Spatiotemporal Biosurveillance Under Nonhomogeneous Population

• Activity Number: 42
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Sung Won Han*+ and Seoung Bum Kim and Kwok-Leung Tsui
• Companies: Georgia Institute of Technology and Korea University and Georgia Institute of Technology
• Keywords: health care surveillance ; scan statistic ; likelihood ratios ; CUSUM ; EWMA ; on-line monitoring

Abstract

Motivated by the applications in health care and bio surveillance, we discuss the spatiotemporal surveillance problem of detecting the mean change of Poisson count data under non-homogeneous population. We review some of the current methods for the non-homogeneous sample size for temporal surveillance and propose new spatiotemporal surveillance methods. Generalized likelihood ratios, weighted likelihood ratio, and normal approximation are used to adjust the population effect. We compare them through Monte Carlo simulation, and apply them to a real case of male thyroid cancer data in New Mexico. We show that Rossi's normal approximation or EWMA technique performs poorly under decreasing or increasing population, respectively. Based on the simulation study, we argue that the weighted likelihood ratio technique should be carefully applied to multivariate monitoring cases.