Patient Segmentation Based on Visit Patterns and Diagnoses Using the K-Means Clustering Algorithm on Medical Records from XYZ Clinic in 2024

Abstract

Outpatient clinics in Indonesia routinely generate extensive health data through patient visits; however, such data remain underutilized for strategic and clinical decision-making. This study aims to segment patients based on visit frequency, diagnosis codes, demographic characteristics, and payment types using three clustering techniques: K-Means, Agglomerative Hierarchical Clustering, and DBSCAN. The objective is to determine the most effective method for patient stratification in a primary healthcare setting. Patient visit data from Klinik Pratama UIN Sunan Kalijaga for the year 2024 were analyzed. K-Means produced the most granular structure with nine clusters, DBSCAN identified seven clusters including a noise group, while Hierarchical Clustering yielded three macro-clusters. Internal validation using Silhouette Score, Davies-Bouldin Index, and Calinski-Harabasz Index revealed Hierarchical Clustering as the optimal model, achieving the highest cluster cohesion and separation with a Silhouette Score of 0.502, Calinski-Harabasz Index of 2134.87, and Davies-Bouldin Index of 0.668. The dendrogram and principal component analysis visualization confirmed the natural separation into three clinically meaningful patient segments. Cluster 0 comprised patients with acute respiratory and digestive conditions exhibiting sporadic visits. Cluster 1 consisted predominantly of male BPJS-insured patients with musculoskeletal and dental complaints and moderate visit frequency. Cluster 2 included female BPJS-insured patients with chronic metabolic and vascular diseases requiring consistent and frequent care. These findings demonstrate the efficacy of hierarchical clustering in producing interpretable patient segments and provide a valuable foundation for targeted healthcare management and resource allocation in outpatient clinics.