Accommodating Rate Instability for Small Areas in Disease Mapping

• Activity Number: 308
• Type: Invited
• Date/Time: Wednesday, August 6, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Daniel E. Wartenberg*+ and Gerald A. Harris and Andrew Booth Lawson
• Companies: University of Medicine and Dentistry of New Jersey and Rutgers University and University of South Carolina
• Keywords: aggregation ; empirical Bayes ; small area analysis ; rate instability ; disease mapping

Abstract

Disease mapping methods are often used to describe and summarize geographic patterns of disease. Theoretically, to achieve the most accurate and precise map, one uses data at the smallest geographic scale available. In practice, both population and disease data are often reported for political units such as towns, zip codes, and counties, and these define the unit of analysis. Some units have small populations and are extremely sensitive to variations of just a few cases of disease. This can result in misleading interpretations and inappropriate statistical comparisons of rates if not adjusted. We examine several methods used to address this concern by either averaging or combining nearby regions. These include methods that aggregate neighboring regions in an arbitrary order until they achieve a minimum population size, ones that achieve a minimum population size by aggregating regions that are similar in terms of possible confounding variables, and also smoothing approaches, such as empirical Bayes. Results are compared for both simulated and real data. Strengths and weaknesses of smoothing versus aggregation are discussed, and recommendations are provided.