Committee on Women in Statistics

SESSION DESCRIPTION: The focus of the session will be the discussion of Bayesian methodology applications in health care policy and outcomes research. Speakers will show that a possibly complex Bayesian methodology can be used to assure cleaner, more straightforward results less subject to misinterpretation. In the information age, computing capabilities are greatly expanding which has enabled the use of Bayesian methods to a much greater extent. The application of Bayesian methodology to health outcomes is a fairly recent development and addresses problems that have not yet been adequately answered. Four women who have specialized in this area, from a new graduate to those in the field for years will discuss a spectrum of views from across the country.

SPONSOR: Committee on Women in Statistics

CO-SPONSOR: Section on Bayesian Statistical Science

SESSION ORGANIZER AND CHAIR:
J. Lynn Palmer
Departments of Symptom Control & Palliative Care and Biostatistics
The University of Texas M.D. Anderson Cancer Center
1515 Holcombe Boulevard, Box 8
Houston, TX 77030
Email: jlp@odin.mdacc.tmc.edu

INVITED SPEAKERS:
Kate Cowles
Department of Statistics and Actuarial Science, 241 SH
The University of Iowa
Iowa City, IA 52242-1409
Email: kcowles@stat.uiowa.edu
Title: "Surrogate endpoints for accelerating drug approval: Bayesian evaluative methods"
The criteria used for approval of new drugs are an important health policy issue. In order to reduce the size and duration of clinical trials, surrogate endpoints, such as laboratory values, are increasingly being used as clinical trial endpoints instead of the clinical events, such as death or worsening clinical condition, that are of true interest. This talk will discuss Bayesian methods for attempting to determine whether the same conclusions are likely to be reached using surrogate endpoints as would have been reached using the true clinical endpoints.

Jennifer Hill
Columbia School of Social Work
622 W. 113th St., Room 803
NY, NY 10025
Email: jh1030@columbia.edu
Title: "Use of a Bayesian approach to resolve long-standing debates about opinion-changing behavior"
Finite mixture models are built which allow us to resolve long-standing debates between the most prominent schools of thought in the opinion-changing-behavior literature. A Bayesian approach facilitates model fitting as well as valid and flexible model comparisons (even in situations where likelihood ratio tests do not satisfy the appropriate regularity conditions) leading to more straightforward conclusions about the merits of competing hypotheses. Results have implications for the process by which policy is formed.

Carolyn Rutter
Group Health Cooperative
Center for Health Studies
1730 Minor Ave, Suite 1600
Seattle, WA 98101
Email: rutter.c@ghc.org
Title: "Evaluating the symptom checklist as a diagnostic tool: A Bayesian method that allows error in true depression status"
We use a Bayesian model that combines latent class models with receiver operating characteristic (ROC) analysis to estimate the ability of the symptom checklist to detect depression while accounting for errors in diagnosis. There are many statistical innovations in this work, including the application of latent class models to ROC curve estimation, the form of the ROC models used for symptom checklist distributions, and the inclusion of longitudinal assessments within subjects. These models provide more realistic estimates of symptom checklist accuracy than frequentist models that assume an error-free depression diagnosis. The models we develop are generally useful for mental health diagnoses with a 'true state' that is based on a constellation of symptoms.

DISCUSSANT:
Cindy Christiansen
Boston University
Edith Nourse Memorial Veterans Hospital
200 Spring Road, Bedford, MA 01730
Email: cindylc@bu.edu
Dr. Christiansen is an expert in the use of Bayesian methodologies for health services research. She will summarize, discuss, and identify the common-threads of the applications and the statistical approaches from these three talks.