Abstracts – by Session (updated 5-8-13)
John M. Lachin, Sc.D.
Professor of Biostatistics, Epidemiology, and Statistics
Former Director, The Biostatistics Center
The George Washington University
Statistical Practice and Challenges in Chronic Disease Clinical Trials – Diabetes and Cardiovascular Diseases
The evaluation of treatments (or treatment strategies) in clinical trials of long-term chronic diseases poses specific statistical challenges. Issues will be discussed related to the design, conduct and analysis of such clinical trials in diabetes and cardiovascular diseases. Opportunities for, and obstacles to, statistical innovation will be discussed.
Brian Claggett, Ph.D.
Department of Biostatistics
Harvard School of Public Health
Hicham Skali, MD, MSc
Instructor, Harvard Medical School
Associate Physician, Cardiovascular Medicine,
Brigham & Women’s Hospital
Clinical and Statistical Perspectives on Personalized Medicine
Randomized clinical trials have traditionally been used to assess the effect of an intervention on a primary (often composite) outcome through a global two-sample comparison, followed by multiple secondary analyses of non-primary endpoints and subgroups. Personalized medicine is based on the concept that an intervention may have varying effects on different individuals and different outcomes, including adverse events. Furthermore, the impact of personalized treatment guidelines may be assessed at both the individual and population levels, each requiring different summary measures. We describe some of these issues based on experience from past RCTs and simulations and suggest some approaches based on ongoing work in progress related to patient management and trial design, including potential modifications of the adaptive trial design.
Keith Goodman, D.B.A.
Chief Technology Officer,
Cancer Research And Biostatistics
The Automatic Clinical Trial: Leveraging the Electronic Medical Record in Multi-site Cancer Clinical Trials
Submission of data into clinical trial electronic data capture (EDC) systems currently requires redundant entry of data that already exist in the electronic medical record (EMR). Being able to automatically transfer data from the EMR to the EDC would save many hours of arduous effort, especially for multi-site data-intensive oncology trials. Designing health record systems with research in mind will provide opportunities for transferring data from EMR to EDC. As new research systems are developed and EMR systems adapt, other technology in the form of data mining or middle tier applications are certain to provide assistance in this effort.
William T. Barry, Ph.D.
Nancy and Morris John Lurie Investigator
Assistant Professor, Harvard Medical School
Department of Biostatistics and Computational Biology
Dana-Farber Cancer Institute
Adaptive trial designs with integral biomarkers: innovation versus feasibility
The role of biomarkers has increased in cancer clinical trials such that designs are needed to efficiently answer questions both of efficacy of novel therapeutics and of biomarker performance. In randomized phase II studies, response-adaptive designs using Bayesian statistics and hierarchical models have been advocated for studies with multiple biomarkers. This approach allows for a gradual and seamless transition from randomized-blocks to marker-enrichment during the trial with careful selection of prior distributions. In this setting, gains in efficiency can be seen with adaptive designs relative to nested staged designs. However several key operational requirements may limit the feasibility of running such trials in multi-institutional clinical trial groups and networks.
Lorenzo Trippa, Ph.D.
Assistant Professor of Biostatistics,
Harvard School of Public Health
Department of Biostatistics and Computational Biology,
Dana-Farber Cancer Institute
Optimal Bayesian Designs and Frequentist Constraints
Most trial designs are developed using frequentist criteria. Concepts such as the type one error probability are established and, in some cases, required by authorities. In contrast, Bayesian designs do not explicitly consider these frequentist quantities. The Bayesian investigator uses simulation methods to derive and control the size and power. Then, tuning parameters are adjusted in order to guarantee predefined of operating characteristics. These adjustments affect the use of prior information and are suboptimal. We consider a Bayesian decision theoretic approach that account for the presence of frequentist authorities. The proposed approach allows to use utility functions and produces designs that satisfy a set of required operating characteristics. We illustrate the method with group-sequential multi-arm Phase II trials and a group-sequential bridging trials. (Joint with Steffen Ventz)
Pamela Shaw, Ph.D.
Biostatistics Research Branch
National Institute of Allergy and Infectious Diseases (NIAID)
A randomized delayed start trial of linezolid in patients with extensively drug-resistant tuberculosis
Incidence of multi-drug resistant (MDR) and extensively drug-resistant tuberculosis (XDR TB) is increasing worldwide. Linezolid, an antimicrobial approved for gram-positive bacterial infections, has been used off-label for drug resistant tuberculosis, despite lack of evidence from clinical trials. In 2008, an NIH-sponsored randomized phase 2a study was initiated in South Korea to evaluate linezolid in patients with chronic, culture-confirmed XDR-TB. A delayed start design was chosen to allow for a controlled comparison, while providing all patients access to the novel treatment regimen. Patients were randomly assigned to an immediate or two-month delayed start of linezolid, with no other change to their background regimen. Patients were followed until one year following end of therapy, defined as 18 months after culture conversion. The primary endpoint was time-to-culture conversion, censored at 4 months. A novel test statistic was developed for the primary analysis that incorporates information on survival, if observed, and otherwise uses the primary (surrogate) endpoint of culture conversion. During the design phase, dose-limiting side effects were a known concern. Upon achieving the primary endpoint, patients underwent a second randomization to continued linezolid therapy at the original or reduced dose. Novel statistical aspects of the trial design are discussed and the trial results are presented.
Ivan S. F. Chan, Ph.D.
Late Development Statistics
Merck Research Laboratories
Assessing Immune Correlates of Protection in Vaccine Studies
One of the important goals of vaccine development is to identify potential immunological markers that predict vaccine efficacy. Having such a predictive marker (correlate of protection) can substantially accelerate the vaccine development as subsequent studies can be conducted efficiently based on these immunological markers instead of efficacy endpoints. In this talk, we will discuss the key statistical issues involved in assessing the correlates of protection in vaccine studies. Statistical approaches based on the Prentice criteria and the causal inference framework will be discussed. These methods were used to analyze an immune correlate for herpes zoster vaccine using data from two large-scale phase III trials. We will discuss the potential implications for subsequent studies with the same vaccine.
Marvin Zelen Leadership Award in Statistical Science
John J. Crowley, Ph.D.
CEO, Cancer Research And Biostatistics
A Brief History of Survival Analysis
In this talk I will cover some of the key historical developments in the field of survival analysis, from the product limit estimator of a survival curve, through nonparametric testing of differences between survival curves, to regression methods and exploratory tools for survival data. Along the way, and in keeping with the occasion, I will indicate the many ways in which Dr. Marvin Zelen has served as a mentor to me, and has inspired me to try to create environments in which statistical scientists can flourish, from the SWOG Statistical Center (think ECOG) to Cancer Research And Biostatistics (CRAB; think Frontier Science).
I will use data from SWOG and CRAB on patients with multiple myeloma as examples of the statistical methods discussed. Myeloma, a malignancy involving the immunoglobulin-producing plasma cells of the bone marrow, is a disease entity once uniformly described as incurable. Incredible progress has been made over the past few decades, in both biology and treatment, as I will illustrate with some analyses of gene expression profile data for risk stratification, and through the application of some statistical cure models to data from some recent clinical trials.