Mapping Disease Risk with Changing Census Boundaries

• Activity Number: 526
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Paul Nguyen*+ and Patrick Brown and Jamie Stafford
• Companies: University of Toronto and Cancer Care Ontario and University of Toronto
• Keywords: local-EM algorithm ; parametric bootstrapping ; risk surface ; tessellation

Abstract

Mapping disease risk often involves working with data aggregated in space and time due to data being collected at an area level, such as postal codes or health regions. Furthermore, studying rare diseases requires using data collected over a long time period, introducing the problem of the boundaries of census regions changing over time. Adopting a local likelihood framework, we present a local-EM algorithm that permits multiple maps with different census boundaries to be combined. This algorithm estimates the risk surface on a tessellation where the cumulative intersections of the various maps are distinct regions. We also implement parametric bootstrapping procedures to make inference on the uncertainty of these risk surface estimates. These methods were demonstrated on mesothelioma lung cancer data collected in the counties of Lambton and Middlesex of southwestern Ontario, Canada between 1985 and 2007. We identified several areas of elevated risk throughout Lambton County and areas of low risk throughout Middlesex County.