Defining health service areas in Korea for research on geographic variations in health service utilization

Abstract

Research on geographic variations in health services utilization is one of the traditional subjects in health care systems. Also, in many developed countries, results of variations research have been actually used for proposing the direction of reformation of health care policy and deriving its detail. For this reason, recently, variations research has gained increasing attention in Korea. Therefore, in this thesis, we proposed the methodology for delineating health service areas and verifying its validity for the very first step of research on variations. Methods can be divided as follows. First of all, methodology of delineating health service areas was proposed. Using hospital location data and patients’ residence information, patients’ medical use matrices were constructed. Then, based on these matrices, health service areas were defined by using hierarchical clustering including constraints such as minimum population, minimum self-containment, and merging distance. Second, statistical approach for verifying the validity of delineating method was applied. To do this, change detection for patients’ medical use matrices, which were used as input data of defining health service areas, have to be conducted. Therefore, we conducted pairwise comparisons for matrices of two consecutive years by using kernel density estimation (KDE) and expectation maximization (EM) algorithm in order to determine whether there was a significant change from one year to the next year in the pattern of hospital usage for acute care admission. The results show that during the ten years period, there were two changes in the acute care admission data, from 2006 to 2007 and from 2010 to 2011. For the other consecutive years, the data were unchanged year to year. After that, we checked whether the health service areas were changed significantly depending on the statistical results. Lastly, to evaluate partial changes of patient origin matrices, we used structural similarity index (SSIM). By using aggregated data from the two unchanged period (2007-2010, 2011-2015) and SSIM, we assessed whether the health service areas were changed appropriately. The results show that the change in the health service areas between the two periods was proper. Hence, the proposed delineating methodology and its resulting areas are suitable for long-term analysis of variations research.