Log-Linear Modeling of Health Services Use in a Sample Survey

• Activity Number: 257
• Type: Topic Contributed
• Date/Time: Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): John S. Preisser*+ and Thomas A. Arcury
• Companies: The University of North Carolina at Chapel Hill and Wake Forest University
• Keywords: Incidence density ; Poisson regression ; SUDAAN software

Abstract

Survey regression techniques for count data can be used to model the number of events such as health care visits among individuals in complex sample surveys. The Geographic Accessibility of Health Care in Rural Areas Study was a cross-sectional stratified multi-stage sample survey conducted to study health care utilization in a rural area. A total of 1059 adults from 12 counties in western North Carolina were interviewed and asked questions regarding personal geographic factors including activity space and transportation use, as well as sociodemographic, cultural and health status characteristics. We discuss the use of the SUDAAN software package, and particularly, its LOGLINK procedure, to quantify associations between individual characteristics and utilization rates, adjusting for features---unequal weighting, stratification, and clustering---of the sample design.