BACKGROUND: Disease rates for geographic areas with small populations may be unstable. Therefore, accurate nonparametric methods for smoothing or stabilizing rates are needed. METHODS: We propose an innovative locally-weighted-average method as an easy tool for disease surveillance. Our approach has several important advantages over existing locally-weighted-average methods. One advantage is that the buffer zone is created based on a polygon rather than centroid. Second, the buffer distance is determined by a user-specified population threshold. Third, a weighting factor that accounts for variability in the rate is used in the smoothing process. We further propose a variance-driven procedure to reduce arbitrariness in selecting the population threshold, and a binary search technique to quickly and precisely find the buffer distance according to the specified population threshold. Lastly, we develop a software tool using ArcObjects (ESRI, Redland, CA) to implement this method. RESULTS: Our method was applied to town-level lung cancer incidence rates for New Hampshire. A comparison with a traditional point-based method indicated that our method produced less under- and over-smoothing. CONCLUSION: Our method and the software tool are suitable for researchers and public health workers who want to apply geographic information systems to map smoothed disease rates for exploratory purposes.
BACKGROUND: Disease rates for geographic areas with small populations may be unstable. Therefore, accurate nonparametric methods for smoothing or stabilizing rates are needed. METHODS: We propose an innovative locally-weighted-average method as an easy tool for disease surveillance. Our approach has several important advantages over existing locally-weighted-average methods. One advantage is that the buffer zone is created based on a polygon rather than centroid. Second, the buffer distance is determined by a user-specified population threshold. Third, a weighting factor that accounts for variability in the rate is used in the smoothing process. We further propose a variance-driven procedure to reduce arbitrariness in selecting the population threshold, and a binary search technique to quickly and precisely find the buffer distance according to the specified population threshold. Lastly, we develop a software tool using ArcObjects (ESRI, Redland, CA) to implement this method. RESULTS: Our method was applied to town-level lung cancer incidence rates for New Hampshire. A comparison with a traditional point-based method indicated that our method produced less under- and over-smoothing. CONCLUSION: Our method and the software tool are suitable for researchers and public health workers who want to apply geographic information systems to map smoothed disease rates for exploratory purposes.