|
Statistical Breakfast Roundtable Discussions
$20 to attend. Registration will be limited so sign up early!
THURSDAY from 7-8am (additional fee applies)
Who is missing? Preparing for the occurrence of missing data in research
Maria Magdalena Llabre, Behavioral Medicine Research Center, University of Miami, Miami, FL
It is now recognized that the methods commonly used to deal with the pervasive problem of missing data in randomized clinical trials produce biased estimates of the parameters of change and may result in a loss of power. The bias results from the requirement of such methods that the missingness be due to a completely random process (MCAR), a requirement seldom satisfied in the real world. Newer methods for analyzing data in the presence of missing observations have less stringent assumptions about the missingness mechanism. The label used to describe the requirements of the newer methods, that data be missing at random (MAR), has led to confusion, as MAR means that missingness can be predicted from data at hand. Analytic methods that require MAR will benefit from the inclusion of variables that can differentiate participants who will have missing data on a particular variable versus those who will not. Data collection at baseline in a clinical trial becomes particularly important, as it can incorporate measures unrelated to the primary outcome, but related to missingness. As an example participants might be asked to report on their level of motivation for completion of the trial.
Participants at this roundtable will learn:
- the difference between missing completely at random, missing at random, and missing not at random
- the assumptions made by the various methods of dealing with missing data
- the general ideas behind multiple imputation and full information maximum likelihood methods
- how to prepare for the occurrence of missing data in a study
FRIDAY from 7-8am (additional fee applies)
Statistical Guidelines for Psychosomatic Medicine
Michael A. Babyak, Department of Psychiatry and Behavioral Science, Duke University Medical Center, Durham, NC and Kenneth E. Freedland, Department of Psychiatry, Washington University School of Medicine, St. Louis, MO
Psychosomatic Medicine implemented a new set of statistical guidelines in 2005. The initial guidelines concern 1) how statistical methods and results are reported, 2) directional hypothesis tests, 3) artificial categorization of continuous variables, 4) automated selection regression models, 5) covariables and covariate adjustment, and 6) model validation. The guidelines have been very well received, although occasional controversies have arisen about some of them. During the year that they have been in effect, they have helped to raise the methodological quality of research reported in the journal. Additional guidelines are planned for testing interactions, subgroup analyses in clinical trials, and multiple testing procedures. The session will be led by two of the journal's associate editors. They will discuss the guidelines, answer questions, and seek feedback from the participants about existing and planned guidelines.
Participants at this roundtable will learn:
- the reasons behind the journal's decision to implement statistical guidelines,
- the rationale for each guideline,
- how to use the guidelines to improve the chances that their articles will be accepted for publication, and
- how to use the guidelines when reviewing manuscripts for the journal.
|
