Spatial clustering of suicide mortality and associated community characteristics in Kanagawa prefecture, Japan, 2011–2017

Among the leading causes of death, suicide was recognized as a critical public health issue by the World Health Organization (WHO) in its Comprehensive Mental Health Action Plan [1, 2]. Although substantial reductions in suicide have been detected globally, suicide remains a leading cause of years of life lost in many areas in Japan [3]. Suicide mortality in Japan was among the ten worst in the world and the second highest among Asian countries following South Korea. Therefore, suicide prevention remains an important health issue due to the magnitude of its impact [4]. Factors related to suicide deaths and global action to prevent suicide were summarized by WHO [2]. However, suicide mortality has remarkable heterogeneity in trends across countries and in demographic subgroups such as those based on sex, age, and other factors that warrant further investigation. Early interventions to solve this problem by taking action according to the local situation should be developed.

To provide the accurate regional status of current suicide mortality would serve as a useful tool for evaluations by communities as well as political decision making regarding comprehensive suicide prevention. For these purposes, spatial statistics are widely used to detect geographical disease clusters according to different types of data [5, 6]. To investigate space-time clustering of suicide mortality, many different statistical tests have been proposed to determine whether the suicide risk is relatively high compared with that in surrounding regions or in subsequent time periods [5]. Kulldorff’s spatial scan statistic [7] assuming circular clusters along with Spatial, temporal, or space-time scan (SaTScan) software [8] and Tango and Takahashi’s flexible spatial scan statistic [9, 10] assuming non-circular clusters along with Flexible scan (FlexScan) software [11] have been utilized in a wide variety of epidemiological studies and disease surveillance. Tango’s test [12, 13] has been used for detecting disease clustering in many epidemiological studies. The specific characteristics of the model or test may result in differences in the detection of spatial-clusters; therefore, it is worthwhile to use several methods in such examinations.

As for the application related to spatial epidemiology, there have been several reports of suicide cluster analyses worldwide [14‐19]. Namely, a study in New South Wales, Australia, between 2005 and 2013 [14] demonstrated the importance of examining geographical variations in spatial clustering of fatal and non-fatal suicide attempts. A study in Central Brazil during 2000–2010 [15] showed spatial-temporal trends and risk of suicide; a study in Sergipe, Brazil, [16] performed spatial analysis and detected temporal trends in suicide mortality for the period from 2000 to 2015; in Kentucky a spatial epidemiologic investigation.

was performed from 1999 to 2008 [17]; and in Idaho in the US [18] spatial clustering of suicide and associated community characteristics for the period 2010 to 2014 were reported. Also, suicide trends were analyzed in Scotland from 1950 to 2014 in comparison with England & Wales [19] and identified ‘vulnerable’ cohorts for providing opportunities to develop suicide prevention strategies. Despite these studies, thus far, detailed analyses to reveal factors affecting suicide death rates as well as differences between the sexes have not been conducted.

In this study, we focused on suicide deaths in Kanagawa prefecture, located next to Tokyo and that has the second largest population in Japan. Although the number of suicides in Kanagawa prefecture has tended to decrease in recent years, it was still the fourth highest in Japan in 2017. Based on the recognition that many suicides can be prevented by social efforts [2], various background issues related to suicide and causative factors in society have been examined from a multi-faceted perspective. Use of surveys and analyses in line with conditions within a specific community are required [4]. In particular, it is warranted to analyze the actual situation of suicides through the use of vital statistics and to summarize and provide the results of these analyses according to individual municipalities.

There has been no previous study that examined statistically significant clusters of suicide mortality in Kanagawa prefecture, Japan, including analyses to adjust for covariates using regional vital statistics. The detection of clusters may be highly useful in surveillance of suicide, finding factors related to suicide, and making suitable policies to control these factors.

This study used spatial epidemiology focusing on deaths in males and females by suicide in Kanagawa prefecture and examined these data by spatial clustering methods. The aim of this study was to examine spatial clustering and clustering over time using spatial data from regional statistics on male and female suicide deaths in Kanagawa prefecture, Japan by looking at clustering over time after the East Japan Earthquake and Tsunami.

The design of this study was a longitudinal follow-up using vital statistics. We used spatial data on regional statistics (58 regions in Kanagawa prefecture) of the National Vital Statistics published by the Cabinet Office, government of Japan, from 2011 to 2017 [20]. These data were downloaded using e-Stat, which is a portal site for Japanese Government Statistics.

Among causes of male suicide deaths age in the 40s (21%) and 50s (18%), being unemployed (50%), living alone (37%), and having family problems (29%) were comparatively frequent causes. For females, age in the 30s, 40s, and 50s (16% for each), being unemployed (79%), living alone (24%), health problems (50%), and family problems (16%) were comparatively frequent causes. (Table 1)

Characteristics of covariates in 58 regions in Kanagawa prefecture are shown in Table 2. Unemployment was high in region No. 19 (Kawasaki ward) and family size, population density, proportion of higher education, average income, and proportion of tertiary industries were higher in small towns in rural areas (regions No. 54,55,57, and 58).

In the maps for SMR and EBSMR, the estimates were classified into 5 levels (low: 0–80; somewhat low: 80–90; moderate: 90–110; somewhat high: 110–12; high: 120-). Since the populations in the regions varied (from 2000 to over 200000), the SMR (Additional files 1: Material 4 and 5) varied largely according to region and year that data were collected compared to EBSMR (Figs. 4 and 5). In the case of total year standardization, the EBSMR showed a temporal tendency toward decreases in suicide mortality more clearly compared to data for each-year standardization (Additional file 1: Materials 6 and 7).

The results of several spatial tests as well as regression models are summarized in Table 3 for males and Table 4 for females. Regions detected as having significant spatial clusters by Tango’s test and FlexScan, were few and varied somewhat according to the method used.

Especially among females, by the 3 methods, only in 2017 were such regions detected. By the CAR model only including the smoothing parameter (without covariates) showed no significant regional cluster for both males and females (results are not shown). However, with CAR models with covariates, regional tendencies became much clearer by considering the spatial parameter for both males and females and more regions were detected compared to the model that did not include a spatial parameter. Specifically Kawasaki ward was detected as being a significantly higher suicide region for both males and females. Figure 6 shows a scatter plot of the estimated suicide death rates by the CAR model with 5 covariates and the unemployment rate for males. The numbers in the scatter plot denote the region number. This figure shows that Kawasaki ward was the highest both for suicide rate and unemployment rate.

As for the factors related to suicide death rates, by the linear regression model, the selected variables were different between the sexes. For males, unemployment (in 2014, 2016, 2017), family size (in 2011, 2012, 2013), and the proportion of higher education (in 2011, 2015) were detected to be significant in two or more years. However, for females, family size (in 2013, 2016) and the divorce rate (in 2012, 2014) were detected to be significant in two or more years. As to the CAR model with 10 covariates, no significant variables were detected except for the proportion of higher education in 2011 for both males and females and the proportion of tertiary industries in 2013 for females. The results by the CAR model with 5 covariates for males showed the significant protective effects of family size (in 2011, 2012, 2013), population density (in 2011), and proportion of higher education in 2015). In the case of females, a protective effect of population density in 2013 was observed and the proportion of tertiary industries (in 2012) and the divorce rate (in 2012) were related to greater suicide deaths.

Using vital statistics and census data from National Vital Statistics published by the Cabinet Office, government of Japan, from 2011 to 2017 [20], this study evaluated suicide mortality among males and females in Kanagawa prefecture by spatial epidemiology. Results of the analyses showed evidence of hotspots of suicide mortality across regions in Kanagawa prefecture and revealed related factors. To our knowledge this is the first study to investigate spatial and temporal patterns of suicide mortality by considering related factors at the regional level in Japan. During this period, the suicide rates in Kanagawa prefecture for each year were lower than those of Japan as a whole (Fig. 1). Although suicide mortality has been gradually decreasing in Kanagawa prefecture since 2011, some regions within the prefecture were still comparatively high-risk regions. The results of this study can be expected to bring a better understanding of where to target resources and preventive efforts at the regional level to reduce the burden of suicide in high-risk regions.

Many studies have used spatial epidemiological technics especially for infectious diseases. These studies employed spatial epidemiology, which was used to describe and analyze geographic variations in disease with respect to demographic, environmental, behavioral, socioeconomic, genetic, and infectious risk factors [26, 27]. These methods are useful not only for studies of infectious diseases but also for studies of non-infectious diseases or events like suicide. Worldwide there have been several reports of suicide cluster analyses [14‐19]. We performed spatial cluster analyses using several methods as well as examined related factors using the CAR model including spatial smoothing. Thus far, detailed analyses such as ours have not been conducted as mentioned in the Background section. That these analyses revealed that factors affecting suicide death rates differed between the sexes and among the years studied might be useful in formulating community countermeasures.

According to the report by WHO [2], in high-income countries three times as many males die by suicide as females. In Kanagawa prefecture, the male-to-female ratio for suicide indicated that twice as many males as females committed suicide (Table 1, Fig. 1), with suicide deaths more frequent in persons in their 40s and 50s. In our study, Kawasaki ward (region number 19) was revealed to be a hotspot irrespective of differences in the models employed. The EMSMRs in Kawasaki ward varied from 110.8 to 131.2 for males and 101.6 to 111.1 for females during 2011 to 2017; these values were the largest among regions within Kanagawa prefecture. Furthermore, the unemployment rate was the largest in Kawasaki ward (5.8%). (see Fig. 6) Kawasaki ward is an urban industrial zone and is located in a coastal area (Kawasaki Port), the Keihin Industrial Area, where information service industries, large oil complexes, and steel works are concentrated as well as many large factories.

The main industry is manufacturing, and daily employment is still present. Furthermore, Kawasaki ward has fewer medical clinics than Kanagawa prefecture as a whole. Regional medical resource data from the Japan Medical Association Regional Medical Information System showed that there are fewer clinics per 100000 people in Kawasaki ward compared to Kanagawa prefecture (Kawasaki ward vs. Kanagawa prefecture: 52.83 vs. 68.14), with few psychiatric clinics per 100000 people (5.37 vs. 6.52). The scarcity of clinics could hamper early detection/continuation of treatment of depression, alcohol dependence, etc. In fact our study results suggested that difficulties in access to clinics might be one reason for the high suicide mortality in Kawasaki ward.

Results of a quasi-experimental study conducted in Japan [28] to examine the effectiveness of the Multimodal Community Intervention Program to Prevent Suicide suggested the effectiveness of interventions for males and the elderly in rural areas to prevent suicide. Suicide prevention has been considered to be an important priority in policy making [29]. The incredibly rapid change in society throughout Japan may have resulted in difficulties in forming social networks such as neighborhood relationships because individuals move into an area primarily because of the workplace rather than to create a permanent household. Also, the welfare utilization rate is high, the foreign population is large, and disparities in the social environments are large even within the ward.

Many suicide attempters belong to highly vulnerable marginalized groups. Especially, the young and elderly are considered most susceptible to suicidal ideation and self-harm. However, in this study, the 40th and 50th were recognized as vulnerable, and men in that age group can be considered to be marginalized with regard to suicide possibly because of working conditions and social roles. Men in midlife tend to devote themselves to work and family and lack outside social connections. One study based on the National death certification, issued by a registrar of vital statistics for official register of deaths, suggested that middle-aged Japanese men among management workers rather than clerks and blue-collar workers tend to commit suicide [30]. These findings imply the necessity of introducing or improving workplace mental health policies and services. The Japanese government set the General Principles of Suicide Prevention Policy in 2007, and that document suggested the use of gatekeepers, volunteers who assist the vulnerable by offering services such as helplines in society for suicide prevention [31]. Further research is necessary to evaluate the impact of policies for middle-aged men to prevent suicide. Policies including those on statistical research and provision of information on suicide prevention measures have been cited as new developments as has analysis of the actual situation of suicide in specific areas.

Suicide prevention addresses a wide range of social factors. Furthermore, it is important to note that since suicide is affected by sociocultural factors, effective interventions in a certain population may not work in another population. Therefore, since preparing profiles of actual local suicides should be important, our study can provide useful information. In addition, our finding that unemployment was frequently associated with suicide deaths was similar to the finding that socioeconomic deprivation has been associated with high-risk suicide clusters detected in the above-mentioned studies.

Statistically significant spatial clusters of populations at high risk of suicide were detected at the regional level. The present results using spatial clustering of suicide data in Kanagawa Prefecture detected some regions within the prefecture that were at high risk. These results would provide important information for policy making on suicide prevention.