Spring 2005 Meeting
Held on March 10,
2005 at the Wyndham Glenview Suites.
The Program consisted of three presentations:
- Confidence Regions for Random-Effects: Calibration
Curves with Heteroscedastic Errors.
Dulal Bhaumik, Ph.D., Center for Health Statistics, University of
Illinois at Chicago
- Data Safety and Monitoring Boards: Some Personal
Robert (Skip) Woolson, Ph.D., Department of Biostatistics,
Bioinformatics & Epidemiology, Medical University of South Carolina
- VA Cooperative Study #364: Long term patency of
saphenous vein and left internal mammary artery grafts after coronary
artery bypass surgery.
Thomas Moritz, M.S., Hines Veterans Affairs Hospital
Dr. Dulal Bhaumik is presently an Associate Professor of Biostatistics and
Psychiatry at the University of Illinois at Chicago (UIC). He joined UIC in
September 2002 after 12 years of teaching and research at the University of
South Alabama Department of Mathematics and Statistics. He obtained his
Master’s degree in Statistics from the Indian Statistical Institute in 1983
and his Ph.D. in Statistics from the University of Maryland (UMBC) in 1988.
In 2002, he won the American Statistical Association’s Youden Prize for
Contributions to Interlaboratory Calibration. Dr. Bhaumik has over 25
peer-reviewed publications including articles in the Journal of the Royal
Statistical Society, Journal of the American Statistical Association, and
Technometrics. He is also Co-Principal Investigator for several current
NIH/NIMH funded research projects.
Confidence bounds are constructed for a random-effects calibration curve
model. An example application is analysis of analytical chemistry data in
which the calibration curve contains measurements y for several
values of known concentration x in each of q laboratories.
Laboratory is considered a random-effect in this design, and the intercept
and slope of the calibration curve are allowed to have laboratory-specific
values. This presentation focuses on: (i) develop an appropriate
inter-laboratory calibration curve for heteroscedastic data of the type
commonly observed in analytical chemistry, (ii) compute a point estimate for
an unknown true concentration x when corresponding measured
concentrations y1, y2, … yq' are provided from q' laboratories
(i.e., a subset of the original q laboratories used to calibrate the model,
where 1 < q' < q), (iii) compute the asymptotic mean and
variance of the estimate, (iv) construct a confidence region for x.
The methods are then illustrated using both simulated and typical
inter-laboratory calibration data.Other relevant applications of the general
approach will be highlighted.
Dr. Robert F. Woolson is presently Professor of Biostatistics,
Bioinformatics and Epidemiology at the Medical University of South Carolina
(MUSC) in Charleston, and Professor Emeritus of Biostatistics and Statistics
at University of Iowa (UI). He joined MUSC in August of 2002 after
concluding 29 years on the medical school/public health school faculty at
UI. At UI he was Professor and Head of Biostatistics and Associate Dean for
Research for the UI College of Public Health. Dr. Woolson conducts both
collaborative clinical research and biostatistical methodologic research. At
UI he founded and directed the Clinical Trials Statistical Data Management
Center, which continues to serve as a resource for the design, conduct,
coordination, and statistical analysis of multi-center clinical trials.He
was principal investigator for several large NINDS grants to coordinate
multi-center trials as well as statistical methodology grants—TOAST, IHAST,
COSS (planning grant).
Dr. Woolson served on the VA Cooperative Studies Evaluation Committee for
six years and served a term as a member of the NIAID Data Safety &
Monitoring Board for AIDS Therapeutic Trials Program. At the present time he
is a member of several DSMB’s including: the NIAID Hematopoietic Stem Cell
Transplant program; and for VA Cooperative Study # 526, a trial evaluating
thyroid hormone for heart failure. He has also served on DSMB’s for a number
of corporate trials, mostly trials involving potential stroke therapies. His
methodological research interests include longitudinal data analysis,
survival analysis, and clinical trial / epidemiological methods. His
statistical research grants have been supported by the National Institute of
Mental Health and the National Cancer Institute. Dr. Woolson has mentored
numerous K30, K08 and other clinical research trainees, as well as graduate
students; and faculty. He is an ASA Fellow, is presently on the editorial
board for Statistics In Medicine, and previously served as an associate
editor for Controlled Clinical Trials and for Statistical Methods for
Safety and Monitoring Boards (DSMB’s) are often appointed for the purpose of
reviewing interim data during the course of a randomized clinical trial.
Such boards are charged with reviewing accumulating evidence to see if there
is sufficient evidence to conclude a trial on the basis of benefit, lack of
benefit, or if there is undue harm to study participants. Such DSMB’s have
held a prominent place in the conduct of federally sponsored trials, for
example NIH, VA and related funding agencies. DSMB’s are also common in
corporate sponsored trials, particularly those sponsored by pharmaceutical
corporations. Indeed, common practice today within any academic health
center is for a data safety and monitoring plan to be in place for a
clinical study, and for a clinical trial this data safety and monitoring
plan would be to have a DSMB with formal guidelines for consideration of
early trial termination.
Biostatisticians have the opportunity to serve on DSMB’s, or to be one of
the key liaisons between a study and an external DSMB. This presentation
discusses general issues, challenges and personal experiences in the context
Thomas Moritz is presently a Biostatistician in the Cooperative Studies
Program Coordinating Center at Hines Veterans Affairs Hospital. He obtained
a B.S. in Mathematics at Marquette University and M.S. degrees in Applied
Statistics and Biostatistics at Iowa State University and the Medical
College of Wisconsin, respectively. During his 20 year tenure at the Hines
Coordinating Center he has managed studies in lung and colon cancer,
pulmonary disease, liver disease, cardiac surgery and vascular surgery. He
has over 50 publications and has also presented at the Society of Clinical
Trials and NIC/ASA meetings.
The VA Cooperative Studies Trial defined long-term (ten-year) saphenous vein
graft (SVG) and left internal mammary artery (IMA) patency in patients
undergoing coronary artery bypass graft (CABG) surgery. Traditionally,
reports of long-term graft patency rates have used coronary angiography
results from a single time point after CABG. Patency rates from a single
time period distant from the original operation may result in biased
estimates if there is no accounting for interceding interventions such as
repeat coronary surgery, percutaneous coronary intervention or death. In
this study, patients had between 1 and 5 serial angiograms.
The data from this study posed two major analytic problems. The first
problem was that the exact time of graft occlusion could not be known. This
problem is addressed by using interval-censored observations in the survival
analysis (PROC LIFETEST in SAS® ). This analysis requires identification of
the time interval in which the occlusion occurred. Comparisons of
Kaplan-Meier product-limit survival curves were made with the log-rank test.
Although time-related analyses of graft patency data used the exact date of
each postoperative angiogram, for convenience of presentation, some
information is presented in arbitrarily defined time frames. The second
analytic problem was related to the fact that there were multiple grafts,
i.e., clustered observations, within a patient. It has previously been
demonstrated that graft patency within a patient is not independent. This
does not affect the estimates for patency rates, but does cause the standard
error terms to be underestimated. Recently, the SAS® macro, IWM, has become
available for the analysis of clustered, interval-censored survival data.
This approach produces robust estimates of the standard error terms by
adjusting for the correlated nature of the clustered observations.
Patient-related risk variables, graft-related risk variables and CABG
surgery processes of care variables were used as candidate independent
variables in the IWM macro to identify the set of variables that jointly
predict ten-year graft patency.