Abstract
Methods for detecting spatial and spatiotemporal clusters of health and disease have advanced significantly in the past decade. This article reviews recent advances in four areas: spatial search processes, network-based methods, statistical analysis and modelling of local clusters and space-time cluster detection. I then turn to a more critical discussion of the implications of hotspot mapping for public health policy and intervention, highlighting the need to incorporate process-based understandings that impact spatial and social inequalities in ill health for particular health issues in particular geographic contexts.
Spatial clusters, or ‘hotspots,’ of disease and health-related behaviours have long been of interest to public health researchers and policymakers. Defined as an unusual number of cases within a population, place, and time period, a disease hotspot is a geographical construct that can be identified, visualized and explored using GIS and spatial analysis methods. For researchers, hotspots can provide clues to disease aetiology and risk behaviours, suggesting local environmental or social characteristics that promote increased risk. For policymakers and planners, selectively targeting interventions to hotspot areas can be an effective public health intervention strategy. In the past two decades, methods for detecting disease hotspots have advanced significantly to incorporate more robust statistical formulations and spatial search processes that are important for accurately detecting disease hotspots; however, less attention has been paid to the conceptual underpinnings which concern places and processes that are important for disease aetiology and transmission. This article selectively reviews recent developments in spatial and spatiotemporal methods for local hotspot mapping of health-related phenomena and then turns to a more critical discussion of the implications of hotspot mapping for public health policy and intervention.
Interest in analysing and detecting local spatial clusters of disease emerged in the 1970s and 1980s in response to two broad, interlinked social trends: growing awareness of the health impacts of environmental contaminants and increasing citizen concerns about disease hotspots in their neighbourhoods. Public health officials and scientists needed reliable tools for mapping disease risk and determining if risk was elevated in particular areas (Neutra, Swan, and Mack Citation1992). At the same time, developments in geographic information systems provided tools for performing large-scale spatial analytical computations and for layering spatial information on disease incidence with data on diverse social, demographic and environmental phenomena. These intersecting sociopolitical and technological trends fuelled an active, multidisciplinary, area of research on spatial cluster detection that continues to the present.
Methods for detecting local spatial clusters typically consist of two interrelated components: First is a geographical search method used to identify local concentrations of disease cases to be tested for clustering. Approaches to searching include field-based, ‘site-side’ and ‘case-side’ methods that scan the study area at regularly spaced intervals to check for local clusters, and object-based, approaches that search by building clusters around disease cases or areal units containing case concentrations (Cromley and McLafferty Citation2011; Shi Citation2010). Spatial searching also involves decisions about the size and configuration of the search window/filter within which local clusters are detected. In some methods, the window consists of a simple geometric form such as a circle or an ellipse, while other methods utilize a predefined spatial weights matrix to identify the zone in which clusters may be detected Second, hotspot detection methods include a statistical model for determining if the local concentration of disease is unusual – significantly higher than expected based on statistical or other criteria. Both frequentist and Bayesian methods have been developed for cluster testing. This testing component also can include spatial and GIS-based models of individual characteristics and/or local population and environment that might influence where clusters occur. Factors like uneven population distribution, transportation networks and topography can be incorporated in both the geographical search and statistical hotspot detection processes.
Early efforts by geographers and statisticians to devise local disease cluster detection methods resulted in tools like the geographical analysis machine (GAM) (Openshaw, Charlton, and Craft Citation1988), the Besag and Newell method (Besag and Newell Citation1991), Disease Mapping and Analysis Program (DMAP; Rushton and Lolonis Citation1996), the spatial scan statistic (Kulldorf Citation1997), Fotheringham and Zhan’s method (Fotheringham and Zhan Citation1996), and local indicators of spatial autocorrelation such as local Moran’s I and G* (Anselin Citation1995). These methods have been widely applied in health-related research, with applications ranging from cancer survival (Wan et al. Citation2012) to prescription drug abuse (King and Essick Citation2013), and many others. In addition, public health departments have adopted these methods in ongoing disease surveillance programmes. The widespread application of these methods in both the research and policy arenas calls for reflection on recent methodological developments and future directions.
The past decade has seen major advances in spatial cluster detection methods. These advances can be categorized into four broad themes: spatial search; network-based methods; statistical analysis and modelling of local clusters; and space-time cluster detection (). Although classifying the methods this way is useful for presentation purposes, it is important to recognize that the four areas are interrelated and that many recent methods address multiple themes. The next section briefly reviews developments in each of these areas. I then discuss future research needs and policy implications.
Table 1. Selected recent developments in methods for spatial and spatiotemporal cluster detection.
Spatial search processes
Innovations in the spatial search process figure prominently in many new methods for spatial cluster detection. These innovations primarily focus on the search window (spatial filter or kernel) used in identifying local areas in which to test for spatial clusters. Most traditional methods such as the spatial scan statistic, DMAP and GAM, relied on simple geometric forms – circles or ellipses – of uniform size in searching for clusters; however, these forms limit our ability to identify irregularly shaped clusters that reflect uneven environmental and settlement patterns. Some new methods address this problem by modifying the spatial search process. For example, the AMOEBA method (Aldstadt and Getis Citation2006) searches for clusters by sequentially joining areas or disease cases outward from case locations. Tango and Takahashi (Citation2005) used a similar approach in their revision of the spatial scan method to detect irregularly shaped clusters. In a later publication, Tango and Takahashi (Citation2012) modified the search process and statistical model used in their method to make it more computationally feasible. Multiobjective optimization methods have also been used in identifying irregularly shaped clusters by trading off objectives related to cluster shape and local relative risk of disease (Duczmal, Cançado, and Takahashi Citation2008).
Taking this work a step further, Murray, Grubesic, and Wei (Citation2014) develop a method to identify the cluster that maximizes the differences between expected and actual cases in the zones comprising the cluster with no constraints on cluster shape and contiguity. They frame the problem as a constrained maximization problem in which all combinations of contiguous zones that contain at least p cases are tested. Clusters of any size and shape can be detected, freeing the cluster detection process from limitations imposed by spatial structure.
Another innovation in the spatial search process involves using variable-size, adaptive, spatial filters. A major challenge in estimating local disease incidence rates or ratios is that estimates at different locations are based on differing levels of population support: rates calculated for areas that have small populations suffer from the small numbers problem and are less reliable than those for areas with large populations. Hotspot methods that rely on fixed-size search windows typically result in widely different levels of population support, as windows in rural areas contain small populations and those in urban areas very large populations (Shi Citation2010). Some early cluster detection methods addressed this issue by incorporating a minimum population threshold for spatial filters (e.g. Turnbull et al. Citation1990), but these concepts were not incorporated in ‘moving window’ methods like the spatial scan statistic and DMAP. Recent enhancements to these methods, however, allow for variable-sized spatial filters that extend the search radius until a minimum population threshold is reached (Tiwari and Rushton Citation2004). The spatial search window adapts in size based on local population density, with larger windows in low-population density areas. Cai, Rushton, and Bhaduri (Citation2012) discuss how one can determine the optimal population threshold value in order to achieve a constant standard error of observed to expected disease cases. While ensuring adequate levels of population support for estimating disease risks and incidence rates, these adaptive windows result in varying levels of geographic support, implying variations in the sizes and characteristics of local areas in which clusters are detected. Implications of this variability are poorly understood and differ among geographic settings.
Network-based cluster detection
Network-based approaches to cluster detection are also attracting attention. In these methods, the search for disease clusters extends along transportation or other networks that channel people’s locations, movements, and interactions, and distances between places are computed through this network space. One of the earliest examples was Black’s adaptation of Moran’s I to detect spatial clustering of flows along a transport network (Black Citation1992). Recent network cluster detection methods include Yamada and Thill’s adaptation of the local k function to networks (Yamada and Thill Citation2007), and their novel recent method that detects hotspots based on network link characteristics using local Moran’s I (Yamada and Thill Citation2010). The latter has been used to identify clusters of pedestrian traffic accidents (Ha and Thill Citation2011). Similarly, researchers have adapted kernel density estimation (Okabe, Satoh, and Sugihara Citation2009) and the spatial scan statistic (Shiode Citation2011) to network spaces. Recently, Vandenbuklcke et al. (Citation2014) proposed a Bayesian, case-control method, for hotspot detection along networks.
Applying network methods to real-world problems is becoming easier with the expanding array of software tools – notably SANET (Okabe, Okunuki, and Shiode Citation2006) and GeoDANet (Hwang and Winslow Citation2012). However, the methods’ very large computational requirements currently limit the size and scope of applications. In addition, topological features of network spaces present methodological challenges that demand research attention. Although adoption of these methods by health researchers and policymakers is still in its infancy, we are likely to see great increases in use of network-based methods as awareness and training expand and as computational and methodological challenges are addressed.
Statistical analysis and modelling of local clusters
Advances in statistical modelling of local disease risk have also occurred in the past decade. Improvements to well-established methods such as SatScan have been developed (Zhang, Assunção, and Kulldorf Citation2010). However, much effort has focused on innovative Bayesian methods. These methods are widely considered to be more robust than frequentist methods to issues such as the multiple testing, small numbers and change of support problems (Best, Richardson, and Thomson Citation2005). Lawson (Citation2006) proposed a Bayesian method for identifying disease clusters based on case event data. The method explicitly models local disease intensity/clustering, a feature absent from frequentist approaches. Other recent Bayesian methods focus on identifying the locations and spatial extents of disease clusters, an essential element in hotspot mapping. Using a risk model similar to that employed in the spatial scan statistic, Wakefield and Kim (Citation2013) developed a Bayesian model that tests the significance of spatial clustering within zones created through a process of spatial growth and trimming. An important benefit of this approach is that multiple hotspots can be detected simultaneously. Anderson, Lee, and Dean (Citation2014) devised a two-stage Bayesian method in which potential hotspot structures are first identified based on ancillary disease data, then the structures are tested and modified in a step-wise fashion based on data for the disease of interest. Although these and other Bayesian methods greatly strengthen the statistical foundation of cluster detection, they pay less attention to the quality and meaning of the spatial relationships that underpin cluster detection.
Spatial cluster detection methods have also been extended to include multifactorial models of health and disease outcomes. Many hotspot methods can easily incorporate individual-level covariates such as age, gender and socioeconomic or behavioural risk factors, or area-level covariates. In some cases, identification of spatial clusters occurs after these covariates are controlled, so the focus is on residual spatial clustering. Other methods include these factors in modelling the expected number of disease cases within a local area. Some recent efforts focus on accurately modelling the geographical distribution of residential population by downscaling population values for large areas to smaller, more detailed areal units (Cai, Rushton, and Bhaduri Citation2012, Shi Citation2009; Shi et al. Citation2013). Identified clusters often shift with changes in residential population distribution.
Researchers are also developing innovative computational approaches that assess spatial uncertainties stemming from the data, models and methods used in identifying clusters. Typically these approaches involve Monte Carlo simulation of disease outcomes within a geographically varied study region based on an appropriate null hypothesis distribution (Goovaerts and Jacquez Citation2005; Cai, Rushton, and Bhaduri Citation2012). Monte Carlo simulation was used in early methods such as GAM and DMAP, and it is an integral component of most recent cluster detection methods, including Bayesian-, Euclidean-, and network-based approaches. Efforts at modelling spatial uncertainties associated with demographic and environmental data, as in Shi (Citation2009), rely on similar computational approaches. Simulating large numbers of possible disease outcomes in complex environmental settings with varying statistical assumptions requires huge geocomputational resources, creating opportunities to take advantage of recent developments in high-performance computing and cyberGIS (Wang et al. Citation2013).
Space-time cluster detection methods
Methods for detecting clusters in space and time have also advanced in the past decade. Space-time variants of established methods like the spatial scan statistic and kernel estimation have existed for many years, but recent work has taken them a step further by adapting them to a network context (e.g., Nakaya and Yano Citation2010; Shiode and Shiode Citation2013). Others have devised methods for detecting spatiotemporal clusters based on changing disease incidence data for interconnected regional systems (Rogerson and Yamada Citation2004). There are also new Bayesian methods for space-time cluster detection on networks that have been developed for analysing hotspots of crime (Li et al. Citation2014). Although these methods have not yet been applied in identifying disease clusters, they have clear applicability in the disease/health context. The vast majority of these methods involve simply adding time as a third dimension to 2-D spatial approaches, without fully considering the movements of people through space and time that affect disease risk. In contrast, novel new approaches view space and time as fully integrated, and assess clustering among dynamic space-time trajectories. A good example is the method of Q-statistics, developed by Jacquez and others (Jacquez et al. Citation2006), that searches for clusters of disease cases compared to controls in space and time based on residential history data. Recent studies have used Q-statistics in identifying space-time clusters of breast cancer in Denmark (Nordsberg et al. Citation2014), for example. These trajectory-based methods offer great possibilities for health research, but their application requires detailed data on people’s migration trajectories that is not widely available in countries like the United States.
Another innovative space-time method works with longitudinal data on health outcomes and residential locations over time (Cook, Gold, and Li Citation2009). By analysing cumulative geographic residuals – higher-than-expected disease outcomes at various residential locations during a person’s life – the method pinpoints locations and times of significant clusters for repeated outcomes such as asthma. As with the Q method, data requirements are high for this innovative approach.
Discussion
Despite significant advances in detecting and modelling disease hotspots in the past two decades, many methods fall short in incorporating process-based understandings that are important in assessing spatial and spatiotemporal patterns of ill health. Although identifying hotspots in the absence of these understandings may be useful for policymakers in determining where spatially to target public health interventions, planning the characteristics of such interventions requires understanding why disease risk is elevated in the target area. We need to view hotspots not as well-defined spatial or spatiotemporal objects, but as fuzzy objects, with permeable boundaries, that change depending on how they are defined. Incorporating process-based understandings involves thinking about the social, biological and environmental processes that may impact disease risk within local areas.
Everyday mobility is an important element in disease processes that has not been well incorporated in spatial cluster detection methods. Kwan (Citation2012) emphasizes the importance of mobility, arguing that the geographic contexts that are relevant for health and well-being are dynamic, changing through space and time as people move through their daily activity spaces. Most spatial cluster detection methods rely on health data geocoded to residential locations, ignoring everyday spaces that may be important for environmental exposure. Common sites of exposure like schools, workplaces and risky spaces can only be detected by analysing activity space information, and as Yiannakoulias (Citation2011) argues, these places are important sites for public health intervention. To identify these places, space-time methods such as the Q method can be adapted and implemented using data on individuals’ daily activity spaces and mobility trajectories. Another approach involves using hierarchical clustering methods such as the one developed by Chen et al. (Citation2011) which search for spatial clustering and similarity within individual space-time activity data.
Process-based cluster detection also entails modelling of environmental factors that influence and constrain disease risk. Environmental factors and processes range from air pollution plumes to hazardous facilities to transportation networks to places of social support and interaction. Many cluster detection methods ignore the environmental setting in which clusters are being analysed and assume that unmodelled environmental and population factors account for observed clustering. However, this is a naïve approach. Developments in network-based cluster detection show how basic changes in the environmental template for spatial cluster detection affect the locations, sizes and shapes of hotspots identified. Similarly, efforts by Shi (Citation2010) and Shi et al. (Citation2013) and others reveal that a more accurate spatial representation of residential population results in improved detection of spatial disease clusters. These kinds of developments pave the way for a stronger geographical foundation for spatial disease hotspot detection.
Novel methods that trace back from spatial cluster information to deduce spatial or environmental sources of disease also present exciting opportunities. A recent article by Verity et al. (Citation2014) uses geographic profiling – a method for identifying locations of serial criminals based on observed crime locations – to identify source areas for malaria infection. The authors propose a multi-step process that uses a Dirichlet process mixture (Bayesian) model to characterize cluster formation and estimate probabilities associated with particular source locations. Of course in the case of malaria, the disease transmission process is well understood. This makes it possible to develop a well-specified model and incorporate appropriate environmental data on, for example, water sources and mosquito flight range. However, the general approach may be useful for many health concerns as a way of working backwards from clusters to identify relevant environmental exposures and places of high risk.
In incorporating environmental factors, it makes sense to adopt a step-wise process similar to that used in incorporating individual-level covariates in cluster detection. After adding environmental factors, one can observe changes in the locations, sizes and spatial extents of identified clusters which suggest whether the environmental factor has an impact on local disease clustering. Policymakers can use this information in creating environmental and place-based strategies for health intervention.
In summary, spatial cluster detection methods have advanced significantly in the past decade, fuelled by improvements in spatial search processes and developments in statistical and computation methods, network-based approaches, and space-time analysis. The next step is to incorporate more process-based understandings that are specific to particular health concerns in particular geographic contexts. Integrating environmental, population and mobility data more fully with disease cluster detection will greatly enhance the utility of the methods for public health planning and intervention.
Disclosure statement
No potential conflict of interest was reported by the author.
References
- Aldstadt, J., and A. Getis. 2006. “Using AMOEBA to Create a Spatial Weights Matrix and Identify Spatial Clusters.” Geographical Analysis 38: 327–343. doi:10.1111/j.1538-4632.2006.00689.x.
- Anderson, C., D. Lee, and N. Dean. 2014. “Identifying Clusters in Bayesian Disease Mapping.” Biostatistics. doi:10.1093/biostatistics/kxu005.
- Anselin, L. 1995. “Local Indicators of Spatial Association-LISA.” Geographical Analysis 27 (2): 93–115. doi:10.1111/j.1538-4632.1995.tb00338.x.
- Besag, J., and J. Newell. 1991. “The Detection of Clusters in Rare Diseases.” Journal of the Royal Statistical Society, Series A 154 (1): 143–155. doi:10.2307/2982708.
- Best, N., S. Richardson, and A. Thomson. 2005. “A Comparison of Bayesian Spatial Models for Disease Mapping.” Statistical Methods in Medical Research 14: 35–59. doi:10.1191/0962280205sm388oa.
- Black, W. R. 1992. “Network Autocorrelation in Transport Network and Flow Systems.” Geographical Analysis 24: 207–222. doi:10.1111/j.1538-4632.1992.tb00262.x.
- Cai, Q., G. Rushton, and B. Bhaduri. 2012. “Validation Tests of an Improved Kernel Density Estimation Method for Identifying Disease Clusters.” Journal of Geographical Systems 14: 243–264. doi:10.1007/s10109-010-0146-0.
- Chen, J., S.-L. Shaw, H. Yu, F. Lu, Y. Chai, and Q. Jia. 2011. “Exploratory Data Analysis of Activity Diary Data: A Space–Time GIS Approach.” Journal of Transport Geography 19: 394–404. doi:10.1016/j.jtrangeo.2010.11.002.
- Cook, A. J., D. R. Gold, and Y. Li. 2009. “Spatial Cluster Detection for Repeatedly Measured Outcomes while Accounting for Residential History.” Biometrical Journal 5: 801–818. doi:10.1002/bimj.200800269.
- Cromley, E., and S. McLafferty. 2011. GIS and Public Health. 2nd ed. New York, NY: Guilford.
- Duczmal, L., A. Cançado, and R. Takahashi. 2008. “Delineation of Irregularly Shaped Disease Clusters through Multiobjective Optimization.” Journal of Computational and Graphical Statistics 17: 243–262. doi:10.1198/106186008X287517.
- Fotheringham, A. S., and F. B. Zhan. 1996. “A Comparison of Three Exploratory Methods for Cluster Detection in Spatial Point Patterns.” Geographical Analysis 28: 200–218. doi:10.1111/j.1538-4632.1996.tb00931.x.
- Goovaerts, P., and G. Jacquez. 2005. “Detection of Temporal Changes in the Spatial Distribution of Cancer Rates Using Local Moran’s I and Geostatistically Simulated Spatial Neutral Models.” Journal of Geographical Systems 7: 137–159. doi:10.1007/s10109-005-0154-7.
- Ha, H., and J.-C. Thill. 2011. “Analysis of Traffic Hazard Intensity: A Spatial Epidemiology Case Study of Urban Pedestrians.” Computers, Environment and Urban Systems 35: 230–240. doi:10.1016/j.compenvurbsys.2010.12.004.
- Hwang, M., and A. Winslow. 2012. “User manual for GeoDaNet: spatial analysis on undirected networks.” In Working paper, GeoDa Center for Geospatial Analysis and Computation Arizona State University, March 22.
- Jacquez, G. M., J. R. Meliker, G. A. Avruskin, P. Goovaerts, A. Kaufmann, M. L. Wilson, and J. Nriagu. 2006. “Case–Control Geographic Clustering for Residential Histories Accounting for Risk Factors and Covariates.” International Journal of Health Geographics 5: 32. doi:10.1186/1476-072X-5-32.
- King, M., and C. Essick. 2013. “The Geography of Antidepressant, Antipsychotic, and Stimulant Utilization in the United States.” Health & Place 20: 32–38. doi:10.1016/j.healthplace.2012.11.007.
- Kulldorf, M. 1997. “A Spatial Scan Statistic.” Communications in Statistics - Theory and Methods 26 (6): 1481–1496. doi:10.1080/03610929708831995.
- Kwan, M.-P. 2012. “The Uncertain Geographic Context Problem.” Annals of the Association of American Geographers 102: 958–968. doi:10.1080/00045608.2012.687349.
- Lawson, A. 2006. “Disease Cluster Detection: A Critique and a Bayesian Proposal.” Statistics in Medicine 25: 897–916. doi:10.1002/sim.2417.
- Li, G., R. Haining, S. Richardson, and N. Best. 2014. “Space–Time Variability in Burglary Risk: A Bayesian Spatio-Temporal Modelling Approach.” Spatial Statistics 9: 180–191. doi:10.1016/j.spasta.2014.03.006.
- Murray, A. T., T. Grubesic, and R. Wei. 2014. “Spatially Significant Cluster Detection.” Spatial Statistics 10: 103–116. doi:10.1016/j.spasta.2014.03.001.
- Nakaya, T., and K. Yano. 2010. “Visualising Crime Clusters in a Space-time Cube: An Exploratory Data-analysis Approach Using Space-time Kernel Density Estimation and Scan Statistics.” Transactions in GIS 14: 223–239. doi:10.1111/j.1467-9671.2010.01194.x.
- Neutra, R., S. Swan, and T. Mack. 1992. “Clusters Galore: Insights about Environmental Clusters from Probability Theory.” The Science of the Total Environment 127 (1–2): 187–200. doi:10.1016/0048-9697(92)90477-A.
- Nordsberg, R. B., J. R. Meliker, A. K. Ersboll, G. M. Jacquez, A. H. Poulsen, and O. Raaschou-Nielsen. 2014. “Space-time Clusters of Breast Cancer Using Residential Histories: A Danish Case–Control Study.” BMC Cancer 14: 255. http://www.biomedcentral.com/1471-2407/14/255
- Okabe, A., K. Okunuki, and S. Shiode. 2006. “SANET: a Toolbox for Spatial Analysis on a Network.” Geographical Analysis 38: 57–66. doi:10.1111/j.0016-7363.2005.00674.x.
- Okabe, A., T. Satoh, and K. Sugihara. 2009. “A Kernel Density Estimation Method for Networks, Its Computational Method and a Gis-based Tool.” International Journal of Geographical Information Science 23 (1): 7–32. doi:10.1080/13658810802475491.
- Openshaw, S., M. Charlton, and A. Craft. 1988. “Searching for Leukaemia Clusters using a Geographical Analysis Machine.” Papers in Regional Science 64: 95–106. doi:10.1111/j.1435-5597.1988.tb01117.x.
- Rogerson, P., and I. Yamada. 2004. “Monitoring Change in Spatial Patterns of Disease: Comparing Univariate and Multivariate Cumulative Sum Approaches.” Statistics in Medicine 23: 2195–2214. doi:10.1002/sim.1806.
- Rushton, G., and P. Lolonis. 1996. “Exploratory Spatial Analysis of Birth Defect Rates in an Urban Population.” Statistics in Medicine 15 (7–9): 717–726. doi:10.1002/(SICI)1097-0258(19960415)15:7/9<717::AID-SIM243>3.0.CO;2-0.
- Shi, X. 2009. “A Geocomputational Process for Characterizing the Spatial Pattern of Lung Cancer Incidence in New Hampshire.” Annals of the Association of American Geographers 99: 521–533. doi:10.1080/00045600902931801.
- Shi, X. 2010. “Selection of Bandwidth Type and Adjustment Side in Kernel Density Estimation over Inhomogeneous Backgrounds.” International Journal of Geographical Information Science 24: 643–660. doi:10.1080/13658810902950625.
- Shi, X., S. D. Miller, K. M. Mwenda, A. Onda, J. R. Rees, T. L. Onega, J. Gui, M. R. Karagas, E. Demidenko, and J. B. Moeschler. 2013. “Mapping Disease at an Approximated Individual Level using Aggregate Data: A Case Study of Mapping New Hampshire Birth Defects.” International Journal of Environmental Research and Public Health 10: 4161–4174. doi:10.3390/ijerph10094161.
- Shiode, S. 2011. “Street-level Spatial Scan Statistic and STAC for Analysing Street Crime Concentrations.” Transactions in GIS 15 (3): 365–383. doi:10.1111/j.1467-9671.2011.01255.x.
- Shiode, S., and N. Shiode. 2013. “Network-Based Space-Time Search-Window Technique for Hotspot Detection of Street-Level Crime Incidents.” International Journal of Geographical Information Science 27: 866–882. doi:10.1080/13658816.2012.724175.
- Tango, T., and K. Takahashi. 2005. “A Flexibly Shaped Spatial Scan Statistic for Detecting Clusters.” International Journal of Health Geographics 4: 11. doi:10.1186/1476-072X-4-11.
- Tango, T., and K. Takahashi. 2012. “A Flexible Spatial Scan Statistic with A Restricted Likelihood Ratio for Detecting Disease Clusters.” Statistics in Medicine 31: 4207–4218. doi:10.1002/sim.5478.
- Tiwari, C., and G. Rushton. 2004. “Using Spatial Adaptive Filters to Map Late-Stage Colorectal Cancer Incidence in Iowa.” In Advances in Spatial Data Handling II, edited by P. Fisher, 125–136. Berlin: Springer.
- Turnbull, B., E. Iwano, W. Burnett, H. L. Howe, and L. Clark. 1990. “Monitoring for Clusters of Disease: Application to Leukaemia Incidence in Upstate New York.” American Journal of Epidemiology 132: S136–S143.
- Vandenbulcke, G., I. Thomas, and L. Panis 2014. “Predicting Cycling Accident Risk in Brussels: A Spatial Case-Control Approach.” Accident Analysis and Prevention 62: 341–357. doi:10.1016/j.aap.2013.07.001.
- Verity, R., M. Stevenson, D. K. Rossmo, R. Nichols, and S. LeComber. 2014. “Spatial Targeting of Infectious Disease Control: Identifying Multiple, Unknown Sources.” Methods in Ecology and Evolution 5: 647–655. doi:10.1111/2041-210X.12190.
- Wakefield, J., and A. Kim. 2013. “A Bayesian Model for Cluster Detection.” Biostatistics 14: 752–765. doi:10.1093/biostatistics/kxt001.
- Wan, N., F. B. Zhan, Y. Lu, and J. Tiefenbacher. 2012. “Access to Healthcare and Disparities in Colorectal Cancer Survival in Texas.” Health and Place 18: 321–329. doi:10.1016/j.healthplace.2011.10.007.
- Wang, S., L. Anselin, B. Bhaduri, C. Crosby, M. Goodchild, Y. Liu, and T. Nyerges. 2013. “Cybergis Software: A Synthetic Review and Integration Roadmap.” International Journal of Geographical Information Science 27: 2122–2145. doi:10.1080/13658816.2013.776049.
- Yamada, I., and J.-C. Thill. 2007. “Local Indicators of Network-Constrained Clusters in Spatial Point Patterns.” Geographical Analysis 39: 268–292. doi:10.1111/j.1538-4632.2007.00704.x.
- Yamada, I., and J.-C. Thill. 2010. “Local Indicators of Network-Constrained Clusters in Spatial Patterns Represented by a Link Attribute.” Annals of the Association of American Geographers 100 (2): 269–285. doi:10.1080/00045600903550337.
- Yiannakoulias, N. 2011. “Spatial Aberration vs. Geographical Substance: Representing Place in Public Health Surveillance.” Health & Place 17: 1242–1248. doi:10.1016/j.healthplace.2011.07.009.
- Zhang, Z., R. Assunção, and M. Kulldorf. 2010. “Spatial Scan Statistics Adjusted for Multiple Clusters”. Journal of Probability and Statistics 2010: 1–11. doi:10.1155/2010/642379.