Pengelompokan Pasien Penyakit Kronis Diabetes, Hipertensi, dan Gagal Ginjal Menggunakan K-Means Berdasarkan Variabel Klinis

Abstract

The increasing number of patients with chronic diseases poses challenges in delivering effective and efficient healthcare services. The heterogeneity of patients’ clinical characteristics makes risk identification and clinical decision-making processes more complex. This study aimed to classify patients based on similarities in clinical characteristics using the K-Means algorithm to identify specific health patterns. The study employed a quantitative approach with a cross-sectional observational design and data mining techniques. Research data were obtained from a public Kaggle dataset which, after the data cleaning process, resulted in 51 patient records with analytical variables including age, blood pressure, blood glucose level, hemoglobin, diabetes mellitus, hypertension, anemia, and red blood cell (RBC) condition. Cluster analysis was performed using the K-Means algorithm, while cluster validity was evaluated using the Silhouette and Dunn indices. The results showed that the optimal number of clusters was three, with a Silhouette score of 0.537 and a Dunn index of 0.612. Cluster 1 (n=38) represented a low-risk profile characterized by relatively normal blood pressure and blood glucose levels without cases of diabetes mellitus or hypertension. Cluster 2 (n=8) was characterized by a high prevalence of anemia (87.5%) and low hemoglobin levels. Meanwhile, Cluster 3 (n=5) exhibited the most severe metabolic profile, with hypertension prevalence reaching 100%, diabetes mellitus 80%, and the highest average blood glucose level 348.8 mg/dL. These findings indicate that the K-Means method is effective in identifying patient segmentation based on distinct clinical characteristics. However, due to the relatively limited sample size and imbalanced cluster distribution, the findings should be interpreted as an exploratory analysis that requires further validation using larger and more diverse datasets.

Keywords: clinical pattern, data mining, health stratification, patient segmentation, unsupervised learning