesday 8:30 – 9:00 REGISTRATION AND CONTINENTAL BREAKFAST 9:00 – 9:10 Welcome Thomas Maeder, Executive Director, MTLI, AdvaMed Daniel Shultz, Director, CDRH, FDA Gregory Campbell, Director Division of Biostatistics, CDRH, FDA 9:10 – 9:30 Statistics in Medical Devices and Diagnostics – An Overview Gregory Campbell, Director Division of Biostatistics, CDRH, FDA General overview of how statistics for devices and diagnostics differs from that in drugs and other areas. How has the field evolved over the years, what are the current issues, and what do we hope to accomplish here during the next two days? 9:30 – 10:50 Session: The Use of Adaptive Design Session planners: Yonghong Gao, Mathematical Statistician, CDRH, FDA Xiaolong Luo, Senior Mathematical Statistician, Cordis A considerable amount of recent research and discussion has focused on statistical and logistical issues associated with adaptive clinical study designs. Presentations in this session address experiences and lessons learned in the design and implementation of such studies – some utilizing a frequentist adaptive approach and some a Bayesian approach. 9:30 – 9:50 Case Study: Sample Size Re-Estimation Roseann White, Director, Global Biostatistics Clinical Data Systems, Abbott Vascular Adaptive sample size re-estimation techniques can be useful when one is unsure of assumptions about the true rates in a randomized or non- randomized trial. Operational challenges arise, however, in avoiding the introduction of bias or when enrollment time is much shorter than the time to follow-up for the interim analysis. This session features a case study in which many of these operational issues are creatively addressed. 9:50 – 10:10 Case Study: The SAPPHIRE Study David Snead, Associate Director, Johnson & Johnson The SAPPHIRE Study used a randomized non-inferiority design with a novel sequential interval censored survival method to compare carotid artery stenting to endarterectomy. The study design and results will be presented along with lessons learned. 10:10 – 10:30 A Consultant’s Perspective on Bayesian Adaptive Device Trials Scott Berry, President, Berry Consulting A consultant discusses aspects of Bayesian approach to device trials that are advantageous to the sponsor, regulator, and to patients. A case study examines an adaptive device trial using predictive probabilities to select one of two devices and the appropriate sample size, in which partial information for subjects is utilized with longitudinal models. 10:30 – 10:50 The FDA Perspective on Adaptive Device Trial Design Yonghong Gao, Mathematical Statistician, CDRH, FDA An FDA statistical reviewer shares experiences in reviewing adaptive design trials in terms of both statistical and regulatory considerations. Discussion points will be raised that may facilitate better communication between industry and FDA in future trials of this type. 10:50 – 11:10 BREAK 11:10 – 12:30 Session: The Application of Bayesian Statistics Session planners: Telba Irony, Chief, General and Surgical Devices Branch, Division of Biostatistics, CDRH, FDA Roseann White, Director, Global Biostatistics Clinical Data Systems, Abbott Vascular In 2006, CDRH issued a draft guidance document on the use of Bayesian statistics to help people developing or reviewing device or diagnostic clinical trials determine whether Bayesian methods would be effective and to apply those methods appropriately. This session provides more detailed information and illustrative case studies on appropriate use of the Bayesian approach. In a concluding panel discussion, the presenters and organizers will be available to answer questions from the audience. 11:10 – 11:15 Introductory Remarks on Bayesian Device and Diagnostics Studies Roseann White, Director, Global Biostatistics Clinical Data Systems, Abbott Vascular 11:15 – 11:30 To Use Bayesian or Not to Use Bayesian, that is the Question Andrew Mugglin, Research Associate Professor, University of Minnesota With any statistical methodology one must understand not only how to use it but also when it makes sense in terms of resources and time. This session examines the key issues to consider before deciding to use a Bayesian approach to the design and analysis of a study. 11:30 – 11:45 Bayesian Statistics for Therapeutic Devices Pablo Bonangelino, Mathematical Statistician, CDRH, FDA Xuefeng Li, Mathematical Statistician, CDRH, FDA Bayesian adaptive designs using predictive probability and regulatory considerations regarding these designs for therapeutic devices will be presented. How much strength could be borrowed in a hierarchical model for a simple case will be discussed. 11:45 – 12:00 Recent Experiences Applying Bayesian Analysis to Medical Device Trials A. James O’Malley, Associate Professor of Statistics, Department of Health Care Policy, Harvard Medical School Applying Bayesian analysis to medical device trials, though challenging, can provide insights into your data that are unattainable with other methods. Assuring that the method is appropriately applied can be challenging, however, and this session addresses the use and interpretation of Bayesian analyses. 12:00 – 12:10 Update on the Bayesian Guidance Document Telba Irony, Chief, General and Surgical Devices Branch, Division of Biostatistics, CDRH, FDA 12:10 – 12:30 Bayesian Panel Discussion Moderator: Roseann White, Director, Global Biostatistics Clinical Data Systems, Abbott Vascular Panel Members: Andrew Mugglin, Research Associate Professor, University of Minnesota A. James O’Malley, Associate Professor of Statistics, Department of Health Care Policy, Harvard Medical School Telba Irony, Chief, General and Surgical Devices Branch, Division of Biostatistics, CDRH, FDA 12:30 – 2:00 LUNCH 2:00 – 3:20 Multiplicity Session planners: Shiowjen Lee, Mathematical Statistician, CDRH, FDA Michael Lu, Edwards Lifesciences Medical device clinical trials often consider multiple endpoints to assess the safety and efficacy of a given treatment. In exchange for distinct advantages, however, trials with multiple endpoints also pose challenges, and interpretation of the results must be considered carefully. 2:00 – 2:30 Rectifying FDA and Bayesian Views on Adjustment for Multiplicity Gene Pennello, Mathematical Statistician, CDRH, FDA Although frequentist properties of Bayesian designs are not routinely evaluated in academic research, FDA can request that the Type I error rate be assessed for proposed Bayesian trials, especially those involving multiple testing. Justifiable under the Agency’s mandate to obtain valid evidence of safety and effectiveness, strict control of the overall Type I error rate can nonetheless negate the advantage of using prior information through Bayesian analysis. How can one relax the Type I error rate while maintaining required standards of evidence? 2:30 – 3:00 Control of Type I Error and the Correlation of Multiple Endpoints in a Medical Device Trial Kevin Najarian, Boston Scientific Multiple univariate analyses can inflate the Type I error rate, necessitating adjustments of the observed p-values. In addition to common techniques for multiplicity adjustments, some others use correlation information and consider the homogeneity of treatment effects to determine the degree of adjustment necessary. An analysis of real life simulated medical device data is considered to assess techniques for multiplicity with added regard to the relationships among endpoints. 3:00 – 3:20 Multiplicity Panel Discussion Panelists: Gene Pennello, Mathematical Statistician, CDRH, FDA Shiowjen Lee, Mathematical Statistician, CDRH, FDA Zengri Wang, Medtronic Andrew Mugglin, Research Associate Professor, University of Minnesota 3:20 – 3:40 BREAK 3:40 – 5:00 Session: Issues with Missing Data Session planners: Jianxiong (George) Chu, Mathematical Statistician, CDRH, FDA John Evans, Senior Biostatistics Manager, Boston Scientific Missing data is a common problem in clinical studies, which causes the usual statistical analysis of complete or all available data to be subject to potential biases. However, there is no universally applicable best method for handling missing data. In this session, speakers/panelists will discuss theoretical considerations, technical issues and case studies with an emphasis on the importance of sensitivity analysis to assess the impact of missing data on statistical inference and interpretation under different scenarios of assumptions. 3:40 – 4:00 Missing Data: An Overview Michael Kenward, Professor of Biostatistics, London School of Hygiene & Tropical Medicine In this introduction and overview to the statistical problem of missing data, the nature and cases of missingness will be considered in a trial setting, and Rubin’s framework for missing value mechanisms will be introduced, together with the implications for subsequent analysis. Both ad hoc (e.g. completers, last observation carried forward) and statistically principled (e.g. likelihood, multiple imputation) approaches to analysis will be discussed. The inherent ambiguity of statistical analyses for incomplete data will be stressed throughout, and the important role of sensitivity analysis considered. 4:00 – 4:20 Case Study: Sensitivity Analysis in DES Clinical Trials from the FDA Reviewer’s Perspective Xu (Sherry) Yan, Mathematical Statistician, CDRH, FDA The impact of missing data has always been a major concern in the interpretation of clinical trial results. Procedures of statistical inference for handling missing data, such as multiple imputation, typically require the specification of missing data mechanisms for the implementation. When the missing data mechanism is unknown, sensitivity analyses may provide helpful information. This presentation addresses several sensitivity analyses, such as worst case scenario analysis and tipping- point analysis that were performed for two drug-eluting stent trials. 4:20 – 4:40 Incomplete Data in Clinical Studies: Analysis & Sensitivity Analysis Geert Molenberghs, Professor of Biostatistics, Hasselt University, Diepenbeek, & Katholieke Universiteit, Leuven, Belgium Incomplete data are common, as are modeling and other data analysis tools for such data, with ever increasing complexities. Model assumptions may directly affect inferences and substantive conclusions without being fully verifiable from the observed data, and it is therefore essential to recognize sensitivities caused by these assumptions before proceeding with the analysis. In particular, one cannot formally distinguish between MAR and MNAR mechanisms. Theoretical considerations and practical illustrations highlight the implications for sensitivity analysis. 4:20 – 5:00 Panel Discussion on Missing Data Moderator: Jason Roy, Research Investigator, Center for Health Research, Geisinger Health System Panelists: Michael Kenward, London School of Hygiene & Tropical Medicine Xu (Sherry) Yan, CDRH, FDA Geert Molenberghs, Hasselt University, Belgium Recai Yucel, Assistant Professor of Biostatistics, Department of Epidemiology & Biostatistics, School of Public Health, SUNY-Albany 5:00 ADJOURNMENT 5:00 – 6:30 RECEPTION April 17, Thursday 8:30 – 9:00 CONTINENTAL BREAKFAST 9:00 – 10:20 Session: Trial Design – No Controls Available or Imperfect Controls Session Planners: Lilly Yue, Chief, Cardiovascular & Ophthalmic Devices Branch, Division of Biostatistics, CDRH, FDA Brandon Sparks, Medtronic Randomized, well-controlled, double-blind clinical trials have been viewed as the gold standard in the evaluation of medical products, but medical device clinical studies often depart from this ideal paradigm for ethical or practical reasons. In studies with imperfect or no controls, the resulting statistical inferences may carry a lower level of scientific assurance. This session provides an overview, illustrated with case studies, of non- randomized or imperfectly controlled medical device clinical trials, considering design and statistical analysis from industry and regulatory perspectives. 9:00 – 9:20 Unique Challenges Regarding Control Groups for Neurostimulation Therapies Steven Broste, Director of Biostatistics & Data Management, Medtronic Neurostimulation Implantable neurostimulation systems are in clinical studies or development for several new indications or anatomical regions. The mechanism of action for these novel, invasive treatments is unclear, yet the optimal blinded, controlled studies may not be possible, whether for ethical reasons or because physical effects of treatment make blinding impossible. This presentation elaborates on the challenges in such a study, and describes attempts to deal with these issues. 9:20 – 9:40 Case Study of a Trial Design with Imperfect Control: Industry View Bryan Randall, Senior Manager, Biostatistics & Clinical Science, Boston Scientific This case study features a proposed trial design with imperfect control to establish the safety and efficacy of carotid stenting compared to carotid surgery in high-risk patients. The essential design feature is a concurrent, non-randomized surgery control arm that poses challenges to minimizing bias from several sources, including operator selection, patient selection, and outcome evaluation. How can the statistical design and analysis and trial execution plans mitigate the impact of these potential biases? 9:40 – 10:00 Trial Design with Imperfect Controls or No Control Available: a Regulatory Perspective Lilly Yue, Chief, Cardiovascular & Ophthalmic Devices Branch, Division of Biostatistics, CDRH, FDA Ethical or practical considerations do not always permit randomized, well- controlled, double-blind clinical trials for the evaluation of medical products. How does a regulatory agency approach the challenge of imperfect or no controls? 10:00 – 10:20 Panel Discussion on No Controls Available or Imperfect Controls Moderator: Brandon Sparks, Medtronic Panelists: Steven Broste, Director of Biostatistics & Data Management, Medtronic Neurostimulation Bryan Randall, Senior Manager, Biostatistics & Clinical Science, Boston Scientific Lilly Yue, Chief, Cardiovascular & Ophthalmic Devices Branch, CDRH, FDA Lei Peng, Abbott Cardiovascular 10:20 – 10:40 BREAK 10:40 – 12:00 Session: Data Poolability Session Planners: Chul H. Ahn, Mathematical Statistician, CDRH, FDA Dennis W. King, President & CEO, STATKING Consulting Poolability comprises a heterogeneous set of problems that need to be addressed on a case by case basis, depending, among other things, on whether the groups in question are investigational sites, geographical regions, clinical studies (or substudies), patient populations, device models, and so on. This session will consist of presentations by both FDA and the industry regarding their perspective on this issue followed by the discussion by the panelist. 10:40 – 11:00 Data Pooling in Medical Device Trials Chul H. Ahn, Mathematical Statistician, CDRH, FDA One often encounters data poolability problems when trying to combine data from different groups to obtain an overall estimate of an outcome variable. Different groups may mean different centers, studies, patient populations, device models, and so on. This presentation addresses some of the issues in data pooling with examples from recent clinical studies. 11:00 – 11:20 Medical Device Case Studies in Poolability John C. Evans, Senior Biostatistics Manager, Boston Scientific Medical device poolability case studies will be presented from 1- and 2- arm studies. Poolability of patients from different centers assumes that all sites follow identical protocols. To determine poolability, the comparability of baseline variables, demographic characteristics, and safety and efficacy endpoints is assessed across sites. In single-arm studies, the effect of study center is investigated to determine the poolability of the primary endpoint from different institutions. For two- or more arm studies, the effect of interaction between treatments and sites is considered to determine the poolability of treatment differences in the primary endpoint across sites. 11:20 – 11:40 Pooling Non-Poolable Clinical Data: So What Jeng Mah, Principal Biostatistician, American Medical Systems The ability of statistics to examine underlying consistent states is based on the assumption that random samples from each state are independently identically distributed (IID). Acceptance of this assumption is not trivial, as evidenced by the stringent requirements for RCTs and the difficulty of developing complicated covariate analyses to accommodate lesser clinical data. Homogeneity of device clinical data, even from well conducted RCTs, may require closer scrutiny than with pharmaceuticals, because observed effects of medical devices show a combination of patient characteristics, device design, and implanter skill. Device studies often prospectively balance treatment assignments within centers and retrospectively try to verify data poolability. This study uses simulation to evaluate the cost of neglecting center effects in terms of false positive rate, assay sensitivity, and degree of bias in estimating treatment effects. 11:40 – 12:00 Poolability Panel Discussion Moderator: Dennis W. King, President & CEO, STATKING Consulting Panelists: Chul H. Ahn, Mathematical Statistician, CDRH, FDA John C. Evans, Senior Biostatistics Manager, Boston Scientific Jeng Mah, Principal Biostatistician, American Medical Systems 12:00 – 1:00 LUNCH Two Simultaneous Tracks 1:00 – 5:00 Track A ; Statistical Issues in Diagnostics and Imaging Session Planners: Lakshmi Vishnuvajjala, Chief, Diagnostic Device Branch, Division of Biostatistics, CDRH, FDA Betty Stephenson, Becton Dickinson Vicki Petrides, Statistician Section Head, Abbott Diagnostics Pat Meyers, Abbott Diagnostics Statistical Issues in Diagnostics and Imaging Studies. Diagnostic studies present unique challenges in the medical device area. For laboratory tests which are regulated by the Office of In Vitro Diagnostics at CDRH, randomized studies are not the rule since samples can be split in most cases to be tested by both the old and the new tests. But certain types of bias are much more common and need to be guarded against. Imaging studies such as mammography, ultrasound and MRI are greatly affected by the case mix and reader variability. 1:00 – 1:10 Diagnostic Devices: A Brief Overview Lakshmi Vishnuvajjala, Chief, Diagnostic Device Branch, Division of Biostatistics, CDRH, FDA There are many challenges that must be addressed in the development of diagnostic and imaging products. This session will explore some of these concerns, including co-development of biomarkers and assays, special problems encountered in diagnostic imaging studies, and quality control issues. 1:10 – 1:35 Statistical Design Challenges in Phase III Biomarker-based Clinical Trials Dan Sargent, Statistician, Mayo Clinic (Biomarkers) This presentation will focus on issues to consider when designing a clinical trial assessing the usefulness of a predictive marker. We present two classes of clinical trial designs, the 'selection design', which enrolls only a specific target population, and an unselected but prospectively specified marker-based design. We also discuss the relative merits of prospectively enrolled clinical trials versus prospectively specified analyses on data from previous clinical trials. 1:35 – 2:00 Design Issues in Radiological Image Reader Studies Tom Gwise, Imaging Studies, CDRH, FDA Medical devices reaching a certain risk threshold must be evaluated as to their safety and efficacy before they can be legally marketed in the United States. Evaluation of radiological imaging devices poses certain complications because some level of subjective reader interpretation is necessary to utilize them. Many sources of bias can affect such studies. This presentation will discuss sources of bias in reader studies and steps that can be taken to mitigate them. 2:00 – 2:25 Considerations When Building Microarray-based Classifiers and Predictors of Clinical Endpoints: Perspectives from the MAQC II Wendell Jones, Senior Director of Statistics and Bioinformatics, Expression Analysis The MicroArray Quality Control consortium is currently focused on the ability of microarrays to classify or predict outcomes for samples and subjects in clinical and toxicogenomic studies. There are particular aspects of microarray data including the magnitude of factors available, quality issues, and technical effects which may have large ramifications to the success of a classifier/predictor. 2:25 – 2:45 Question and Discussion 2:45 – 3:00 BREAK 3:45 – 5:00 Expert Panel on Guidance for Laboratory Tests Moderator: Betty Stephenson, Becton Dickinson Speakers: Jan Krouwer, Krouwer Consulting, (CLSI Guidelines) Kyungsook Kim, CDRH, FDA (CLIAI Guidance) Marina Kondratovich, CDRH, FDA,(Migration Guidance, FDA perspective) Angel DeGuzman, Abbott Diagnostics, (Migration Guidance, industry perspective) There are multiple aspects of diagnostic products that need to be evaluated by manufacturers, the FDA, and laboratories. To assist in evaluating some of these characteristics, guidance documents have been published through the joint efforts of interested parties. This session will provide a forum during which these guidance documents will be discussed. 1:00 – 3:00 Track B: Postmarket Statistical Issues Session Planners: Greg Campbell, Director, Division of Biostatistics, CDRH, FDA An Liu, Principal Statistician, Medtronic Cardiac Rhythm Disease Management This session will concentrate on statistical issues that arise in the post- market arena. It will include statistical concerns arising in condition of approval studies, as well as other post-approval studies to enable a change in the indication for use for a marketed device. Another postmarket issue in the session is the use of statistical methodologies in the analysis of data from FDA’s Medical Device Reporting (MDR) system. 1:00 – 1:30 General Statistical Methodologies in the Analysis of Spontaneously Reporting Post-Marketing Medical Device Reports (MDRs) and Design for Post-Approval Studies Chang S. Lao, Senior Mathematical Statistician, Division of Biostatistics, CDRH, FDA The sources of Medical Device Reports (MDRs) and related required regulations are introduced. Attention is then focused on the statistical methods for MDR spike detection based on modeling and non-modeling approaches as well as non-parametric regression smoothing for exploratory analysis and non-parametric trend analysis, with two real data examples. Related statistical issues for analysis of spontaneously reporting MDRs will be briefly discussed. Finally, a few real examples for Post-approval studies and general statistical issues for study design of Post-approval studies will be described. 1:30 – 2:00 Using Seattle Angina Questionnaire (SAQ) to Evaluate Patient Reported Outcome in a Post-marketing Setting Author: Vivian Mao, Clinical Research Manager, Abbot Vascular Presenter: Roseann White, Director, Global Biostatistics Clinical Data Systems, Abbott Vascular Patient-reported quality of life outcome instrument can be used to measure the impact of an intervention on patient health status in medical device clinical trials. Using a valid and reliable instrument to systematically evaluate patient’s perspective on the impact of drug eluting stents may provide valuable information that can be lost through a clinicians’ interpretation. This talk discusses the usage of the Seattle Angina Questionnaire (SAQ) to evaluate patient health status in patients receiving XIENCE™ V Everolimus Eluting Coronary Stent System (EECSS) in a post-marketing environment and its statistical considerations. 2:00 – 2:30 Statistical Challenges in the Post-Approval Registry for Drug Eluting Stents Peter S. Lam, Director, Biostatistics & Clinical Sciences, Boston Scientific The statistical challenges in the mandatory post-approval registry for Drug Eluting Stents (DES) to address the safety endpoints of death, myocardial infarction, and stent thrombosis will be presented using a case example. These endpoints were not sufficiently powered in the pre-approval trials due to their low incidence rates. Additional challenges to attempt to investigate the optimal duration of dual anti-platelet therapy for DES patients will also be explored in this presentation. 2:30 – 2:45 BREAK 2:45 – 3:15 A Bayesian Design Applied to an Endovascular Post-Market Study Yuqing Dai, Principal Statistician, Medtronic Cardiovascular Feng Tang, Principal Statistician, Medtronic Cardiovascular Often prior knowledge, which contains valuable information, on an existing device is available when we conduct a new post-market study. Appropriately using such prior information could lead to sample size reduction, which is very desirable when the accruement of the sufficient number of patients could be challenging due to the relatively low incidence of a rare outcome. In this talk, we use an endovascular trial as an example, illustrating a Bayesian design on a post-market study. Also, we will study the relationship between the different operation characteristics, the decision criteria and the sample size. 5:00 ADJOURNMENT Important Notice The information provided in this course represents the personal opinions of the instructors and does not necessarily represent the opinions of AdvaMed staff. Companies relying on the information do so at their own risk and assume the risk of any subsequent liability that results from relying on the information. The information does not constitute legal advice.