Geographical Distribution and Regional Disparity of Suicide Risks in Korea, 2008-2012

Abstract

The suicide problem has been grave national public health issue. This study aims to explore geographical distribution of suicide risks in regional scale and to measure the regional disparity. The study object is complete suicide, so the national mortality data, cause of death statistics, was used for this study. From the explorative analysis, significant differences of suicide rates among different sex and age strata were found. The associations between categorical socio-demographic variables and violent suicide means selection, holiday effect, and copycat effect were identified with log-linear model. Spatio-temporally comparable suicide risks and statistically robust relative risks were estimated by direct standardization and Bayesian estimation, respectively. Various visualizing techniques, such as building legend scheme, statistical significance mapping, and cartogram, were applied for mapping the risks and the results were produced with interactive maps, utilizing the interface of Google Maps and Google Earth. In metropolitan areas, bulls eye pattern was found. On the national scale, low suicide risks in southern areas were clearly distinguished with high risks in northern parts. Taking into account statistical significance, the regions with high relative risks were mainly distributed in Gangwon, Chungnam, and Chungbuk. The regional disparity of suicide risks was measured with spatial dissimilarity index, which integrates two divergent dimensions: aspatial unevenness and spatial clustering. First, the characteristics of the distribution of suicide risks were measured in each dimension. Regional unevenness was measured with aspatial inequality indices and global/local spatial autocorrelation was explored with spatial cluster analysis. The regional inequality decreased from 2008 to 2010, and increased consistently after 2010. Moreover, strong spatial autocorrelation was observed throughout the study period. Based on these observations, the global spatial dissimilarity index was computed with Gaussian kernel. It did not make any change in the pattern of temporal variation derived with aspatial measures due to the stable spatio-temporal distribution of suicide risks. However, when the administrative units are ignored with kernel smoothing, temporal pattern changed, which implies the possibility of MAUP. Lastly, cumulative frequency legend and 3D mapping was applied to suicide atlas to reflect the information on the computed regional disparity of suicide risks.