A Comparison of Likelihood-Based Spatiotemporal Surveillance Methods in Non-Homogeneous Population

• Activity Number: 675
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Sung Won Han*+ and Lianjie Shu and Wei Jiang and Tsui Kwok-Leung
• Companies: University of Pennsylvania and University of Macau and Hong Kong University of Science and Technology and City University of Hong Kong
• Keywords: spatiotemporal surveillance ; generalized likelihood ratios ; weighted likelihood ; non-homogeneous Poisson ; change point detection ; scan statistics

Abstract

Motivated by the applications in healthcare surveillance, this paper discusses the spatiotemporal surveillance problem of detecting the mean change of Poisson count data in non-homogeneous population environment. Through Monte Carlo simulations, we investigate several likelihood ratio-based approaches and compare them under various scenarios depending on the factors such as the population trend, the change magnitude, the change coverage, and the change time. Our simulation study shows that no method is uniformly better than others in all scenarios. The performance of the methods based on single radius or variable radius depends on only the two of the factors, not all of them. In addition, the difference between generalized likelihood ratios (GLR) approach and weighted likelihood ratios (WLR) approach depends mainly on population size, not the change coverage, change magnitude, or change time. We find that the part (time period or spatial region) with a small population benefits the WLR approach, but that with a large population benefits the GLR under any population trends.