5203.0: Wednesday, October 24, 2001: 2:30 PM-4:00 PM | ||||
Oral Session | ||||
| ||||
| Epidemiological data pose many statistical challenges due to the frequent need to cope with observational data that is often highly variable and/or skewed, incomplete and/or unreliable. This session includes papers dealing with statistical methodologies for handling epidemiological data with one or more of these problems. When observers vary in assessing a categorical attribute, an inappropriate use and associated incorrect package-based computation of the common kappa statistic is discussed. Bayesian and frequentist latent variable models are proposed for data where observers may vary due to specific intent to conceal sensitive information, such as substance abuse. In papers on missing data, a modified product-binomial likelihood analysis is presented for cohort or case-control studies with non-ignorably missing data, particularly where supplementary data collection can be used to obtain further information about non-response, and methods are compared for dealing with missing-at-random (MAR) predictors in logistic regression analyses of complex survey data. Finally, logistic regression versions of common nonparametric tests for matched data are discussed and shown to be useful in analyses with extremely variable predictors, such as many studies in occupational epidemiology. | ||||
| See individual abstracts for presenting author's disclosure statement. | ||||
| Learning Objectives: At the conclusion of the session, the participant should be able to: 1. Recognize conceptual difficulties with kappa statistics, and approaches to overcoming them. 2. Recognize, and identify strengths and weaknesses, of new methods for estimating rates in the presence of possibly discordant reports from multiple informants. 3. Describe analyses of 2´2 tables with non-ignorable missingness and possibly with supplementary data, using a modified product-binomial likelihood. 4. Recognize three approaches to handling missing-at-random covariate data in logistic regression analyses of data from complex sample surveys. 5. Describe the use of conditional logistic regression to obtain covariate adjusted rank tests for use with ill-conditioned epidemiological exposure data. | ||||
| Vicki S. Hertzberg, PhD | ||||
| Difficulties of using kappa statistics in epidemiologic studies Sunny Kim, PhD, Stanley Lemeshow, PhD | ||||
| Latent variable models for estimating rates of sensitive behaviors subject to discordant reporting Jeffrey A. Welge, PhD, John Schafer, PhD | ||||
| Informatively missing data in the 2 by 2 table: Likelihood-based analysis for case-control and cohort studies Robert H. Lyles, PhD, Andrew S. Allen, MA | ||||
| Logistic regression with incomplete covariate data in complex survey sampling Charity G Moore, MSPH, PhD, Stuart R Lipsitz, ScD, Cheryl L Addy, PhD, James R Hussey, PhD, Donald G Edwards | ||||
| Conditional logistic regression rank tests Peter B. Imrey, PhD | ||||
| Sponsor: | Statistics | |||
| CE Credits: | CME, Health Education (CHES), Nursing, Pharmacy, Social Work | |||