- Open Access
Variations in societal characteristics of spatial disease clusters: examples of colon, lung and breast cancer in Japan
© Fukuda et al; licensee BioMed Central Ltd. 2005
- Received: 28 April 2005
- Accepted: 14 June 2005
- Published: 14 June 2005
Spatial analyses and ecological studies are essential for epidemiology and public health. The present study combining these two methods was performed to identify spatial clusters of selected types of cancer in Japan and to determine their societal characteristics focusing on homogeneity among clusters.
Spatial clusters of high mortality rates of male colon and lung cancer and of female breast cancer were identified by the spatial scan statistic using Japanese municipal data (N = 3360) from 1993 to 1998 and also municipalities were divided into four societal clusters based on socioeconomic indicators and population density (urban-rich, suburban, rural-poor, and clutter). Five, seven, and four mortality clusters were identified for lung, colon and breast cancer, respectively. For colon and breast cancer, most municipalities of all except one cluster were included in a single societal cluster (urban-rich). The municipalities associated with mortality clusters for lung cancer belonged to various societal clusters.
Increased mortality rates of colon and breast cancer can be explained by homogenous societal characteristics related to urbanisation, although there were exceptional areas with higher mortality rates. The regional variation in lung cancer mortality rate appeared to be due to heterogeneous factors. These findings and the analysis performed in the present study will contribute to both nationwide and region-specific cancer prevention strategies.
- Breast Cancer
- Principle Component Analysis
- Spatial Cluster
- Standardize Mortality Ratio
- Female Breast Cancer
Health levels vary substantially between different regions, and it is essential to characterise these regional variations and identify areas with an accumulation of health problems for epidemiologic research and to allow appropriate public health policy decisions [1, 2]. Recent advancements in technologies, such as geographic information systems (GIS), have allowed the application of not only disease mapping but also spatial analyses, such as spatial clustering and cluster detection, in epidemiological research [3–6]. In this context, clusters are defined as unusual concentrations of health events in both space and time .
Ecological studies examining the relationships between regional health levels and various characteristics represent another essential approach in epidemiology and public health. Although such studies have a number of limitations, especially confounding factors and ecological fallacy, factors that may contribute to regional health variations can be identified and hypotheses can be formulated for further research [7, 8]. Several ecological studies have demonstrated relationships between mortality and regional characteristics related to the environment, health-related behaviour, and economic and demographic factors in Japan [9–12]. Recent systematic studies using municipal data regarding all causes and cause-specific mortality along with large numbers of societal indicators showed substantial relationships between a region's mortality rate and societal characteristics [13–15].
In general, the relations between health levels and regional characteristics are examined by correlation and regression analyses [8–14]. These methods can reveal factors correlated with regional variations in a specific health issue across study areas. However, if various factors contribute separately to a health issue for different areas, or where there is an exceptional factor contributing to a health issue in a limited area, such analyses would not be effective in identifying the contributing factors and may overlook exceptional factors.
The present study was performed to determine whether areas with a specific health problem have homogeneous regional characteristics or different patterns of characteristics. We first identified spatial clusters of three common types of cancer (colon, lung and breast) using the spatial scan statistic. The societal characteristics of the clusters were then elucidated, focusing on heterogeneity in the characteristics among clusters, using municipal data across Japan.
Result of principle component analysis.
Rotated component Matrix from the principle component analysis of seven societal indicators.
Unemployment rate (women)
Unemployment rate (men)
Number of rooms per household
Dwelling area per capita
Education level (women)
Education level (men)
Income per capita
Characteristics of societal clusters.
Comparison of societal indices and population density among societal clusters (SC).
SC1 (N = 507)
SC2 (N = 1483)
SC3 (N = 1246)
SC4 (N = 124)
Index 1: high unemployment and overcrowding
0.89 ± 0.66
-0.08 ± 0.63
-0.54 ± 0.74
2.69 ± 1.23
Index 2: high income and educational level
1.56 ± 1.06
-0.06 ± 0.64
-0.45 ± 0.62
-1.13 ± 0.69
Population density (log)
8.04 ± 0.84
5.75 ± 0.63
3.68 ± 0.83
6.44 ± 1.31
The mortality clusters of male lung cancer are illustrated in Figure 3 (b). The primary cluster (MC1) was located in an area including Osaka, with RR of 1.17. A total of six secondary clusters were also identified. The municipalities of MC2 belonged mainly to Hokkaido prefecture, and those of MC3 belonged to Okinawa prefecture consisting of the southern islands. MC4, MC6, and MC7 included the metropolitan areas of Nagoya, Fukuoka and Tokyo, respectively, while MC5 was located in the mountainous area on Shikoku Island.
As shown in Figure 4 (b), of the four mortality clusters identified for female breast cancer, three were located in metropolitan areas: MC1 in Tokyo, MC2 in Osaka and MC4 in Nagoya. The centre of the remaining cluster, MC3, was located in Hokkaido.
Relative risk of societal indices, population density and societal cluster.
Relative risk was estimated by the hierarchical Poisson regression.
Male colon cancer
Male lung cancer
Female breast cancer
Population density (log)
Relationship between mortality clusters and societal clusters.
Figures show the number of municipalities and the percentage in parenthesis by mortality cluster.
Mortality cluster a
Societal cluster b
SC1 (N = 507)
SC2 (N = 1483)
SC3 (N = 1246)
SC4 (N = 124)
Male colon cancer
MC1 (N = 53)
MC2 (N = 228)
MC3 (N = 61)
MC4 (N = 27)
MC5 (N = 1)
Male lung cancer
MC1 (N = 159)
MC2 (N = 289)
MC3 (N = 48)
MC4 (N = 54)
MC5 (N = 12)
MC6 (N = 162)
MC7 (N = 8)
Female breast cancer
MC1 (N = 51)
MC2 (N = 68)
MC3 (N = 168)
MC4 (N = 15)
Comparisons of crude and adjusted mortality clustering
Male colon cancer
Male lung cancer
Female breast cancer
The results of the present study identified spatial clusters with high mortality rates of colon and lung cancer in men, and of breast cancer in women in Japan. The societal characteristics of the municipalities belonging to these clusters were determined by the relationships between mortality clusters and societal clusters. A single dominant societal cluster was detected for colon and breast cancer, although one mortality cluster was exclusive for each cancer. In contrast, we did not detect a dominant societal cluster for lung cancer.
The detection of a single dominant societal cluster for colon and breast cancer, SC1, suggested that there were homogeneous area characteristics for increased mortality due to these types of cancer. This societal cluster had a high Index 2 representing high income and education level and high population density, which were urban characteristics. These findings were consistence with those of a previous study indicating a positive relationship between mortality from these cancers and socioeconomic index of urbanisation . The relationship between mortality from colon and breast cancer and urban residence is plausible considering risk factors of these cancers, such as westernised dietary habits and low birth rate .
One mortality cluster for each of colon and breast cancer (MC2 and MC 3, respectively) showed different characteristics from other mortality clusters. It is possible that factors other than those related to urbanisation contributed to the increased mortality in these areas, and further studies are required to elucidate these unique factors. This observation suggests that the factors contributing to the increased mortality in these exceptional areas may be overlooked in conventional ecological studies.
Unlike colon and breast cancer, no dominant societal cluster was observed for lung cancer. The prevalence of smokers was not included in the set of indicators in the present study because municipal data concerning smoking were not available. It is possible that socioeconomic factors used in this study are surrogates of factors related to colon and breast cancers (e.g., dietary habits), while they might not be surrogates of smoking. The higher mortality in Hokkaido prefecture, as identified by MC2, could be explained by the slightly higher smoking rate reported in this area . However, the prefectural data of smoking did not found that other clusters were not related to areas with higher smoking rate . Previous studies showed small variation in male smoking rate and little relationship between smoking rate and regional socioeconomic conditions, and there was no correlation of male smoking rate with lung cancer mortality [17, 18]. Thus, it seems that the difference of smoking rate does not thoroughly explain the regional variation in lung cancer mortality rate, although there is no doubt in contribution of smoking to lung cancer, which is the leading cause of cancer deaths in Japanese men .
A number of possible contributors to increased mortality from lung cancer in addition to smoking have been reported . Air pollution is an important factor among these possible contributors, and the observation that several clusters of lung cancer were located in metropolitan areas may be explained by the increase in lung cancer due to air pollution. In Okinawa prefecture, local brand cigarettes with a higher tar yield and the prevalence of human papilloma virus infection were suspected to contribute to the increased mortality from lung cancer in this area [21, 22]. If multiple factors: i.e., smoking, air pollution and other specific local factors, contribute to the regional variation in lung cancer mortality, it is reasonable that no uniform characteristics of mortality clusters were detected in the present study.
We found a similarity of mortality clusters among three types of cancers. Three metropolitan areas (Tokyo, Osaka, and Nagoya) were detected as mortality clusters for all cancers. Urban areas recently show a decrease of the relative health level in Japan, and cancer mortality largely attributes to the decreased health level among urban populations [13–15]. Mortality from several types of cancers seems to be concurrently increased by factors related to urban areas such as health risk behaviour and fewer attendances in cancer screening [23, 24]. On the other hand, our findings suggested that the northern part including lots of rural-poor municipalities (SC3) appeared to be another area with higher mortality from some types of cancer. The possible causes of higher mortality in this area should be carefully investigated focusing on differences from those in urban areas.
Several methodological issues about mortality and societal clusters and their relationship should be mentioned. There are several alternative methods for mortality clustering such as Openshaw's and Begas and Newell's methods [25, 26]. Although the spatial scan statistic has been widely applied, some possible limitations remain, especially about setting of maximum spatial cluster size and detecting and meaning of the secondary clusters [25, 26]. The comparison of SMR mappings and mortality clusters might suggest that municipalities with higher mortality were not necessarily accumulated with circular shape, and thus the non-circular shaped spatial scan statistic  could detect more accurate mortality clusters. Due to the use of mortality, instead of incidence, the result in the present study could be influenced by not only cancer incidence but also regional differences in health care qualities and others. The incidence data from such as cancer registration could detect more accurate disease clusters in restricted local areas [28–30], but the incidence data of cancer across the country was not available in Japan.
The societal indicators used in this study were restricted. We used indicators that were demonstrated previously to be critically associated with health level [13, 15, 31], although some indicators of potential cancer risks may not have been included, especially with regard to health-related behaviour. Second, the societal clustering of municipalities was an important issue in the present study. In contrast to other countries [32–34], as there are no established area classifications or societal indices representing regional characteristics in Japan, we formulated societal indices and classified municipalities by the principle component analysis and the cluster analysis. Different combinations of indicators may result in different figures of societal clusters. Especially, the principle component analysis has been the subject of a variety of criticisms including sensitivity of indicator selection and meaning of the components extracted [35, 36], although it has been used to reduce socioeconomic indicators and to obtain one or a few composite index [29, 32, 37]. In addition, unlike mortality data, societal data were not treated by spatial statistics. Spatial methods such as using population potential  instead population density and data smoothing for unstability in the municipalities with small population could contribute to more accurate societal classifications of municipality.
The relation between societal characteristics and mortality was mainly examined using societal clusters and mortality clusters. Societal indices and population density showed the significant relation to mortality according to types of cancer, and they might be more sensitive than societal clusters. The statistical comparisons of societal indices and population density among mortality clusters showed significant differences for most pairs of mortality clusters (data not shown). Thus, the analyses with these variables appeared to be too sensitive to examine homogeneity and heterogeneity among mortality clusters. Since the number of societal clusters was arbitrary in the cluster analysis, an increase of the number of societal clusters would show more complicated variations in the societal characteristics among mortality clusters. Significantly, in the present study even when simple societal clustering was applied, both heterogeneity and homogeneity in societal characteristics among mortality clusters were observed. In addition, the comparison of mortality clusters before and after adjustment for societal characteristics quantitatively supported these heterogeneity and homogeneity.
The combination of spatial analysis and investigation of the relationships between mortality and societal factors revealed areas in Japan with higher mortality rates and their societal characteristics. The spatial clusters of colon and breast cancer showed homogeneous societal characteristics, with the exception of one cluster. However, the societal characteristics of clusters of lung cancer varied. The homogeneous characteristics of areas with higher mortality rates require strategies across the country or common between higher mortality areas, while exclusive clusters, such as those seen for colon and breast cancer, and variations in societal characteristics for lung cancer imply the need of strategies specific for selected areas with higher mortality.
Study unit and period
Local public entities in Japan are divided into two categories: the first consists of municipalities (i.e., cities, towns and villages), while the second consists of prefectures. All districts in the country belong to one of the municipalities and fall within the boundaries of one of the prefectures. Tokyo prefecture (Tokyo Metropolis) includes 23 special wards ("ku") in addition to cities, towns and villages. Twelve large cities (cities designated by ordinance), such as Osaka and Nagoya, consist of wards ("ku"). In 1995, there were a total of 3372 municipalities (23 Tokyo special ward cities, 127 wards of 12 cities designated by ordinance, 651 cities, 1994 towns and 577 villages) nested within 47 prefectures .
The study was performed from 1993 to 1998 during which time several municipalities were annexed or divided, and therefore the aggregated data from these municipalities could not be used. Thus, the final number of municipalities analyzed in the present study was 3360.
In this study, we examined the mortality rates of three high priority cancers: male lung and colon cancer and female breast cancer. Lung and colon cancer were the first and fourth leading causes of cancer death, respectively, in men, and breast cancer was the fourth one in women in the Japanese population in 1995 . The rates of colon and breast cancer have both increased steadily over the last several decades in Japan. Classification was based on the 9th and 10th versions of the International Classification of Diseases (ICD-9 in 1993–94 and ICD-10 in 1995 to 1998): colon cancer, ICD-9 153–154 and ICD-10 C18-C21; lung cancer, ICD-9 162 and ICD-10 C33-C34; and breast cancer, ICD-9 174 and ICD-10 C50 [40, 41].
As our focus was on premature mortality, which is more closely related to regional societal characteristics, we examined deaths in the population aged under 75 years old . The numbers of cause-specific deaths by municipality from 1993 to 1998 were compiled. The data regarding deaths in 1995 were excluded to avoid the influence of the Hanshin-Awaji earthquake . Total number of deaths during 5 years was 57,109 for colon cancer, 101,515 for lung cancer, and 32,290 breast cancer. The nationwide age-and cause-specific mortality rates and census municipal age-specific population in 1995 were used as data sources . The aggregated data using macrofiles of the vital statistics were drawn from a database of previous studies .
Municipal SMR was calculated and disease mapping was drawn. For calculation of SMR, the hierarchical Poisson regression analysis [13, 14, 43] was applied since this analysis could correct the unstability in mortality due to heterogeneity of population size: there was marked variation in the population size among municipalities, ranging from a few hundred to a few hundred thousand, and municipalities with a small population showed statistical fluctuation in mortality. The secondary medical care zone (SMCZ), which is defined by prefectural governments for medical care planning according to the Medical Service Law, was used as a higher level. There were 344 SMCZs across Japan in 1995, each of which consisted of neighbouring municipalities and covered a population of 300,000 on average. Bayesian standardized mortality ratio of municipalities was estimated using the iterative generalized least squares (IGLS) and the Markov chain Monte Carlo method . Relative risks (RRs) of societal indices, population density, and societal clusters for cancer morality were estimated using the hierarchical Poisson regression with IGLS. In addition to crude RRs, societal and population density were included in the model to estimate adjusted RRs. For societal clusters, SC1 was used as the reference category. The details of hierarchical Poisson regression are described in previous studies [13, 14, 44]
Mortality clusters: spatial scan statistic
The spatial scan statistic was used to detect and evaluate the statistical significance of spatial clusters. The details of the spatial scan statistic were reported previously [4–6, 36] and SaTScan ver. 4.0.3 http://www.satscan.org was used for the analysis. The numbers of deaths in each municipality were modelled as Poisson distributions. Under the null hypothesis, the expected number of deaths calculated using age-specific national mortality rates and the age-specific municipal population from the 1995 census  was proportional to the indirectly age-adjusted population at risk. An infinite number of circles were superimposed on the map, using the municipal centroid as the centre. The municipal centroid (latitude and longitude) was computed with the map of Japan (geographic coordinate system, GRS 1980; http://www.esrij.com) using ArcGIS 8.3 (ESRI Japan, Tokyo). The radii of the circles were set to vary continuously from zero to a maximum including at most 10% of the total population at risk, to obtain a certain number of potential clusters. The data for an entire circle contained different sets of neighbouring municipalities, and each circle represented a potential mortality cluster. For each circle, the likelihood was calculated for observing the number of deaths occurring within that circle, and the circle with the maximum likelihood was taken as the primary cluster. The distribution of maximum likelihood under the null hypothesis was evaluated using the Monte Carlo hypothesis testing set with 999 simulations. In addition to the primary cluster, the spatial scan statistic identified the secondary clusters, and ordered them according to the likelihood ratio test statistics. In the present study, secondary clusters were identified using no geographical overlap procedure and those with p-values of less than 0.05 were significant. Mortality clusters were mapped using ArcGIS 8.3 (ESRI, Japan).
Based on the findings of previous studies [13, 15, 31], seven socioeconomic indicators were chosen as potential factors related to mortality (Table 1). These indicators were obtained and calculated using the System of Social and Demographic Statistics consisting of governmental statistics including mainly census data . Unemployment rate reflected the percentage of unemployed persons aged 15–65 years in the total workforce. Educational level reflected the age-adjusted educational level, using the percentage of those who had graduated from college or a higher level among the population aged 20 and over, and was standardised by nationwide sex-and age-specific populations as for standardisation of age-adjusted mortality rate. Income per capita was calculated by aggregating the annual taxable income per household by municipality, and dividing it by the total municipal population.
To reduce the number of variables and identify dimensional societal factors, the principal component analysis with correlation matrix analysis and varimax rotation was performed. The principle components for which the correlation matrix eigenvalues were more than 1.0 were selected as significant dimensions. The component score for the extracted component was assigned to municipalities as a composite societal index: consequently two indices were obtained as shown in Table 1. Then, municipalities were classified into four societal clusters using the K-means cluster analysis with two societal indices and population density (log-transformed). The principle component analysis and the cluster analysis were performed using SPSS 11.0 (SPSS Inc., Chicago, IL, USA).
Relationships between mortality clusters and societal clusters
The relationships between societal characteristics and mortality clusters identified by the spatial scan statistic were examined by the cross-tabular analysis of mortality clusters and societal clusters.
Furthermore, cluster detections with the spatial scan statistic were performed adjusting for societal indices and population density or societal clusters to determine whether mortality clusters would change before and after adjustment for these variables. The risks of municipalities were calculated using RRs from the hierarchical Poisson regression and used as the adjustment file in the spatial scan statistic [29, 44].
This research was supported by Grant-in-Aid for Scientific Researches by the Japan Society for the Promotion of Science (Grant No. 14570326 and 16590497).
- Cromley EK, McLafferty SL: GIS and public health. 2002, New York. Guilford PressGoogle Scholar
- Ellioitt P, Wakefield JC, Best NG, Briggs DJ: Spatial epidemiology. 2000, New York: Oxford University PressGoogle Scholar
- Rosenberg MS, Sokal RR, Oden NL, DiGiovann D: Spatial autocorrelation of cancer in Western Europe. Eur J Epidemiol. 1999, 15: 15-22. 10.1023/A:1007559728848.PubMedView ArticleGoogle Scholar
- Kulldorff M, Feuer EJ, Miller BA, Freedman LS: Breast cancer clusters in the northeast United States: a geographic analysis. Am J Epidemiol. 1997, 146: 161-170.PubMedView ArticleGoogle Scholar
- Green C, Hoppa RD, Young TK, Blanchard JF: Geographic analysis of diabetes prevalence in an urban area. Soc Sci Med. 2003, 57: 551-560. 10.1016/S0277-9536(02)00380-5.PubMedView ArticleGoogle Scholar
- Jemal A, Kulldorff M, Devesa SS, Hayes RB, Fraumeni F: A geographic analysis of prostate cancer mortality in the United States, 1970–89. Int J Cancer. 2002, 101: 168-174. 10.1002/ijc.10594.PubMedView ArticleGoogle Scholar
- Diez Roux AV, Schwartz S, Susser E: Ecological variables, ecological studies, and multilevel studies in public health research. Oxford textbook of public health. Edited by: Detels R, McEwen J, Beaglehole R, Tanaka H. 2002, New York: Oxford University Press, 493-507.Google Scholar
- Morgentern H: Ecological study. Modern epidemiology. Edited by: Rothman KJ, Greenland S. 1998, Philadelphia: Lippincott Williams & Wilkins, 459-480.Google Scholar
- Uehara M, Takahashi M, Hoshuyama T, Pan G, Feng Y: Geographical correlation between ambient UVB level and mortality risk of leukemia in Japan. Environ Res. 2003, 92: 78-84. 10.1016/S0013-9351(03)00034-3.PubMedView ArticleGoogle Scholar
- Tominaga S, Kuroishi T: An ecological study on diet/nutrition and cancer in Japan. Int J Cancer. 1997, Suppl 10: 2-6. 10.1002/(SICI)1097-0215(1997)10+<2::AID-IJC2>3.0.CO;2-C.PubMedView ArticleGoogle Scholar
- Aihara H, Iki M: An ecological study of the relations between the recent high suicide rates and economic and demographic factors in Japan. J Epidemiol. 2003, 13: A56-61.View ArticleGoogle Scholar
- Nagata C: Ecological study of the association between soy product intake and mortality from cancer and heart disease in Japan. Int J Epidemiol. 2000, 29: 832-836. 10.1093/ije/29.5.832.PubMedView ArticleGoogle Scholar
- Fukuda Y, Nakamura K, Takano T: Municipal socioeconomic status and mortality in Japan: sex and age difference, and trends of 1973–1998. Soc Sci Med. 2004, 59: 2435-2445. 10.1016/j.socscimed.2004.04.012.PubMedView ArticleGoogle Scholar
- Fukuda Y, Nakamura K, Takano T: Cause-specific mortality differences across socioeconomic position of municipalities in Japan, 1973 to 1998: increased importance of injury and suicide to inequality for ages under 75. Int J Epidemiol. 2005, 34: 100-109. 10.1093/ije/dyh283.PubMedView ArticleGoogle Scholar
- Fukuda Y, Nakamura K, Takano T: Wide range of socioeconomic factors associated with mortality among cities in Japan. Health Promot Int. 2004, 19: 177-187. 10.1093/heapro/dah205.PubMedView ArticleGoogle Scholar
- Kanda A, Ojima T, Miura Y, et al: Prefectural variation in alcohol consumption, smoking, physical exercise, and obesity and the trend. Koseino Shihyo. 2002, 49 (15): 7-15.Google Scholar
- Fukuda Y, Nakamura K, Takano T: Socioeconomic pattern of smoking in Japan: income inequality and gender and age differences. Ann Epidemiol. 2005, 15: 365-372. 10.1016/j.annepidem.2004.09.003.PubMedView ArticleGoogle Scholar
- Asahi S, Watanabe M, Tajimi M, et al: Relationships between prefectural smoking and alcohol consumption rates and cause-specific mortality. Koseino Shihyo. 2003, 50 (1): 1-7.Google Scholar
- Health and Welfare Statitics Association: Kokumin ensei no doko. 2002, Tokyo: Health and Welfare Statistics AssociationGoogle Scholar
- Colditz GA, Atwood KA, Emmons K, et al: Harvard report on cancer prevention volume 4: Harvard Cancer Risk Index. Cancer Causes Control. 2000, 11: 477-488. 10.1023/A:1008984432272.PubMedView ArticleGoogle Scholar
- Wakai K, Ohno Y, Genka K, et al: Smoking habits, local brand cigarettes and lung cancer risk in Okinawa, Japan. J Epidemiol. 1999, 7: 99-105.View ArticleGoogle Scholar
- Nakazato I, Hirayasu T, Kamada Y, Tsuhako K, Iwamasa T: Carcinoma of the lung in Okinawa, Japan: with special reference to squamous cell carcinoma and squamous metaplasia. Pathology International. 1997, 47: 659-672.PubMedView ArticleGoogle Scholar
- Fukuda Y, Nakamura K, Takano T: Accumulation of health risk behaviors is associated with lower socioeconomic status and women's urban residence: a multilevel analysis in Japan. BMC Public Health. 2005, 5: 53-10.1186/1471-2458-5-53.PubMedPubMed CentralView ArticleGoogle Scholar
- Fukuda Y, Nakamura K, Takano T: Reduced likelihood of cancer screening among women in urban areas and with low socioeconomic status: a multilevel analysis in Japan. Public Health.Google Scholar
- Wakefield JC, Kelsall JE, Morris SE: Clustering, cluster detection, and spatial variation in risk. Spatial Epidemiology. Edited by: Elliot P, Wakefield JC, Best NG, Briggs DJ. 2000, NY: Oxford University Press, 128-152.Google Scholar
- Waller LA, Gotway CA: Spatial clustering of health events: regional count data. Applied Spatial Statistics for Public Health Data. 2004, New Jersey: John Wiley & Sons, 200-271.View ArticleGoogle Scholar
- Tango T, Takahashi K: A flexibly shaped spatial scan statistic for detecting clusters. Int J Health Geogr. 2005, 4: 11-10.1186/1476-072X-4-11.PubMedPubMed CentralView ArticleGoogle Scholar
- Buntinx F, Gey H, Lousbergh D, et al: Geographical differences in cancer incidence in the Belgian province of Limburg. Eur J Cancer. 2003, 39: 2058-2072. 10.1016/S0959-8049(02)00734-7.PubMedView ArticleGoogle Scholar
- Sheehan TJ, DeChelo LM, Kulldorff M, Gregorio DI, Gershman S, Mroszczyk M: The geographic distribution of breast cancer incidence in Massachusetts 1988 to adjusted for covariates. Int J Health Geogr. 2004, 3: 17-10.1186/1476-072X-3-17.View ArticleGoogle Scholar
- Kulldorff M, Nagarwalla N: Spatial disease clusters: detection and inference. Stat Med. 1995, 14: 799-810.PubMedView ArticleGoogle Scholar
- Takano T, Nakamura K: An analysis of health levels and various indicators of urban environments for Healthy Cities projects. J Epidemiol Community Health. 2001, 55: 263-270. 10.1136/jech.55.4.263.PubMedPubMed CentralView ArticleGoogle Scholar
- Krieger N, Chen JT, Waterman PD, Soobader MJ, Subramanian SV, Carson R: Geocoding and monitoring of US socioeconomic inequalities in mortality and cancer incidence: does the choice of area-based measure and geographic level matter?; The Public Health Disparities Geocoding Project. Am J Epidemiol. 2002, 156: 471-482. 10.1093/aje/kwf068.PubMedView ArticleGoogle Scholar
- House JM, Lepkowski JM, Williams DR, et al: Excess mortality among urban residents: how much, for whom, and why?. Am J Public Health. 2000, 90: 1898-1904.PubMedPubMed CentralView ArticleGoogle Scholar
- Morris R, Carstairs V: Which deprivation? A comparison of selected deprivation indexes. J Public Health Med. 1991, 13: 318-326.PubMedGoogle Scholar
- Armitage P, Berrry G, Matthews JNS: Mutivariate methods. Statitical Methods in Medical Research. 2002, Oxford: Blackwell Sceince, 455-484.View ArticleGoogle Scholar
- Kachigan SK: Factor analysis. Multvariate Statitical Analysis. 1991, New York: Radium, 236-260.Google Scholar
- Folwell K: Single measures of deprivation. J Epidemiol Community Health. 1995, 49: S51-S56.PubMedPubMed CentralView ArticleGoogle Scholar
- Middleton N, Gunnell D, Frankel S, Whitley E, Dorling D: Urban-rural differences in suicide trends in young adults: England and Wales, 1981–1998. Soc Sci Med. 2003, 57: 1183-1194. 10.1016/S0277-9536(02)00496-3.PubMedView ArticleGoogle Scholar
- Society for self-government of municipalities: Municipal handbook. 2000, Tokyo: Daiichi HokiGoogle Scholar
- Ministry of Health and Welfare: Vital statistics 1994. 1997, Tokyo: Health and Welfare Statistics AssociationGoogle Scholar
- Ministry of Health, Labour, and Welfare: Vital statistics 1995. 1997, Tokyo: Health and Welfare Statistics AssociationGoogle Scholar
- Statistics Bureau, Management and Coordination Agency: System of socioeconomic and demographic statistics. 2000, Tokyo: Statistical Information Institute for Consulting and AnalysisGoogle Scholar
- Leyland AH, McLeod A: Mortality in England and Wales, 1979–1992. 2000, Glasgow: MRC Social & Public Health Sciences UnitGoogle Scholar
- Rasbash J, Browne W, Goldstein H, et al: A user's guide to MLwiN. 2002, London: Centre for Multilevel Modeling, Institute of Education, University of LondonGoogle Scholar
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 cited.