The purpose of this paper is to address the urgent need for guiding the construction of information systems for disease surveillance and early warning, given the latest efforts of the report system of public health information over China.
A system framework for disease surveillance and early warning, based on disease clustering test and cluster detection techniques, geographical information system, network and communication is conceived. Through geographical surveillance analysis of severe acute respiratory syndrome occurring in Beijing in 2003, an application example of the framework is illustrated.
Through approaches such as integrating spatial‐time clustering test and cluster detection algorithms, spatial visualization, computer network, wireless communication, it is feasible to build a systematic, automatic, real‐time surveillance and early warning system for prevention and control of disease.
The present study provides an underlying framework for the development of disease surveillance and early warning system enabling data acquisition, data analysis and alarm publishing.
The framework integrates report system of public health information, GIS and disease clustering test and cluster detection techniques into an application, which will significantly enhance the resilience of healthcare facilities. It is supposed to be implemented in near future and provides fundamental support for nation‐wide disease surveillance and early warning.
Achour, N., Price, A.D.F., Zhong, S., Sun, Z., Huang, Q. and Cao, C. (2011), "A framework for geographical surveillance of disease in China", International Journal of Disaster Resilience in the Built Environment, Vol. 2 No. 3, pp. 256-267. https://doi.org/10.1108/17595901111167123Download as .RIS
Emerald Group Publishing Limited
Copyright © 2011, Emerald Group Publishing Limited
Owing to environmental deterioration and ecological degeneration, some emerging and re‐emerging diseases have frequently attacked us in recent decades. Besides, after some major natural disasters such as flood and earthquake, the affected people are vulnerable to some malignant epidemics. Surveillance and early warning are two important approaches to disease prevention and control, which continuously collect disease cases and related data through various techniques of data acquisition and analysis and determine whether and how an alarm should be signalled.
There are three phases for the process of surveillance and early warning of disease: data acquisition and transmission, data processing and analysis and alarm representation and signalling:
Data acquisition and transmission. Data acquisition and transmission is focused on continuous collection and transmission of disease occurrences and related information including environmental, economic and social data. This phase relies on systematic construction and normal operation of disease monitoring network and communication infrastructure. To efficiently gather data, the monitoring network must be planned and designed scientifically. Monitoring network can be constituted with several kinds of nodes such as medical agency, hospital, social organization and news media.
Data processing and analysis. The raw data acquired from the monitoring network are probably with some problems such as errors, missing items and non‐uniform format which need to be handled before further analysis. Then effective analysis methods will be employed to answer some questions of the underlying situation of disease. For example, whether there a clustering of disease cases or not, where the clusters of disease in space, time, or space and time, and whether the clustering or clusters are significantly higher than normal level or not.
Alarm representation and signalling. The essential factors of the early warning message to be published include disease situation, alert receivers, publishing extent, prevention proposal, starting time and end time, etc. With multimedia and some visualization techniques, these factors can be presented properly so that the public can easily identify and understand.
Around the three phases of surveillance and early warning of disease, some related groundwork and applications in China are presented and further development aspects of them are discussed. Then a system framework for surveillance and early warning is proposed. Next, a case study of Beijing severe acute respiratory syndrome (SARS) is explored which illustrates the promising application of the framework with a specific cluster detection technique (space‐time permutation scan statistic) for disease surveillance and early warning.
2 Preliminary discussion
2.1 Public health information network
Up to now, a public health information network (PHIN) (Liu et al., 2008) over the whole country has been built up in China and thereby disease cases can be collected rapidly and accurately once happening. In case of a catastrophe, some temporary surveillance sites can be set up rapidly and real‐time disease event information can be acquired.
National PHI comes from various sources, including hospitals, pharmacies, clinics, medical laboratories and so on. The information is not directly received from the bottom health institutions but aggregated level by level, through a hierarchical report system setup as shown in Figure 1.
2.2 Geographical surveillance of disease
Alternative techniques and tools for data processing include but not limited to statistical methods, error processing theory and geographical information system (GIS). Temporal and space‐time clustering test and cluster detection techniques are the core approaches to disease data analysis. With these methods, the potential clustering or clusters can be identified and then whether or not an alarm should be published may be determined given a pre‐defined threshold.
Tobler (1970) presented the idea that, in 1954, “everything is related to everything else, but near things are more related to each other”, which is referred to as the “first law of geography”. In fact, prior to Tobler, an associated concept spatial autocorrelation has been coined. Spatial autocorrelation is caused by spatial dependency which, like temporal autocorrelation, violates classical statistical techniques that assume independence among observations. To test or detect spatial dependency, some indexes measuring the spatial autocorrelation are proposed. These indexes are observation data‐based statistics and can be classified into two types: global clustering test and local cluster detection. The former is to infer whether there exists clustering and the latter is to infer where the most likely cluster(s).
As one of the spatial data, disease data are inherently spatially dependent. For example, the occurrence of an endemic disease is correlated with a certain environment which is subject to the first of geography. Point data and aggregated data are the two main forms of disease data (Pfeiffer, 2008). There are some data‐specific methods for measuring spatial autocorrelation. Autocorrelation statistics for aggregated data provide an estimate of the degree of spatial similarity observed among neighbouring values of an attribute over a study area. There are various autocorrelation statistics for aggregated data such as global Geary's c (Geary, 1954), global Moran's I (Moran, 1950), general G(d) (Getis and Ord, 1992), local Moran's Ii (Anselin, 1995), local Gi(d) and local Gi*(d) (Ord and Getis, 1995) and so on. For point data, some alternative methods include nearest neighbour analysis (Cuzick and Edwards, 1990), Ripley's (1976, 1977) K‐function and Local K‐function (Getis, 1992).
In essence, these indexes are some quantitative forms of the first law of geography. They can be used in epidemiology to determine whether global clustering or local clusters in space exist in disease data. Nonetheless, these general methods have some weakness for disease data analysis since they ignore some specific consideration about the epidemiological characteristics of diseases such as background population distribution heterogeneity and population interaction. Furthermore, the inherent temporal feature in disease analysis requires us to explore new methods which can measure the autocorrelation in space and time simultaneously. Thus, some disease specific spatial autocorrelation analysis methods are required for better disease data analysis.
While spatial patterns of disease are of great interest, space‐time interactions are also important, particularly so when trying to determine whether a disease is infectious. In such instances, it is necessary to evaluate whether cases that are close in space are also close in time and vice versa.
After going through several decades of research work, some algorithms used for detecting spatial, temporal or space‐time clustering and clusters have been developed. For instances, Knox proposed a statistical technique for testing space‐time interaction (Knox, 1964). Tango's (1984) index is a test for temporal clustering and Tango's (2008) spatial scan statistic can be used to detect the most likely cluster in space. Kulldorff developed some scan statistic methods for identifying space/time clusters (Kulldorff, 1997, 2001; Kulldorff et al., 1997, 1998, 2005, 2006). And Rogerson proposed CUSUM algorithms suitable for ongoing surveillance of disease (Rogerson, 1997, 2001; Rogerson and Yamada, 2004a, b). In general, these methods may be classified into two categories: retrospective and prospective. Prospective methods are appropriate for real‐time surveillance of some space‐time events (disease and crime are two typical examples). Among the prospective methods, Kulldorff's space‐time scan statistic is one of the most promising methods. Furthermore, some researchers systematically compare the power of some disease clustering test or cluster detection methods with each another (Kulldorff et al., 2003; Song and Kulldorff, 2003; Asmodt et al., 2006). Nevertheless, it is imperative to study efficient and effective methods and integrated use of them, with which the outbreak and prevalence of disease can be confirmed definitely and opportunely.
2.3 Decision support system
Generally speaking, an early warning should be conveyed to the public as early as possible. However, some studies have argued that literally publishing disease information could not always be the best manners (Gong and Xiao, 2007). In this case, clarity in the broadcasting of news is essential. Sometimes when journalism access to information they are not in condition to digest, a sentiment of panic emerges because of serious problems in the dissemination of news. So, the strategies and manners of signalling of an alarm are important. Publishing strategy in society should be analysed and decided to avoid the general panic in case of virus outbreaks.
To help make publishing strategy and propose scientific prevention action to the public, a decision support system (DSS) is required which can assist in decision makers to work out optimal or feasible solutions through providing necessary information and proposal.
DSS is a kind of expert system built on model base, knowledge base and case base. There are three core tasks in constructing a DSS of disease publishing strategy, i.e. the implementation of model for news publishing sensitivity analysis, the design and building of the rule set of publishing and the typical cases of historical solutions. Among the three tasks, model is the most challenging one where some current work is done mainly around theoretical studies.
2.4 Signalling of disease alarm
Currently, internet, wireless communication and broadcasting and television networks are three most frequently used approaches to information dissemination. Owing to the rapid advance in integration of the three networks in China, alarm signalling for disease outbreaks can be made better efforts in future. Some emerging techniques can also aid the signalling procedures of a disease alarm. Among them Web GIS is the most popular one, which provides high performance map service on the internet. With Web GIS, the publishing information of disease can be shown in text, chart, table, map and animation, etc. these easy to understand forms can provide more efficient convey of the published information. Google Maps and Bing Maps are among the most well‐known Web GIS applications for the public. During Wenchuan earthquake of China (May 12, 2008) and Fukuoka earthquake of Japan (March 11, 2011), Google Maps provide strong map support and news publishing for disaster rescue and recovery. What is more, as a kind of information system capable of capturing, editing, storing, integrating, managing, analysing, sharing, and displaying geographically referenced information, in the study of epidemics, GIS can be used as an appropriate tool for processing, analysis, representation of epidemic data. One can collect and locate disease data and relevant geographical, climatic, socio‐economic information, detect the outbreak of a disease, and investigate the relationship between epidemic occurrences and all kinds of potential factors that may affect the occurrence of the disease.
After health information is aggregated to the national centre for disease control and prevention (CDC), some methods are required to detect the potential cluster hidden in these disease data. The core approaches to this work are some geographical surveillance methods for disease data including spatial, temporal and space‐time clustering test and cluster detection techniques. Surveillance and early warning are logically interrelated processes and some other advanced techniques can be used to construct an information system for support of the core activities. These techniques include GIS, computer network, wireless communication, broadcast network, etc. as well as the cluster detection algorithms. To provide extra support of spatial data manipulation, some techniques from GIS such as layer overlay, projection, visualization and data mining are necessary for the preparation and processing of input data and presentation of output results of the cluster algorithms.
In Figure 2, a system framework is conceived with those techniques. The framework is composed of three tiers: data foundation, methodological support and applications. The data foundation tier is responsible for data store and data access through spatial data engine (for spatial data) and data access object (for attribute data). Methodological support tier implements core algorithms for detecting space‐time clustering, and the applications tier provides business support for end‐users. Moreover, it has a connection with the report system of PHI from which the disease data are extracted with some techniques of enterprise application integration such as web service, simple object access protocol, eXtensible Markup Language and message middleware.
From Figure 2, computer network mainly is used to connect the report system of PHI and continuously collect the disease occurrences and related data which come from all kinds of sources including temporary monitoring station, hospital, medical agency, etc. GIS provides fundamental support for the implementation of the space/time clustering and cluster algorithms. This is due to the involvement of algorithms in the processing, storage, analysis and presentation of geographically referenced data (including disease outbreaks and related environmental data). GIS is the optimal tool for these tasks. Furthermore, GIS is also able to aid the signalling process of a disease alarm, which can conveniently mark the epicentre, draw the area of influence and count the population. In collaboration with broadcasting and television network and wireless communication network, these functions can greatly enhance the operations of signalling an alarm.
4 Case study
Kulldorff et al. (1997) developed space scan statistic that construct a series of circles with different radius centred at a specified location. The radius continuously increases starting at zero until the population within the circle reaches the pre‐defined proportion. A hypothesis test is taken for every circle. The null hypothesis means the risk inside the circle has not significant difference from that outside the circle, whereas the alternative hypothesis says the difference is significant.
The space scan statistic is calculated as follows: Equation 1 where Zc denotes all circles but Z itself. O and p are the observed case number and population number, respectively. I is an indicator function.
Through including time dimension, the space scan statistic can be extended into a space‐time scan statistic which can detect the clusters in space and time dimensions simultaneously. Kulldorff et al. (1998) proposed a space‐time scan statistic that can be used to detect space‐time clusters with replacing the circle with cylinder whose height denotes the time interval of a cluster. The emerging clusters (include their location and size) can be identified by periodically carrying out the space‐time scan statistic. Based on his space‐time scan statistic, Kulldorff et al. (2005) further developed space‐time permutation scan statistic. The improved algorithm does not need the risk population distribution any more, and may carry out the detection process with only the case data.
In a method for detecting space‐time clusters, the time frame parameter is a quantity that determines the scan period. The height of the cylinder (the time size of a cluster) will vary in the period. For example, with a time frame of two weeks, the starting and end time is taken values in (two weeks before, now), and the period between starting time and end time is not greater than 14 days. The time frame parameter determines the sensitivity to detect a cluster. The larger (smaller) this value is, the less (more) frequent the alarm is signalled and with less (more) false alarms.
4.2 Materials and tools
The disease data used for analysis illustration are from SARS in Beijing. SARS is a new infectious disease first reported in November 2002 in the Guangdong Province of China (WHO, 2010). In Beijing, the first SARS case was confirmed on March 5, 2003. From then on, SARS spread rapidly over the city. The disease data are aggregated by their reported hospitals due to the unavailability of individual residence. So the following analyses are all based on hospital location in space dimension. Figure 3(a) is a distribution map of hospitals where SARS cases are admitted during March 5‐May 29, 2003 and Figure 3(b) shows the change curve of the new cases by day in the period.
Based on their space‐time scan statistics, Kulldorff (2005) has developed a software program called SaTScan, which is free software that analyses spatial, temporal and space‐time data using the spatial, temporal, or space‐time scan statistics. It is designed for any of the following interrelated purposes:
perform geographical surveillance of disease, to detect spatial or space‐time disease clusters, and to see if they are statistically significant;
test whether a disease is randomly distributed over space, over time or over space and time;
evaluate the statistical significance of disease cluster alarms; and
perform repeated time‐periodic disease surveillance for early detection of disease outbreaks.
4.3 Results and analyses
With the space‐time permutation statistic implemented in SaTScan, the most likely clusters, which are detected with time frame parameters one week and two weeks, respectively, are shown as Figure 4. With the results, the alarms of the clusters that have quantitative time and space size will be signalled on May 29, 2003.
Figure 4 shows the time interval of the most likely two clusters (the circle with dash line and the circle with solid line) are May 16‐29 and May 24‐29, 2003. However, from Figure 3(b), the most likely cluster should be around April 25, 2003 when the outbreak peak occurs. The reason for the difference is because the most likely clusters in both of space and time are got with the space‐time permutation statistic while the peak shown in Figure 3(b) only appears when exploring the disease data in pure time. In fact, a disease cluster is identified only when plenty of disease cases are observed in small space and short time simultaneously.
The result is not confident if we use only one method. After identifying the most likely clusters, the result is validated by the CUSUM algorithm (Rogerson, 1997, 2001). In this case, the two methods give out similar results. Thus, the clusters can be held and the alarm can be prepared for publishing.
As mentioned above, a simulation analysis should be done to understand the effect of the news publishing on the public. If a panic can come together with the publishing and worse situation will appear in society, not publishing the disease news or making a strategy for publishing should be considered alternatively. For SARS case, China Government made a fairly successful crisis response. Especially in the later period, the timely and accurate publishing of disease data is of great importance for the whole country to defeat SARS.
4.4 Resilience implications
Within the last decade many major hazards took place causing significant damage to the healthcare service and creating the ideal environment for disease outbreaks. Outbreaks are also expected to increase due to the effect of climate change (Science Daily, 2009) and could “be one of the earliest biological expressions of climate instability” (Epstein, 2002). Consequently, there is an urgent need to develop tools able to identify disease outbreaks at an early stage in order to mitigate their impact on society and economy.
Following the 2010 Haiti Earthquake, an outbreak of cholera was confirmed on October 21, 2010 (282 days after the country is hit by the earthquake). One month later, the Haitian Ministry of Public Health and Population reported 60,240 cumulative cholera cases including 1,415 deaths (WHO, 2010). The risk of cholera could perhaps have been reduced if more hygienic environment, clean supplies (e.g. water) and appropriate medical care were available; however, this could be limited by the economic pressure that the country was going through, at the time, which brings the issue of more efficient management to the front. Efficiency was also brought to attention during the Beijing SARS when over 5,000 cases were reported. Substantial health resources (e.g. doctors, nurses and medicines) were allocated, in designated hospitals and clinics, to guarantee remedial containment of SARS which resulted in a significant resources waste (e.g. human power, wards and medicine) that could have been saved. The proposed framework provides an excellent tool to manage healthcare resources more effectively and efficiently by providing incipient indication of endemic or pandemic outbreaks. The ideal implementation of the framework is through building a network comprised of temporary and permanent stations setup in health and care facilities, such as hospitals and clinics, to monitor diseases and report to local and national health institutions continuously to be analysed in real time. This will perceive potential risks at an early stage and thus reduce cost of health resources and recover their normal operations as well as decrease any potential losses.
A more comprehensive resilience strategy, with the potential for implementation, is therefore needed to respond to high risks resulting from climate change and other hazards. Researchers such as Achour and Price (2010) have developed a climate change adaptation model for sustainability and resilience. This framework could be integrated with this model to develop a more comprehensive strategic model that is able to respond to the increasing risks of disease outbreaks.
5 Conclusions and prospects
Surveillance and early warning is of great importance in disease prevention and control. With advances in spatial, temporal and space‐time clustering test and cluster detection techniques as well as disease monitoring, computer network, wireless communication and software development. It is promising to construct an information system that enables the integration of disease data acquisition and transmission, processing and analysis and presentation and alarm signaling.
In this paper, in collaboration with advanced information techniques, a system framework based on the hierarchical reporting system of PHI that is supposed to support the surveillance and early warning of disease is conceived. We also explored the Kulldorff's space‐time scan statistic and illustrated their application in Beijing SARS.
In China, the PHIN has been built up with which daily disease data can be reported to the national CDC level by level. This lays a solid foundation for disease surveillance and early warning over the country. Currently China is boosting the aggregation of the three networks (internet, broadcasting and television network and wireless communication network). This will greatly facilitate the publication of disease alarm.
Going through several decades, disease spatial, temporal and space‐time clustering test and cluster detection techniques has achieved marvelous efforts. These algorithms are the core of analysis for disease monitoring data. Several promising algorithms have been proposed by professionals who have devoted themselves to the study of spatial epidemiology. Nevertheless, existing methods have their merits and shortcomings. Therefore, it is also imperative to study more efficient and effective methods and integrated use of them, with which the outbreak and prevalence of disease can be confirmed definitely and opportunely.
Data quality is a key for correctly detection of the potential clusters. Consequently, the monitoring network of disease need to have a complete coverage over the target area and can work stably and efficiently. Furthermore, the current methods of cluster detection still have some disadvantages and methodological research is still urgent for better results. Except for these technical aspects, some management factors including the standard and rule of disease report, education and training also need solving for implementation and effective use of the system framework.
With some groundwork being studied in this paper, in near future plan, the framework is going to be put into practice. And the implementation of the framework, as complex information system engineering, will be focused on some issues of software development, hardware integration, and interface between software and hardware.
The integration of the framework with resilience and sustainability strategies will also be targeted in order to develop a universal model to improve the resilience, sustainability and disaster prevention techniques (e.g. this framework).
Shaobo Zhong and can be contacted at: firstname.lastname@example.org
Aamodt, G., Samuelsen, S.O. and Skrondal, A. (2006), “A simulation study of three methods for detecting disease clusters”, International Journal of Health Geographics, available at: http://ij‐healthgeographics.com/content/5/1/15 (accessed 21 June 2011).
Achour, N. and Price, A.D.F. (2010), “Resilience strategies of healthcare facilities: present and future”, International Journal of Disaster Resilience in the Built Environment, Vol. 1 No. 3, pp. 264‐76.
Anselin, L. (1995), “The local indicators of spatial association‐LISA”, Geographical Analysis, Vol. 27, pp. 93‐115.
Cuzick, J. and Edwards, R. (1990), “Spatial clustering for inhomogeneous populations (with discussion)”, Journal of the Royal Statistical Society Series B, Vol. 52, pp. 73‐104.
Epstein, P.R. (2002), “Climate change and infectious disease: stormy weather ahead?”, Epidemiology, Vol. 13 No. 4, pp. 373‐5.
Geary, R.C. (1954), “The contiguity ratio and statistical mapping”, The Incorporated Statistician, Vol. 5 No. 3, pp. 115‐45.
Getis, A. and Ord, J.K. (1992), “The analysis of spatial association by use of distance statistics”, Geographical Analysis, Vol. 24, pp. 189‐206.
Gong, X. and Xiao, R. (2007), “Research on multi‐agent simulation of epidemic news spread characteristics”, Journal of Artificial Societies and Social Simulation, available at: http://jasss.soc.surrey.ac.uk/10/3/1.html (accessed 30 March 2011).
Knox, E.G. (1964), “The detection of space‐time interaction”, Applied Statistics, Vol. 13, pp. 25‐9.
Kulldorff, M. (1997), “A spatial scan statistic”, Communications in Statistics – Theory and Methods, Vol. 26, pp. 1481‐96.
Kulldorff, M. (2001), “Prospective time periodic geographical disease surveillance using a scan statistic”, Journal of the Royal Statistical Society Series A (Statistics in Society), Vol. 164, pp. 61‐72.
Kulldorff, M., Tango, T. and Park, P.J. (2003), “Power comparisons for disease clustering tests”, Computational Statistics & Data Analysis, Vol. 42, pp. 665‐84.
Kulldorff, M., Feuer, E., Miller, B. and Freedman, L. (1997), “Breast cancer clusters in the northeast United States: a geographic analysis”, American Journal of Epidemiology, Vol. 146, pp. 161‐70.
Kulldorff, M., Huang, L., Pickle, L. and Duczmal, L. (2006), “An elliptic spatial scan statistic”, Statistics in Medicine, Vol. 25, pp. 3929‐43.
Kulldorff, M., Athas, W., Feuer, E., Miller, B. and Key, C. (1998), “Evaluating cluster alarms: a space‐time scan statistic and brain cancer in Los Alamos, New Mexico”, American Journal of Public Health, Vol. 88 No. 9, pp. 1377‐80.
Kulldorff, M., Heffernan, R., Hartman, J., Assuncao, R. and Mostashari, F. (2005), “A space‐time permutation scan statistic for disease outbreak detection”, PLoS Medicine, Vol. 3, pp. 216‐24.
Liu, D., Wang, X., Pan, F., Xu, Y., Yang, P. and Rao, K. (2008), “Web‐based infectious disease reporting using XML forms”, International Journal of Medical Informatics, Vol. 77, pp. 630‐40.
Moran, P.A.P. (1950), “Notes on continuous stochastic phenomena”, Biometrika, Vol. 37, pp. 17‐23.
Ord, J.K. and Getis, A. (1995), “Local spatial autocorrelation statistics: distribution issues and an application”, Geographical Analysis, Vol. 27 No. 4, pp. 286‐306.
Pfeiffer, D.U., Robinson, T.P., Stevenson, M., Stevens, K.B., Rogers, D.J. and ACA Clements (2008), Spatial Analysis in Epidemiology, Oxford University Press, Oxford.
Ripley, B.D. (1976), “The second‐order analysis of stationary point patterns”, Journal of Applied Probability, Vol. 13, pp. 255‐66.
Ripley, B.D. (1977), “Modeling spatial patterns (with discussion)”, Journal of the Royal Statistical Society, Series B, Vol. 39, pp. 172‐212.
Rogerson, P. (1997), “Surveillance systems for monitoring the development of spatial patterns”, Statistics in Medicine, Vol. 16, pp. 2081‐93.
Rogerson, P. (2001), “Monitoring point patterns for the development of space‐time clusters”, Journal of the Royal Statistical Society Series A (Statistics in Society), Vol. 164, pp. 87‐96.
Rogerson, P. and Yamada, I. (2004a), “Geosurveillance: software for monitoring change in geographic patterns”, Computational Science and its Applications – ICCSA 2004, Part – 2, pp. 1096‐106.
Rogerson, P. and Yamada, I. (2004b), “Monitoring change in spatial patterns of disease: comparing univariate and multivariate cumulative sum approaches”, Statistics in Medicine, Vol. 23, pp. 2195‐214.
Science Daily (2009), “Role of climate change in disease spread examined”, available at: http://sciencedaily.com/releases/2009/02/090205142203.htm (accessed 19 June 2011).
Song, C.H. and Kulldorff, M. (2011), “Power evaluation of disease clustering tests”, International Journal of Health Geographics, Vol. 21, June, available at: http://ij‐healthgeographics.com/content/2/1/9 (accessed 21 June 2011).
Tango, T. (1984), “The detection of disease cluster in time”, Biometrics, Vol. 40, pp. 15‐26.
Tango, T. (2008), “A spatial scan statistic with restricted likelihood ratio”, Japanese Journal of Biometrics, Vol. 29, pp. 75‐95.
Tobler, W. (1970), “A computer movie simulating urban growth in the detroit region”, Economic Geography, Vol. 46 No. 2, pp. 234‐40.
WHO (2010), “Cholera in Haiti – update 4”, available at: http://who.int/csr/don/2010_11_24/en/index.html (accessed 21 June 2011).