- Research article
- Open Access
- Open Peer Review
This article has Open Peer Review reports available.
Bowel disorders and its spatial trend in Manitoba, Canada
- Mahmoud Torabi^{1}Email author
https://doi.org/10.1186/1471-2458-14-285
© Torabi; licensee BioMed Central Ltd. 2014
Received: 17 September 2013
Accepted: 19 March 2014
Published: 27 March 2014
Abstract
Background
Bowel disorders have destructive impacts on the patients social and mental aspects of life and can cause emotional distress. The risk of developing bowel incontinence also increases with age. The rate of incidence of inflammatory bowel disease in Manitoba, Canada, has been unusually raised. Therefore, it is important to identify trends in the incidence of bowel disorders that may suggest further epidemiological studies to identify risk factors and identify any changes in important factors.
Methods
An important part of spatial epidemiology is cluster detection as it has the potential to identify possible risk factors associated with disease, which in turn may lead to further investigations into the nature of diseases. To test for potential disease clusters many methods have been proposed. The focused detection methods including the circular spatial scan statistic (CSS), flexible spatial scan statistic (FSS), and Bayesian disease mapping (BYM) are among the most popular disease detection procedures. A frequentist approach based on maximum likelihood estimation (MLE) has been recently used to identify potential focused clusters as well. The aforementioned approaches are studied by analyzing a dataset of bowel disorders in the province of Manitoba, Canada, from 2001 to 2010.
Results
The CSS method identified less regions than the FSS method in the south part of the province as potential clusters. The same regions were identified by the BYM and MLE methods as being potential clusters of bowel disorders with a slightly different order of significance. Most of these regions were also detected by the CSS or FSS methods.
Conclusions
Overall, we recommend using the methods BYM and MLE for cluster detection with the similar population and structure of regions as in Manitoba. The potential clusters of bowel disorders are generally located in the southern part of the province including the eastern part of the city of Winnipeg. These results may represent real increases in bowel disorders or they may be an indication of other covariates that were not adjusted for in the model used here. Further investigation is needed to examine these findings, and also to explore the cause of these increases.
Keywords
- Bayesian computation
- Bowel disorders
- Geographic epidemiology
- Spatial cluster detection
Background
Bowel disorders consist of a variety of diseases and syndromes including Crohn’s disease, ulcerative colitis, irritable bowel syndrome and bowel incontinence. Inflammatory bowel disease (IBD) includes Crohn’s disease and ulcerative colitis and is distinguished by the presence of chronic immunoinflammatory lesions in the large intestine wall (ulcerative colitis) or anywhere in the gastrointestinal tract (Crohn’s disease) [1]. These diseases are often diagnosed in early adulthood and are characterized by a relapsing and remitting course. Treatment for the recurrent episodes of abdominal pain, diarrhea and bleeding are necessary throughout the patients’ lifetime [2, 3]. Since IBD is often diagnosed when the patients are still fairly young and these diseases have recurrent episodes, IBD can have a large negative affect on the patients’ quality of life [2]. Although the cause of IBD is unknown, potential risk factors include a family history of IBD (i.e., genetics), a history of smoking, age (early adulthood) and people who have been exposed to microorganisms early in life. Individuals who were breastfed as a baby are less likely to develop IBD. As well, people who have had an appendectomy prior to diagnosis are less likely to develop ulcerative colitis [4, 5].
Functional gastrointestinal disorders are characterized by situations where there are recurrent symptoms, however, when examined there appears to be nothing wrong [6]. Irritable bowel syndrome (IBS) is a functional bowel disorder. IBS affects approximately 9% to 23% of the general population. The symptoms of IBS are abdominal pain or discomfort with a change in bowel function. Possible risk factors of IBS include genetics, stress, infection and a poor diet [7]. The treatment is aimed at reducing the symptoms of IBS and may be treated by dietary and lifestyle changes recommended by a doctor, pharmacotherapy, and psychosocial interventions [6, 7].
Bowel incontinence is defined as the involuntary act of having a bowel movement. This disorder has destructive impacts on the patients social and mental aspects of life and can cause emotional distress [8]. The risk of developing bowel incontinence increases with age. It is estimated that between 2% and 18% of the general population and 50% of the people in nursing homes are affected by bowel incontinence [8, 9].
In addition to the deterioration of the patients’ quality of life, bowel disorders including Crohn’s disease, ulcerative colitis, irritable bowel syndrome, and bowel incontinence have large negative economic impacts. An article from 1999 stated that Manitoba had unusually high rates of incidence of inflammatory bowel disease [10]. Therefore, it is important to identify trends in the incidence of bowel disorders that may suggest further epidemiological studies to identify risk factors and identify any changes in important factors. Trends may occur over a region and the primary outcome measure of our paper is to examine geographical variation in the number of people diagnosed as having a bowel disorder during 2001 to 2010 in the province of Manitoba, Canada.
A spatial cluster is a small region within the entire study area which has a high number of disease cases relative to the respective population [11]. Possible factors associated with disease may be found through the identification of disease clusters. This may lead to an improved understanding of etiology, which in turn may lead to further studies to find the link between exposures and disease interventions [12].
There are two main groups of statistical cluster detection methods, focused and non-focused (general). Focused cluster detection methods identify regions with a high number of disease occurrences in an area around a potential cause (i.e., a toxic waste site) [13, 14]. Non-focused cluster detection methods are implemented in order to find regions, in general, with high number of disease [15–17]. Focused cluster detection methods include the circular spatial scan statistic (CSS) [18], flexible spatial scan statistic (FSS) [19], and Bayesian disease mapping (BYM) [15]. The Besag and Newell (BN) [20, 21] test and the maximizing excess event test (MEET) [22] are general cluster detection methods. Non-focused tests are used to discover possible clusters in the study area, while focused tests are used to test the null hypothesis of no spatial cluster against the alternative hypothesis that a spatial cluster exists. Hence, the test statistics of focused tests (CSS, FSS and BYM) are designed to detect a potential cluster in a specific area of interest and the goal of non-focused tests (BN and MEET) is to capture any significant cluster in the entire study region without identifying a specific area of interest. A comparison of these methods is given in [23] with an application to childhood cancer in Alberta, Canada.
This paper is centered around the focused cluster detection methods. The non-informative Bayesian approach has become quite popular with advances in computational power. The Bayesian approach can be used as a modeling approach to identify the potential clusters. Data Cloning (DC), which was proposed by Lele et al. [24], is a computing algorithm to obtain maximum likelihood estimates (MLE) and their standard errors for general hierarchical models. Lele et al. [25] outlined a method to calculate the prediction and prediction intervals for the random effects in the class of generalised linear mixed models. The MLE approach, via DC, was then proposed to identify the possible clusters [26].
In this paper, the aforementioned focused approaches (CSS, FSS, BYM, and MLE) are used to analyze a real dataset of bowel disorders in the province of Manitoba, Canada, from 2001 to 2010.
Methods
Study subjects
This study was based on the Canadian Community Health Survey (CCHS) [27] from Statistics Canada. Information is gathered from the Canadian population regarding health status, health care utilization and health determinants from the cross-sectional CCHS. In order to provide reliable estimates at the health region level, the CCHS collects data from individuals aged twelve and older [27]. The number of people with bowel disorders in the province of Manitoba, Canada, from 2001 to 2010 is the focus of this study. The province of Manitoba is divided into five Regional Health Authorities which are further sub-divided into 67 Regional Health Authority Districts (RHADs). The geographic units used in our model are the RHADs and all of the data used in the study are related to these geographic boundaries. For simplicity, the RHADs are labelled 1,2,…,67. A population-based was also provided for each RHAD. Since the bowel disorder data used in the study came from a survey, appropriate weights established by Statistics Canada [27] were applied to the data, which was then aggregated over the study period from 2001 to 2010.
The province of Manitoba’s population was steady over the study period with approximately 1.15 million people in 2001 to 1.20 million people in 2010. The average population sizes varied across the regions with 920 people being the smallest population in region 38 and 91,633 people being the largest population size in region 62. The mean and median population sizes in Manitoba were 17,471 and 9,466, respectively. The total number of people with bowel disorders was 138,296 with a mean and median of 2064 and 858 people, respectively. These observations were based on the weighted results of people with bowel disorders across the 67 regions in Manitoba.
Important factors required for focused spatial detection approaches are the observed number of people with a bowel disorder and the expected number of people with a bowel disorder or the population size of each region. If the expected number of people with bowel disorders varies by different variables such as year, age, or gender, adjustments may be made. In our application, the expected number of disease cases was adjusted by year (1-10), age group ((0-5),(6-20),(21-40),(41,88),(89+)) and gender (male, female). The CSS, FSS, BYM, and MLE spatial focused cluster detection methods are outlined in the Appendix.
There are different assumptions for each of these four focused spatial cluster detection methods. The CSS and FSS approaches are distribution free, whereas, in the BYM and MLE methods it is assumed that the number of disease cases follows a Poisson distribution. As well, in the CSS and FSS approaches, the number of regions to be included in a cluster needs to be specified, however, this is not required for the BYM and MLE approaches. For the model-based cluster identification methods (BYM and MLE), if the model does not fit the data well, the result can be misleading. So, the deviance residual [28] should be also checked.
Specific hypotheses
Results
The results of the four different cluster detection techniques when applied to a bowel disorder dataset in the province of Manitoba, Canada, from 2001 to 2010 are shown and compared in this section.
Based on the 67 regions, four different clusters were tested: (1) a case of no clusters (called A), (2) seven regions from the north part of the province (called B), (3) seven regions from south-central part of the province (called C), and (4) 12 regions which consist Winnipeg region (called D). For A, no region was specified as a potential cluster. The regions for scenarios B, C, and D are B = {31,33,34,36,38,40,41},C = {27,28,29,30,50,51,52},D = {56,57,58,59,60,61,62,63,64,65,66,67}, respectively.
The order of significant regions of the methods CSS, FSS, BYM, and MLE for cluster A
Method | ||||||||
---|---|---|---|---|---|---|---|---|
Region | RR > 1.0 | RR > 1.5 | RR > 2.0 | |||||
CSS | FSS | BYM | MLE | BYM | MLE | BYM | MLE | |
1 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
2 | - | 3 | 19 | 19 | - | - | - | - |
3 | - | 3 | - | - | - | - | - | - |
5 | - | - | 5 | 5 | 5 | 5 | - | - |
10 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
11 | 1 | 1 | 7 | 8 | 7 | 8 | - | - |
12 | 1 | 1 | 6 | 6 | 6 | 6 | - | - |
13 | 1 | - | - | - | - | - | - | - |
15 | 1 | - | - | - | - | - | - | - |
16 | 1 | - | - | - | - | - | - | - |
17 | 1 | 1 | 16 | 15 | - | - | - | - |
18 | 1 | - | - | - | - | - | - | - |
20 | - | 3 | 9 | 9 | - | - | - | - |
22 | 1 | 1 | 13 | 12 | - | - | - | - |
27 | - | 3 | 14 | 14 | - | - | - | - |
28 | - | 3 | - | - | - | - | - | - |
43 | - | 2 | 4 | 4 | 4 | 4 | - | - |
46 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 |
47 | - | 2 | - | - | - | - | - | - |
50 | - | - | 10 | 10 | - | - | - | - |
58 | - | - | 18 | 18 | - | - | - | - |
59 | 1 | 1 | 8 | 7 | 8 | 7 | - | - |
60 | 1 | 1 | 17 | 17 | - | - | - | - |
61 | 1 | 1 | 11 | 11 | - | - | - | - |
62 | 1 | 1 | 12 | 13 | - | - | - | - |
67 | - | - | 15 | 16 | - | - | - | - |
The CSS and FSS methods identified some similar regions as being potential clusters. In particular, the CSS method identified 15 regions as possible clusters while the FSS approach detected 18 regions as potential clusters of bowel disorders. The same 19 regions were identified by the BYM and MLE methods as being potential clusters of bowel disorders with a slightly different order of significance. Most of these regions were also detected by the CSS or FSS methods. Note that evaluating the criterion of the RR values from greater than 1 to 1.5 or even 2, the number of potential clusters decreases (Table 1). Based on the deviance residual plots for both methods BYM and MLE, we found that there is no serious lack of fit in the model.
For the case of cluster B, the methods BYM and MLE were only able to detect the region 31 as a potential cluster while the methods CSS and FSS also detected the region 38 as a potential cluster, noting that none of these four methods detected the other five regions (33, 34, 36, 40, 41) as a potential cluster.
For cluster C, the CSS and FSS methods detected 13 regions in addition to the cluster C as a potential cluster. The BYM and MLE methods also detected 17 regions in addition to the cluster C (except the region 52) as a potential cluster.
For cluster D, the all four methods detected the D cluster as a potential cluster. In addition to the regions in Winnipeg (cluster D), the methods BYM and MLE were also able to detect some neighbours of Winnipeg (13 regions) as potential clusters. However, the method CSS only detected three regions 1,10, and 46 as a potential cluster while the method FSS also detected the region 12 as a potential cluster in addition to regions detected by the CSS method.
Discussion and conclusion
The four methods CSS, FSS, BYM, and MLE were studied with potential for detecting clusters with high ratio of bowel disorders in the province of Manitoba, Canada. These four methods have been extensively used in the literature and are relatively comprehensive. These methods use different approaches (semi-parametric to parametric) to test for significant clusters.
We considered four different alternative hypotheses to compare the results of different methods. In general, the CSS method identified a lower number of regions combined as a potential cluster compared to FSS method, due to non-circular shape of some regions in the province of Manitoba. It also seems that the bowel disorder cases tend to constitute the potential clusters in south-central part of the province. Note that we used four different alternative hypotheses (with low and high dense regions in terms of population) to compare these four methods, however, one can also use an extensive simulation study to compare the performance of these four methods.
The methods BYM and MLE did good jobs for dispersed population (cluster B) and also for dense population (clusters C and D) compared to the methods CSS and FSS. Also, in our study, the method FSS did a better job compared to the method CSS to detect potential clusters. Overall, we recommend using the methods BYM and MLE for cluster detection for the similar population and structure of regions as in Manitoba.
A region was identified as a potential cluster if the credible/prediction interval of the estimated relative risk was larger than one for the BYM and MLE approaches. Different decision rules may be defined where the estimated relative risk (in terms of the credible/prediction interval) would be larger or smaller than one [29]. One could also consider the exceedance probability Pr(RR _{ i } > b) > c, where b can be 1, 2 or 3 and c might be a large value such as 0.90 [30].
Here, three important factors, age, gender and year were used to adjust the expected number of bowel disorders in the province of Manitoba. Unlike the methods CSS and FSS methods, we can extend the model (2)-(3), for both BYM and MLE methods, to include other covariates directly which may be required for some applications.
We also note that the methods have different settings and assumptions which motivate our comparison. User-chosen settings are part of all cluster tests and different choices could lead to different results. All four methods have been proposed for local clusters. Under the null hypothesis, the number of bowel disorder cases follows a Poisson distribution for the BYM and MLE methods, while the test statistic for the CSS and FSS methods has an asymptotically χ ^{2} distribution. These features motivated us to consider these important methods and apply them to our bowel disorder cases.
As limitations of study, we assumed that our bowel disorder cases are rare cases to be able to use Poisson model in our BYM and MLE methods. We used survey data (weighted to the population level) in our study. Strengths of the study include the evaluation of multiple cluster detection methods.
The potential clusters of bowel disorders are generally located in the southern part of the province including the eastern part of the city of Winnipeg (cluster A). These results may represent real increases in bowel disorders or they may be an indication of other covariates that were not adjusted for in the model used here. Further investigation is needed to examine these findings, and also to explore the cause of these increases.
Appendix
The CSS, FSS, BYM, and MLE spatial focused cluster detection methods are outlined below.
Circular spatial scan statistic (CSS)
where the observed number of cases and expected number of cases inside a circle are denoted by C _{ i } and E _{ i }, respectively and the observed number and expected number of cases outside a circle are denoted by (N - C _{ i }) and (N-E _{ i }), respectively. The indicator function I(C _{ i } > E _{ i }) is equal to 1 when C _{ i } > E _{ i } and 0 otherwise. Potential clusters are identified by circles with high likelihood ratio statistics [18].
This method can be conducted using SaTScan [32] or FleXScan [33] software. The J is usually chosen to encompass at most 50% of the population at risk, however, we used J = 15, which is the FleXScan default. The region centroid had to be included in the radius of the circle in order for the region to be part of the circle.
Flexible spatial scan statistic (FSS)
The flexible spatial scan statistic is similar to the method of the CSS except now the detected cluster is flexible in shape while still being bound to a small neighbourhood of each region. An irregularly shaped window S is placed on each region by the flexible scan statistic. This is done by connecting its adjacent regions. For any region i, the set of irregularly shaped windows of length j, which contains j connected regions including region i, can vary from 1 to the pre-specified maximum J, where J is the maximum length of a cluster. In order to avoid unlikely cluster shapes, the joined regions are confined to the subsets of the set of regions i and (J - 1)-th nearest neighbours of region i. The set of all windows to be scanned by the flexible spatial scan statistic is then S _{2} = {S _{ i:j(k)};i = 1,…,m;j = 1,…,J;k = 1,…,k _{ ij }}. The size of S _{2} is much larger than S _{1} which is at most mJ. This is because for each region i the flexible scan statistic studies J circles plus all the sets of connected regions whose centroids are found within the J -th largest concentric circle, whereas the circular scan statistic considers only J circles for each region i. The likelihood ratio in (1) can be used for the flexible spatial scan statistic where the circle defined in (1) now refers to S _{2} rather than S _{1}. As with the CSS method, circles with high likelihood ratio values are considered to be possible areas of disease clusters [19]. The FSS method is conducted using the FleXScan software [33], with J = 15, which is the FleXScan default.
Bayesian disease mapping (BYM)
where P is a m × m diagonal matrix with elements P _{ ii } = 1/E _{ i };D is a m × m matrix with elements D _{ ij } = (E _{ j }/E _{ i })^{1/2} if region i and j are adjacent and D _{ ij } = 0 otherwise; ${\sigma}_{\eta}^{2}$ is the spatial dispersion parameter; λ _{ η } measures the spatial autocorrelation, λ _{ min } ≤ λ _{ η } ≤ λ _{ max }, where ${\lambda}_{\mathit{\text{min}}}^{-1}$ and ${\lambda}_{\mathit{\text{max}}}^{-1}$ are the smallest and largest eigenvalues of P ^{-1/2} DP ^{1/2}; and I _{ m } is an identity matrix of dimension m. We refer to [36] for details of this proper CAR model. Within the Bayesian framework (MCMC) the parameters can be estimated using vague prior distributions. This produces posterior distributions for the parameters in the model given in (2)-(3) [15].
In terms of their credibility sets, when the estimated relative risk is significantly larger than one (i.e., the lower level of the credible set is larger than one) the region is considered to be a disease cluster [37]. WinBUGS software [36] was used to conduct this method and to calculate the relative risk values.
Frequentist approach using MLE for disease mapping (MLE)
where the prior distribution on the parameter space is π(α) and $H\left({\mathit{C}}^{\left(L\right)}\right)=\int {\left\{L\right(\mathit{\alpha};\mathit{C}\left)\right\}}^{L}\pi \left(\mathit{\alpha}\right)d\mathit{\alpha}$ is the normalizing constant. The likelihood for L copies of the original data is denoted by {L(α;C)}^{ L }. It was shown by Lele et al. [24, 25] that when L is large enough, π _{ L }(α|C ^{(L)}) will converge to a multivariate Normal distribution with the mean given by the MLE of the model parameters and variance-covariance matrix equal to 1/L times the inverse of the Fisher information matrix for the MLE. For large L, this distribution is almost degenerate at the MLE α. In addition, an estimate of the MLE is given by the sample mean vector of the generated random numbers and L times their sample variance-covariance matrix is an estimate of the asymptotic variance-covariance matrix for the MLE $\widehat{\mathit{\alpha}}$. Different tests to determine the adequate number of clones L were also provided by Lele et al. [25].
Prediction of relative risk
where ${\alpha}_{1}=\mu ,{\mathit{\alpha}}_{2}={({\lambda}_{\eta},{\sigma}_{\eta}^{2})}^{\prime},f(\xb7)$ and g(·) are Poisson and Normal distributions, ϕ(.,ξ,Σ) denotes a multivariate Normal density with mean ξ and variance-covariance Σ and $H\left(\mathit{C}\right)=\int L(\mathit{\alpha};\mathit{C})\pi \left(\mathit{\alpha}\right)d\mathit{\alpha}$ is the normalizing constant. Using the density given in equation (5) and MCMC sampling, the prediction of the r is found. Similar to the BYM method, a region where the estimated relative risk (in terms of their prediction interval) is significantly larger than one is considered to be a disease cluster. In order to calculate the relative risk values, the dclone package [39] is used in the R software [40].
Declarations
Acknowledgements
I would like to thank two referees for constructive comments and suggestions, which led to an improved version of the manuscript. This work was supported by grants from the Manitoba Health Research Council (MHRC) and the Natural Sciences and Engineering Research Council of Canada (NSERC). The author thanks Katie Galloway for manuscript preparation.
Authors’ Affiliations
References
- Casellas F, Ginard D, Vera I, Torrejón A: Satisfaction of the health care professionals managing patients with inflammatory bowel disease. J Crohns Colitis. 2013, 7: e249-e255. 10.1016/j.crohns.2012.10.003.View ArticlePubMedGoogle Scholar
- Feagan BG: Review article: economic issues in Crohn’s disease - assessing the effects of new treatments on health-related quality of life. Ailment Phrarmocol Ther. 1999, 13 (Suppl. 4): 29-37.View ArticleGoogle Scholar
- Longobardi T, Jacobs P, Wu L, Bernstein CN: Work losses related to inflammatory bowel disease in Canada: results from a national population health survey. Am J Gastroenterol. 2003, 98 (4): 844-849. 10.1111/j.1572-0241.2003.07378.x.View ArticlePubMedGoogle Scholar
- Bernstein CN, Rawsthorne P, Cheang M, Blanchard JF: A population-based case control study of potential risk factors for IBD. Am J Gastroenterol. 2006, 101: 993-1002. 10.1111/j.1572-0241.2006.00381.x.View ArticlePubMedGoogle Scholar
- Molodecky NA, Kaplan GG: Environmental risk factors for inflammatory bowel disease. Gastroenterol Hepatol. 2010, 6 (5): 339-346.Google Scholar
- Hyams JS: Functional gastrointestinal disorders. Curr Opin Pediatr. 1999, 11: 375-378. 10.1097/00008480-199910000-00001.View ArticlePubMedGoogle Scholar
- Schwetz I, Chang L: Irritable bowel syndrome. Encyclopedia of Gastroenterology. Edited by: Johnson L. 2004, New York: Elsevier, 467-475.View ArticleGoogle Scholar
- Norton C, Dibley LB, Bassett P: Faecal incontinence in inflammatory bowel disease: associations and effect on quality of life. J Crohns Colitis. 2013, 7: e302-e311. 10.1016/j.crohns.2012.11.004.View ArticlePubMedGoogle Scholar
- Crowell MD, Schettler VA, Brian E. Lacy BE, Lunsford TN, Harris LA, DiBaise JK, Jones MP: Impact of anal incontinence on psychosocial function and health-related quality of life. Dig Dis Sci. 2007, 52: 1627-1631. 10.1007/s10620-006-9249-3.View ArticlePubMedGoogle Scholar
- Bernstein CN, Blanchard JF, Rawsthorne P, Wajda A: Epidemiology of Crohn’s disease and ulcerative colitis in a central Canadian province: a population-based study. Am J Epidemiol. 1999, 149 (10): 916-924. 10.1093/oxfordjournals.aje.a009735.View ArticlePubMedGoogle Scholar
- Lawson AB: Statistical Methods in Spatial Epidemiology, 2nd edition. 2006, London: John Wiley & Sons, LtdView ArticleGoogle Scholar
- Jennings JM, Curriero FC, Celentano D, Ellen JM: Geographic identification of high gonorrhea transmission areas in Baltimore, Maryland. Am J Epid. 2005, 161: 73-80. 10.1093/aje/kwi012.View ArticleGoogle Scholar
- Elliott P, Briggs D, Morris S, de Hoogh C, Hurt C, Jensen TK, Maitland I, Richardson S, Wakefield J, Jarup L: Risk of adverse birth outcomes in populations living near landfill sites. Br Med J. 2001, 323: 363-368. 10.1136/bmj.323.7309.363.View ArticleGoogle Scholar
- Lawson AB, Biggeri A, Williams FLR: A review of modeling approaches in health risk assessment around putative sources. Disease Mapping and Risk Assessment for Public Health. Edited by: Bertollini R, Viel J, Lesaffre E, Lawson AB, Biggeri A BöhningD. 1999, New York: Wiley, 231-245.Google Scholar
- Besag JE, York JC, Mollìe A: Bayesian image restoration with two applications in spatial statistics (with discussion). Ann Inst Statist Math. 1991, 43: 1-59. 10.1007/BF00116466.View ArticleGoogle Scholar
- Clayton D, Bernardinelli L: Bayesian methods for mapping disease risk. Geographical and environmental epidemiology: methods for small-area studies. Edited by: Stern R, English D, Cuzick J, Elliott P. 1996, Oxford University Press, 205-220.View ArticleGoogle Scholar
- Clayton D, Kaldor J: Empirical Bayes estimates of age-standardized relative risks for use in disease mapping. Biometrics. 1987, 43: 671-681. 10.2307/2532003.View ArticlePubMedGoogle Scholar
- Kulldorff M: A spatial scan statistics. Comm Statist: Theor Meth. 1997, 26: 1481-1496. 10.1080/03610929708831995.View ArticleGoogle Scholar
- Tango T, Takahashi K: A flexibly shaped spatial scan statistic for detecting clusters. Int J Health Geogr. 2005, 4 (11): 1-15.Google Scholar
- Besag JE, Newell J: The detection of clusters in rare diseases. J Roy Statist Soc Ser A. 1991, 154: 143-155. 10.2307/2982708.View ArticleGoogle Scholar
- Torabi M, Rosychuk RJ: Spatial event cluster detection using an approximate normal distribution. Int J Health Geogr. 2008, 7 (61): 1-22.Google Scholar
- Tango T: A test for spatial disease clustering adjusted for multiple testing. Statist Med. 2000, 19: 191-204. 10.1002/(SICI)1097-0258(20000130)19:2<191::AID-SIM281>3.0.CO;2-Q.View ArticleGoogle Scholar
- Torabi M, Rosychuk RJ: An examination of five spatial disease clustering methodologies for the identification of childhood cancer clusters in Alberta, Canada. Spat Spatio-tempor Epidemiol. 2011, 2: 321-330. 10.1016/j.sste.2011.10.003.View ArticleGoogle Scholar
- Lele SR, Dennis B, Lutscher F: Data cloning: easy maximum likelihood estimation for complex ecological models using Bayesian Markov chain Monte Carlo methods. Ecol Lett. 2007, 10: 551-563. 10.1111/j.1461-0248.2007.01047.x.View ArticlePubMedGoogle Scholar
- Lele SR, Nadeem K, Schmuland B: Estimability and likelihood inference for generalized linear mixed models using data cloning. J Am Statist Assoc. 2010, 105: 1617-1625. 10.1198/jasa.2010.tm09757.View ArticleGoogle Scholar
- Torabi M: Spatial disease cluster detection: an application to childohhod asthma in Manitoba, Canada. J Biomet Biostat. 2012, S7: 1010-doi:10.4172/2155-6180.S7-010Google Scholar
- Statistics Canada: Canadian Community Health Survey User Guide. Ottawa ON: Statistics Canada, 2001–2010Google Scholar
- McCullagh P, Nelder JA: Generalized Linear Models, 2nd ed. 1989, London: Chapman and HallView ArticleGoogle Scholar
- Richardson S, Thomson A, Best N, Elliott P: Interpreting posterior risk estimates in disease-mapping studies. Environ Health Persp. 2004, 112 (9): 1016-1025. 10.1289/ehp.6740.View ArticleGoogle Scholar
- Banerjee S, Gelfand AE, Carlin BP: Hierarchical Modeling and Analysis for Spatial Data. 2004, London: Chapman and HallGoogle Scholar
- Fukuda Y, Umezaki M, Nakamura K, Takano T: ariations in social characteristics of spatial disease clusters: examples of colon, lung and breast cancer in Japan. Int J Health Geogr. 2005, 4 (16): 1-13.Google Scholar
- Kulldorff M, Rand K, Gherman G, Williams G, DeFrancesco D, SaTScan V2. 1: Software for the Spatial and Space-time Scan Statistics. 1998, Bethesda: National Centre InstituteGoogle Scholar
- Takahashi K, Yokoyama T, Tango T: FleXScan: Software for the Flexible Scan Statistic. 2006, Japan: National Institute of Public HealthGoogle Scholar
- Bernardinelli L, Montomoli C: Empirical Bayes versus fully Bayesian analysis of geographical variation in disease risk. Statist Med. 1992, 11: 983-1007. 10.1002/sim.4780110802.View ArticleGoogle Scholar
- Gilks WR, Richardson S, Spielhalter DJ (Eds): Markov chain Monte Carlo in Practice. 1996, New York: Springer-VerlagGoogle Scholar
- Spiegelhalter D, Thomas A, Best N, Lunn D: WinBUGS version 1.4 User Manual. 2004, London: MRC Biostatistics unit, Institute of Public HealthGoogle Scholar
- Aamodt G, Samuelsen SO, Skrondal A: A simulated study of three methods for detecting disease clusters. Int J Health Geogr. 2006, 5 (15): 1-11.Google Scholar
- Hamilton JD: A standard error for the estimated state vector of a state-space model. J Economet. 1986, 33: 387-397. 10.1016/0304-4076(86)90004-7.View ArticleGoogle Scholar
- Sólymos P: dclone: Data cloning in R. The R J. 2010, 2: 29-37.Google Scholar
- R Core Team: R: A Language and Environment for Statistical Computing. 2013, Vienna: Austria: R Foundation for Statistical Computing, http://www.R-project.org,Google Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2458/14/285/prepub
Pre-publication history
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.