Skip to main content
  • Research article
  • Open access
  • Published:

Does place of residence affect risk of suicide? a spatial epidemiologic investigation in Kentucky from 1999 to 2008

Abstract

Background

Approximately 32,000 people take their own lives every year in the United States. In Kentucky, suicide mortality rates have been steadily increasing since 1999. Few studies in the United States have assessed spatial clustering of suicides. The purpose of this study was to identify high-risk clusters of suicide at the county level in Kentucky and assess the characteristics of those suicide cases within the clusters.

Methods

A spatial epidemiological study was undertaken using suicide data for the period January 1, 1999 to December 31, 2008, obtained from the Kentucky Office of Vital Statistics. Descriptive analyses using Pearson's chi-square test and t-test were performed to determine whether differences existed in age, marital status, year, season, and suicide method between males and females, and between cases inside and outside high-risk spatial clusters. Annual age-adjusted cumulative incidence rates were also calculated. Suicide incidence rates were spatially smoothed using the Spatial Empirical Bayesian technique. Kulldorff's spatial scan statistic was applied on all suicide cases at the county level to identify counties with the highest risks of suicide. Temporal cluster analysis was also performed.

Results

There were a total of 5,551 suicide cases in Kentucky from 1999 to 2008, of which 5,237 (94%) were included in our analyses. The majority of suicide cases were males (82%). The average age of suicide victims was 45.4 years. Two statistically significant (p < 0.05) high-risk spatial clusters, involving 15 counties, were detected. The county level cumulative incidence rate in the most likely high-risk cluster ranged from 12.4 to 21.6 suicides per 100,000 persons. The counties inside both high-risk clusters had relative risks ranging from 1.24 to 1.38.

Conclusions

Statistically significant high-risk spatial clusters of suicide were detected at the county level. This study may be useful for guiding future research and intervention efforts. Future studies will need to focus on these high-risk clusters to investigate reasons for these occurrences.

Peer Review reports

Background

Between 2000 and 2007, there were 256,085 deaths in the United States (US) attributed to suicide. Approximately 32,000 people take their own lives every year in the US [1]. In 2009, suicide was the tenth leading cause of death for all ages, the second leading cause of death among 25-34-year olds, and the third leading cause of death among 15-24-year olds [1]. Firearms, suffocation, and poisoning are the most common methods of suicide; however, men and women differ in the methods used. In the same year, firearms were the most commonly used methods of suicide among males, while poisoning was the most commonly used mechanism in females. Males died by suicide at nearly four times the rate of females and represented 78.8% of all US suicides. During their lifetime though, women attempt suicide about two to three times as often as men [1].

In Kentucky, suicide mortality rates have been steadily increasing since 1999. In 2006, suicides rose to 14.4 per 100,000 persons from the 1999 rate of 11.3 per 100,000 persons, a 27% increase. With an average of 13.4 suicides per 100,000 people annually (2000-2006), Kentucky ranks 16th highest for suicide in the US [1]. Additionally, medical costs and lost wages associated with suicide also take their toll on communities. In 2005, suicide cost society $26.7 billion in combined medical and work loss costs, while in Kentucky it was estimated to cost $481 million [2].

A combination of demographic, individual, relational, community, and societal factors contribute to the risk of suicide. According to the World Health Organization's (WHO) report on violence and health, demographic factors such as age and sex, psychiatric, biological, social and environmental factors, as well as factors related to an individual's life history might play a role in making people more likely to attempt or commit suicide [3].

Although much is understood about suicide at the individual level, including multiple factors associated with increased risk of suicide [3], little has been done at the ecologic level to identify counties or neighborhoods with the greatest risk of suicide. Just as determining individual-level risk factors for suicide is vital for suicide prevention efforts, identifying high-risk areas and investigating spatial patterns for suicide provides a richer understanding of the determinants of suicide than the individual-level risk factors alone. Thus, identifying high-risk counties using spatial statistics may allow for a better targeting of resources and suicide intervention efforts so as to prevent future suicides.

Several studies have used spatial statistical techniques in assessing the presence of high-risk clusters including a brain cancer cluster study [4], a study on networks of sexually transmitted infection [5], and on breast cancer mortality disparities [6]. Other studies incorporating similar methodologies have assessed high-risk clusters of La Crosse virus in West Virginia [7], clusters of giardiasis in Canada [8], and clustering of lung cancer in Italy [9]. However, spatial studies of suicide clusters have been limited. Exeter and Boyle (2007) found a significant geographical cluster of suicide among young adults in east Glasgow across three time periods (1980 to 1982, 1990 to 1992, and 1999 to 2001), which were attributed to socioeconomic deprivation [10]. Another study investigating suicide clusters in Queensland, Australia, found clusters in low socioeconomic areas [11]. These studies provide some support that regions at high risk for suicide are those with greater socioeconomic deprivation.

Though suicide studies have used spatial statistical techniques in other countries, little has been published in the US. The present study has the potential to bridge the gap between suicide research and targeted prevention. The primary purpose of this study was to identify counties at the highest risk for suicide. Secondarily, this study also tests whether suicides are clustered temporally. Thus, the current study provides an analytical framework, using spatial statistics, to identify and target areas with the highest risk of suicide. Further, identification of counties at a greater risk for suicide is expected to guide resources and assist in policy decision making at the county level.

Methods

Study design

This is an ecological study of suicides in Kentucky counties (n = 120) from January 1, 1999 to December 31, 2008. All the cases of suicide identified in the study occurred in Kentucky. This research was approved by the University of Kentucky Institutional Review Board (IRB #: 02-0441-p6h).

The Kentucky Office of Vital Statistics provided electronic injury-related death certificate data files from January 1, 1999 to December 31, 2008. A subset of the data was then generated using The International Classification of Diseases, 10th revision (ICD-10) External Causes of Death Codes (X60-X84, Y10-Y34, Y87.0), which meet the Center's for Disease Control and Prevention's (CDC) National Violent Death Reporting System's definition of suicide [12]. Data elements included gender, age, marital status, date of suicide, county of suicide, and suicide method. Several exclusion criteria were applied: 1) cases were first excluded if the underlying cause of death did not match the manner of death; 2) from this set, cases with unknown county of death were excluded; 3) thirdly, any cases where age was unknown were excluded; 4) lastly, cases where the date of death was unknown were also excluded (see Exclusion Criteria in Figure 1). After these cases were excluded, the remaining data set did not have any missing values.

Figure 1
figure 1

Exclusion criteria used in selection of cases for this study.

Descriptive statistical analyses

We used Pearson's chi-square test and two-tailed unequal variance t-tests to determine whether there were significant differences between cases by gender for age, marital status, year, season, and suicide method. Pearson's chi square tests and t-tests were calculated in WinPepi v11.4 [13] and SAS v9.3 (SAS Institute Inc, Cary, NC) [14], respectively.

Age adjusted male and female cumulative incidence rates were directly adjusted using the decennial 2000 United States Census Population. Cumulative incidence rates were expressed as the number of suicides per 100,000 persons for all of Kentucky from 1999 to 2008, and calculated in Microsoft Excel 2007 [15].

Geographic analysis

Smoothed cumulative incidence suicide rates were mapped at the county level (n = 120). The heterogeneity of variances and spatial autocorrelation of suicide rates were adjusted for by smoothing using Spatial Empirical Bayesian (SEB) rate smoothing [16]. This technique is appropriate when population sizes of areas of aggregation vary and there is spatial autocorrelation in the data. Given that population sizes vary by county, suicide rates from counties of small population have greater variance than counties with higher populations [17]. This technique was implemented in GeoDa v0.95i [18] using a first-order queen contiguity spatial weights matrix [19].

The geographic boundary file used in this study was downloaded from the United States Census, TIGER, Geodatabase [20]. ArcGIS v10 (ESRI, Inc, Redlands, CA.) [21] was used to create the cartographic displays of smoothed incidence rates, with single hue color schemes generated by http://ColorBrewer.org[22]. The SEB smoothed incidence rates were grouped into five classes/categories using the Jenks natural breaks optimization algorithm [23].

Spatial and temporal scan statistical cluster analyses

The detection of high-risk local spatial clusters of suicide was performed using Kulldorff's 2-dimensional spatial scan statistic [24], and implemented in SaTScan v9.1 [25]. The advantages of using the scan statistic include controlling for covariates, adjusting for multiple comparisons and various population sizes, and limiting preselection bias by not specifying a priori the observed set of cases within a cluster [24, 26, 27]. Significance testing is performed using Monte Carlo simulation where the null hypothesis of no cluster is rejected at an α level of 0.05 if the simulated p-value is less than or equal to 0.05 [28]. A significant high-risk cluster is interpreted as having an increased risk of suicide within the circular window relative to outside [27]. We also used Pearson's chi-square tests and two-tailed unequal variance t-tests to determine whether the characteristics of suicide victims that fell within the identified spatial clusters were significantly different from the state distribution. For the purely spatial analysis, a discrete Poisson probability model was used to scan for non-overlapping geographical areas (counties) with statistically significant high rates of suicide, using a maximum spatial cluster size of 50% of the total population at risk, as recommended by Kulldorff [24].

Further, high-risk temporal trends were investigated using the temporal scan statistic [25]. The temporal scan statistic uses a window that moves in one dimension, time, to identify a time period where suicide risk was higher than outside that time period. For the purely temporal analysis, we used a discrete Poisson probability model using a maximum temporal cluster size of 1-year and a 1-year time aggregation.

For statistical inference, 19,999 Monte Carlo replications were performed for all analyses, and the null hypothesis of no clusters was rejected at a probability (p) value of ≤ 0.05. All analyses were implemented in SatScan and adjusted for age (10-14, 15-19, 20-24,... ≥ 85) and gender distributions at the county level for each individual year from 1999 to 2008 using population estimates derived from the US Census Bureau's Population Estimates Program obtained from the Surveillance, Epidemiology, and End Results (SEER) Program at the National Cancer Institute [29]. Cartographic displays of spatial clusters were made using ArcGIS 10 [21].

Results

Descriptive epidemiology of suicide mortality

Of the 5,551 suicide deaths abstracted from death certificates from the Kentucky Office of Vital Statistics between 1999 and 2008, 5,237 were used in our analyses after applying our exclusion criteria (see Exclusion Criteria in Figure 1). The excluded cases where the underlying cause of death did not match the manner of death (153 cases) appeared to be differentially distributed by year, where 128 (84%) of those excluded cases occurred in 2000 and 2001. Among the other excluded cases, no apparent differential patterns were present.

Differences in age, marital status, year, seasons, and suicide method among suicide victims by gender are shown in Table 1. There were 4,313 (82%) male suicide cases. The average age of suicide victims was 45.4 years, with significant differences between males (45.6 years old) and females (44.3 years old). Among age categories, males and females also showed significant differences (p < 0.001). There were also significant differences by marital status, as 31% of females were divorced versus 22% of males (p < 0.001). Overall, 42% of all cases were married, 26% were single or never married, 24% were divorced, and 7% were widowed. Males and females differed significantly in the suicide method (p < 0.001), where 34.1% of female victims used self poisoning versus 7.7% of male victims. Male suicide victims used firearms with a greater frequency than females (71.1% vs 48.8%). Overall, 3,517 (67.2%) suicides were firearm related.

Table 1 Characteristics of suicide cases in Kentucky, 1999-2008

Figure 2 displays age adjusted male and female cumulative suicide incidence rates by year along with total suicides by year. The average male cumulative incidence rate from 1999 to 2008 was 20.9 per 100,000 persons, while the average female cumulative incidence rate was 4.3 per 100,000 persons.

Figure 2
figure 2

Age adjusted male and female cumulative incidence by year in Kentucky, 1999-2008*. *Average male cumulative incidence: 20.9 per 100,000 persons. Average female cumulative incidence: 4.3 per 100,000 persons.

Spatial distribution of suicide

The annual median SEB smoothed rate was 12.7 per 100,000 persons (range: 1-20.9) (Figure 3). The counties with the highest SEB suicide mortality rates (greater than 17.6 suicides per 100,000) included Carlisle (17.7 suicides per 100,000), Hickman (17.9 suicides per 100,000), Ballard (18.4 suicides per 100,000), Wolfe (20.7 suicides per 100,000), and Gallatin (20.9 suicides per 100,000) (Figure 3).

Figure 3
figure 3

The smoothed cumulative incidence of suicide at the county level. The distribution of spatial empirical Bayesian (SEB) smoothed suicide incidence rates at the county level for Kentucky between 1999 and 2008.

Spatial and temporal clusters of suicide

Two significant (p < 0.05) high-risk spatial clusters of suicide were identified (Table 2 & Figure 4). The most likely high-risk spatial cluster of suicide comprised seven counties (Ballard, Carlisle, Graves, Hickman, Livingston, Marshall, and McCracken). Populations within the counties of the most likely cluster are at a 38% greater risk (relative risk, RR = 1.38; p = 0.0029) of suicide than outside the cluster (Table 2 & Figure 4). Additionally, the most likely high-risk spatial cluster had an estimated 19.8 suicides per 100,000 persons annually. A secondary high-risk spatial cluster was found among eight counties (Breckinridge, Daviess, Grayson, Hancock, Henderson, McLean, Meade, Ohio). This cluster had a 24% increased risk (RR = 1.24, p = 0.0324) of suicide than outside the cluster. This secondary cluster of high risk had an estimated 17.8 suicides per 100,000 persons annually (Table 2 & Figure 4).

Table 2 Purely spatial and temporal significant clusters of suicides in Kentucky counties, 1999-2008
Figure 4
figure 4

Significant spatial clusters of high suicide mortality risks from 1999 to 2008 in Kentucky. These maps show the statistically significant (p < 0.05) high-risk spatial clusters of suicide in Kentucky (A) at the county level (B) from 1999-2008 detected by Kulldorff's spatial scan statistic, controlling for age and gender. RR = relative risk.

The purely temporal cluster analysis revealed that 2006 experienced a greater than expected number of suicides, with a marginal significantly higher risk of 12% compared to the other years (RR = 1.12, p = 0.04415) (Table 2).

Descriptive epidemiology of high risk spatial clusters

Table 3 compares the most likely and secondary spatial clusters to the rest of Kentucky. Only variables that had significant differences inside versus outside clusters are included in the table. The average age within the most likely spatial cluster was statistically different from the rest of Kentucky (48.9 yrs vs. 45.2 yrs) (p < 0.01), as was suicide cases by age categories (p = 0.02). This was also the case for the average age of males within the most likely cluster (49.3 yrs) versus outside the cluster (45.4 yrs) (p < 0.01). Also, 19% of suicides within the secondary spatial cluster used poisoning as the method of suicide, while only 12% used poisoning outside the cluster (p < 0.001). Though average age was not significantly different within versus outside the secondary cluster, suicide cases by age categories for within versus outside the cluster was significant (p = 0.009).

Table 3 Descriptive characteristics of suicide cases inside and outside both significant spatial high-risk suicide clusters

The cumulative incidence inside the most likely cluster was 17.6 per 100,000 persons (range 12.4-21.6) while outside the cluster the rate was 12.4 per 100,000 persons (range 0-27.6). The cumulative incidence inside the secondary high-risk spatial cluster was 14.5 per 100,000 persons (range 10.2-18.2) (Table 3).

Discussion

We investigated the spatial epidemiology of suicides in Kentucky as reported in death certificate data files from the Kentucky Office of Vital Statistics between 1999 and 2008 using scan statistics and descriptive epidemiological methods. The results show evidence of hotspots of suicides across Kentucky counties, and describe the differences in suicide characteristics between genders, and for cases inside and outside high-risk spatial clusters. Further, when the purely spatial cluster analysis, SEB map, and cumulative incidence rates are jointly examined, they bolster the evidence of the existence of a high-risk suicide cluster in western Kentucky. To our knowledge this is the first study to investigate spatial and temporal patterns of suicide mortality at the county level in the US. This study allows for a better understanding of where to target resources and prevention efforts at the county level to reduce the burden of suicide in areas of greatest risk [30, 31].

Moreover, suicide victim characteristics within the two spatial clusters allow for more targeted prevention efforts. For example, there is a moderately older age group of victims in the most likely cluster, and a greater proportion of 40-44 year old suicide victims inside the secondary cluster compared to outside the cluster. In addition, suicide cases inside the secondary cluster are more likely to self-poison than outside the cluster. Although our results reveal that females are more likely to self-poison than males (and this agrees with national data) [1], there were not significantly more female suicides inside the secondary cluster than outside. These divergent characteristics of suicide cases inside versus outside clusters can inform interventions and guide future studies. Suicide risk was also found to be highest in 2006, providing evidence that suicide has been gradually increasing in Kentucky since 1999.

Our study differs from Exeter and Boyle's (2007) Scotland suicide cluster analysis in several ways [10]. Exeter and Boyle's study, which was limited to young adult suicides (15 to 44 years old), found high risk clusters in Glasgow, Scotland, across three time periods with relative risks ranging from 1.53 to 2.41, while our study revealed clusters with relative risks ranging from 1.24 to 1.38. Unlike our study, where 67% of suicides were firearm related (65% for adults 15 to 44 years old), Exeter and Boyle found that 63% of suicides among young adults were from poisonings, hangings, strangulation and suffocation. In addition, Scotland's suicide rate among adults aged 15 to 44 years in 1999 to 2001 was 24.3 per 100,000 persons, which was markedly higher than Kentucky's suicide rate of 13.5 per 100,000 persons among adults aged 15 to 44 years during the same period [1]. This difference in suicide method and rates suggests that suicide may be better explained by other factors-such as low socioeconomic status [3, 10], less access to mental health care [32], sudden unemployment [33], depression [34], and alcoholism [35]-than by firearm access alone [36], given that Scotland's suicide rate is higher than Kentucky's despite there being lower access to firearms in Scotland than Kentucky [37]. Thus, this comparison suggests that firearm availability may not be the sole driving force behind increased regional suicide rates. Both studies reveal that suicide does not occur randomly in space, and that the characteristics of suicide cases inside clusters tend to be different than outside; specifically in Scotland where suicide clusters have been explained by the concentration of socioeconomic deprivation [10].

Strengths, limitations, and future research

In addition to the previously mentioned advantages of using scan statistics over other epidemiology methods, scan statistics do not assume that observations are spatially independent, but rather test for spatial randomness, or in other words, spatial independence of the observations. An additional strength of this study is that it uses novel spatial techniques to provide ecological information on suicide risk, thus having the potential to guide interventions in those high-risk counties.

This study used data from death certificates which can be affected by errors. Pierce and Denison (2006) assessed place-of-residence errors on death certificates in only two Texas counties and found a 14% error rate in recording county of residence for deaths [38]. Within our dataset, cases where the underlying cause of death and the manner of death did not match were excluded. Further, excluded cases due to this discrepancy came mainly from 2000 and 2001. This differential distribution in excluded cases may have biased the temporal analysis results away from the null hypothesis. Although 2006 has the highest suicide incidence rate, this introduced bias may have inflated that excess risk because of the seemingly lower rates in 2000 and 2001. Additionally, suicide mortality may not reflect current prevention needs. This study also used county-level data, which does not discriminate among suicide mortality risk in different parts of the county.

We are inclined to recommend future studies be undertaken in the high-risk counties to identify reasons for the high rates observed. Although Exeter and Boyle's (2007) study found high-risk suicide clusters to be explained by greater social deprivation [10], implying that greater suicide risk is dependent on a relatively stable regional risk factor, several interventions such as physician education in depression recognition and treatment, along with greater restriction of access to lethal methods have been shown to be effective in reducing suicide rates [39]. It is important to note that since suicide is affected by sociocultural factors, effective interventions in a certain population may not work elsewhere [40]. Regardless, our study does offer evidence to support increasing availability and access to mental health care facilities, and targeted prevention efforts across the identified high-risk clusters in western Kentucky. Given that suicide is highly associated with poor mental health and depression [3, 41], it is appropriate to make mental health facilities more available to provide services to those populations that most need it. We believe that future studies assessing suicide risk may provide more insight into regional-specific interventions that are most appropriate for western Kentucky.

In guiding future suicide spatial research, we recommend that individual-level circumstance data from the National Violent Death Reporting System (NVDRS) in Kentucky be linked to socioeconomic and demographic data, and vital statistics data to offer a richer understanding of those persons who define the cluster. It is also recommended that spatial analyses be performed using NVDRS suicide data at the census-tract level to allow comparisons between the county-level analysis and to offer a lower level (i.e., finer) visualization of suicide risk. This study also guides future small area analysis research: i.e., given that two high-risk suicide clusters have been identified, we recommend a spatial analysis limited to each cluster, with those high-risk counties divided into smaller census-tracts to identify those tracts within high-risk counties that are at greatest risk for suicide. Moreover, a further investigation determining the factors associated with high-risk clusters of suicide is recommended using regression analysis by linking various sociodemographic and environmental county (or census-tract) characteristics to the vital statistics data. This approach would offer an ecological understanding of the county-level (or census-tract) characteristics that explain suicide risk. We hypothesize that similar results may be found in Kentucky as in Glasgow, UK, and Queensland, Australia, where socioeconomic deprivation has been associated with high-risk suicide clusters [10, 11]. Without more rigorous regression analyses, however, we would only be speculating as to what explains high-risk county clusters of suicide in Kentucky.

Beyond regional social deprivation, several other factors may be contributing to the higher rates of suicide in these relatively rural regions (i.e., all seven counties in the most likely cluster are rural, and three out of eight counties in the secondary cluster are rural) [42]. Specifically, rurality has been found to be a likely regional risk factor for suicide in the United States and Australia [43, 44]. Rural populations have less access to mental health care facilities, putting those with mental health disorders at a greater risk for suicide than their urban counterparts [44, 45]. Furthermore, data collected from the Kentucky Violent Death Reporting System (KVDRS) from 1999-2008 suggest that county-level unemployment may be statistically associated with higher rates of suicide (Sabrina Walsh, unpublished data).

Conclusion

This study found high-risk clusters of suicide in western Kentucky and demonstrated the usefulness of the combination and complementary nature of spatial statistics, cumulative incidence, relative risk, and SEB smoothed rates in identifying areas at highest risk for suicide. The findings can guide intervention efforts at the county level in Kentucky to reduce suicide in those areas at greatest risk.

Abbreviations

US:

United States

WHO:

World health organization

IRB:

Institutional review board

ICD-10:

International classification of diseases: 10th revision

CDC:

Center's for disease control and prevention

SEB:

Spatial empirical Bayesian

SEER:

Surveillance: epidemiology: and end results

NVDRS:

National violent death reporting system.

References

  1. Kochanek KD, Xu J, Murphy SL, Minino AM, Kung HC: Deaths: preliminary data for 2009. National Vital Statistics Reports. Centers for Disease Control and Prevention. 2011, 59: 1-51.

    Google Scholar 

  2. Centers for Disease Control and Prevention (CDC): Web-based Injury Statistics Query and Reporting System (WISQARS). Data & Statistics (WISQARS™): Cost of Injury Reports. 2005, Retrieved August 8, 2011, from http://wisqars.cdc.gov:8080/costT/

    Google Scholar 

  3. Krug EG, Dahlberg LL, Mercy JA, Zwi AB, Lozano R: World report on violence and health. Self-directed violence. 2002, Geneva, Switzerland: World Health Organization

    Google Scholar 

  4. Kulldorff M, Athas WF, Feuer EJ, Miller BA, Key C: Evaluating cluster alarms: a space-time scan statistic and brain cancer in Los Alamos, New Mexico. Am J Public Health. 1998, 88: 1377-1380.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  5. Wylie JL, Cabral T, Jolly AM: Identification of networks of sexually transmitted infection: a molecular, geographic, and social network analysis. J Iinfect Dis. 2005, 191 (6): 899-906.

    Article  Google Scholar 

  6. Hsu CE, Jacobson H, Mas FS: Evaluating the disparity of female breast cancer mortality among racial groups - a spatiotemporal analysis. Int J Health Geogr. 2004, 3: 4-

    Article  Google Scholar 

  7. Haddow AD, Bixler D, Odoi A: The spatial epidemiology and clinical features of reported cases of La Crosse Virus infection in West Virgina from 2003 to 2007. BMC Infect Dis. 2011, 11: 29-

    Article  PubMed  PubMed Central  Google Scholar 

  8. Odoi A, Martin SW, Michel P, Holt J, Middleton D, Wilson J: Geographical and temporal distribution of human giardiasis in Ontario, Canada. Int J Health Geogr. 2003, 2 (1): 5-

    Article  PubMed  PubMed Central  Google Scholar 

  9. Bilancia M, Fedespina A: Geographical clustering of lung cancer in the province of Lecce, Italy: 1992-2001. Int J Health Geographics. 2009, 8: 40-

    Article  Google Scholar 

  10. Exeter DJ, Boyle PJ: Does young adult suicide cluster geographically in Scotland?. J Epidemiol Community Health. 2007, 61 (8): 731-736.

    Article  PubMed  PubMed Central  Google Scholar 

  11. Qi X, Tong S, Hu W: Spatial distribution of suicide in Queensland, Australia. BMC Psychiatry. 2010, 10: 106-

    Article  PubMed  PubMed Central  Google Scholar 

  12. Centers for Disease Control and Prevention (CDC): Uniform definitions for self-directed violence. 2011, Available at http://www.cdc.gov/ViolencePrevention/pub/selfdirected_violence.html. Accessed August 20, 2011

    Google Scholar 

  13. Abramson JH: WINPEPI updated: computer programs for epidemiologists, and their teaching potential. Epidemiologic Perspectives & Innovations. 2011, 8: 1-

    Article  Google Scholar 

  14. SAS Institute: SAS 9.3. 2011, Cary, North Carolina, USA: SAS Institute Inc

    Google Scholar 

  15. Microsoft Cooporation: Microsoft Office Excel 2007. Microsoft Office, USA

  16. Waller L, Gotway C: Mapping smoothed rates and probabilities. Applied Statial Statistics for Public Health Data. 2004, Hoboken, New Jersey: John Wiley & Sons, Inc, 86-104.

    Book  Google Scholar 

  17. Clayton D, Bernardinelli L: Bayesian methods for mapping disease risk. Geographical and Environmental Epidemiology. Edited by: Elliott CJ, English D, Stern R. 1997, Oxford: Oxford University Press, 206-220.

    Google Scholar 

  18. Anselin L: GeoDa Version 0.95i. 2003, IL, USA: Spatial Analysis Lab, University of Illinois Urbana-Champaign

    Google Scholar 

  19. Anselin L, Syabri I, Kho Y: GeoDa: an introduction to spatial data analysis. Geogr Anal. 2006, 38 (1): 5-22.

    Article  Google Scholar 

  20. US Census Bureau: Census TIGER/Line® Shapefiles. 2010, Available at http://www.census.gov/cgi-bin/geo/shapefiles2010/main. Accessed May 20, 2011

    Google Scholar 

  21. ESRI: ArcInfo GIS v10. 2011, Redlands, CA, USA: Environmental Systems Research Institute, Inc

    Google Scholar 

  22. Brewer CA: ColorBrewer 2.0. Available at http://colorbrewer2.org. Accessed May 20, 2011

  23. Jenks G: The data concept in statistical mapping. The International Yearbook of Cartography. 1967, 7: 186-190.

    Google Scholar 

  24. Kulldorff M: A spatial scan statistic. Communications in Statistics: Theory and Methods. 1997, 26: 1481-1496.

    Article  Google Scholar 

  25. Kulldorff M: SatScan v9.1.1: Software for the spatial and space-time scan statistics. 2011, Information Management Services, Inc, Silver Spring, Maryland, USA

    Google Scholar 

  26. Kulldorff M, Athas WF, Feuer EJ, Miller BA, Key CR: Evaluating cluster alarms: a space-time scan statistic and brain cancer in Los Alamos, New Mexico. Am J Public Health. 1998, 88: 1377-1380.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Kulldorff M, Nagarwalla N: Spatial disease clusters: detection and inference. Stat Med. 1995, 14: 799-810.

    Article  CAS  PubMed  Google Scholar 

  28. Dwass M: Modified randomization tests for non-parametric hypothesis. Ann Math Stat. 1957, 28: 181-187.

    Article  Google Scholar 

  29. US Census Bureau Population Estimates Program: Surveillance, Epidemiology, and End Results (SEER) Program at the National Cancer Institute: US Population Data 1969-2009. 2011, Available at http://seer.cancer.gov/popdata/. Accessed May 20, 2011

    Google Scholar 

  30. Holowaty EJ, Norwood TA, Wanigaratne S, Abellan JJ, Beale L: Feasibility and utility of mapping disease risk at the neighbourhood level within a Canadian public health unit: an ecological study. Int J Health Geogr. 2010, 9: 21-

    Article  PubMed  PubMed Central  Google Scholar 

  31. Odoi A, Martin SW, Michel P, Middleton D, Holton J, Wilson J: Investigation of clusters of giardiasis using GIS and a spatial scan statistic. Int J Health Geogr. 2004, 3: 11-

    Article  PubMed  PubMed Central  Google Scholar 

  32. Tondo L, Albert MJ, Baldessarini RJ: Suicide rates in relation to health care access in the United States: an ecological study. J Clin Psychiatry. 2006, 67 (4): 517-523.

    Article  PubMed  Google Scholar 

  33. Beautrais AL, Joyce PR, Mulder RT: Unemployment and serious suicide attempts. Psychol Med. 1998, 28: 209-218.

    Article  CAS  PubMed  Google Scholar 

  34. Beck AT, Brown G, Berchick RJ, Stewart BL, Steer RA: The relationship between hopelessness and ultimate suicide: a replication with psychiatric outpatients. Am J Psychiatry. 1990, 147: 190-195.

    Article  CAS  PubMed  Google Scholar 

  35. Murphy GE, Wetzel RD: The life-time risk of suicide in alcoholism. Arch Gen Psychiatry. 1990, 47: 383-392.

    Article  CAS  PubMed  Google Scholar 

  36. Kates DB, Mauser G: Would banning firearms reduce murder and suicide? a review of international and some domestic evidence. Harvard Journal of Law & Public Policy. 2007, 30 (2): 649-694.

    Google Scholar 

  37. Handguns to be banned in the UK: BBC on this day: October 16. 1996, Available at: http://news.bbc.co.uk/onthisday/hi/dates/stories/october/16/newsid_3110000/3110949.stm. Accessed October 7, 2011

    Google Scholar 

  38. Pierce RJ, Denison AV: Place-of-residence errors on death certificates for two contiguous U.S. counties. Popul Health Metr. 2006, 4: 6-

    Article  PubMed  PubMed Central  Google Scholar 

  39. Mann JJ, Apter A, Bertolote J, Beatrais A, Currier D, Haas A, et al: Suicide prevention strategies: a systematic review. JAMA. 2005, 294 (16): 2064-2074.

    Article  CAS  PubMed  Google Scholar 

  40. Bertolote JM: Suicide prevention: at what level does it work?. World Psychiatry. 2004, 3 (3): 147-151.

    PubMed  PubMed Central  Google Scholar 

  41. Blumenthal SJ: Suicide: a guide to risk factors, assessment, and treatment of suicidal patterns. Med Clin North Am. 1988, 72: 937-971.

    Article  CAS  PubMed  Google Scholar 

  42. United States Department of Agriculture, Economic Research Service: Rural-urban continuum codes for KY. 2003, Available at http://www.ers.usda.gov/data/ruralurbancontinuumcodes/2003. Accessed December 26, 2011

    Google Scholar 

  43. Centers for Disease Control and Prevention (CDC): Web-based Injury Statistics Query and Reporting System (WISQARS). National Center for Injury Prevention and Control. Fatal injury mapping, United States Counties. 2000, Retrieved December 26, 2011, from http://wisqars.cdc.gov:8080/cdcMapFramework/mapModuleInterface.jsp, -2006

    Google Scholar 

  44. Singh GP, Siahpush M: Increasing rural-urban gradients in US suicide mortality, 1970-1997. Am J Public Health. 2002, 92: 1161-1167.

    Article  PubMed  PubMed Central  Google Scholar 

  45. Caldwell TM, Jorm AF, Dear KBG: Suicide and mental health in rural, remote and metropolitan areas in Australia. Med J Aust. 2004, 181: S10-S14.

    PubMed  Google Scholar 

Pre-publication history

Download references

Acknowledgements

This study was supported by Cooperative Agreement CE09-904 from the Centers for Disease Control and Prevention. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of CDC. We would also like to thank Amanda Thaxton for assisting with data acquisition.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Agricola Odoi.

Additional information

Competing interests

The authors declare that they have no competing interests.

Authors' contributions

DMS conceived the research idea, study design, performed all analyses, managed data, interpreted results, and wrote the manuscript. AO was involved in study design, results interpretation, and editing of the manuscript. AB drafted the Introduction and was involved in data acquisition. SW was involved in data acquisition, results interpretation, as well as review and editing of the manuscript. All authors certify that they have participated sufficiently in the research to believe in its overall validity and have read and approved the final manuscript.

Authors’ original submitted files for images

Rights and permissions

Open Access This article is published under license to BioMed Central Ltd. This is an Open Access article is distributed under the terms of the Creative Commons Attribution License ( https://creativecommons.org/licenses/by/2.0 ), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Reprints and permissions

About this article

Cite this article

Saman, D.M., Walsh, S., Borówko, A. et al. Does place of residence affect risk of suicide? a spatial epidemiologic investigation in Kentucky from 1999 to 2008. BMC Public Health 12, 108 (2012). https://doi.org/10.1186/1471-2458-12-108

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/1471-2458-12-108

Keywords