Locating irregularly shaped clusters of infection intensity Niko Yiannakoulias 1 , Shona Wilson 2 , H. Curtis Kariuki 3 , Joseph K. Mwatha 4 , John H. Ouma 4 , Eric Muchiri 3 , Gachuhi Kimani 4 , Birgitte J. Vennervald 5 , David W. Dunne 2 1 School of Geography and Earth Sciences, McMaster University, Hamilton, Ontario L8S4K1, Canada; 2 Department of Pathology, University of Cambridge, UK; 3 Division of Vector Borne Diseases, Ministry of Health, Nairobi, Kenya; 4 Kenya Medical Research Institute, Nairobi, Kenya; 5 DBL-Centre for Health Research and Development, Faculty of Life Science, University of Copenhagen, Copenhagen, Denmark Abstract. Patterns of disease may take on irregular geographic shapes, especially when features of the physical envi- ronment influence risk. Identifying these patterns can be important for planning, and also identifying new environ- mental or social factors associated with high or low risk of illness. Until recently, cluster detection methods were lim- ited in their ability to detect irregular spatial patterns, and limited to finding clusters that were roughly circular in shape. This approach has less power to detect irregularly-shaped, yet important spatial anomalies, particularly at high spatial resolutions. We employ a new method of finding irregularly-shaped spatial clusters at micro-geographical scales using both simulated and real data on Schistosoma mansoni and hookworm infection intensities. This method, which we refer to as the “greedy growth scan”, is a modification of the spatial scan method for cluster detection. Real data are based on samples of hookworm and S. mansoni from Kitengei, Makueni district, Kenya. Our analysis of simulat- ed data shows how methods able to find irregular shapes are more likely to identify clusters along rivers than methods constrained to fixed geometries. Our analysis of infection intensity identifies two small areas within the study region in which infection intensity is elevated, possibly due to local features of the physical or social environment. Collectively, our results show that the “greedy growth scan” is a suitable method for exploratory geographical analysis of infection intensity data when irregular shapes are suspected, especially at micro-geographical scales. Keywords: disease clusters, schistosomiasis, hookworm, spatial scan, Kenya. Introduction Geographic cluster detection methods can be employed to identify localised geographic patterns of disease, and have been applied in a variety of decision support and research contexts. These meth- ods are used to find geographic “hot-spots” where illness and/or infection is highly concentrated in the population. One of the most commonly used meth- ods is the spatial scan (Kulldorff, 1997). In its gen- eral form, the spatial scan enumerates a large num- ber of potential geographic clusters in a study area in order to determine which is the most likely to have caused the rejection of a null hypothesis of constant risk. This cluster is referred to as a “most- likely cluster”, and is tested for significance using Monte Carlo methods. In most instances, there are a large number of potential clusters; in a study area of 40 locations, there are over one-thousand billion different possible cluster sets. However, geographic clusters usually require some sort of geographic structure to be of interest, so the original spatial scan method is restricted to circularly shaped clus- ters, which also reduces the total number of possible clusters searched. The drawback of this and similar fixed search geometries is that they are less able to detect hot-spots of irregular shape; for example, a Corresponding author: Niko Yiannakoulias School of Geography and Earth Sciences McMaster University, 1280 Main Street West Hamilton Ontario L8S4K1, USA Tel. +1 905 525 9140 ext. 20117; Fax +1 905 546 0463 E-mail: yiannan@mcmaster.ca Geospatial Health 4(2), 2010, pp. 191-200