Summer Institute for Training in Biostatistics
Instructors
Karen Pieper received her Master's degree at the University of North Carolina in Chapel Hill in 1987. She has since worked at Duke University as a cardiovascular clinical trials statistician, at the University of Virginia as a general medical center statistician, and as a private statistical consultant. She returned to the Duke Clinical Research Institute in 1999. At present, she is the manager of the manuscripts for all completed clinical trials and registries, primarily in the field of cardiology. She is also beginning a program of mentoring and education for faculty and physicians in training as well as beginning statisticians.
I got my PhD from Iowa State University. Since then I have been in the Statistics Department at NC State. I am currently the co-Director of the Statistics Graduate Program. In this role, I recruit and mentor graduate students, especially in the transition from undergraduate to graduate study. My current research is in the area of statistics education. I work on assessing student learning and developing materials for other instructors to use. I have also done research in survey sampling. It is common to see examples in statistics classes where we assume the data comprise a simple random sample from the population of interest. In survey sampling, we study how to select samples using methods other than simple random sampling. Specialized methods must be used in analyzing data from such a sampling design.
My research considers nonlinear models in which the predictor variables are measured with error. For example, suppose we want to use blood pressure to predict heart disease. If blood pressure is measured with error, we are interested in how the error effects the parameter estimates and how we can correct for these effects. I also work with undergraduate students on a variety of research projects that develop regression models for air pollution, specifically ozone and particulate matter.
I work in a variety of areas that intersect with biomedical research. Currently my major focus is on model selection and model checking. For example, given a study of 100,000 subjects where the response variable Y is presence (=1) or absence (=0) of a particular disease and 500 independent X variables, what is the best model relating the probability of disease P(Y=1) to some function of a subset of the 500 X variables? Then using the selected model, estimate the probability that a particular person (not in the study) will have the disease and give a measure of reliability of the estimate.
I have worked extensively in the area of pharmacokinetics, the study of "what the body does to the drug." Pharmacologists need to understand the pharmacokinetic processes that take place in the body after a drug is administered and how these processes vary across people in order to develop dosing regimens that will be effective for different types of patients. This is facilitated by fitting a statistical model describing concentrations of drug in the body over time in terms of quantities characterizing how the body absorbs, metabolizes, and eliminates drug and how these quantities vary across people. I also work on general methods for analyzing longitudinal data (data collected over time). For example, the effectiveness of two anti-hypertensive medications may be compared on the basis of how and to what extent they reduce diastolic blood pressure over the period of study. Data collected over time on human subjects are often missing, because subjects fail to show up for their scheduled study visits, and I have worked on methods for analysis that take this into account.
My major research interest is in developing models and methods for analyzing longitudinal data arising in many biomedical studies, especially in epidemiology. For example, in the Study of Women's Health Across the Nation (SWAN), annual bone mineral density (BMD) is to be collected for women transitioning to menopause and the research interest is in identifying risk factors associated with BMD loss. The results from the study will help biomedical investigators better understand the BMD loss process and hopefully will eventually lead to preventive measures for BMD-related disease such as osteoporosis. Compared to the traditional design, a longitudinal study will be more efficient to detect the treatment difference of interest and offer more evidence for possible causal inference.
Most of my research is motivated by problems I have encountered while working as a clinical trials statistician for cooperative groups in cancer and AIDS and more recently on clinical trials in cardiology. Since the primary endpoint in such chronic disease clinical trials is time to an event, often right censored, I have studied specialized methods for the design of such studies and the analysis of the data. Because of ethical as well as practical reasons, data are monitored periodically and, if there is sufficient evidence of treatment effect (or lack thereof), a clinical trial may be stopped early. Much of my research has considered the statistical consequences of stopping rules. Recently, I have been working on efficient designs for stopping studies which will, on average, stop a study as early as possible while still preserving the statistical accuracy desired.
In clinical research, the randomized trial is considered the gold-standard for comparing treatments. Randomization guarantees that the treatment groups are balanced on average. My interest is in the more common situation where patients are not randomized to treatments. In these situations, the treatment decisions are determined by the patient's characteristics and preferences and the groups of patients are not necessarily similar. I'm interested in a variety of statistical techniques that reduce bias caused by non-random treatment selection. In addition to hard clinical end-points such as mortality and heart attacks, I'm interested in the analysis of medical costs and quality-of-life.
I have published extensively in the area of meta-analysis (combining results from multiple studies). One of my papers, with Dr. David won the FHP Foundation prize as the outstanding paper published in the International Journal of Technology Assessment in Health Care in 1990. We published a book entitled Meta-Analysis by the Confidence Profile Method (Academic Press, 1992). I have also co-authored several other articles on meta-analysis with other researchers. In addition, I have published several papers on fitting mixtures of probability distributions to data. For example, we have used distribution fitting as a tool to measure disease progression in human blood cells. I have also worked on methods for dose-response analysis in the area of environmental toxicology. The methods involve extrapolating dose-response curves to very low doses in order to insure environmental safety.
My current research focuses on mapping genes for complex diseases using linkage disequilibrium methods. These methods take advantage of associations between DNA sequences at chromosomal locations that are in close proximity to one another due to their common evolutionary history. A common design in genetic studies of complex disease collects genetic information from diseased individuals and their nuclear (or extended) family members. The strategy in such studies is to look at how genetic information is passed from parent to offspring and if certain genetic variants are transmitted more often to diseased offspring that would be expected under Mendelian law. Missing data is common in these studies (missing genotype data, missing family members, haplotype phase ambiguity, etc.). A major component of my research aims at developing statistical methods for family-based designs that are appropriate when confronted with different types of missing data.
Robert Harrington, M.D., has been head of the DCRI's Cardiovascular Clinical Trials division since 1999, and in 2006 became Director of DCRI. He has helped design and lead some of the largest cardiology trials of the past several years, including PURSUIT, PARAGON-A, and PARAGON-B. Before becoming a professor of medicine at Duke University Medical Center, Dr. Harrington taught at the University of Massachusetts Medical Center, where he was selected as an Outstanding Medical Educator. He is a fellow of the American College of Cardiology and of the Society for Cardiovascular Angiography and Intervention, as well as an associate editor of the American Heart Journal. He has authored numerous peer-review publications, review articles, book chapters and editorials, and co-edited a recent textbook on antiplatelet therapy.
My recent research has focused on group testing, which has applications in many areas. In public health work, for example, it has been used in AIDS screening, and to estimate HIV-seropositivity rates within various populations and subpopulations. Cross-over designs are sometimes used in clinical trials for chronic conditions; in a cross-over design each subject is switched from one treatment to another and often back to the first one during the course of a long-term trial.