Disease mapping and cluster detection methods are statistical approaches to identify and summarize geographic patterns of disease occurrence. With the widespread availability and use of geographic information systems (GIS) and disease and environmental data, it is important that the analytic tools provide clear, accurate, precise and interpretable results. Toward that end, the proposed project will address three critical issues in geographical analysis. First, we will investigate a variety of methods for accommodating instability in rates from regions with small populations at risk. If not address adequately, maps may display spurious peaks (i.e., clusters) and valleys that can lead to misinterpretation. Traditional approaches include empirical Bayes mapping (i.e., smoothing) and grouping of neighboring geographical units. Second, we will continue our work on the development of geographic surveillance tools. One main goal of surveillance is to identify important changes in rates or patterns of disease occurrence for disease prevention and control activities by reviewing routinely collected data on an on-going basis. However, most approaches consider only temporal changes. Yet, perceptions of clusters are often spatial, and environmental pollutants typically are described in terms of the spatial or space-time distribution. This project will extend surveillance methods for use with spatial and space-time data, and will develop approaches for prospective rather than only retrospective evaluation. Third, we will extend our work on methods for analyzing geographic data when information is missing for some of the geographic units. Often, due to administrative or jurisdictional limitations, data are not available for an entire study region. However, there may be a need to be able to estimate rates for the entire region, or for those geographical units for which data are unavailable. We will explore methods for imputing values and adjusting for edge effects using methods including Markov Chain Monte Carlo simulations.
