SUBDISTRICT CLUSTERING IN WEST JAVA PROVINCE BASED ON DISEASE INCIDENCE OF JKN PARTICIPANTS PRIMARY SERVICES

Abstract

One of the efforts that can be done to optimize health services and the distribution of facilities and infrastructure efficiently in a wide scope is by profiling and clustering areas in the province of West Java to the scope of sub-districts that have similar characteristics of disease category. The methods that will be compared to get the best clustering are hierarchical clustering and ensemble clustering. The data used as the object of research is the BPJS Kesehatan capitation primary service sample data for the 2017-2018 period. Some of the important variables used include: primary disease diagnosis data (ICD-10) of patients at the puskesmas, service time, type of visit, and location of service sub-district. This study uses several evaluation metrics Silhouette coefficient, Dunn index, Davies-Bouldin index, and C-index to determine the optimal number of clusters formed. In addition, descriptive analysis and visualization of the clustering results are also used as considerations in selecting the optimal cluster. Based on the evaluation results, the optimal method is hierarchical clustering with complete linkage. This method produces three clusters: cluster 1 consists of 5 sub-districts that have a high/dominant mean value in almost all disease categories, cluster 2 consists of 26 sub-districts that have a medium mean value, and cluster 3 consists of 589 sub-districts that have a low mean value. Most of the members of clusters 1 and 2 are sub-districts located in the districts/cities around the national capital (DKI Jakarta) and the provincial capital (Bandung) while the members of cluster 3 are mostly sub-districts located in suburban districts/cities or far from the central government.