Stat4Onc
2019 Symposium Short Courses
April 25,
2019
Morning courses 8:30 AM to 12:00
noon with one break
Afternoon
courses 1:30 PM to 5:00
PM
with one break
Course 3 (Afternoon
Course) Statistical Remedies for Flawed Conventions in Medical Research
Instructor – Dr. Peter Thall,
University of Texas, M.D. Anderson Cancer Center
Abstract
Many statistical methods commonly used for data analysis or
clinical trial design by medical researchers are deeply flawed.
Unfortunately, many of these dysfunctional statistical conventions and
paradigms are deeply embedded in the medical research community, and have
become standard or even required practice. Ultimately, the consequence is that
practicing physicians are misled to choose inferior or harmful treatments for
their patients. In this half day short course, I will identify and describe, by
example, severe problems with a variety of statistical practices commonly used
by medical statisticians and physician researchers. For each flawed practice, I
will provide at least one practical alternative.
Prerequisites: The short course
will include limited mathematical detail, with each example presented from a
practical viewpoint. Attendees should have some knowledge of elementary
probability and statistics, and clinical trial design.
Target Audience: Attendees may be
anyone involved in statistical analysis of medical data, or clinical design and
conduct, including statisticians, physicians, research nurses, professionals in
the pharmaceutical industry, and federal employees in the NIH or FDA.
Topics Covered, as time permits
Misinterpreting tests of hypotheses
P-values, strength of evidence, and Bayesian inference
Relationships between treatment response and survival
time
Being misled by single-arm trials
Futile futility stopping rules
Unsafe safety monitoring rules
Simpson’s paradox
Sexy biomarkers and stratification
Simple graphical methods to avoid disaster
Randomization and causality
Inverse probability of treatment weighted estimation
Outcome adaptive randomization
Cherry picking
Treatment screening and phase II-III designs
Estimating treatment-covariate interactions
Avoiding overfitting regression models
Salvage therapy and dynamic treatment regimes
SMART designs
Choosing doses the wrong and right way
Biography
Peter
F. Thall is the Anise J. Sorrell Professor in the
Department of Biostatistics at M.D. Anderson Cancer Center, and an adjunct
professor in the Department of Statistics at Rice University. He is a
Fellow of the American Statistical Association and the Society for Clinical
Trials, and received the Don Owen Award in 2014. Dr. Thall
has published over 250 papers and book chapters in the statistical and medical
literature, and co-authored the 2016 book Bayesian Designs for Phase I-II
Clinical Trials. His latest book, Statistical Remedies for Medical
Researchers, is in press and should become available in the summer of 2019.
Dr. Thall’s research areas include clinical trial
design, precision medicine, Bayesian nonparametric statistics, incorporating
expert opinion into Bayesian inference, and dynamic treatment regimes. He has
presented over 200 invited talks and 30 short courses, served as an associate
editor for Journal of the National Cancer Institute, Statistics in
Medicine, Statistics in Biosciences, Clinical Trials, and
Biometrics, and is an ASA media expert.