Past Biostatistics Seminars

Principles and Challenges for Ethical Biostatistical Practice in Clinical and Translational Research: An Illustrated Panel Discussion

Shelley Hurwitz, PhD.
Director of Biostatistics, Center for Clinical Investigation, Brigham and Women's Hospital
Assistant Professor (Biostatistics), Harvard Medical School
Chair, American Statistical Association Committee on Professional Ethics

Jonathan Gelfond, MD PhD
Assistant Professor, Department of Epidemiology & Biostatistics
University of Texas Health Science Center at San Antonio

Peter Imrey, PhD
Professor, Cleveland Clinic Lerner College of Medicine, Case Western Reserve University
Past President, International Biometric Society Eastern North American Region
Member, American Statistical Association Committee on Professional Ethics

Wednesday, January 11, 2012
Time: 3:00pm-5:00pm
Reception: 5:00pm-5:30pm
Brigham and Women's Hospital
Carrie Hall
Peter Bent Brigham building
15 Francis Street
Boston, MA

Three panelists and the audience will interactively explore ethical issues in biostatistical practice. Dr. Hurwitz will introduce the topic and present the historical context of ethics in statistical practice. Dr. Gelfond will review ethical principles recently proposed specifically to guide data analysis by clinical and translational researchers (Stat Med. 2011, 30:2785-92), with examples that will ring true for many biostatisticians. Dr. Imrey will add examples, comment from a systems perspective on 21st century medical research integrity concerns, and the statistical profession's responses to medical society reform efforts, and moderate audience discussion, including comments, questions, suggestions, and objections.

Issues in the Analysis of Progression-Free Survival from a Cancer Clinical Trial

Dianne M. Finkelstein, PhD, Massachusetts General Hospital
David Schoenfeld, PhD, Massachusetts General Hospital
Paul Goss, MD, Massachusetts General Hospital

Wednesday, November 16, 2011, 3:00pm-5:00pm
Massachusetts General Hospital
Room: Thier 101

This seminar will be geared to a statistical audience. Dr. Finkelstein will introduce the issues involved in PFS analysis in a cancer trial and Paul Goss will describe breast cancer in general, the specific issues in early stage BC and one specific trial (MA-17) that highlights the issues discussed in this talk. David Schoenfeld will give a talk about some methods work on the topic we are doing. There will then be open discussion with the audience.

Abstract:
The use of PFS as a primary endpoint in cancer trials must consider several issues to ensure validity of this outcome as a surrogate for survival. First, although a trial is designed to evaluate progression at regular prescribed time points, recurrence can be recorded at times outside these times because visits are missed, resulting in interval censored data. Second, the patient may die of the disease before the time of progression is recorded. Third, patients may go off (or change) therapy or withdraw from the study, which could bias the analysis. In addition, the real endpoint of interest in a trial is cancer mortality, but for early stage patients, it is not feasible to design a trial on this endpoint, and showing a treatment is superior on PFS is sometimes not sufficient to change practice. We will discuss these issues and methodology that can be used to refine the analysis of PFS in cancer clinical trials.

Neuropsychological Profiles in Alzheimer's Disease and Cerebral Infarction: A Longitudinal MIMIC Model

Frances Yang, Ph.D., Hebrew Senior Life
Richard Jones, Sc.D., Hebrew Senior Life
Alex Grigorenko, Department of Biostatistics

Wednesday, October 5, 2011, 3:30-5:30 PM, reception to follow
Harvard School of Public Health
Kresge G2

This seminar will describe a longitudinal extension of the Multiple Indicators Multiple Causes (MIMIC) model to characterize associations between cognitive decline and findings of Alzheimer's disease (AD) or cerebral infarction at death. The data come from the Religious Orders Study, a longitudinal study of priests, monks, and nuns who agreed prospectively to autopsy.

The speakers will describe statistical methods for identifying a specific neuropsychological profile characteristic of emerging AD and cerebral infarction. They hypothesized that specific neuropsychological functions are preferentially impaired in the presence of AD and vascular neuropathology. The seminar will cover three topics, (1) Background, (2) an extension of the MIMIC model to the longitudinal setting and its implementation in Mplus, and (3) results.

The study used data from the Religious Orders Study (ROS), a large prospective study of cognitive aging and neuropathology. The sample included 502 ROS participants followed from enrollment to death with an annual neuropsychological battery and brain autopsy. The analytic approach involved the use of Mplus software to estimate a measurement model for neuropsychological performance assessed with 17 neuropsychological tests, extended to accommodate repeated assessments over 10 years. Preliminary results will be presented describing the general pattern of cognitive decline and impairments specific to individual tests in the presence of AD neuropathology or cerebrovascular infarction.

Slides from this presentation.

Comparative Effectiveness Clinical Trials: Methodological and Practical Considerations

Peter Peduzzi, PhD
Professor, Yale School of Public Health
Director, Yale Center for Analytical Sciences
VA Cooperative Studies Program

Wednesday, May 4, 2011, 3:30-4:45 PM
Beth Israel Deaconess Medical Center
Kirstein Living Room, Kirstein Building - 1st Floor
330 Brookline Avenue

Comparative effectiveness clinical trials are comparisons of treatments (usually randomized) designed to determine which treatment options are superior in order to help better inform decision makers. Treatment options could be similar, such as a comparison of different drugs or surgical techniques, or could be very different, such as comparisons of open surgery versus a device or behavioral therapy versus pharmacological therapy. Comparative effectiveness studies have a long history in clinical trials. Schwartz and Lellouch (1967) made the distinction between pragmatic (effectiveness) and explanatory (efficacy) trials. The distinction between effectiveness and efficacy is not always clear. Many trials have elements of both types of studies, i.e., hybrid designs. In this talk, some methodological and practical considerations for designing and conducting these types of studies will be presented and illustrated with data from actual clinical trials. Some of these considerations relate to managing risk factors, maintaining studywide clinical equipoise, accounting for patient preferences, accommodating evolving technology, and using usual care as a comparator. Future directions are discussed and include making comparative effectiveness clinical trials more efficient and generalizable and strengthening the research infrastructure.

The Biostatistician's Role in Managing Clinical Translational Research Data

Brad H. Pollock, MPH, PhD
Department of Epidemiology and Biostatistics
School of Medicine
University of Texas Health Science Center

Wednesday, March 30, 2011, 3:30-4:45 PM
Dana-Farber Cancer Institute
CLSB Building 11081, 3 Blackfan Circle, 11th Floor

Computation has played a pivotal role in modern biostatistical practice with a major emphasis on the development and application of new analytic methods. Less computational attention has been focused on the data management component of biostatistics units. Biomedical informatics has an increasingly prominent role in the clinical translational research enterprise, especially with the growth of the Clinical Translational Science Award program; however, interactions between biostatistics and those in computational disciplines have not been fully exploited. Clinical translational research is likely to be strengthened through synergistic interaction between biostatisticians, informaticians, and information technology experts.

For research data operations, we will discuss the: 1) infrastructure and technologies; 2) personnel responsibilities and oversight; 3) human subjects and security considerations; and 4) opportunities to promote interactions between disciplines. Examples will be given of systems and projects that bring biostatistics together with other experts in order to optimize biostatistical core operations.

Slides from Dr. Pollock's presentation.

Dense longitudinal data analysis for counts of self-reported events

Ronald Thisted, PhD
Department of Health Studies
The University of Chicago

Tuesday, February 8, 2011, 3:30-4:45 PM
Brigham & Women's Hospital
OBC Room 4-002B

Pseudobulbar affect is a condition manifested by socially debilitating outbursts of uncontrollable laughing or crying that can occur multiple times per day. An effective drug for treating this condition will reduce the number of reported episodes, but making that simple idea operational is surprisingly difficult. We examine challenges (and some approaches) that arise in describing, modeling, and making inferences in the context of a randomized clinical trial for which the outcome consists of dense longitudinal observations of daily episode counts. We illustrate these ideas using data from a recently completed clinical trial (Pioro, et al., Ann Neurol 2010; 68: 693-702).

More flexible linear mixed effects models for longitudinal data analysis

Garrett Fitzmaurice, ScD
Professor in the Department of Biostatistics
Harvard School of Public Health

Wednesday, January 19th, 2011, 3:30-4:45 PM
Massachusetts General Hospital
Yawkey 7-980

Linear mixed effects models have become established and enduring methods for longitudinal analyses. However, linear mixed effects models have an important potential limitation: they assume that the shape of the functional relationship between the mean of the longitudinal response and the covariates is known. In this talk we briefly review linear mixed effects models and then discuss a simple extension that allows greater flexibility for the form of the relationship. Specifically, we review the connection between penalized splines and linear mixed effects models and show how a mixed effects model representation of penalized splines makes their extension to the longitudinal setting relatively straightforward.

The main ideas are illustrated using longitudinal data on progesterone metabolite concentration from a study of early pregnancy loss.

The Design and Analysis of Studies of Diagnostic Tests: The Methodologist as Medical Decision Maker

Christopher Lindsell, PhD
Associate Professor and Director of Research, Department of Emergency Medicine
University of Cincinnati

Tuesday, November 9, 2010, 3:30-4:45 PM
Children's Hospital
CLSB, 3 Blackfan Circle, 12th Floor, Room 12007

Physicians rely on diagnostic tests to help with their decision making, and must weigh the strength of information provided against the risks of making an incorrect diagnosis. In designing the research surrounding the diagnostic test, the methodologist has significant impact on subsequent medical decision making. A poorly designed study, or a well designed study poorly analyzed and reported, can compromise a physician's diagnosis and may have a direct effect on patients' lives. It is imperative that every methodologist facilitating research in Academic Health Centers have a thorough understanding of the fundamental components of diagnostic test research, and can apply sound statistical principles to the analysis of these studies. This talk will discuss the primary considerations for the design of diagnostic test research, including choice of index tests and criterion standards, avoiding work up bias, and eliminating circular reasoning. The statistical methods used to analyze diagnostic studies will be reviewed. Approaches that the methodologist can use to extend the analysis to provide information useful to the clinical decision maker evaluating an individual patient will be explored.

Statistical Strategies for Comparative Effectiveness Research Using Observational Data

Sharon-Lise Normand, PhD
Professor of Health Care Policy (Biostatistics)
Harvard Medical School
Professor, Department of Biostatistics
Harvard School of Public Health

Tuesday, October 19, 2010, 3:30-4:45 PM
Kresge G2
Harvard School of Public Health

Substantial attention is focused on comparative effectiveness research, that is, "comparing different medical interventions and strategies to prevent, diagnose, treat, and monitor health conditions" in order to improve health outcomes. The assumption that comparative effectiveness research will provide timely, relevant evidence rests on changing the current framework for assembling evidence. Divergence from this framework involves the recognition that randomized trials that often serve as the basis for new technology approval are small and short-term, and post-market studies are often voluntary and difficult to implement. These problems have become increasingly important over the last decade because technology is changing at a rapid pace, therapies are utilized outside their intended populations, and more representative groups of patients are likely to have differential responses to the same therapy.

In this talk, I will discuss three questions: (1) why use observational data analysis for comparative effectiveness research; (2) how to use observational data for comparative effectiveness research; and (3) what new statistical methodologies will be required for comparative effectiveness research. Key statistical issues, such as defining causal effects, justification for lack of randomization, as well as design and analytical strategies (use of multiple control groups and matching strategies), will be discussed. Examples involving the safety and effectiveness of direct thrombin inhibitors compared to heparin, and of metal-on-metal total hip replacement systems compared to other bearing surface hips, illustrate methods.

Data Monitoring of Clinical Trials Using Prediction

Scott Evans, PhD
Senior Research Scientist in the Department of Biostatistics at the Harvard School of Public Health
(with Lingling Li, Hajime Uno, and LJ Wei)

Wednesday, April 14, 2010, 3:30-5:00 PM
East Campus, Reisman Lecture Hall, Feldberg/Reisman Complex
Beth Israel Deaconess Medical Center

We present use of prediction as an informative and flexible tool for quantitative monitoring of Clinical Trials (CTs). Prediction can be used to assist in the evaluation of efficacy, futility, or to evaluate design assumptions. Prediction provides information regarding effect size estimates and associated precision with trial continuation; can be used in superiority or noninferiority CTs; can be used for binary, continuous, and time-to-event endpoints; provides flexibility in the decision making process; and can be used in tandem with repeated confidence intervals to control error rates. We will describe predicted interval plots (PIPs) and show examples from CTs in which these methods have been utilized by protocol teams, Data Safety and Monitoring Boards (DSMBs), and planners of development programs in decision-making.

Slides to Dr. Evans's presentation.

Adaptive Designs for Clinical Trials: Insightfully Innovative or Irrelevantly Impractical

Stuart Pocock, PhD
Professor of Medical Statistics
London School of Hygiene & Tropical Medicine

Wednesday, March 10, 2010, 3:30-5:00 PM
Dana-Farber Cancer Institute
CLSB Building, 3 Blackfan Circle, 11th Floor, Room 11081A & B

There is much interest in the role of adaptive designs in major Phase III trials with a view to making the development of new treatments more flexible and efficient. This talk will review the main types of Adaptive Design (e.g., unblinded sample size re-estimation, seamless Phase II/III trials) from both statistical and practical perspectives. The methodology will be illustrated by several real examples of adaptive designs.

Slides to Dr. Pocock's presentation.

How many participants? How many measurements? The design of longitudinal studies

Donna Spiegelman, ScD
Professor of Epidemiologic Methods
Harvard School of Public Health

Tuesday, February 23, 2010, 3:00-4:30pm
Yawkey Room 10-660
Massachusetts General Hospital

Longitudinal studies follow N participants, and data on variables of interest are collected r more times after baseline for each participant. In some studies, the number of participants is fixed and the investigator needs to determine the minimum number of additional measurements subject to a pre-specified power constraint. In other studies, the number of times measurements are taken is fixed and the investigator needs to determine how many participants are needed to attain a fixed power. And in some studies, both N and r are free, and the investigator may choose the combination that minimizes study cost for a fixed power, or that maximizes power for a fixed cost. In a longitudinal study, the investigator must specify features of the correlation matrix that describe the relationship between repeated measures from the same person in addition to the usual design inputs. Methods previously developed in the context of clinical trials are extended to allow for exposure prevalence to vary, allowing for primary time metrics other than duration of follow-up, for time-varying exposures, and to allow for correlation between the primary time metric and exposure. Software is available to implement these methods.

Slides to Dr. Spiegelman's presentation.

Overview of methods for analyzing cluster-correlated data

Garrett Fitzmaurice, ScD
Professor in the Department of Biostatistics
Harvard School of Public Health

Wednesday, January 27, 2010, 3:30-5:00pm
(Please note different time)
Kresge G2
Harvard School of Public Health

Many studies in the health sciences give rise to data that are clustered or cluster- correlated. For example, clustered data commonly arise when intact groups are randomized to health interventions or when naturally occurring groups in the population are randomly sampled. With clustered data, we might reasonably expect that measurements on units within a cluster are more similar than measurements on units in different clusters. The degree of clustering can be expressed in terms of correlation and this correlation invalidates the crucial assumption of independence that is the cornerstone of so many standard statistical techniques. In this seminar we (i) present numerous examples where cluster-correlated data arise, (ii) discuss the consequences of clustering for statistical analysis, (iii) review some of the dominant approaches for analyzing cluster-correlated data, and (iv) illustrate, via two case-studies, the application of methods for analyzing cluster-correlated data.

Slides from Dr. Fitzmaurice's presentation.

Musings about missing data: Challenges for the analysis of observational and randomized studies

Nicholas Horton, ScD
Associate Professor, Department of Mathematics & Statistics
Smith College

Tuesday, December 15, 2009, 3:00–4:30pm
Ledge Room 4-002B, One Brigham Circle
Brigham and Women's Hospital

Missing data arise in almost all real-world situations and can cause bias or lead to inefficient analyses. The development of statistical methods to address missingness has been actively pursued in recent years. This talk will (1) address complications in observational studies when there are many patterns of missing values for categorical and continuous predictors, (2) discuss issues in implementing analyses that are consistent with the intention to treat principle in randomized trials, and (3) demonstrate how these methods can be implemented through detailed discussion of examples.

Slides from Dr. Horton's presentation.

Translating research to practice: An introduction to causal inference, with extensions to longitudinal data

Tyler VanderWeele, PhD
Associate Professor of Epidemiology
Harvard School of Public Health

Wednesday, November 18, 2009, 3:00–4:30pm
Trustman Boardroom
East Campus, Feldberg / Reisman Complex - 2nd Floor
Beth Israel Deaconess Medical Center

The first talk in the series, to be presented by Tyler VanderWeele, PhD, will discuss causal inference in the context of longitudinal data. The lecture will give a brief overview of how the "counterfactual" or "potential outcomes" framework can be useful in distinguishing association from causation. Issues concerning time-dependent confounding that can arise in longitudinal data will be discussed, and an introduction to causal methods to handle time-dependent confounding will be given. The ideas will be illustrated by a detailed discussion of an example using longitudinal data to distinguish the relative persistence of the effect of loneliness on depression versus on subjective well-being.

Slides from Dr. VanderWeele's presentation.