Northeastern Illinois Chapter
American Statistical Association

Fall 2010 Meeting


To be held on Wednesday, October 20, 2010 at the Crowne Plaza Hotel in Northbrook, IL. For registration information, click here. The deadline to register is October 13th.

The Program will consist of three presentations:

  1. Mixed Models For Longitudinal Ordinal And Nominal Data
    Don Hedeker, Professor of Biostatistics, University of Illinois at Chicago
  2. Analysis of Competing Risks
    James Dignam, Associate Professor of Biostatistics, University of Chicago
  3. Meta Analysis for Rare Event Studies
  4. Dulal K Bhaumik, Professor of Biostatistics and Psychiatry, University of Illinois at Chicago




Mixed Models For Longitudinal Ordinal And Nominal Data by Don Hedeker, Professor of Biostatistics, University of Illinois at Chicago

Abstract:
Ordinal and nominal outcomes are common in many areas of research. It can often be the case that these outcomes are repeatedly measured from individuals. This presentation will focus on generalizations of the logistic regression model for categorical longitudinal data, within a mixed model framework.
Specifically, the following models will be described and compared: mixed-effects logistic regression for nominal outcomes, and mixed-effects proportional odds and non-proportional odds models for ordinal outcomes. The latter models are useful because the proportional odds assumption of equal covariate effects across the cumulative logits of the model is often unreasonable. Instead, allowing the covariates to have varying influences across the cumulative logits of the ordinal responses provides a very flexible approach to the modeling of ordinal outcomes.
To illustrate and compare model features, analyses will be presented of a dataset where psychiatric homeless subjects were repeatedly assessed in terms of their housing situation (classified as either independent, community, or variable housing). Treatment of these repeated housing status classifications as ordinal and nominal outcomes will be compared. The use of SAS PROC NLMIXED for these models will be described and illustrated.

Biographical Background:
Dr. Hedeker is a Professor of Biostatistics in the Division of Epidemiology and Biostatistics, School of Public Health, where has been since 1993. He received his Ph.D. in Quantitative Psychology from The University of Chicago in 1989. Dr. Hedekerís main expertise is in the development and use of advanced statistical methods for clustered and longitudinal data, with particular emphasis on mixed-effects models. He is the primary author of several freeware computer programs for mixed-effects analysis and is published, with co-author Robert D. Gibbons, the text "Longitudinal Data Analysis" (Wiley, 2006). In 2000, Dr. Hedeker was named a Fellow of the American Statistical Association. Dr. Hedeker is an Associate Editor for Statistics in Medicine and Journal of Statistical Software, a past Associate Editor for Journal of Educational and Behavioral Statistics. He has also been the PI, co-PI, or co-I on over twenty NIH or CDC research grants.



Analysis of Competing Risks by James Dignam, Associate Professor of Biostatistics, University of Chicago

Abstract:
In clinical cancer research, competing risks are frequently encountered. For example, individuals undergoing treatment for surgically resectable disease may experience recurrence near the removed tumor, metastatic recurrence at other sites, occurrence of second primary cancer, or death from non-cancer causes prior to any of these events. Two quantities, the cause-specific hazard function and the cumulative incidence function, are commonly used to summarize outcomes by event type. Tests for event-specific differences between treatment groups may thus be based on comparison of a) cause-specific hazards via a logrank or related test, or b) the cumulative incidence functions via one of several available tests. Similarly, modeling of covariate effects may be based on either metric. Inferential results for these approaches can differ considerably for the same cause-specific endpoint. Depending on the questions of principal interest, one or both metrics may be appropriate to consider.
Simulation study results will be presented along with a discussion of examples from cancer clinical trials to illustrate these points and provide guidance for analysis when competing risks are present.

Biographical Background:
Dr. Dignam is an Associate Professor of Biostatistics in the Department of Health Studies at the University of Chicago and an Investigator in the University of Chicago Cancer Research Center. He has an extensive background in multi-center cancer clinical trial research, having been a statistician for the National Surgical Adjuvant Breast and Bowel Cancer Project (NSABP) and the American College of Surgeons Oncology Group (ACoSOG) before recently becoming the Group Statistician for the Radiation Therapy Oncology Group (RTOG). He has also been the principal investigator of National Institutes of Health and Susan G. Komen Breast Cancer Foundation grants in breast cancer research. He currently teaches courses in introductory Biostatistics, Design and Analysis of Clinical Trials, and Survival Analysis, and his research interests include cancer clinical trial design and interim monitoring, racial/ethnic background and cancer prognosis, and competing risks and multiple endpoints in survival analysis. Dr. Dignam holds a Ph.D. in biostatistics from the University of Pittsburgh.



Meta Analysis for Rare Event Studies by Dulal K Bhaumik, Professor of Biostatistics and Psychiatry, University of Illinois at Chicago

Abstract:
This presentation will discuss how to estimate the treatment effect when heterogeneity is present for rare event studies. Several moment-based estimates will be compared with the marginal maximum-likelihood estimate. A new method is proposed to test the heterogeneity parameter and show that it controls the type I error rate to the pre-specified nominal level. A definite guideline will be provided to select an appropriate model when model violations are expected. The role of sample size is also investigated in the context of both estimation and testing. The results are illustrated with a real data example.

Biographical Background:
Dr. Bhaumik is a Professor of Psychiatry, Biostatistics and Bioengineering at the University of Illinois at Chicago. He received his BS in Statistics from Calcutta University, his MS in Statistics from the Indian Statistical Institute, and his PhD in the same field from the University of Maryland. Before joining UIC, he served as a Professor of Statistics in the Department of Mathematics and Statistics at University of South Alabama.



Last updated: 09/24/10