Incorporating retrospective clustering into a prospective cusum methodology for anthrax: Evaluating the effects of disease expectation
Introduction
The most common techniques for examining the presence of spatial or spatio-temporal clusters are retrospective analyses, which are often carried out as a onetime analysis of past data, after the onset, and often after the completion of an outbreak (Tango and Takashi, 2005). The prevalence of these techniques in the literature provides a valuable reference resource for analysing the spatial (Moran, 1950, Anselin, 1995, Ord and Getis, 1995, Kulldorff, 1997) and/or temporal distribution (Knox, 1964, Mantel, 1967, Wallenstein, 1980) of past health events.
Yet, in human and veterinary epidemiology the objective is often to detect the onset of health events as quickly as possible. Statistical applications that allow for the continual evaluation of a disease status over time are advantageous since they may be able to identify the onset of clusters in a timelier manner. In this area of health analyses retrospective techniques may incur specific limitations due to issues of multiple hypotheses testing that occur when these methodologies are used to measure a disease status continuously over time (Tango, 2000). Alternatively, prospective statistical techniques such as the cumulative sum (CUSUM) approach (Page, 1954), originally developed for process control, can be used in a continuous detection system to monitor the status of disease over time in an attempt to detect the onset of clusters (Rogerson, 1997). A more comprehensive review of prospective techniques has been described elsewhere (Sonesson and Bock, 2003, Woodall et al., 2008).
For the purposes of this study we are particularly interested in the application of clustering techniques used in the monitoring of veterinary health. The application of space–time clustering using both retrospective techniques (Carpenter et al., 1996, Hoar et al., 2003, Orazi et al., 2007) and prospective techniques (Mostashari et al., 2003, Hohle et al., 2009) has been shown to be successful in research examining the spatial and temporal distribution of diseases in veterinary epidemiology. A study by Ward et al. (1996) found clustering of bluetongue virus serotypes among cattle herds in Queensland, Australia using the Cuzick and Edward’s test. Research using retrospective space–time clustering techniques in the investigation of acute respiratory infections in cattle suggests that the identification of clusters link multiple illnesses to a single pathogen (Norstrom et al., 2000). Several studies have also utilized the SaTScan software to retrospectively identify the clustering of bovine spongiform encephalopathy in cattle (Doherr et al., 2002, Sheridan et al., 2005, Allepuz et al., 2007, Heres, 2008). Ward and Carpenter (2000) and Carpenter (2001) provide reviews of additional methods used to investigate the distribution of health events in veterinary epidemiology.
In the field of prospective surveillance Rogerson (1997) employed a one-sided CUSUM approach using a modified Tango’s statistic to reanalyse data from Williams et al. (1978) in order to identify the presence of emerging space–time patterns of Burkitt’s lymphoma in Uganda. This study found the CUSUM method detected the emergence of additional clusters previously unidentified by retrospective analyses. Research applying early detection methods in livestock surveillance incorporated a log-linear regression method to derive expected counts from a baseline dataset, which was constructed using historical case records, in order to elucidate anomalies in the distribution of Salmonella spp. infections (Kosmider et al. 2006). Additionally, Gilbert et al. (2005) illustrates the efficacy of the monitoring of livestock diseases by deriving model parameters from historical data, in conjunction with biotic and abiotic variables to predict a shifting geographic distribution of Bovine Tuberculosis (BTB) on a yearly basis. Like in the aforementioned case of BTB there is a crucial need to monitor other zoonotic livestock/wildlife pathogens (those transferrable from animals to humans) such as anthrax that threaten not only animal populations, but human populations as well.
Bacillus anthracis, the causative agent of anthrax, is a gram-positive spore-forming bacterium, that affects livestock and wildlife (primarily herbivorous ungulates), and secondarily humans (Van Ness, 1971). Outbreaks of the disease in Central Asia, including Kazakhstan (Woods et al., 2004), have increased in recent years due to inadequacies in public health and veterinary surveillance (Hugh-Jones, 1999).
Several studies have described the spatial and temporal distribution of anthrax infections in livestock (Dragon et al., 1999, Turner et al., 1999, Parkinson et al., 2003, Clegg et al., 2007, Himsworth and Argue, 2008, Mongoh et al., 2008). Van Ert et al. (2007) showed through mapping the phylogeography of B. anthracis that its global distribution may be influenced by its genetic variation. Research has also used GIS mapping in conjunction with ecological niche modeling to predict the potential geographic distribution of B. anthracis in the US (Blackburn et al., 2007) and in Kazakhstan (Joyner et al., 2010). However, few of these studies have applied spatio-temporal techniques to quantitatively describe the distribution of anthrax infections. This is also true in research looking at human infections of the disease, which have either focused on, the bioterrorist event in the US in 2001(Jernigan et al., 2002, Webb and Blaser, 2002), syndromic studies related to potential bioterrorism (Kleinman et al., 2005, Buckeridge et al., 2006), or the accidental release of weaponized anthrax in Sverdlovsk, Russia in 1979 (Meselson et al., 1994, Wilkening, 2006).
The few studies that have utilized spatio-temporal statistical techniques to analyse the distribution of anthrax infections have illustrated the potential usefulness of these tools. Initial research by Smith et al. (1999) identified three anthrax isolates responsible for wildlife epidemics in Kruger National Park (KNP), South Africa and found using the Mantel’s test they were clustered in both space and time. A subsequent study by Smith et al. (2000) indicated through the use of SaTScan that there was distinct spatio-temporal clustering of two major anthrax strain types, A and B within KNP, due to possible differences in soil composition that may have exerted an influence on the location of each strain. Current research on the distribution of anthrax outbreaks is limited to retrospective analyses allowing for the implementation of prospective methodologies to add to the current body of anthrax literature.
The purpose of this current study was to conduct an exploratory analysis of the spatial and temporal distribution of historical anthrax outbreaks among livestock in Kazakhstan utilizing a prospective CUSUM approach. Specifically this study had two objectives: (1) examine the methods for deriving a baseline rate of disease for use in a CUSUM methodology when no population data are available and (2) to evaluate the influence that various derived expectations of disease have on the detection of clusters in space and time in an annual CUSUM methodology. This study represents one of the first prospective statistical examinations of anthrax in livestock.
Section snippets
Methodology
As part of a larger effort to map and model the geographic distribution of anthrax and its control in Kazakhstan, the Kazakh Science Center for Quarantine and Zoonotic Disease developed a spatial database of database totaling 3963 outbreaks that were reported over a 74-year period from 1933–2006 (Aikembayev et al., 2010). This current study employs a selection of that historical record. A subset of the data representing livestock outbreaks (a combination of small and large ruminant outbreaks)
Prospective analysis
The CUSUM analysis using an ARL of 100 (h = 7.9), shows the spatial relationship of each of the three different methods for calculating z-scores MWA, AVG, and LISA during the period 1960–2006 (Fig. 3). During the 47-year period the MWA methodology had the lowest number of rayons eliciting alarm signals n = 3, while the LISA methodology showed the highest number of rayons with alarm events n = 16 and the AVG methodology had n = 11 rayons signal an alarm. The presence of alarm events was consistent
Discussion
The methods set forth in this paper introduce techniques for selecting a baseline rate of disease for use in a prospective CUSUM. While prospective methodologies have been widely used in the analysis of human disease data, their application in veterinary health is far less extensive (Kosmider et al. 2006). This is probably due to the difficulties encountered when attempting to analyse the spatial and/or temporal patterns of livestock diseases when population figures are not available, or when
Acknowledgements
This project was funded by the United States Defense Threat Reduction Agency (DTRA) as part of the Biological Threat Reduction Program’s Cooperative Biological Research Program in Kazakhstan. Y. Sansyzbayev, M.E. Hugh-Jones, A. Curtis, and T.A. Joyner assisted with the original database development. We would like to thank two anonymous reviewers for strengthening this manuscript.
References (63)
- et al.
Spatial analysis of bovine spongiform encephalopathy in Galicia, Spain (2000–2005)
Prev Vet Med
(2007) - et al.
A space-time investigation of the epidemiology of fowl cholera
Prev Vet Med
(1996) Methods to investigate spatial and temporal clustering in veterinary epidemiology
Prev Vet Med
(2001)- et al.
Spatial analysis of Yersinia pestis and Bartonella vinsonii subsp. Berkhoffii seroprevalence in California coyotes (Canis latrans)
Prev Vet Med
(2003) - et al.
Statistical approaches to the monitoring and surveillance of infectious diseases fro veterinary public health
Prev Vet Med
(2009) - et al.
A space-time cluster investigation of an outbreak of acute respiratory disease in Norwegian cattle herds
Prev Vet Med
(2000) - et al.
Techniques for analysis of disease clustering in space and in time in veterinary epidemiology
Prev Vet Med
(2000) - et al.
Historical distribution and molecular diversity of Bacillus anthracis, Kazakhstan
Emerg Infect Dis
(2010) Local indicators of spatial autocorrelation
Geographic Analalysis
(1995)- et al.
GeoDa: an introduction to spatial data analysis
Geographic Analalysis
(2006)
Modeling the geographic distribution of Bacillus anthracis, the causative agent of anthrax disease, for the contiguous United States using predictive ecologic niche modeling
Am J Trop Med Hyg
Evaluating detection of an inhalation anthrax outbreak
Emerg Infect Dis
Evaluation and extension of the cusum technique with an application to Salmonella surveillance
J Vet Diagn Invest
Massive outbreak of anthrax in wildlife in the Malilangwe wildlife reserve, Zimbabwe
Vet Rec
Geographical clustering of cases of bovine spongiform encephalopathy (BSE) born in Switzerland after the feed ban
Vet Rec
Spatial distribution of brucellosis in sheep and goats in Sicily from 2001 to 2005
Vet Ital
A review of anthrax in Canada and implications for research on the disease in northern bison
J Appl Microbiol
Attle movements and bovine tuberculosis in Great Britian
Nature
Spatial analysis of BSE cases in The Netherlands
BMC Vet Res
Anthrax in Saskatchewan 2006: an outbreak overview
Can Vet J
1996–97 global anthrax report
J Appl Microbiol
A simulation model for assessing aberration detection methods used in public health surveillance for systems with limited baselines
Statistic in Med
Evaluation of a syndromic surveillance for the early detection of outbreaks among military personnel in a tropical country
J Pub Health.
Investigation of bioterrorism-related anthrax, United States, 2001: epidemiologic findings
Emerg Infect Dis
Modeling the potential distribution of Bacillus anthracis under multiple climate change scenarios for Kazalhstan
PLoS ONE
A model-adjusted space-time scan statistic with an application to syndromic surveillance
Epidemiol Infect
A statistical system for detecting salmonella outbreaks in British livestock
Epidemiol Infect
The detection of space-time interactions
Applied Statistics
A spatial scan statistic
Commun Stat-theor Meth
Cited by (13)
A systematic review of aberration detection algorithms used in public health surveillance
2019, Journal of Biomedical InformaticsCitation Excerpt :The total number of analysis methods identified exceeds the total number of identified studies, as a single study could explore multiple detection methods. In this review, we identified 32 studies (22%) of SPC or SPC-inspired methods and most of these papers described the application or evaluation of a previously described method (25/32) [25,32,41,47,49,51,64,67,9,71,72,77,85–87,90,91,98–101,105,108,114,115,118,119,129,144–146,150]. SPC methods were first developed in 1920s for use in manufacturing.
Differential detection of a surrogate biological threat agent (Bacillus globigii) with a portable surface plasmon resonance biosensor
2016, Biosensors and BioelectronicsCitation Excerpt :Portable SPR biosensors may be used in manned (Naimushin et al., 2005) or unmanned (Palframan et al., 2014) biodetection operations in a variety of environments (Usachev et al., 2014) (Fig. 1). Bacillus anthracis is a gram positive, rod-shaped, spore-forming bacterium that can cause the disease anthrax in humans and animals (Blackburn, 2010; Kracalik et al., 2011; Read et al., 2003; Spencer, 2003). This bacterium has been used as a biological warfare agent, and is still considered to be a threat to biosecurity (Webb, 2003).
Spatio-temporal patterns of livestock anthrax in Ukraine during the past century (1913-2012)
2014, Applied GeographyCitation Excerpt :A very large spike in outbreak reports occurred in the late 1940s; immediately after World War II (WWII), in 1946–1947 (large spike reaching a maximum of 821 sites), which then plummeted during the 1950s (Fig. 2). A similar pattern of increased livestock anthrax post-WWII was also apparent in a recent analysis of historical data from Kazakhstan by Kracalik et al. (2011). The Soviet Union was heavily impacted during WWII with much of the war in Europe fought on Soviet lands (Harrison, 2011).
Plasmonic Sensors for Detection of Chemical and Biological Warfare Agents
2021, Plasmonic Sensors and their ApplicationsMapping areas of asynchronous-temporal interaction in animal-telemetry data
2020, Transactions in GIS